diff --git a/tests/integration/_support/actors.py b/tests/integration/_support/actors.py index 6a1e53bda31..0d628cc3b88 100644 --- a/tests/integration/_support/actors.py +++ b/tests/integration/_support/actors.py @@ -15,7 +15,7 @@ from pydantic import JsonValue from litellm.proxy.utils import PrismaClient, hash_token from tests.integration._support.client import Gateway, Scenario, string_value from tests.integration._support.prisma_clients import connected_proxy_prisma -from tests.integration._support.process import owned_proxy +from tests.integration._support.process import OwnedState, owned_proxy TEAM_ALPHA: Final = "behavior-pin-team-alpha" TEAM_BETA: Final = "behavior-pin-team-beta" @@ -124,7 +124,7 @@ def owned_settings_client( config: Final = {**base, "litellm_settings": {**base["litellm_settings"], **litellm_settings}} path: Final = directory / "owned-settings.yaml" path.write_text(yaml.safe_dump(config)) - with owned_proxy(client.gateway, directory, {}, config=path) as candidate: + with owned_proxy(client.gateway, directory, {}, config=path, state=OwnedState.SHARED_WITH_WORKER) as candidate: yield GatewayClient(candidate, client.replacements) diff --git a/tests/integration/_support/bedrock_discovery.py b/tests/integration/_support/bedrock_discovery.py index 5b3a091dc17..058e0850068 100644 --- a/tests/integration/_support/bedrock_discovery.py +++ b/tests/integration/_support/bedrock_discovery.py @@ -23,7 +23,7 @@ from integration._support.aws_control_plane import ( verified_scope, ) from integration._support.client import JSON_OBJECT, Gateway, Scenario, eventually, held, object_value, string_value -from integration._support.process import OwnedProxy, owned_proxy_process +from integration._support.process import OwnedProxy, OwnedState, owned_proxy_process from litellm.litellm_core_utils.aws_partition import get_aws_dns_suffix from pydantic import JsonValue @@ -115,6 +115,7 @@ def discovery_proxy( config=discovery_config(parent, directory, check_provider_endpoint=check_provider_endpoint), remove_environment=aws_environment_names(), workers=workers, + state=OwnedState.SHARED_WITH_WORKER, ) as owned: wait_for_every_worker(owned.log, workers) yield owned diff --git a/tests/integration/_support/client.py b/tests/integration/_support/client.py index 053318c2451..6f407295259 100644 --- a/tests/integration/_support/client.py +++ b/tests/integration/_support/client.py @@ -35,9 +35,9 @@ def list_value(value: JsonValue) -> Sequence[JsonValue]: def delete_key_if_present(candidate: Gateway, key: str) -> None: digest: Final = sha256(key.encode()).hexdigest() - if read_rows('SELECT token FROM "LiteLLM_VerificationToken" WHERE token=%s', (digest,)): + if candidate.rows('SELECT token FROM "LiteLLM_VerificationToken" WHERE token=%s', (digest,)): candidate.post("/key/delete", {"keys": [key]}) - assert read_rows('SELECT token FROM "LiteLLM_VerificationToken" WHERE token=%s', (digest,)) == [] + assert candidate.rows('SELECT token FROM "LiteLLM_VerificationToken" WHERE token=%s', (digest,)) == [] def eventually( @@ -95,6 +95,10 @@ class Gateway: client: httpx.Client key: str upstream_url: str + database_url: str | None = None + + def rows(self, query: str, parameters: tuple[str, ...]) -> list[dict[str, JsonValue]]: + return read_rows(query, parameters, database_url=self.database_url) def request( self, @@ -175,7 +179,7 @@ class Scenario: def delete_team(self, identity: str) -> None: self.gateway.post("/team/delete", {"team_ids": [identity]}) - assert read_rows('SELECT team_id FROM "LiteLLM_TeamTable" WHERE team_id = %s', (identity,)) == [] + assert self.gateway.rows('SELECT team_id FROM "LiteLLM_TeamTable" WHERE team_id = %s', (identity,)) == [] def project(self, team_id: str, **fields: JsonValue) -> str: created: Final = self.gateway.post( @@ -188,7 +192,9 @@ class Scenario: def delete_project(self, identity: str) -> None: response: Final = self.gateway.request("DELETE", "/project/delete", {"project_ids": [identity]}) assert response.status_code == 200, response.text - assert read_rows('SELECT project_id FROM "LiteLLM_ProjectTable" WHERE project_id = %s', (identity,)) == [] + assert ( + self.gateway.rows('SELECT project_id FROM "LiteLLM_ProjectTable" WHERE project_id = %s', (identity,)) == [] + ) def budget(self, **fields: JsonValue) -> str: created: Final = self.gateway.post("/budget/new", fields) @@ -198,7 +204,7 @@ class Scenario: def delete_budget(self, identity: str) -> None: self.gateway.post("/budget/delete", {"id": identity}) - assert read_rows('SELECT budget_id FROM "LiteLLM_BudgetTable" WHERE budget_id = %s', (identity,)) == [] + assert self.gateway.rows('SELECT budget_id FROM "LiteLLM_BudgetTable" WHERE budget_id = %s', (identity,)) == [] def organization(self, **fields: JsonValue) -> str: created: Final = self.gateway.post( @@ -212,7 +218,9 @@ class Scenario: response: Final = self.gateway.request("DELETE", "/organization/delete", {"organization_ids": [identity]}) assert response.status_code == 200, response.text assert ( - read_rows('SELECT organization_id FROM "LiteLLM_OrganizationTable" WHERE organization_id = %s', (identity,)) + self.gateway.rows( + 'SELECT organization_id FROM "LiteLLM_OrganizationTable" WHERE organization_id = %s', (identity,) + ) == [] ) self.delete_budget(budget_id) @@ -236,7 +244,7 @@ class Scenario: def delete_user(self, identity: str) -> None: response: Final = self.gateway.request("POST", "/user/delete", {"user_ids": [identity]}) assert response.status_code == 200 and response.json() == 1, response.text - assert read_rows('SELECT user_id FROM "LiteLLM_UserTable" WHERE user_id = %s', (identity,)) == [] + assert self.gateway.rows('SELECT user_id FROM "LiteLLM_UserTable" WHERE user_id = %s', (identity,)) == [] def member(self, team_id: str, role: str = "user") -> str: """Create an internal user and add them to ``team_id``; deleting the user later removes the membership.""" @@ -247,7 +255,7 @@ class Scenario: def delete_key(self, token: str) -> None: self.gateway.post("/key/delete", {"keys": [token]}) hashed: Final = sha256(token.encode()).hexdigest() - assert read_rows('SELECT token FROM "LiteLLM_VerificationToken" WHERE token = %s', (hashed,)) == [] + assert self.gateway.rows('SELECT token FROM "LiteLLM_VerificationToken" WHERE token = %s', (hashed,)) == [] info: Final = object_value(self.gateway.get("/key/info", {"key": hashed})["info"]) assert info["status"] == "deleted", f"Deleted key still served as live: {info['status']}" @@ -265,7 +273,9 @@ class Scenario: lambda served: identity not in served, seconds=2 * PROXY_CONFIG_RELOAD_INTERVAL_SECONDS, ) - assert read_rows('SELECT model_id FROM "LiteLLM_ProxyModelTable" WHERE model_id = %s', (identity,)) == [] + assert ( + self.gateway.rows('SELECT model_id FROM "LiteLLM_ProxyModelTable" WHERE model_id = %s', (identity,)) == [] + ) def model(self, *, model_info: Mapping[str, JsonValue] | None = None, **parameters: JsonValue) -> str: name: Final = f"integration-{uuid.uuid4().hex}" diff --git a/tests/integration/_support/process.py b/tests/integration/_support/process.py index d5ce6f916a2..b08037e80db 100644 --- a/tests/integration/_support/process.py +++ b/tests/integration/_support/process.py @@ -7,9 +7,9 @@ import subprocess import sys import time import uuid -from collections.abc import Generator, Iterable, Iterator, Mapping +from collections.abc import Callable, Generator, Iterable, Iterator, Mapping from concurrent.futures import ThreadPoolExecutor -from contextlib import contextmanager, suppress +from contextlib import AbstractContextManager, contextmanager, suppress from dataclasses import dataclass from enum import Enum from datetime import UTC, datetime, timedelta @@ -17,7 +17,7 @@ from ipaddress import IPv4Address from itertools import accumulate from pathlib import Path from types import MappingProxyType -from typing import TYPE_CHECKING, Final +from typing import TYPE_CHECKING, Final, TypeAlias from urllib.parse import urlsplit import httpx @@ -33,6 +33,8 @@ from tests.integration._support.budget import import_seconds, readiness_for, rea from tests.integration._support.client import GATEWAY_LIMITS, Gateway, gateway_from_environment from tests.integration._support.database import scratch_database from tests.integration._support.ports import allocate_port +from tests.integration._support.redis_process import owned_redis +from tests.integration._support.workers import gateway_profile, replica_urls if TYPE_CHECKING: import pytest @@ -43,6 +45,75 @@ LEGACY_MIGRATE_DEPLOY: Final = ("--use_legacy_migration_resolver",) CONFIGURATION: Final = TypeAdapter(dict[str, object]) _DEPLOYMENT_IDS: Final = TypeAdapter(dict[str, list[str]]) MCP_COVERAGE_CONFIG: Final = Path(__file__).resolve().parents[1] / "mcp_coverage.toml" +OWNED_STATE_VARIABLE: Final = "INTEGRATION_OWNED_STATE" + + +class OwnedState(Enum): + PRIVATE = "private" + SHARED_WITH_WORKER = "shared_with_worker" + + +@dataclass(frozen=True, slots=True) +class PrivateState: + database_url: str + environment: Mapping[str, str] + + +PrivateStateFactory: TypeAlias = Callable[[Path, Mapping[str, str]], AbstractContextManager[PrivateState]] + + +def boots_private_state( + state: OwnedState | None, + overrides: Mapping[str, str], + remove_environment: tuple[str, ...], + inherited: Mapping[str, str], +) -> bool: + chosen: Final = state or OwnedState(inherited.get(OWNED_STATE_VARIABLE, OwnedState.PRIVATE.value)) + return chosen is OwnedState.PRIVATE and "DATABASE_URL" not in overrides and "DATABASE_URL" not in remove_environment + + +@contextmanager +def private_state(directory: Path, inherited: Mapping[str, str]) -> Iterator[PrivateState]: + with scratch_database("owned") as database, owned_redis(directory, inherited=inherited) as cache: + replica: Final = replica_urls(database, inherited) if gateway_profile(inherited).replica else None + yield PrivateState( + database, + MappingProxyType( + { + "DATABASE_URL": database if replica is None else replica["INTEGRATION_PROXY_DATABASE_URL"], + **( + {} + if replica is None + else {"DATABASE_URL_READ_REPLICA": replica["INTEGRATION_PROXY_READ_REPLICA_URL"]} + ), + "REDIS_HOST": cache.host, + "REDIS_PORT": str(cache.port), + } + ), + ) + + +@dataclass(frozen=True, slots=True) +class OwnedEnvironment: + overrides: Mapping[str, str] + database_url: str | None + + +@contextmanager +def owned_environment( + directory: Path, + overrides: Mapping[str, str], + remove_environment: tuple[str, ...], + state: OwnedState | None, + inherited: Mapping[str, str], + private: PrivateStateFactory = private_state, +) -> Iterator[OwnedEnvironment]: + if not boots_private_state(state, overrides, remove_environment, inherited): + yield OwnedEnvironment(overrides, None) + return + with private(directory, inherited) as booted: + kept: Final = {name: value for name, value in booted.environment.items() if name not in remove_environment} + yield OwnedEnvironment(MappingProxyType({**kept, **overrides}), booted.database_url) def proxy_database_environment(inherited: Mapping[str, str] = os.environ) -> Mapping[str, str]: @@ -368,6 +439,7 @@ def owned_proxy( database_setup: tuple[str, ...] = DB_PUSH, coverage_file: Path | None = None, inherited: Mapping[str, str] = os.environ, + state: OwnedState | None = None, ) -> Iterator[Gateway]: with owned_proxy_process( gateway, @@ -379,20 +451,21 @@ def owned_proxy( database_setup=database_setup, coverage_file=coverage_file, inherited=inherited, + state=state, ) as owned: yield owned.gateway @contextmanager def isolated_gateway(directory: Path, environment: "pytest.MonkeyPatch") -> Iterator[Gateway]: - with gateway_from_environment() as shared, scratch_database() as database: - environment.setenv("DATABASE_URL", database) - environment.setenv("INTEGRATION_PROXY_DATABASE_URL", database) + with gateway_from_environment() as shared, owned_proxy(shared, directory, {}, state=OwnedState.PRIVATE) as owned: + assert owned.database_url is not None + environment.setenv("DATABASE_URL", owned.database_url) + environment.setenv("INTEGRATION_PROXY_DATABASE_URL", owned.database_url) environment.delenv("INTEGRATION_PROXY_READ_REPLICA_URL", raising=False) environment.delenv("DATABASE_URL_READ_REPLICA", raising=False) - with owned_proxy(shared, directory, {}) as owned: - owned.post("/config/update", {"router_settings": {"num_retries": 0}}) - yield owned + owned.post("/config/update", {"router_settings": {"num_retries": 0}}) + yield owned def _stop(process: subprocess.Popen[bytes]) -> None: @@ -634,6 +707,32 @@ def _proxy_root() -> Path: return Path(os.environ.get("INTEGRATION_PROXY_ROOT") or Path(__file__).resolve().parents[3]) +def _configuration_copy(config: Path, directory: Path, prefix: str) -> Path: + checkouts: Final = (_harness_root(), _proxy_root().resolve()) + inside_checkout: Final = any(config.resolve().is_relative_to(checkout) for checkout in checkouts) + target: Final = directory if inside_checkout else config.parent + target.mkdir(parents=True, exist_ok=True) + return target / f"{prefix}-{uuid.uuid4().hex}.yaml" + + +def private_configuration(config: Path, directory: Path) -> Path: + loaded: Final[object] = yaml.safe_load(config.read_text()) + if not isinstance(loaded, dict): + return config + settings: Final = CONFIGURATION.validate_python(loaded) + router: Final = settings.get("router_settings") or {} + if not isinstance(router, dict) or "num_retries" in router: + return config + destination: Final = _configuration_copy(config, directory, "private-config") + destination.write_text( + yaml.safe_dump( + {**settings, "router_settings": {**CONFIGURATION.validate_python(router), "num_retries": 0}}, + sort_keys=False, + ) + ) + return destination + + def worker_configuration(config: Path, directory: Path, inherited: Mapping[str, str] = os.environ) -> Path: if inherited.get("INTEGRATION_WORKER_ISOLATION") != "1": return config @@ -644,11 +743,7 @@ def worker_configuration(config: Path, directory: Path, inherited: Mapping[str, return f"http://127.0.0.1:{inherited['INTEGRATION_PROVIDER_PORT']}" contents: Final = re.sub(r"http://127[.]0[.]0[.]1:(8190|8191)(?![0-9])", endpoint, config.read_text()) - checkouts: Final = (_harness_root(), _proxy_root().resolve()) - inside_checkout: Final = any(config.resolve().is_relative_to(checkout) for checkout in checkouts) - target: Final = directory if inside_checkout else config.parent - target.mkdir(parents=True, exist_ok=True) - destination: Final = target / f"worker-config-{uuid.uuid4().hex}.yaml" + destination: Final = _configuration_copy(config, directory, "worker-config") loaded: Final[object] = yaml.safe_load(contents) if not isinstance(loaded, dict): destination.write_text(contents) @@ -753,42 +848,58 @@ def owned_proxy_process( healthcheck: WorkerHealthcheck = WorkerHealthcheck.STRICT, inherited: Mapping[str, str] = os.environ, client_timeout: float = 15, + state: OwnedState | None = None, ) -> Iterator[OwnedProxy]: assert "--timeout_worker_healthcheck" not in extra_arguments, "Choose the worker healthcheck with `healthcheck`" root: Final = _proxy_root() - environment: Final = _proxy_environment( - gateway, - {**overrides, **({"COVERAGE_FILE": str(coverage_file)} if coverage_file else {})}, - remove_environment, - inherited, - ) - output: Final = Path(inherited.get("INTEGRATION_RESULTS_DIR", str(directory))) - output.mkdir(parents=True, exist_ok=True) - command: Final = ( - sys.executable, - "-m", - *(("coverage", "run", f"--rcfile={MCP_COVERAGE_CONFIG}", "-m") if coverage_file else ()), - "integration._support.proxy", - "--config", - str(worker_configuration(config or root / "tests/integration/proxy_config.yaml", directory, inherited)), - "--host", - "127.0.0.1", - "--num_workers", - str(workers), - "--timeout_worker_healthcheck", - str(healthcheck_seconds(healthcheck)), - *database_setup, - *extra_arguments, - ) - launch: Final = _launch_until_ready(command, root, environment, output, workers) - process: Final = launch.process - try: - with httpx.Client( - base_url=f"http://127.0.0.1:{launch.port}", timeout=client_timeout, trust_env=False, limits=GATEWAY_LIMITS - ) as client: - yield OwnedProxy(Gateway(client, gateway.key, gateway.upstream_url), process, launch.log, healthcheck) - finally: - _stop_launch(launch) + with owned_environment(directory, overrides, remove_environment, state, inherited) as owned: + configured: Final = worker_configuration( + config or root / "tests/integration/proxy_config.yaml", directory, inherited + ) + owned_configuration: Final = ( + configured if owned.database_url is None else private_configuration(configured, directory) + ) + environment: Final = _proxy_environment( + gateway, + {**owned.overrides, **({"COVERAGE_FILE": str(coverage_file)} if coverage_file else {})}, + remove_environment, + inherited, + ) + output: Final = Path(inherited.get("INTEGRATION_RESULTS_DIR", str(directory))) + output.mkdir(parents=True, exist_ok=True) + command: Final = ( + sys.executable, + "-m", + *(("coverage", "run", f"--rcfile={MCP_COVERAGE_CONFIG}", "-m") if coverage_file else ()), + "integration._support.proxy", + "--config", + str(owned_configuration), + "--host", + "127.0.0.1", + "--num_workers", + str(workers), + "--timeout_worker_healthcheck", + str(healthcheck_seconds(healthcheck)), + *database_setup, + *extra_arguments, + ) + launch: Final = _launch_until_ready(command, root, environment, output, workers) + process: Final = launch.process + try: + with httpx.Client( + base_url=f"http://127.0.0.1:{launch.port}", + timeout=client_timeout, + trust_env=False, + limits=GATEWAY_LIMITS, + ) as client: + yield OwnedProxy( + Gateway(client, gateway.key, gateway.upstream_url, owned.database_url), + process, + launch.log, + healthcheck, + ) + finally: + _stop_launch(launch) @contextmanager diff --git a/tests/integration/_support/prometheus_series.py b/tests/integration/_support/prometheus_series.py index 3a406a97d11..6f54f2a8f7f 100644 --- a/tests/integration/_support/prometheus_series.py +++ b/tests/integration/_support/prometheus_series.py @@ -19,7 +19,7 @@ import httpx import yaml from integration._support.client import Gateway, Scenario, eventually, gateway_from_environment, object_value from integration._support.database import read_rows -from integration._support.process import OwnedProxy, owned_proxy_process +from integration._support.process import OwnedProxy, OwnedState, owned_proxy_process from integration._support.responses_vendor import ResponsesVendor, same_response from integration._support.wire import Reply, Request, Wire, wire_server from prometheus_client.mmap_dict import MmapedDict @@ -207,6 +207,7 @@ def series_cap_rig( config=config, remove_environment=() if shared_samples else ("PROMETHEUS_MULTIPROC_DIR",), workers=workers, + state=OwnedState.SHARED_WITH_WORKER, ) ) scenario: Final = stack.enter_context(owned.gateway.scenario()) diff --git a/tests/integration/_support/safeguards_rig.py b/tests/integration/_support/safeguards_rig.py index 8ba6f14f71c..f0acf956e5e 100644 --- a/tests/integration/_support/safeguards_rig.py +++ b/tests/integration/_support/safeguards_rig.py @@ -30,7 +30,7 @@ from integration._support.client import ( string_value, ) from integration._support.database import read_rows -from integration._support.process import OwnedProxy, owned_proxy_process +from integration._support.process import OwnedProxy, OwnedState, owned_proxy_process from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import JsonValue, TypeAdapter @@ -313,6 +313,7 @@ def safeguards_rig( config=path, remove_environment=remove_environment, workers=workers, + state=OwnedState.SHARED_WITH_WORKER, ) as owned: yield Rig(wire, scripts, owned) diff --git a/tests/integration/_support/workers.py b/tests/integration/_support/workers.py index e75b0b3c6bb..f1958ef6990 100644 --- a/tests/integration/_support/workers.py +++ b/tests/integration/_support/workers.py @@ -86,9 +86,9 @@ def worker_endpoints() -> Mapping[str, str]: ) -def _replica_urls(database: str) -> Mapping[str, str]: - writer: Final = urlsplit(os.environ["INTEGRATION_PROXY_DATABASE_URL"]) - reader: Final = urlsplit(os.environ["INTEGRATION_PROXY_READ_REPLICA_URL"]) +def replica_urls(database: str, inherited: Mapping[str, str] = os.environ) -> Mapping[str, str]: + writer: Final = urlsplit(inherited["INTEGRATION_PROXY_DATABASE_URL"]) + reader: Final = urlsplit(inherited["INTEGRATION_PROXY_READ_REPLICA_URL"]) database_path: Final = urlsplit(database).path assert writer.username is not None and reader.username is not None with psycopg.connect(database, autocommit=True) as admin: @@ -164,7 +164,7 @@ def worker_services( "INTEGRATION_PROXY_READ_REPLICA_URL": "", "REDIS_HOST": cache.host, "REDIS_PORT": str(cache.port), - **(_replica_urls(database) if replica else {}), + **(replica_urls(database, inherited) if replica else {}), } ) diff --git a/tests/integration/authorization/test_access_group_budget.py b/tests/integration/authorization/test_access_group_budget.py index 0db05a649e3..71302be2f37 100644 --- a/tests/integration/authorization/test_access_group_budget.py +++ b/tests/integration/authorization/test_access_group_budget.py @@ -22,6 +22,8 @@ from integration._support.process import owned_proxy from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import BaseModel, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + _INPUT_RATE: Final = 0.000001 _OUTPUT_RATE: Final = 0.00001 _MAX_BUDGET: Final = 0.0001 diff --git a/tests/integration/authorization/test_alias_chain_budget_waiver.py b/tests/integration/authorization/test_alias_chain_budget_waiver.py index 0d64415d445..0908fa0691c 100644 --- a/tests/integration/authorization/test_alias_chain_budget_waiver.py +++ b/tests/integration/authorization/test_alias_chain_budget_waiver.py @@ -62,7 +62,7 @@ from integration.authorization._hidden_alias_budget import ( ) from pydantic import JsonValue, TypeAdapter -pytestmark: Final = pytest.mark.timeout(240) +pytestmark: Final = [pytest.mark.timeout(240), pytest.mark.shares_worker_state] PAID_INPUT_COST_PER_TOKEN: Final = 0.001 PAID_OUTPUT_COST_PER_TOKEN: Final = 0.002 diff --git a/tests/integration/authorization/test_claude_code_gateway_oauth.py b/tests/integration/authorization/test_claude_code_gateway_oauth.py index acab042bfc9..384bf3a2918 100644 --- a/tests/integration/authorization/test_claude_code_gateway_oauth.py +++ b/tests/integration/authorization/test_claude_code_gateway_oauth.py @@ -23,6 +23,8 @@ from integration._support.process import owned_proxy_process from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import BaseModel, ConfigDict, JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + _CLIENT_ID: Final = "scripted-cli-client" _CLIENT_SECRET: Final = "scripted-cli-secret" _AUTHORIZATION_CODE: Final = "scripted-auth-code" diff --git a/tests/integration/authorization/test_cli_sso_login_ui_disabled.py b/tests/integration/authorization/test_cli_sso_login_ui_disabled.py index 315a0b20334..a53fbfacf4c 100644 --- a/tests/integration/authorization/test_cli_sso_login_ui_disabled.py +++ b/tests/integration/authorization/test_cli_sso_login_ui_disabled.py @@ -34,7 +34,7 @@ from tests.integration._support.process import OwnedProxy, group_members, owned_ from tests.integration._support.provider import SharedProvider from tests.integration._support.wire import Reply, Request, Wire, wire_server -pytestmark: Final = pytest.mark.timeout(300) +pytestmark: Final = [pytest.mark.timeout(300), pytest.mark.shares_worker_state] CLI_SOURCE: Final = "litellm-cli" CLI_STATE_PREFIX: Final = "litellm-session-token" diff --git a/tests/integration/authorization/test_end_user_block.py b/tests/integration/authorization/test_end_user_block.py index 8b18bddfd00..fd283380d8f 100644 --- a/tests/integration/authorization/test_end_user_block.py +++ b/tests/integration/authorization/test_end_user_block.py @@ -12,6 +12,8 @@ from integration._support.database import read_rows from integration._support.process import owned_proxy from pydantic import BaseModel, JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + class _BlockedCustomerInfo(BaseModel): user_id: str diff --git a/tests/integration/authorization/test_hidden_alias_budget_bypass.py b/tests/integration/authorization/test_hidden_alias_budget_bypass.py index 493db4c3e62..51b36c145d3 100644 --- a/tests/integration/authorization/test_hidden_alias_budget_bypass.py +++ b/tests/integration/authorization/test_hidden_alias_budget_bypass.py @@ -62,7 +62,7 @@ from integration.authorization._hidden_alias_budget import ( ) from pydantic import JsonValue -pytestmark: Final = pytest.mark.timeout(240) +pytestmark: Final = [pytest.mark.timeout(240), pytest.mark.shares_worker_state] # $0 comes from litellm's own cost-map row (model_prices_and_context_window.json, added 2025-10-14 in 900a57f5b1f): # Lemonade is a local inference server, so there is no vendor price for the row to drift from. A row that gains a diff --git a/tests/integration/authorization/test_hidden_alias_budget_bypass_chaos.py b/tests/integration/authorization/test_hidden_alias_budget_bypass_chaos.py index 0a9b1cb3c4e..3e6acfc72de 100644 --- a/tests/integration/authorization/test_hidden_alias_budget_bypass_chaos.py +++ b/tests/integration/authorization/test_hidden_alias_budget_bypass_chaos.py @@ -32,7 +32,7 @@ from integration.authorization._hidden_alias_budget import ( upstream_requests, ) -pytestmark: Final = pytest.mark.timeout(240) +pytestmark: Final = [pytest.mark.timeout(240), pytest.mark.shares_worker_state] _STARTED_WORKER: Final = re.compile(r"Started server process \[(\d+)\]") _WAVE: Final = 8 diff --git a/tests/integration/authorization/test_jwt_auto_register_map_existing_key.py b/tests/integration/authorization/test_jwt_auto_register_map_existing_key.py index 5215054a364..62eda977438 100644 --- a/tests/integration/authorization/test_jwt_auto_register_map_existing_key.py +++ b/tests/integration/authorization/test_jwt_auto_register_map_existing_key.py @@ -24,7 +24,7 @@ from tests.integration._support.wire import Reply, Request, wire_server KEY_ID: Final = "integration-jwt-map-existing-key" MAPPING_INSERT: Final = b'INSERT INTO "public"."LiteLLM_JWTKeyMapping"' -pytestmark = pytest.mark.timeout(240) +pytestmark: Final = [pytest.mark.timeout(240), pytest.mark.shares_worker_state] def _hash(key: str) -> str: diff --git a/tests/integration/authorization/test_jwt_default_team_provisioning.py b/tests/integration/authorization/test_jwt_default_team_provisioning.py index 04715d128b0..a8c00d50ee9 100644 --- a/tests/integration/authorization/test_jwt_default_team_provisioning.py +++ b/tests/integration/authorization/test_jwt_default_team_provisioning.py @@ -14,6 +14,8 @@ from tests.integration._support.database import read_rows from tests.integration._support.process import owned_proxy from tests.integration._support.wire import Reply, Request, wire_server +pytestmark: Final = pytest.mark.shares_worker_state + KEY_ID: Final = "integration-jwt-signing-key" TEAM_BUDGET: Final = 25.0 diff --git a/tests/integration/authorization/test_jwt_key_mapping_lifecycle.py b/tests/integration/authorization/test_jwt_key_mapping_lifecycle.py index cc1d5c35788..715f3731175 100644 --- a/tests/integration/authorization/test_jwt_key_mapping_lifecycle.py +++ b/tests/integration/authorization/test_jwt_key_mapping_lifecycle.py @@ -36,7 +36,7 @@ _KEY_ID: Final = "integration-signing-key" _UPSTREAM_KEY: Final = "synthetic-openai-key" _CHANNEL: Final = "litellm_proxy.auth_cache_invalidation" _CHAT_BODY: Final = TypeAdapter(dict[str, JsonValue]) -pytestmark: Final = pytest.mark.timeout(180) +pytestmark: Final = [pytest.mark.timeout(180), pytest.mark.shares_worker_state] class _Usage(BaseModel): diff --git a/tests/integration/authorization/test_jwt_mapped_key_email_backfill.py b/tests/integration/authorization/test_jwt_mapped_key_email_backfill.py index 47c8ad5f115..b676ddad494 100644 --- a/tests/integration/authorization/test_jwt_mapped_key_email_backfill.py +++ b/tests/integration/authorization/test_jwt_mapped_key_email_backfill.py @@ -14,6 +14,8 @@ from integration._support.process import owned_proxy from integration._support.wire import Reply, Request, wire_server from jwt.algorithms import RSAAlgorithm +pytestmark: Final = pytest.mark.shares_worker_state + AUDIENCE: Final = "litellm-integration" KEY_ID: Final = "integration-signing-key" CLIENT_CLAIM: Final = "client_id" diff --git a/tests/integration/authorization/test_key_guardrail_opt_out.py b/tests/integration/authorization/test_key_guardrail_opt_out.py index 6591b0b3993..0f47ea269f9 100644 --- a/tests/integration/authorization/test_key_guardrail_opt_out.py +++ b/tests/integration/authorization/test_key_guardrail_opt_out.py @@ -3,6 +3,7 @@ from pathlib import Path from typing import Final import httpx +import pytest import yaml from integration._support.client import Gateway, Scenario, object_value, string_value @@ -16,6 +17,8 @@ from integration.authorization._guardrail_opt_out import ( stored_metadata, ) +pytestmark: Final = pytest.mark.shares_worker_state + _KEY_ROUTES: Final = ["/key/generate", "/key/update", "/key/regenerate", "/v1/chat/completions"] diff --git a/tests/integration/authorization/test_key_guardrail_opt_out_chaos.py b/tests/integration/authorization/test_key_guardrail_opt_out_chaos.py index e136691aa2e..3ceae540ba9 100644 --- a/tests/integration/authorization/test_key_guardrail_opt_out_chaos.py +++ b/tests/integration/authorization/test_key_guardrail_opt_out_chaos.py @@ -9,6 +9,7 @@ from queue import SimpleQueue from typing import Final import httpx +import pytest from integration._support.client import Gateway, eventually, string_value from integration._support.database import read_rows @@ -24,6 +25,8 @@ from integration.authorization._guardrail_opt_out import ( upstream_observations, ) +pytestmark: Final = pytest.mark.shares_worker_state + class _GuardrailSink: """Test-owned guardrail endpoint that can be stopped and restarted on the same port.""" diff --git a/tests/integration/authorization/test_key_guardrail_opt_out_runtime.py b/tests/integration/authorization/test_key_guardrail_opt_out_runtime.py index dc549d3be8d..2b6f1860ea9 100644 --- a/tests/integration/authorization/test_key_guardrail_opt_out_runtime.py +++ b/tests/integration/authorization/test_key_guardrail_opt_out_runtime.py @@ -5,6 +5,7 @@ from pathlib import Path from typing import Final import httpx +import pytest from anthropic import Anthropic from openai import AsyncOpenAI, OpenAI @@ -20,6 +21,8 @@ from integration.authorization._guardrail_opt_out import ( upstream_hits, ) +pytestmark: Final = pytest.mark.shares_worker_state + def _wire_hits(wire: Wire, marker: str) -> int: return sum(1 for request in wire.drain() if marker.encode() in request.body) diff --git a/tests/integration/authorization/test_rag_query_vector_store_allowlist.py b/tests/integration/authorization/test_rag_query_vector_store_allowlist.py index f4b5063441f..c1777c66c39 100644 --- a/tests/integration/authorization/test_rag_query_vector_store_allowlist.py +++ b/tests/integration/authorization/test_rag_query_vector_store_allowlist.py @@ -25,6 +25,8 @@ from openai.types.chat import ChatCompletion from pydantic import JsonValue, TypeAdapter from redis import Redis +pytestmark: Final = pytest.mark.shares_worker_state + CONFIG_STORE_ID: Final = "vs_integration_config_store" PROXY_CONFIG: Final = Path(__file__).resolve().parents[1] / "proxy_config.yaml" REMOVE_OPENAI_API_BASE: Final = ("OPENAI_API_BASE",) diff --git a/tests/integration/authorization/test_rag_team_provider_credentials.py b/tests/integration/authorization/test_rag_team_provider_credentials.py index 8d556b89107..61bab74ecaf 100644 --- a/tests/integration/authorization/test_rag_team_provider_credentials.py +++ b/tests/integration/authorization/test_rag_team_provider_credentials.py @@ -89,7 +89,7 @@ AWS_AMBIENT_CREDENTIALS: Final = ( "AWS_WEB_IDENTITY_TOKEN_FILE", ) -pytestmark: Final = pytest.mark.timeout(OWNED_PROXY_CELL_SECONDS) +pytestmark: Final = [pytest.mark.timeout(OWNED_PROXY_CELL_SECONDS), pytest.mark.shares_worker_state] Signature = tuple[str, str, str, str | None] diff --git a/tests/integration/authorization/test_responses_compact_id_ownership.py b/tests/integration/authorization/test_responses_compact_id_ownership.py index e57118da129..b5a6215dd15 100644 --- a/tests/integration/authorization/test_responses_compact_id_ownership.py +++ b/tests/integration/authorization/test_responses_compact_id_ownership.py @@ -50,6 +50,8 @@ from integration.authorization._responses_compact import ( ) from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + _CLIENTS: Final = ("sync", "async", "httpx") diff --git a/tests/integration/authorization/test_responses_compact_id_ownership_chaos.py b/tests/integration/authorization/test_responses_compact_id_ownership_chaos.py index d6ff66dc7d1..00327c955ac 100644 --- a/tests/integration/authorization/test_responses_compact_id_ownership_chaos.py +++ b/tests/integration/authorization/test_responses_compact_id_ownership_chaos.py @@ -40,6 +40,8 @@ from integration.authorization._responses_compact import ( vendor_id, ) +pytestmark: Final = pytest.mark.shares_worker_state + _Kind = Literal["owner", "other_user", "other_team", "unmanaged"] _KINDS: Final[tuple[_Kind, ...]] = ("owner", "other_user", "other_team", "unmanaged") _REFUSING: Final[tuple[_Kind, ...]] = ("other_user", "other_team", "unmanaged") diff --git a/tests/integration/authorization/test_responses_id_ownership.py b/tests/integration/authorization/test_responses_id_ownership.py index 9a88349cca0..2ab26e75f50 100644 --- a/tests/integration/authorization/test_responses_id_ownership.py +++ b/tests/integration/authorization/test_responses_id_ownership.py @@ -19,6 +19,8 @@ from integration._support.worker_readiness import wait_for_worker_readiness from openai.types.responses import ResponseInputParam from pydantic import BaseModel, ConfigDict, JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + _MODEL: Final = "gpt-5" _API_KEY: Final = "synthetic-responses-key" _JSON_OBJECT: Final = TypeAdapter(dict[str, JsonValue]) diff --git a/tests/integration/authorization/test_sso_blank_identity.py b/tests/integration/authorization/test_sso_blank_identity.py index 6d9b3b50cc0..4383240f0ba 100644 --- a/tests/integration/authorization/test_sso_blank_identity.py +++ b/tests/integration/authorization/test_sso_blank_identity.py @@ -30,7 +30,7 @@ from tests.integration._support.database_relay import dropped_connection_relay from tests.integration._support.process import OwnedProxy, owned_proxy_process from tests.integration._support.wire import Reply, Request, Wire, wire_server -pytestmark: Final = pytest.mark.timeout(300) +pytestmark: Final = [pytest.mark.timeout(300), pytest.mark.shares_worker_state] CLIENT_ID: Final = "integration-entra-client" CLIENT_SECRET: Final = "integration-entra-secret" diff --git a/tests/integration/authorization/test_tool_search_replay_guardrails.py b/tests/integration/authorization/test_tool_search_replay_guardrails.py index 2f0ca383201..5b9e29a6d9e 100644 --- a/tests/integration/authorization/test_tool_search_replay_guardrails.py +++ b/tests/integration/authorization/test_tool_search_replay_guardrails.py @@ -18,7 +18,7 @@ from pydantic import JsonValue from tests.integration.caching.response_cache_case import await_entry, redis_store -pytestmark: Final = pytest.mark.timeout(2 * graceful_stop_seconds() + 120) +pytestmark: Final = [pytest.mark.timeout(2 * graceful_stop_seconds() + 120), pytest.mark.shares_worker_state] _POLICY: Final = "integration-tool-policy" _BLOCK: Final = "tool-permission-block" diff --git a/tests/integration/caching/test_cache_max_messages.py b/tests/integration/caching/test_cache_max_messages.py index 2079761d489..32c71e887fe 100644 --- a/tests/integration/caching/test_cache_max_messages.py +++ b/tests/integration/caching/test_cache_max_messages.py @@ -19,6 +19,8 @@ from pydantic import JsonValue, TypeAdapter from redis import Redis from redis.exceptions import ResponseError +pytestmark: Final = pytest.mark.shares_worker_state + _Turns = Callable[[str], tuple[list[JsonValue], list[JsonValue]]] _INDEXES: Final = TypeAdapter(tuple[bytes, ...]) _COUNT: Final = TypeAdapter(tuple[int]) diff --git a/tests/integration/caching/test_response_cache_chaos.py b/tests/integration/caching/test_response_cache_chaos.py index 17637c088a0..7f330658a65 100644 --- a/tests/integration/caching/test_response_cache_chaos.py +++ b/tests/integration/caching/test_response_cache_chaos.py @@ -61,6 +61,8 @@ from tests.integration.caching.response_cache_case import ( slowed, ) +pytestmark: Final = pytest.mark.shares_worker_state + RecordProperty = Callable[[str, object], None] WORKERS: Final = 2 diff --git a/tests/integration/compatibility/_status_code_audit.py b/tests/integration/compatibility/_status_code_audit.py index 59513891ec0..4ba1ee6b102 100644 --- a/tests/integration/compatibility/_status_code_audit.py +++ b/tests/integration/compatibility/_status_code_audit.py @@ -248,7 +248,7 @@ def owned_gateway( gateway_from_environment() as gateway, owned_proxy_process(gateway, directory, environment or {}, config=path, workers=workers) as owned, ): - yield Gateway(owned.gateway.client, owned.gateway.key, gateway.upstream_url) + yield owned.gateway def stream_finished(endpoint: str, lines: Sequence[str]) -> bool: diff --git a/tests/integration/compatibility/test_codex_model_catalog.py b/tests/integration/compatibility/test_codex_model_catalog.py index f1eea96a268..a3ba032e612 100644 --- a/tests/integration/compatibility/test_codex_model_catalog.py +++ b/tests/integration/compatibility/test_codex_model_catalog.py @@ -32,6 +32,8 @@ from tests.integration._support.database import read_rows from tests.integration._support.process import owned_proxy from tests.integration._support.wire import Reply, Request, Wire, wire_server +pytestmark: Final = pytest.mark.shares_worker_state + T = TypeVar("T") CLIENT_VERSION: Final = "0.159.3" diff --git a/tests/integration/compatibility/test_missing_body_param_status.py b/tests/integration/compatibility/test_missing_body_param_status.py index dd98ff6dea1..673fb0acc51 100644 --- a/tests/integration/compatibility/test_missing_body_param_status.py +++ b/tests/integration/compatibility/test_missing_body_param_status.py @@ -43,6 +43,8 @@ from tests.integration.cost_calculation.cost_tracking_case import ( SseResponse, ) +pytestmark: Final = pytest.mark.shares_worker_state + _Route = tuple[str, str, tuple[str, ...], dict[str, JsonValue], str] _EVAL_DATA_SOURCE_CONFIG: Final[dict[str, JsonValue]] = { "type": "custom", diff --git a/tests/integration/compatibility/test_openai_sdk_client_paths_chaos.py b/tests/integration/compatibility/test_openai_sdk_client_paths_chaos.py index e9ee4a323cd..9710d8a1b5c 100644 --- a/tests/integration/compatibility/test_openai_sdk_client_paths_chaos.py +++ b/tests/integration/compatibility/test_openai_sdk_client_paths_chaos.py @@ -26,6 +26,8 @@ from pydantic import JsonValue from tests.integration.cost_calculation.cost_tracking_case import JsonResponse, RoutedResponse, SseResponse +pytestmark: Final = pytest.mark.shares_worker_state + _Kind: TypeAlias = Literal["chat", "chat_stream", "messages", "responses", "completions", "moderations"] _ALL_KINDS: Final[tuple[_Kind, ...]] = ("chat", "chat_stream", "messages", "responses", "completions", "moderations") diff --git a/tests/integration/compatibility/test_persisted_toolsets.py b/tests/integration/compatibility/test_persisted_toolsets.py index 80f059b732d..156b28c5cc6 100644 --- a/tests/integration/compatibility/test_persisted_toolsets.py +++ b/tests/integration/compatibility/test_persisted_toolsets.py @@ -12,6 +12,8 @@ from integration._support.database import read_rows from integration._support.mcp import call_tool, mcp_peer, register_mcp, tool_names from integration._support.process import owned_proxy +pytestmark: Final = pytest.mark.shares_worker_state + @pytest.mark.covers("other.compatibility.mcp.persisted_tool_names_survive_candidate_startup") def test_existing_toolset_format_loads_before_start_and_keeps_sibling_denied(gateway: Gateway, tmp_path: Path) -> None: diff --git a/tests/integration/compatibility/test_status_code_followups.py b/tests/integration/compatibility/test_status_code_followups.py index 466eeeddde1..662f37aed1a 100644 --- a/tests/integration/compatibility/test_status_code_followups.py +++ b/tests/integration/compatibility/test_status_code_followups.py @@ -207,7 +207,7 @@ def router_default_proxy(tmp_path_factory: pytest.TempPathFactory) -> Iterator[G }, ) with owned_proxy_process(gateway, directory, {}, config=config) as owned: - yield Gateway(owned.gateway.client, owned.gateway.key, gateway.upstream_url) + yield owned.gateway def test_ocr_multipart_without_file_returns_400(gateway: Gateway) -> None: diff --git a/tests/integration/compatibility/test_status_code_followups_chaos.py b/tests/integration/compatibility/test_status_code_followups_chaos.py index 1cb7f45e681..f1ea9e60330 100644 --- a/tests/integration/compatibility/test_status_code_followups_chaos.py +++ b/tests/integration/compatibility/test_status_code_followups_chaos.py @@ -41,7 +41,7 @@ from pydantic import JsonValue from tests.integration.cost_calculation.cost_tracking_case import StoredResponse -pytestmark: Final = pytest.mark.timeout(480) +pytestmark: Final = [pytest.mark.timeout(480), pytest.mark.shares_worker_state] _STARTED_WORKER: Final = re.compile(r"Started server process \[(\d+)\]") _MARKER_HEADER: Final = "x-litellm-spend-logs-metadata" diff --git a/tests/integration/configuration/test_codex_catalog_boot.py b/tests/integration/configuration/test_codex_catalog_boot.py index 4868e43b855..1e691f0caaf 100644 --- a/tests/integration/configuration/test_codex_catalog_boot.py +++ b/tests/integration/configuration/test_codex_catalog_boot.py @@ -16,6 +16,8 @@ from pydantic import JsonValue from tests.integration._support.client import JSON_OBJECT, Gateway, Scenario, eventually, object_value, string_value from tests.integration._support.process import owned_proxy, owned_proxy_process +pytestmark: Final = pytest.mark.shares_worker_state + CLIENT_VERSION: Final = "0.159.3" FRESH_CONNECTION: Final = {"Connection": "close"} CONSISTENT_READS: Final = 8 diff --git a/tests/integration/configuration/test_effective_settings.py b/tests/integration/configuration/test_effective_settings.py index 014cfd3a866..029a5c7220e 100644 --- a/tests/integration/configuration/test_effective_settings.py +++ b/tests/integration/configuration/test_effective_settings.py @@ -23,6 +23,8 @@ from tests.integration._support.database import read_rows, scratch_database from tests.integration._support.process import owned_proxy from tests.integration._support.wire import Reply, Request, Wire, wire_server +pytestmark: Final = pytest.mark.shares_worker_state + _JSON_OBJECT: Final = TypeAdapter(dict[str, JsonValue]) diff --git a/tests/integration/conftest.py b/tests/integration/conftest.py index 92fed7200e1..672ab7b6c6a 100644 --- a/tests/integration/conftest.py +++ b/tests/integration/conftest.py @@ -5,7 +5,7 @@ import json import os import re from collections.abc import AsyncIterator, Generator, Iterator, Mapping, Sequence -from contextlib import ExitStack +from contextlib import ExitStack, contextmanager from copy import copy from functools import partial from importlib.metadata import version @@ -23,6 +23,7 @@ from tests.integration._support.client import Gateway, Scenario, eventually, gat from tests.integration._support.generation import LIFECYCLE_SETTINGS from tests.integration._support.manifest import OWNED_DIRECTORIES from tests.integration._support.ordering import collected_case, fixture_groups, scheduling_groups +from tests.integration._support.process import OWNED_STATE_VARIABLE, OwnedState from tests.integration._support.provider import SharedProvider, shared_provider from tests.integration._support.routing import RoutingPlugin from tests.integration._support.sdk_models import build_sdk_models @@ -69,6 +70,7 @@ def pytest_addoption(parser: pytest.Parser) -> None: def pytest_configure(config: pytest.Config) -> None: config.addinivalue_line("markers", "integration: owned real-service integration contracts") config.addinivalue_line("markers", "covers(*ids): legacy contract IDs kept for existing tests, not enforced") + config.addinivalue_line("markers", "shares_worker_state: owned proxies here share the worker DB and Redis") config.stash[REPORTS] = [] config.stash[SHARD_FILES] = _shard_files(SHARD_PATH.validate_python(config.getoption("integration_shard_files"))) config.pluginmanager.register(IntegrationReportPlugin(config)) @@ -295,6 +297,32 @@ def pytest_sessionfinish(session: pytest.Session, exitstatus: int) -> None: session.exitstatus = pytest.ExitCode.OK +@contextmanager +def _owned_state(item: pytest.Item) -> Iterator[None]: + with pytest.MonkeyPatch.context() as environment: + if item.get_closest_marker("shares_worker_state") is not None: + environment.setenv(OWNED_STATE_VARIABLE, OwnedState.SHARED_WITH_WORKER.value) + yield + + +@pytest.hookimpl(wrapper=True) +def pytest_runtest_setup(item: pytest.Item) -> Generator[None, None, None]: + with _owned_state(item): + return (yield) + + +@pytest.hookimpl(wrapper=True) +def pytest_runtest_call(item: pytest.Item) -> Generator[None, None, None]: + with _owned_state(item): + return (yield) + + +@pytest.hookimpl(wrapper=True) +def pytest_runtest_teardown(item: pytest.Item, nextitem: pytest.Item | None) -> Generator[None, None, None]: + with _owned_state(item): + return (yield) + + @pytest.fixture def gateway() -> Iterator[Gateway]: with gateway_from_environment() as value: diff --git a/tests/integration/cost_calculation/test_base_model_cross_provider_spend.py b/tests/integration/cost_calculation/test_base_model_cross_provider_spend.py index 02d8ca9cc73..c39ac5345c4 100644 --- a/tests/integration/cost_calculation/test_base_model_cross_provider_spend.py +++ b/tests/integration/cost_calculation/test_base_model_cross_provider_spend.py @@ -16,7 +16,7 @@ from integration._support.process import OwnedProxy, owned_proxy_process from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import JsonValue -pytestmark: Final = pytest.mark.timeout(240) +pytestmark: Final = [pytest.mark.timeout(240), pytest.mark.shares_worker_state] _MODEL: Final = "cc-subscription-sol" _INPUT_TOKENS: Final = 1000 diff --git a/tests/integration/cost_calculation/test_bridged_previous_response_id_spend.py b/tests/integration/cost_calculation/test_bridged_previous_response_id_spend.py index 2a8c63fff3d..fcedfcf38ec 100644 --- a/tests/integration/cost_calculation/test_bridged_previous_response_id_spend.py +++ b/tests/integration/cost_calculation/test_bridged_previous_response_id_spend.py @@ -17,7 +17,7 @@ from integration._support.workers import cost_map_environment from integration.cost_calculation.cost_tracking_case import COST_MAP from pydantic import JsonValue, TypeAdapter -pytestmark: Final = pytest.mark.timeout(240) +pytestmark: Final = [pytest.mark.timeout(240), pytest.mark.shares_worker_state] _JSON_OBJECT: Final = TypeAdapter(dict[str, JsonValue]) _MESSAGES: Final = TypeAdapter(list[dict[str, JsonValue]]) diff --git a/tests/integration/cost_calculation/test_responses_websocket_guardrail_spend.py b/tests/integration/cost_calculation/test_responses_websocket_guardrail_spend.py index 8af70d39186..6377be6d095 100644 --- a/tests/integration/cost_calculation/test_responses_websocket_guardrail_spend.py +++ b/tests/integration/cost_calculation/test_responses_websocket_guardrail_spend.py @@ -21,7 +21,7 @@ from integration._support.responses_ws_peer import PROVIDER_MODEL, USAGE, create from integration._support.tls import write_self_signed_cert from pydantic import JsonValue -pytestmark: Final = pytest.mark.timeout(2 * graceful_stop_seconds() + 120) +pytestmark: Final = [pytest.mark.timeout(2 * graceful_stop_seconds() + 120), pytest.mark.shares_worker_state] _INPUT_COST_PER_TOKEN: Final = 0.002 _OUTPUT_COST_PER_TOKEN: Final = 0.004 diff --git a/tests/integration/database/test_reader_writer_regeneration.py b/tests/integration/database/test_reader_writer_regeneration.py index 62c152167ab..fc5dbe87e75 100644 --- a/tests/integration/database/test_reader_writer_regeneration.py +++ b/tests/integration/database/test_reader_writer_regeneration.py @@ -17,6 +17,8 @@ from integration._support.wire import Reply, Request, wire_server from psycopg import sql from pydantic import BaseModel, JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + @pytest.mark.covers("other.database.regeneration.writer_updates_dependent_grants") def test_key_regeneration_uses_writer_with_a_real_readonly_reader(gateway: Gateway, tmp_path: Path) -> None: diff --git a/tests/integration/management/test_batch_file_usage_caps.py b/tests/integration/management/test_batch_file_usage_caps.py index 683425f3b1d..2d8186d87ae 100644 --- a/tests/integration/management/test_batch_file_usage_caps.py +++ b/tests/integration/management/test_batch_file_usage_caps.py @@ -59,7 +59,7 @@ from integration.management._batch_file_caps import ( from pydantic import JsonValue from redis import Redis -pytestmark = pytest.mark.timeout(600) +pytestmark: Final = [pytest.mark.timeout(600), pytest.mark.shares_worker_state] YAML_RECORDS: Final = 3 YAML_UPLOADS: Final = 4 diff --git a/tests/integration/management/test_batch_file_usage_caps_chaos.py b/tests/integration/management/test_batch_file_usage_caps_chaos.py index 5f31fa94230..36730278948 100644 --- a/tests/integration/management/test_batch_file_usage_caps_chaos.py +++ b/tests/integration/management/test_batch_file_usage_caps_chaos.py @@ -46,7 +46,7 @@ from integration.management._batch_file_caps import ( ) from pydantic import JsonValue -pytestmark = pytest.mark.timeout(600) +pytestmark: Final = [pytest.mark.timeout(600), pytest.mark.shares_worker_state] WORKERS: Final = 2 UPLOAD_CAP: Final = 5 diff --git a/tests/integration/management/test_budget_reset_windows.py b/tests/integration/management/test_budget_reset_windows.py index 34a0823cea2..999e42bf13d 100644 --- a/tests/integration/management/test_budget_reset_windows.py +++ b/tests/integration/management/test_budget_reset_windows.py @@ -49,7 +49,7 @@ from integration.cost_calculation.cost_tracking_case import JsonResponse, SseRes from pydantic import JsonValue, TypeAdapter from redis import Redis -pytestmark: Final = pytest.mark.timeout(240) +pytestmark: Final = [pytest.mark.timeout(240), pytest.mark.shares_worker_state] DAY: Final = 86400 HOUR: Final = 3600 diff --git a/tests/integration/management/test_budget_updates.py b/tests/integration/management/test_budget_updates.py index 7a79e35d2af..6939b447866 100644 --- a/tests/integration/management/test_budget_updates.py +++ b/tests/integration/management/test_budget_updates.py @@ -9,6 +9,8 @@ from tests.integration._support.client import Gateway, string_value from tests.integration._support.database import read_rows from tests.integration._support.process import owned_proxy +pytestmark: Final = pytest.mark.shares_worker_state + def _persisted_reset_at(budget_id: str) -> datetime: rows: Final = read_rows( diff --git a/tests/integration/management/test_claude_code_plugin_installation_preference.py b/tests/integration/management/test_claude_code_plugin_installation_preference.py index b8afd3ed65d..aaf6f02ff4e 100644 --- a/tests/integration/management/test_claude_code_plugin_installation_preference.py +++ b/tests/integration/management/test_claude_code_plugin_installation_preference.py @@ -14,6 +14,8 @@ from tests.integration._support.client import JSON_OBJECT, Gateway, Scenario, ev from tests.integration._support.database import read_rows, write_rows from tests.integration._support.process import OwnedProxy, owned_proxy_process +pytestmark: Final = pytest.mark.shares_worker_state + PREFERENCES: Final = ("available", "auto_install", "required") INVALID_PREFERENCES: Final[tuple[JsonValue, ...]] = ( "", diff --git a/tests/integration/management/test_credential_federation_values.py b/tests/integration/management/test_credential_federation_values.py index 99be9f0cb8e..2ca9c0cb47c 100644 --- a/tests/integration/management/test_credential_federation_values.py +++ b/tests/integration/management/test_credential_federation_values.py @@ -48,6 +48,8 @@ from tests.integration._support.database import read_rows from tests.integration._support.process import OwnedProxy, graceful_stop_seconds, owned_proxy_process from tests.integration._support.wire import Reply, Request, Wire, wire_server +pytestmark: Final = pytest.mark.shares_worker_state + T = TypeVar("T") diff --git a/tests/integration/management/test_fallback_management_public_names.py b/tests/integration/management/test_fallback_management_public_names.py index 75dad3ae428..a5b311e38eb 100644 --- a/tests/integration/management/test_fallback_management_public_names.py +++ b/tests/integration/management/test_fallback_management_public_names.py @@ -29,7 +29,7 @@ from tests.integration._support.process import ( ) from tests.integration._support.redis_process import owned_redis -pytestmark: Final = pytest.mark.timeout(300) +pytestmark: Final = [pytest.mark.timeout(300), pytest.mark.shares_worker_state] PROVIDER_KEY: Final = "integration-provider-key" EVICTION_WARNING: Final = "config cache eviction of router_settings failed" diff --git a/tests/integration/management/test_project_lifecycle.py b/tests/integration/management/test_project_lifecycle.py index d503daf6a0f..54989c67bea 100644 --- a/tests/integration/management/test_project_lifecycle.py +++ b/tests/integration/management/test_project_lifecycle.py @@ -31,6 +31,8 @@ from pydantic import JsonValue from litellm.models.user import LiteLLM_UserTable from litellm.proxy.auth.auth_checks import ExperimentalUIJWTToken +pytestmark: Final = pytest.mark.shares_worker_state + _OWNED_PROXY_SALT_KEY: Final = "sk-integration-salt" _OWNERSHIP_POOL_LIMIT: Final = 4 _CONNECTION_BUDGET: Final = re.compile( diff --git a/tests/integration/management/test_ptu_share_guards.py b/tests/integration/management/test_ptu_share_guards.py index 580750bcdef..1b804d1ae7e 100644 --- a/tests/integration/management/test_ptu_share_guards.py +++ b/tests/integration/management/test_ptu_share_guards.py @@ -17,7 +17,7 @@ from integration._support.process import OwnedProxy, graceful_stop_seconds, owne from integration._support.wire import Reply, Request, wire_server from pydantic import JsonValue, TypeAdapter -pytestmark: Final = pytest.mark.timeout(4 * graceful_stop_seconds() + 240) +pytestmark: Final = [pytest.mark.timeout(4 * graceful_stop_seconds() + 240), pytest.mark.shares_worker_state] _RELOAD_SECONDS: Final = 5 _PROPAGATION_SECONDS: Final = 6 * _RELOAD_SECONDS diff --git a/tests/integration/management/test_scim_admin_group.py b/tests/integration/management/test_scim_admin_group.py index b7bf3cfe968..3aeed45d186 100644 --- a/tests/integration/management/test_scim_admin_group.py +++ b/tests/integration/management/test_scim_admin_group.py @@ -13,6 +13,8 @@ from integration._support.database import read_rows from integration._support.process import owned_proxy_process from pydantic import BaseModel, ConfigDict, JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + SCIM_HEADERS: Final = {"Content-Type": "application/scim+json"} SCIM_CORE_USER_SCHEMA: Final = "urn:ietf:params:scim:schemas:core:2.0:User" SCIM_CORE_GROUP_SCHEMA: Final = "urn:ietf:params:scim:schemas:core:2.0:Group" diff --git a/tests/integration/management/test_scim_group_bulk_roster.py b/tests/integration/management/test_scim_group_bulk_roster.py index 2af6ae0a83d..6d8e657c5ce 100644 --- a/tests/integration/management/test_scim_group_bulk_roster.py +++ b/tests/integration/management/test_scim_group_bulk_roster.py @@ -36,6 +36,8 @@ from integration._support.process import OwnedProxy, graceful_stop_seconds, owne from psycopg import sql from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + GROUP_SCHEMA: Final = "urn:ietf:params:scim:schemas:core:2.0:Group" PATCH_OP_SCHEMA: Final = "urn:ietf:params:scim:api:messages:2.0:PatchOp" SMALL_PUSH: Final = 5 diff --git a/tests/integration/management/test_scim_group_pathless_patch.py b/tests/integration/management/test_scim_group_pathless_patch.py index efd3afba5b3..d3bdfb840c3 100644 --- a/tests/integration/management/test_scim_group_pathless_patch.py +++ b/tests/integration/management/test_scim_group_pathless_patch.py @@ -15,6 +15,8 @@ from integration._support.process import owned_proxy_process from integration.authorization._guardrail_opt_out import upstream_hits from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + PATCH_OP_SCHEMA: Final = "urn:ietf:params:scim:api:messages:2.0:PatchOp" GROUP_SCHEMA: Final = "urn:ietf:params:scim:schemas:core:2.0:Group" BURST_TEAMS: Final = 10 diff --git a/tests/integration/management/test_search_tool_crud.py b/tests/integration/management/test_search_tool_crud.py index d7f161431c6..f46da964e14 100644 --- a/tests/integration/management/test_search_tool_crud.py +++ b/tests/integration/management/test_search_tool_crud.py @@ -5,12 +5,15 @@ from types import MappingProxyType from typing import Final import yaml +import pytest from integration._support.client import Gateway, string_value from integration._support.database import read_rows from integration._support.process import owned_proxy from integration._support.wire import Reply, Request, wire_server from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + _PROXY_CONFIG: Final = Path(__file__).resolve().parents[1] / "proxy_config.yaml" _ORIGINAL_KEY: Final = "exa-original-provider-key" _ROTATED_KEY: Final = "exa-rotated-provider-key" diff --git a/tests/integration/management/test_sso_sign_in.py b/tests/integration/management/test_sso_sign_in.py index 628f1807d55..c0c03571a13 100644 --- a/tests/integration/management/test_sso_sign_in.py +++ b/tests/integration/management/test_sso_sign_in.py @@ -18,7 +18,7 @@ from tests.integration._support.database import read_rows from tests.integration._support.process import owned_proxy from tests.integration._support.wire import Reply, Request, wire_server -pytestmark: Final = pytest.mark.timeout(300) +pytestmark: Final = [pytest.mark.timeout(300), pytest.mark.shares_worker_state] CLIENT_ID: Final = "integration-sso-client" CLIENT_SECRET: Final = "integration-sso-secret" diff --git a/tests/integration/management/test_team_budget_duration_defaults.py b/tests/integration/management/test_team_budget_duration_defaults.py index fd459f6a7ef..4c3eb71603a 100644 --- a/tests/integration/management/test_team_budget_duration_defaults.py +++ b/tests/integration/management/test_team_budget_duration_defaults.py @@ -10,6 +10,8 @@ from tests.integration._support.client import Gateway, string_value from tests.integration._support.database import read_rows from tests.integration._support.process import owned_proxy +pytestmark: Final = pytest.mark.shares_worker_state + def _budget_row(team_id: str) -> dict[str, JsonValue]: rows: Final = read_rows( diff --git a/tests/integration/management/test_team_delete_prometheus.py b/tests/integration/management/test_team_delete_prometheus.py index b93e14a0e8d..a089f8ca70d 100644 --- a/tests/integration/management/test_team_delete_prometheus.py +++ b/tests/integration/management/test_team_delete_prometheus.py @@ -24,6 +24,8 @@ from tests.integration._support.client import Gateway, Scenario, eventually, obj from tests.integration._support.database import read_rows from tests.integration._support.process import graceful_stop_seconds, owned_proxy_process +pytestmark: Final = pytest.mark.shares_worker_state + METRIC: Final = "litellm_team_members_metric" METRICS_ROUTE: Final = "/metrics/" MEMBERS: Final = 3 diff --git a/tests/integration/management/test_tool_policy_user.py b/tests/integration/management/test_tool_policy_user.py index 6c0daf50d37..8bc77523aff 100644 --- a/tests/integration/management/test_tool_policy_user.py +++ b/tests/integration/management/test_tool_policy_user.py @@ -22,6 +22,8 @@ from pydantic import BaseModel, JsonValue from litellm.repositories.chunked_in import IN_LIST_CHUNK_SIZE +pytestmark: Final = pytest.mark.shares_worker_state + AUDIENCE: Final = "litellm-integration" KEY_ID: Final = "integration-signing-key" CLIENT_CLAIM: Final = "client_id" diff --git a/tests/integration/management/test_vector_store_file_list_managed_ids.py b/tests/integration/management/test_vector_store_file_list_managed_ids.py index d5765827969..16d33ee83af 100644 --- a/tests/integration/management/test_vector_store_file_list_managed_ids.py +++ b/tests/integration/management/test_vector_store_file_list_managed_ids.py @@ -26,6 +26,8 @@ from integration._support.ports import allocate_port from openai import APIStatusError, AsyncOpenAI, OpenAI from pydantic import JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + MANAGED_PREFIX: Final = "litellm_proxy:" CARRIED_PROVIDER_FILE_ID: Final = re.compile(r"(?:^|;)llm_output_file_id,([^;]+)") UPLOAD_FILENAME: Final = re.compile(rb'filename="([^"]+)"') diff --git a/tests/integration/mcp/test_mcp_accounting_guardrails.py b/tests/integration/mcp/test_mcp_accounting_guardrails.py index 72d6c860d2f..aff67146ce4 100644 --- a/tests/integration/mcp/test_mcp_accounting_guardrails.py +++ b/tests/integration/mcp/test_mcp_accounting_guardrails.py @@ -36,6 +36,8 @@ from integration._support.process import owned_proxy from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + DEFAULT_COST: Final = 0.25 ADD_COST: Final = 0.5 FORBIDDEN: Final = "forbidden-integration-word" diff --git a/tests/integration/mcp/test_mcp_agent_365_guardrail.py b/tests/integration/mcp/test_mcp_agent_365_guardrail.py index ee746bf3dbf..2ee3b5e6b89 100644 --- a/tests/integration/mcp/test_mcp_agent_365_guardrail.py +++ b/tests/integration/mcp/test_mcp_agent_365_guardrail.py @@ -27,6 +27,8 @@ from integration._support.mcp import ( from integration._support.process import owned_proxy_process from integration._support.wire import Reply, Request, wire_server +pytestmark: Final = pytest.mark.shares_worker_state + TENANT: Final = "00000000-0000-4000-8000-0000000a3650" REJECTED: Final = "Agent 365 guardrail rejected the tool call" GUARDRAIL_ROWS: Final = ( diff --git a/tests/integration/mcp/test_mcp_akto_logging_only.py b/tests/integration/mcp/test_mcp_akto_logging_only.py index 5b3fd84aa83..9b891fcfba2 100644 --- a/tests/integration/mcp/test_mcp_akto_logging_only.py +++ b/tests/integration/mcp/test_mcp_akto_logging_only.py @@ -20,6 +20,8 @@ from integration._support.mcp import EntryPoint, McpCaller, McpPeer, echo_tool, from integration._support.process import owned_proxy_process from integration._support.wire import Reply, Request, Wire, wire_server +pytestmark: Final = pytest.mark.shares_worker_state + AKTO_KEY: Final = "synthetic-akto-mcp-log-key" BLOCK_MARK: Final = "SYNTHETIC-AKTO-BLOCK" RESPONSE_CHECK: Final = {"akto_connector": "litellm", "response_guardrails": "true", "ingest_data": "true"} diff --git a/tests/integration/mcp/test_mcp_llm_endpoints.py b/tests/integration/mcp/test_mcp_llm_endpoints.py index e288cef3699..9f6101423bf 100644 --- a/tests/integration/mcp/test_mcp_llm_endpoints.py +++ b/tests/integration/mcp/test_mcp_llm_endpoints.py @@ -41,7 +41,7 @@ from openai.types.responses.tool_param import Mcp from pydantic import JsonValue, TypeAdapter -pytestmark: Final = pytest.mark.xdist_group("mcp_llm_private_gateway") +pytestmark: Final = [pytest.mark.xdist_group("mcp_llm_private_gateway"), pytest.mark.shares_worker_state] @pytest.fixture(scope="session") diff --git a/tests/integration/mcp/test_mcp_management.py b/tests/integration/mcp/test_mcp_management.py index 9775221c6a6..90435f49b34 100644 --- a/tests/integration/mcp/test_mcp_management.py +++ b/tests/integration/mcp/test_mcp_management.py @@ -27,6 +27,8 @@ from pydantic import BaseModel, JsonValue from litellm.models.user import LiteLLM_UserTable from litellm.proxy.auth.auth_checks import ExperimentalUIJWTToken +pytestmark: Final = pytest.mark.shares_worker_state + ADD: Final = {"a": 4, "b": 5} diff --git a/tests/integration/mcp/test_mcp_oauth_flows.py b/tests/integration/mcp/test_mcp_oauth_flows.py index 563d35bf794..327b053d333 100644 --- a/tests/integration/mcp/test_mcp_oauth_flows.py +++ b/tests/integration/mcp/test_mcp_oauth_flows.py @@ -43,6 +43,8 @@ from integration._support.process import owned_proxy from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + ADD: Final = {"a": 2, "b": 3} CLIENT_REDIRECT: Final = "http://127.0.0.1:9/cb" ACCEPT: Final = {"Accept": "application/json, text/event-stream"} diff --git a/tests/integration/mcp/test_mcp_protocol_errors.py b/tests/integration/mcp/test_mcp_protocol_errors.py index 257e90e9d0e..4930258ec07 100644 --- a/tests/integration/mcp/test_mcp_protocol_errors.py +++ b/tests/integration/mcp/test_mcp_protocol_errors.py @@ -9,6 +9,8 @@ from integration._support.client import Gateway from integration._support.mcp import McpPeer, call_tool, register_mcp, tool_names from integration._support.wire import Reply, Request, wire_server +pytestmark: Final = pytest.mark.shares_worker_state + @pytest.mark.covers("other.mcp.errors.protocol_and_malformed_results_cannot_be_empty_success") def test_jsonrpc_error_and_malformed_tool_result_remain_errors(gateway: Gateway) -> None: diff --git a/tests/integration/mcp/test_mcp_resilience.py b/tests/integration/mcp/test_mcp_resilience.py index d821181962f..c335012d787 100644 --- a/tests/integration/mcp/test_mcp_resilience.py +++ b/tests/integration/mcp/test_mcp_resilience.py @@ -32,6 +32,8 @@ from integration._support.process import owned_proxy_process from integration._support.wire import Reply from pydantic import BaseModel, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + _ECHO: Final = "catalog-echo:" _PROBE: Final = "catalog-probe" _DESCRIPTION: Final = "Look up one record" diff --git a/tests/integration/mcp/test_mcp_user_env_vars.py b/tests/integration/mcp/test_mcp_user_env_vars.py index 75bb9ede182..246e6ec85d5 100644 --- a/tests/integration/mcp/test_mcp_user_env_vars.py +++ b/tests/integration/mcp/test_mcp_user_env_vars.py @@ -14,6 +14,8 @@ from integration._support.mcp import McpPeer, call_tool, mcp_peer, register_mcp, from integration._support.process import owned_proxy_process from pydantic import JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + TOKEN: Final = "USER_TOKEN" WORKSPACE: Final = "WORKSPACE" METHODS: Final = ("GET", "POST", "DELETE") diff --git a/tests/integration/mcp/test_oauth_configuration.py b/tests/integration/mcp/test_oauth_configuration.py index e3a42d630cb..9fd26461286 100644 --- a/tests/integration/mcp/test_oauth_configuration.py +++ b/tests/integration/mcp/test_oauth_configuration.py @@ -19,6 +19,8 @@ from integration._support.process import owned_proxy from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + _Upstream = Callable[[Request], Reply] CLIENT_REDIRECT: Final = "http://127.0.0.1:9/cb" diff --git a/tests/integration/messages_endpoint/chat_bridge/test_chat_bridge_multi_tool_call_blocks_wire.py b/tests/integration/messages_endpoint/chat_bridge/test_chat_bridge_multi_tool_call_blocks_wire.py index 1ee19fae339..b7766db946e 100644 --- a/tests/integration/messages_endpoint/chat_bridge/test_chat_bridge_multi_tool_call_blocks_wire.py +++ b/tests/integration/messages_endpoint/chat_bridge/test_chat_bridge_multi_tool_call_blocks_wire.py @@ -26,6 +26,8 @@ from openai.types.chat.chat_completion_chunk import ChoiceDeltaToolCall from openai.types.responses import FunctionToolParam, ResponseFunctionToolCall, ResponseOutputItemDoneEvent from pydantic import JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + _BACKEND: Final = "mistral-multi-tool" _PROVIDER_MODEL: Final = f"mistral/{_BACKEND}" _PROVIDER_KEY: Final = "integration-provider-key" diff --git a/tests/integration/messages_endpoint/providers/anthropic/test_anthropic_gateway_wire.py b/tests/integration/messages_endpoint/providers/anthropic/test_anthropic_gateway_wire.py index b13c34fe2d6..e159573e9fd 100644 --- a/tests/integration/messages_endpoint/providers/anthropic/test_anthropic_gateway_wire.py +++ b/tests/integration/messages_endpoint/providers/anthropic/test_anthropic_gateway_wire.py @@ -14,6 +14,8 @@ from integration._support.process import owned_proxy_process from integration._support.wire import Reply, Request, wire_server from pydantic import JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + _BACKEND: Final = "claude-sonnet-4-6" _API_KEY: Final = "synthetic-anthropic-key" _JSON_OBJECT: Final = TypeAdapter(dict[str, JsonValue]) diff --git a/tests/integration/messages_endpoint/providers/anthropic/test_anthropic_messages_pre_content_retry_chaos.py b/tests/integration/messages_endpoint/providers/anthropic/test_anthropic_messages_pre_content_retry_chaos.py index 8944d1990d9..9dd1a2d2c18 100644 --- a/tests/integration/messages_endpoint/providers/anthropic/test_anthropic_messages_pre_content_retry_chaos.py +++ b/tests/integration/messages_endpoint/providers/anthropic/test_anthropic_messages_pre_content_retry_chaos.py @@ -45,7 +45,7 @@ _STARTED_WORKER: Final = re.compile(r"Started server process \[(\d+)\]") _OUTAGE_STATUSES: Final = (529, 429, 503) _ROUTING_ENCODED_ID: Final = re.compile(r"resp_([A-Za-z0-9+/]+=*)") -pytestmark = pytest.mark.timeout(240) +pytestmark: Final = [pytest.mark.timeout(240), pytest.mark.shares_worker_state] Endpoint = Literal["messages", "chat", "responses"] diff --git a/tests/integration/messages_endpoint/providers/anthropic/test_anthropic_safeguards_beta_wire.py b/tests/integration/messages_endpoint/providers/anthropic/test_anthropic_safeguards_beta_wire.py index cabde4f1e92..cd11644eec0 100644 --- a/tests/integration/messages_endpoint/providers/anthropic/test_anthropic_safeguards_beta_wire.py +++ b/tests/integration/messages_endpoint/providers/anthropic/test_anthropic_safeguards_beta_wire.py @@ -20,6 +20,8 @@ from integration._support.vertex import service_account_json from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + BETA: Final = "dangerous-tool-use-2026-09-03" SAFEGUARDS: Final[list[JsonValue]] = [ {"type": "dangerous_tool_use", "classifier_context": {"v": 1, "permission_mode": "auto"}} diff --git a/tests/integration/messages_endpoint/providers/anthropic/test_passthrough_migration_wire.py b/tests/integration/messages_endpoint/providers/anthropic/test_passthrough_migration_wire.py index 834e10e32b8..cff6d5ef9b2 100644 --- a/tests/integration/messages_endpoint/providers/anthropic/test_passthrough_migration_wire.py +++ b/tests/integration/messages_endpoint/providers/anthropic/test_passthrough_migration_wire.py @@ -6,11 +6,15 @@ from hashlib import sha256 from pathlib import Path from typing import Final +import pytest + from integration._support.client import Gateway, JsonValue, eventually from integration._support.database import read_rows from integration._support.process import owned_proxy from integration._support.wire import Reply, Request, Wire, wire_server +pytestmark: Final = pytest.mark.shares_worker_state + _MODEL: Final = "claude-sonnet-4-5-20250929" _KEY: Final = "synthetic-anthropic-key" diff --git a/tests/integration/messages_endpoint/providers/anthropic/test_pdf_document_count_tokens_chaos.py b/tests/integration/messages_endpoint/providers/anthropic/test_pdf_document_count_tokens_chaos.py index 63325e7a005..6e4842966a8 100644 --- a/tests/integration/messages_endpoint/providers/anthropic/test_pdf_document_count_tokens_chaos.py +++ b/tests/integration/messages_endpoint/providers/anthropic/test_pdf_document_count_tokens_chaos.py @@ -25,6 +25,8 @@ from integration._support.wire import Reply, Request, wire_server from integration.providers._cache_control_marks_support import anthropic_peer, owned_config from pydantic import JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + _STARTED_WORKER: Final = re.compile(r"Started server process \[(\d+)\]") _MODEL: Final = "burst-claude" _PROVIDER_KEY: Final = "synthetic-provider-key" diff --git a/tests/integration/messages_endpoint/providers/anthropic/test_websearch_interception_wire.py b/tests/integration/messages_endpoint/providers/anthropic/test_websearch_interception_wire.py index a6f098cf64f..4e5ef521c0e 100644 --- a/tests/integration/messages_endpoint/providers/anthropic/test_websearch_interception_wire.py +++ b/tests/integration/messages_endpoint/providers/anthropic/test_websearch_interception_wire.py @@ -180,6 +180,8 @@ import httpx import pytest from integration._support.client import Gateway, eventually +pytestmark: Final = pytest.mark.shares_worker_state + _QUERY: Final = "integration capped search" _TEXT_BLOCK: Final = {"type": "text", "text": "searching once more"} _NOT_INTERCEPTED: Final = "native tool reached the provider" diff --git a/tests/integration/messages_endpoint/test_anthropic_messages_caller_tools_wire.py b/tests/integration/messages_endpoint/test_anthropic_messages_caller_tools_wire.py index 15c0593f80d..70131aa1468 100644 --- a/tests/integration/messages_endpoint/test_anthropic_messages_caller_tools_wire.py +++ b/tests/integration/messages_endpoint/test_anthropic_messages_caller_tools_wire.py @@ -15,6 +15,8 @@ from integration._support.process import owned_proxy from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + CLAUDE_MODEL: Final = "caller-tools-claude" OPENAI_MODEL: Final = "caller-tools-openai" GEMINI_MODEL: Final = "caller-tools-gemini" diff --git a/tests/integration/observability/_openinference_support.py b/tests/integration/observability/_openinference_support.py index 61def2a355b..a57e25f88ae 100644 --- a/tests/integration/observability/_openinference_support.py +++ b/tests/integration/observability/_openinference_support.py @@ -5,7 +5,7 @@ import json import uuid from collections.abc import Callable, Iterator, Mapping, Sequence from contextlib import AbstractContextManager, ExitStack, contextmanager, nullcontext -from dataclasses import dataclass +from dataclasses import dataclass, replace from itertools import chain from pathlib import Path from typing import Final @@ -854,11 +854,7 @@ def _rig( if fresh_client_connections else owned.gateway.client ) - proxy: Final = Gateway( - client=proxy_client, - key=owned.gateway.key, - upstream_url=owned.gateway.upstream_url, - ) + proxy: Final = replace(owned.gateway, client=proxy_client) scenario: Final = resources.enter_context(proxy.scenario()) model: Final = scenario.model( model=model_name, diff --git a/tests/integration/observability/azure_dispatch_support.py b/tests/integration/observability/azure_dispatch_support.py index 2f1d758ead4..c54293e90c8 100644 --- a/tests/integration/observability/azure_dispatch_support.py +++ b/tests/integration/observability/azure_dispatch_support.py @@ -10,7 +10,7 @@ from typing import Final import yaml from integration._support.client import Gateway, object_value -from integration._support.process import OwnedProxy, owned_proxy_process +from integration._support.process import OwnedProxy, OwnedState, owned_proxy_process from integration._support.redis_process import OwnedRedis, owned_redis from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import JsonValue @@ -528,6 +528,7 @@ def dispatch_proxy( }, config=config, workers=workers, + state=OwnedState.SHARED_WITH_WORKER, ) as owned: yield owned diff --git a/tests/integration/observability/test_akto_logging_only.py b/tests/integration/observability/test_akto_logging_only.py index 31adacdec75..e6cb5b5c8df 100644 --- a/tests/integration/observability/test_akto_logging_only.py +++ b/tests/integration/observability/test_akto_logging_only.py @@ -26,6 +26,8 @@ from integration._support.database import read_rows from integration._support.process import owned_proxy_process from integration._support.wire import Reply, Request, Wire, wire_server +pytestmark: Final = pytest.mark.shares_worker_state + LOG_KEY: Final = "synthetic-akto-log-key" INPUT_KEY: Final = "synthetic-akto-input-key" OUTPUT_KEY: Final = "synthetic-akto-output-key" diff --git a/tests/integration/observability/test_azure_content_safety_audit.py b/tests/integration/observability/test_azure_content_safety_audit.py index 85ced5f719b..dc0346aee75 100644 --- a/tests/integration/observability/test_azure_content_safety_audit.py +++ b/tests/integration/observability/test_azure_content_safety_audit.py @@ -15,6 +15,8 @@ from integration._support.process import OwnedProxy, owned_proxy_process from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + _ATTACK_MARKER: Final = "synthetic-attack-marker" _MODERATION_MARKER: Final = "synthetic-moderation-marker" diff --git a/tests/integration/observability/test_azure_content_safety_endpoints.py b/tests/integration/observability/test_azure_content_safety_endpoints.py index cd52122267f..dd2efb6ad87 100644 --- a/tests/integration/observability/test_azure_content_safety_endpoints.py +++ b/tests/integration/observability/test_azure_content_safety_endpoints.py @@ -13,6 +13,8 @@ from integration._support.process import owned_proxy from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + _ATTACK_MARKER: Final = "synthetic-attack-marker" _AZURE_TARGET_PREFIX: Final = "/contentsafety/text:shieldPrompt?api-version=" diff --git a/tests/integration/observability/test_azure_storage_client_ttl.py b/tests/integration/observability/test_azure_storage_client_ttl.py index f8f32820daa..1890fad3589 100644 --- a/tests/integration/observability/test_azure_storage_client_ttl.py +++ b/tests/integration/observability/test_azure_storage_client_ttl.py @@ -4,6 +4,8 @@ from collections.abc import Callable from pathlib import Path from typing import Final +import pytest + from _azure_storage_support import ( SINK_HOSTS, RecordingDataLakeSink, @@ -18,6 +20,8 @@ from integration._support.process import owned_proxy from integration._support.tls import server_context, write_self_signed_cert from integration._support.wire import Reply, Request, wire_server +pytestmark: Final = pytest.mark.shares_worker_state + WORKERS: Final = 2 FLUSH_SECONDS: Final = "1" diff --git a/tests/integration/observability/test_bedrock_converse_stream_post_call_guardrail.py b/tests/integration/observability/test_bedrock_converse_stream_post_call_guardrail.py index c55bd181ecb..9546b9549ff 100644 --- a/tests/integration/observability/test_bedrock_converse_stream_post_call_guardrail.py +++ b/tests/integration/observability/test_bedrock_converse_stream_post_call_guardrail.py @@ -36,6 +36,8 @@ from integration._support.process import owned_proxy_process from integration._support.upstream import aws_event_stream_frame from integration._support.wire import Reply, Request, Wire, wire_server +pytestmark: Final = pytest.mark.shares_worker_state + MODEL_ID: Final = "anthropic.claude-sonnet-5-v1:0" ROUTER_MODEL: Final = "converse-relay" EVENT_STREAM: Final = "application/vnd.amazon.eventstream" diff --git a/tests/integration/observability/test_bedrock_invoke_stream_post_call_guardrail.py b/tests/integration/observability/test_bedrock_invoke_stream_post_call_guardrail.py index 3ef73f613fe..ae243324e27 100644 --- a/tests/integration/observability/test_bedrock_invoke_stream_post_call_guardrail.py +++ b/tests/integration/observability/test_bedrock_invoke_stream_post_call_guardrail.py @@ -38,6 +38,8 @@ from integration._support.process import owned_proxy_process from integration._support.upstream import aws_event_stream_frame from integration._support.wire import Reply, Request, Wire, wire_server +pytestmark: Final = pytest.mark.shares_worker_state + CLAUDE_ID: Final = "anthropic.claude-sonnet-5-v1:0" LLAMA_ID: Final = "meta.llama4-maverick-17b-instruct-v1:0" PROFILE_ID: Final = "arn:aws:bedrock:us-east-1:000000000000:application-inference-profile/relayprofile" diff --git a/tests/integration/observability/test_cache_hit_guardrail_metrics.py b/tests/integration/observability/test_cache_hit_guardrail_metrics.py index 2ca9e0c6410..c52649c46df 100644 --- a/tests/integration/observability/test_cache_hit_guardrail_metrics.py +++ b/tests/integration/observability/test_cache_hit_guardrail_metrics.py @@ -18,6 +18,8 @@ from integration._support.process import OwnedProxy, owned_proxy_process from integration._support.wire import Reply, Request, Wire, wire_server from prometheus_client.parser import text_string_to_metric_families +pytestmark: Final = pytest.mark.shares_worker_state + GUARDRAIL_PATH: Final = "/beta/litellm_basic_guardrail_api" DEPLOYMENT_FAILURE: Final = "litellm_deployment_failure_responses_total" DEPLOYMENT_REQUESTS: Final = "litellm_deployment_total_requests_total" diff --git a/tests/integration/observability/test_cache_hit_guardrail_metrics_chaos.py b/tests/integration/observability/test_cache_hit_guardrail_metrics_chaos.py index d4b62b36e8b..b6014c72a2f 100644 --- a/tests/integration/observability/test_cache_hit_guardrail_metrics_chaos.py +++ b/tests/integration/observability/test_cache_hit_guardrail_metrics_chaos.py @@ -7,6 +7,7 @@ from pathlib import Path from typing import Final import httpx +import pytest from integration._support.client import Gateway, eventually, object_value, string_value from integration._support.process import owned_proxy_process from integration._support.redis_process import owned_redis @@ -24,6 +25,8 @@ from test_cache_hit_guardrail_metrics import ( _rig, ) +pytestmark: Final = pytest.mark.shares_worker_state + BURST: Final = 10 diff --git a/tests/integration/observability/test_callback_delivery.py b/tests/integration/observability/test_callback_delivery.py index e786e655a13..f3e8daab014 100644 --- a/tests/integration/observability/test_callback_delivery.py +++ b/tests/integration/observability/test_callback_delivery.py @@ -14,6 +14,8 @@ from integration._support.database import read_rows from integration._support.process import graceful_stop_seconds, owned_proxy from integration._support.wire import Reply, Request, wire_server +pytestmark: Final = pytest.mark.shares_worker_state + @pytest.mark.covers( "other.observability.callbacks.credentials_stay_out_of_event_bodies", diff --git a/tests/integration/observability/test_content_filter_logging_only.py b/tests/integration/observability/test_content_filter_logging_only.py index ab1779c497d..a21c380c177 100644 --- a/tests/integration/observability/test_content_filter_logging_only.py +++ b/tests/integration/observability/test_content_filter_logging_only.py @@ -27,6 +27,8 @@ from integration._support.otlp_sink import AuditConfigWriter, SpanSinks, recorde from integration._support.process import owned_proxy_process from integration._support.wire import Reply, Request, Wire, wire_server +pytestmark: Final = pytest.mark.shares_worker_state + AWS_KEY: Final = "AKIA" + "SYNTHETIC0000000" GITHUB_TOKEN: Final = "ghp_" + "a1" * 18 FORBIDDEN_WORD: Final = "synthetic-forbidden-word" diff --git a/tests/integration/observability/test_custom_logger_redaction.py b/tests/integration/observability/test_custom_logger_redaction.py index adc0eb9103c..2abaa85df25 100644 --- a/tests/integration/observability/test_custom_logger_redaction.py +++ b/tests/integration/observability/test_custom_logger_redaction.py @@ -28,7 +28,7 @@ INSTRUCTIONS: Final = "Reply in French. redaction-instructions-canary" BRIDGED: Final = "bridged-redaction-canary" USAGE_DEPLOYMENT: Final = f"redaction-usage-{uuid.uuid4().hex}" REDACT: Final = {"x-litellm-enable-message-redaction": "true"} -pytestmark = pytest.mark.timeout(240) +pytestmark: Final = [pytest.mark.timeout(240), pytest.mark.shares_worker_state] def _provider(request: Request) -> Reply: diff --git a/tests/integration/observability/test_guardrail_effects.py b/tests/integration/observability/test_guardrail_effects.py index 4e0c3d45cdf..b38ad2a3873 100644 --- a/tests/integration/observability/test_guardrail_effects.py +++ b/tests/integration/observability/test_guardrail_effects.py @@ -27,6 +27,8 @@ from openai import APIError, AsyncOpenAI, OpenAI from openai.types.chat import ChatCompletionChunk from pydantic import BaseModel, JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + @pytest.mark.covers("other.observability.guardrails.rewrite_reaches_correct_anthropic_positions") def test_guardrail_rewrites_system_and_user_in_actual_anthropic_request(gateway: Gateway, tmp_path: Path) -> None: diff --git a/tests/integration/observability/test_guardrail_stream_scope.py b/tests/integration/observability/test_guardrail_stream_scope.py index 8b832f07aa4..071ae2dc5f9 100644 --- a/tests/integration/observability/test_guardrail_stream_scope.py +++ b/tests/integration/observability/test_guardrail_stream_scope.py @@ -25,6 +25,8 @@ from pydantic import TypeAdapter import litellm +pytestmark: Final = pytest.mark.shares_worker_state + JSON_OBJECT: Final = TypeAdapter(dict[str, JsonValue]) EndpointScope: TypeAlias = Literal["streaming", "non_streaming"] ModelKind: TypeAlias = Literal["router", "direct"] diff --git a/tests/integration/observability/test_guardrail_stream_scope_chaos.py b/tests/integration/observability/test_guardrail_stream_scope_chaos.py index 3ada2fb7753..eb534161ab4 100644 --- a/tests/integration/observability/test_guardrail_stream_scope_chaos.py +++ b/tests/integration/observability/test_guardrail_stream_scope_chaos.py @@ -25,6 +25,8 @@ from integration._support.wire import Reply, Request, Wire, wire_server from integration._support.ports import allocate_port from pydantic import TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + ChaosEndpoint: TypeAlias = Literal["chat", "messages", "responses"] CHAOS_ENDPOINTS: Final[tuple[ChaosEndpoint, ...]] = ("chat", "messages", "responses") CHAOS_MODELS: Final = MappingProxyType( diff --git a/tests/integration/observability/test_guardrail_stream_scope_matrix.py b/tests/integration/observability/test_guardrail_stream_scope_matrix.py index ec8eae66d10..d100e44889c 100644 --- a/tests/integration/observability/test_guardrail_stream_scope_matrix.py +++ b/tests/integration/observability/test_guardrail_stream_scope_matrix.py @@ -38,6 +38,8 @@ from integration.cost_calculation.cost_tracking_case import RealtimeResponse from pydantic import TypeAdapter from websockets.asyncio.server import ServerConnection, serve +pytestmark: Final = pytest.mark.shares_worker_state + Endpoint: TypeAlias = Literal["chat", "messages", "responses"] Mode: TypeAlias = Literal["pre_call", "during_call", "post_call", "logging_only"] Scope: TypeAlias = Literal["streaming", "non_streaming"] diff --git a/tests/integration/observability/test_langfuse_delivery.py b/tests/integration/observability/test_langfuse_delivery.py index 7fc2ced63e8..6c66fb0f864 100644 --- a/tests/integration/observability/test_langfuse_delivery.py +++ b/tests/integration/observability/test_langfuse_delivery.py @@ -33,6 +33,8 @@ from opentelemetry.proto.common.v1.common_pb2 import KeyValue from opentelemetry.proto.trace.v1.trace_pb2 import Span from pydantic import BaseModel, JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + PUBLIC_KEY: Final = "pk-lf-integration" SECRET_KEY: Final = "sk-lf-integration" PROJECTS_PATH: Final = "/api/public/projects" diff --git a/tests/integration/observability/test_langtrace_delivery.py b/tests/integration/observability/test_langtrace_delivery.py index f81780f9f72..4709049ef92 100644 --- a/tests/integration/observability/test_langtrace_delivery.py +++ b/tests/integration/observability/test_langtrace_delivery.py @@ -7,7 +7,7 @@ import uuid from collections.abc import Callable, Iterator, Sequence from concurrent.futures import ThreadPoolExecutor from contextlib import contextmanager -from dataclasses import dataclass, field +from dataclasses import dataclass, field, replace from itertools import repeat from pathlib import Path from queue import SimpleQueue @@ -659,7 +659,7 @@ def test_langtrace_survives_a_killed_worker(gateway: Gateway, tmp_path: Path) -> limits=httpx.Limits(max_keepalive_connections=0), ) as fresh_connections, ): - proxy: Final = Gateway(fresh_connections, owned.gateway.key, owned.gateway.upstream_url) + proxy: Final = replace(owned.gateway, client=fresh_connections) with proxy.scenario() as scenario: rig: Final = _Rig(proxy, scenario.model(api_base=provider.url + "/v1"), provider, _Sink(sink, key)) _assert_kill_and_recovery(owned, rig) diff --git a/tests/integration/observability/test_logging_only_scope_chaos.py b/tests/integration/observability/test_logging_only_scope_chaos.py index 48e9930d948..bc8f7bb9da2 100644 --- a/tests/integration/observability/test_logging_only_scope_chaos.py +++ b/tests/integration/observability/test_logging_only_scope_chaos.py @@ -49,6 +49,8 @@ from integration._support.ports import allocate_port from openai import APIConnectionError as OpenAIAPIConnectionError from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + _EDGE_DELAY_SECONDS: Final = 5 diff --git a/tests/integration/observability/test_logging_only_scope_config.py b/tests/integration/observability/test_logging_only_scope_config.py index 354e3b464fe..7c976ed25b3 100644 --- a/tests/integration/observability/test_logging_only_scope_config.py +++ b/tests/integration/observability/test_logging_only_scope_config.py @@ -48,6 +48,8 @@ from integration._support.upstream import delete_scenario, register_scenario from integration._support.wire import Reply, Request from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + def test_G7_database_invalid_scope_reads_preserve_pre_call_blocking(gateway: Gateway, tmp_path: Path) -> None: identity: Final = f"logging-scope-g7-{uuid.uuid4().hex}" diff --git a/tests/integration/observability/test_logging_only_scope_runtime.py b/tests/integration/observability/test_logging_only_scope_runtime.py index b637859da52..83bebb64960 100644 --- a/tests/integration/observability/test_logging_only_scope_runtime.py +++ b/tests/integration/observability/test_logging_only_scope_runtime.py @@ -58,6 +58,8 @@ from pydantic import JsonValue from tests.integration.cost_calculation.cost_tracking_case import JsonResponse +pytestmark: Final = pytest.mark.shares_worker_state + @pytest.mark.parametrize( ("row_id", "endpoint", "stream", "client_kind", "scope", "include_scope", "block_directions"), diff --git a/tests/integration/observability/test_otel_cache_hit_cost.py b/tests/integration/observability/test_otel_cache_hit_cost.py index 76c21b529ef..7f1ba11f566 100644 --- a/tests/integration/observability/test_otel_cache_hit_cost.py +++ b/tests/integration/observability/test_otel_cache_hit_cost.py @@ -19,6 +19,8 @@ from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import JsonValue, TypeAdapter from redis import Redis +pytestmark: Final = pytest.mark.shares_worker_state + INPUT_PRICE: Final = 0.001 OUTPUT_PRICE: Final = 0.002 PROMPT_TOKENS: Final = 7 diff --git a/tests/integration/observability/test_otel_conversation_id.py b/tests/integration/observability/test_otel_conversation_id.py index 6d0e5ea197d..af893abab48 100644 --- a/tests/integration/observability/test_otel_conversation_id.py +++ b/tests/integration/observability/test_otel_conversation_id.py @@ -23,6 +23,8 @@ from integration._support.process import OwnedProxy, owned_proxy_process, stop_r from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + MARKER: Final = re.compile(rb"otelconv-[0-9a-f]{32}") CONVERSATION: Final = "gen_ai.conversation.id" diff --git a/tests/integration/observability/test_otel_trace_response_header.py b/tests/integration/observability/test_otel_trace_response_header.py index e84aeab48cf..5cb66fded99 100644 --- a/tests/integration/observability/test_otel_trace_response_header.py +++ b/tests/integration/observability/test_otel_trace_response_header.py @@ -23,6 +23,8 @@ from integration._support.process import OwnedProxy, graceful_stop_seconds, owne from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + MARKER: Final = re.compile(rb"traceresp-[0-9a-f]{32}") TRACERESPONSE: Final = re.compile(r"00-([0-9a-f]{32})-([0-9a-f]{16})-01") SERVER: Final = 2 diff --git a/tests/integration/observability/test_passthrough_upstream_error_chaos.py b/tests/integration/observability/test_passthrough_upstream_error_chaos.py index 68ad3d50a80..cd063c469f1 100644 --- a/tests/integration/observability/test_passthrough_upstream_error_chaos.py +++ b/tests/integration/observability/test_passthrough_upstream_error_chaos.py @@ -18,6 +18,8 @@ from integration._support.wire import Reply, Request, wire_server from integration._support.ports import allocate_port from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + _GENERATE_CONTENT: Final[dict[str, JsonValue]] = {"contents": [{"role": "user", "parts": [{"text": "hi"}]}]} _NOT_FOUND_BODY: Final = json.dumps( { diff --git a/tests/integration/observability/test_passthrough_upstream_error_visibility.py b/tests/integration/observability/test_passthrough_upstream_error_visibility.py index 83560f863ed..47ccf45f829 100644 --- a/tests/integration/observability/test_passthrough_upstream_error_visibility.py +++ b/tests/integration/observability/test_passthrough_upstream_error_visibility.py @@ -18,6 +18,8 @@ from integration._support.wire import Reply, Request, Wire, wire_server from openai import AsyncOpenAI, NotFoundError, OpenAI from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + _UPSTREAM_ERROR: Final[dict[str, JsonValue]] = { "error": { "code": 404, diff --git a/tests/integration/observability/test_s3_v2_flush_surfaces.py b/tests/integration/observability/test_s3_v2_flush_surfaces.py index 5ee117a4b80..1e373171de6 100644 --- a/tests/integration/observability/test_s3_v2_flush_surfaces.py +++ b/tests/integration/observability/test_s3_v2_flush_surfaces.py @@ -22,6 +22,8 @@ from integration._support.client import Gateway, eventually from integration._support.process import owned_proxy from integration._support.wire import wire_server +pytestmark: Final = pytest.mark.shares_worker_state + PER_REQUEST_KEY: Final = re.compile(rf"^/{BUCKET}/{PREFIX}/\d{{4}}-\d{{2}}-\d{{2}}/.+\.json$") BATCH_KEY: Final = re.compile( rf"^/{BUCKET}/{PREFIX}/\d{{4}}-\d{{2}}-\d{{2}}/batch_\d{{2}}-\d{{2}}-\d{{2}}_[0-9a-f]{{32}}\.jsonl$" diff --git a/tests/integration/observability/test_s3_v2_partition_granularity.py b/tests/integration/observability/test_s3_v2_partition_granularity.py index 382463eb46c..8703c1f318c 100644 --- a/tests/integration/observability/test_s3_v2_partition_granularity.py +++ b/tests/integration/observability/test_s3_v2_partition_granularity.py @@ -33,6 +33,8 @@ from integration._support.database_relay import database_relay from integration._support.process import OwnedProxy, owned_proxy_process from integration._support.wire import Reply, Request, wire_server +pytestmark: Final = pytest.mark.shares_worker_state + FLUSH: Final = {"DEFAULT_S3_FLUSH_INTERVAL_SECONDS": "1"} HOUR: Final = {"s3_partition_granularity": "hour"} ANTHROPIC_MODEL: Final = "anthropic/claude-sonnet-4-5-20250929" diff --git a/tests/integration/observability/test_signoz_delivery.py b/tests/integration/observability/test_signoz_delivery.py index 674087eadf3..6e50a9bedf5 100644 --- a/tests/integration/observability/test_signoz_delivery.py +++ b/tests/integration/observability/test_signoz_delivery.py @@ -26,6 +26,8 @@ from opentelemetry.proto.collector.trace.v1.trace_service_pb2 import ExportTrace from opentelemetry.proto.common.v1.common_pb2 import AnyValue from pydantic import JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + MARKER: Final = re.compile(rb"signoz-[0-9a-f]{32}") JSON: Final[TypeAdapter[JsonValue]] = TypeAdapter(JsonValue) RESPONSE_ID: Final = "gen_ai.response.id" diff --git a/tests/integration/observability/test_straiker_v3_platform.py b/tests/integration/observability/test_straiker_v3_platform.py index c078cbb0c61..c74b094ee95 100644 --- a/tests/integration/observability/test_straiker_v3_platform.py +++ b/tests/integration/observability/test_straiker_v3_platform.py @@ -31,6 +31,8 @@ from integration._support.process import OwnedProxy, WorkerHealthcheck, owned_pr from integration._support.wire import Reply, Request, wire_server from integration._support.ports import allocate_port +pytestmark: Final = pytest.mark.shares_worker_state + V3_KEY: Final = "sk_agt_synthetic_integration_key" V1_KEY: Final = "synthetic-v1-collection-key" V3_PATH: Final = "/api/v3/detect" diff --git a/tests/integration/observability/test_stream_error_alert_api_base.py b/tests/integration/observability/test_stream_error_alert_api_base.py index 449e378dfc7..d7f325da8c7 100644 --- a/tests/integration/observability/test_stream_error_alert_api_base.py +++ b/tests/integration/observability/test_stream_error_alert_api_base.py @@ -34,7 +34,7 @@ from openai.types.chat import ChatCompletionChunk from openai.types.responses import ResponseFailedEvent, ResponseTextDeltaEvent from pydantic import JsonValue -pytestmark: Final = pytest.mark.timeout(2 * graceful_stop_seconds() + 120) +pytestmark: Final = [pytest.mark.timeout(2 * graceful_stop_seconds() + 120), pytest.mark.shares_worker_state] AZURE_KEY: Final = "synthetic-azure-key" API_VERSION: Final = "2025-04-01-preview" diff --git a/tests/integration/observability/test_wingback_guardrail.py b/tests/integration/observability/test_wingback_guardrail.py index e93d873345c..9206acce81d 100644 --- a/tests/integration/observability/test_wingback_guardrail.py +++ b/tests/integration/observability/test_wingback_guardrail.py @@ -13,6 +13,8 @@ from integration._support.database import read_rows from integration._support.process import owned_proxy_process from integration._support.wire import Reply, Request, Wire, wire_server +pytestmark: Final = pytest.mark.shares_worker_state + API_KEY: Final = "wbk_eg_synthetic" APP_ID: Final = "synthetic-app" DENIED: Final = "synthetic wingback denied marker" diff --git a/tests/integration/pricing/test_configured_prices.py b/tests/integration/pricing/test_configured_prices.py index b7eaadc05f9..8e97c9c8322 100644 --- a/tests/integration/pricing/test_configured_prices.py +++ b/tests/integration/pricing/test_configured_prices.py @@ -20,6 +20,8 @@ from tests.integration._support.upstream import delete_scenario, register_scenar from tests.integration.cost_calculation.cost_tracking_case import RealtimeResponse from tests.integration.pricing.test_realtime_cached_audio_pricing import one_realtime_turn +pytestmark: Final = pytest.mark.shares_worker_state + REALTIME_MODEL: Final = "gpt-realtime-2" REALTIME_INPUT_TEXT_TOKENS: Final = 10 REALTIME_INPUT_AUDIO_TOKENS: Final = 20 diff --git a/tests/integration/pricing/test_price_data_reload_prices.py b/tests/integration/pricing/test_price_data_reload_prices.py index bb24130aa6c..0aa3a3bcba2 100644 --- a/tests/integration/pricing/test_price_data_reload_prices.py +++ b/tests/integration/pricing/test_price_data_reload_prices.py @@ -18,6 +18,8 @@ from integration._support.process import graceful_stop_seconds, owned_proxy from integration._support.wire import Reply, Request, wire_server from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + _ROW: Final = "anthropic.claude-audit-reload-v1:0" _GEO: Final = f"au.{_ROW}" _FINETUNE: Final = "ft:gpt-audit-reload-2025-01-01" diff --git a/tests/integration/providers/_idle_upstream_support.py b/tests/integration/providers/_idle_upstream_support.py index 5fccd9804a6..a10017425f1 100644 --- a/tests/integration/providers/_idle_upstream_support.py +++ b/tests/integration/providers/_idle_upstream_support.py @@ -24,7 +24,7 @@ from integration._support.client import Gateway, eventually, held, list_value, o from integration._support.database import read_rows from integration._support.idle_gateway import MARKER, Arrival, Drop, IdleGateway, idle_gateway from integration._support.openai_wire import chat_reply, chat_stream, responses_reply -from integration._support.process import OwnedProxy, WorkerHealthcheck, owned_proxy_process +from integration._support.process import OwnedProxy, OwnedState, WorkerHealthcheck, owned_proxy_process from integration._support.wire import Reply, Request from openai.types.chat import ChatCompletionChunk from openai.types.responses import ResponseCompletedEvent, ResponseTextDeltaEvent @@ -704,7 +704,13 @@ def proxy_for( directory, tuple(row.deploy(row.model, gateways[row.name].url, steady) for row in rows) ) with owned_proxy_process( - gateway, directory, overrides, config=config, workers=2, healthcheck=WorkerHealthcheck.LENIENT + gateway, + directory, + overrides, + config=config, + workers=2, + healthcheck=WorkerHealthcheck.LENIENT, + state=OwnedState.SHARED_WITH_WORKER, ) as owned: yield UpstreamProxy(owned, two_workers(owned), MappingProxyType({row.name: row for row in rows}), gateways) diff --git a/tests/integration/providers/test_anthropic_refusal_stop_details_wire.py b/tests/integration/providers/test_anthropic_refusal_stop_details_wire.py index 461839ccbcc..b47dd242429 100644 --- a/tests/integration/providers/test_anthropic_refusal_stop_details_wire.py +++ b/tests/integration/providers/test_anthropic_refusal_stop_details_wire.py @@ -61,6 +61,7 @@ from pydantic import BaseModel, JsonValue pytestmark: Final = [ pytest.mark.timeout(2 * graceful_stop_seconds() + 120), pytest.mark.xdist_group("anthropic_refusal_stop_details_wire"), + pytest.mark.shares_worker_state, ] _ANTHROPIC_BASE: Final = "http://api.anthropic.com" diff --git a/tests/integration/providers/test_anthropic_thinking_signature_logging_wire.py b/tests/integration/providers/test_anthropic_thinking_signature_logging_wire.py index 6622bbb1c9a..21971335b42 100644 --- a/tests/integration/providers/test_anthropic_thinking_signature_logging_wire.py +++ b/tests/integration/providers/test_anthropic_thinking_signature_logging_wire.py @@ -37,7 +37,7 @@ from integration._support.process import owned_proxy from integration._support.wire import Wire, wire_server from pydantic import JsonValue -pytestmark = pytest.mark.timeout(240) +pytestmark: Final = [pytest.mark.timeout(240), pytest.mark.shares_worker_state] _ANTHROPIC_KEY: Final = "scripted-anthropic-key" _ANTHROPIC_BASE: Final = "http://api.anthropic.com" diff --git a/tests/integration/providers/test_anthropic_tool_search_result_passthrough_wire.py b/tests/integration/providers/test_anthropic_tool_search_result_passthrough_wire.py index 9bae4fe0052..0ce409fb89d 100644 --- a/tests/integration/providers/test_anthropic_tool_search_result_passthrough_wire.py +++ b/tests/integration/providers/test_anthropic_tool_search_result_passthrough_wire.py @@ -27,7 +27,7 @@ from integration.providers._tool_search_support import ( ) from pydantic import JsonValue -pytestmark = pytest.mark.timeout(2 * graceful_stop_seconds() + 120) +pytestmark: Final = [pytest.mark.timeout(2 * graceful_stop_seconds() + 120), pytest.mark.shares_worker_state] _ANTHROPIC_KEY: Final = "scripted-anthropic-key" _ANTHROPIC_BASE: Final = "http://api.anthropic.com" diff --git a/tests/integration/providers/test_bedrock_converse_lookaround_regex_chaos.py b/tests/integration/providers/test_bedrock_converse_lookaround_regex_chaos.py index 2031be5f3be..bb7f685da02 100644 --- a/tests/integration/providers/test_bedrock_converse_lookaround_regex_chaos.py +++ b/tests/integration/providers/test_bedrock_converse_lookaround_regex_chaos.py @@ -22,6 +22,8 @@ from integration._support.upstream import _aws_event_frame from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + _KIMI: Final = "global.moonshotai.kimi-k3" _NOVA: Final = "us.amazon.nova-lite-v1:0" _AWS: Final[dict[str, JsonValue]] = { diff --git a/tests/integration/providers/test_bedrock_converse_missing_content_wire.py b/tests/integration/providers/test_bedrock_converse_missing_content_wire.py index b0a19bbe817..3baeebcd74a 100644 --- a/tests/integration/providers/test_bedrock_converse_missing_content_wire.py +++ b/tests/integration/providers/test_bedrock_converse_missing_content_wire.py @@ -34,6 +34,8 @@ from integration._support.wire import Reply, Request, Wire, wire_server from litellm.proxy.common_utils.encrypt_decrypt_utils import decrypt_if_encrypted_with from pydantic import JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + _MODEL_ID: Final = "anthropic.claude-3-haiku-20240307-v1:0" _CONVERSE_MODEL: Final = f"bedrock/converse/{_MODEL_ID}" _INVOKE_MODEL: Final = f"bedrock/invoke/{_MODEL_ID}" diff --git a/tests/integration/providers/test_bedrock_converse_stream_event_frames_wire.py b/tests/integration/providers/test_bedrock_converse_stream_event_frames_wire.py index 2e881849dc8..a1432ffd7a0 100644 --- a/tests/integration/providers/test_bedrock_converse_stream_event_frames_wire.py +++ b/tests/integration/providers/test_bedrock_converse_stream_event_frames_wire.py @@ -28,6 +28,8 @@ from pydantic import JsonValue, TypeAdapter from litellm.proxy.common_utils.encrypt_decrypt_utils import decrypt_if_encrypted_with +pytestmark: Final = pytest.mark.shares_worker_state + _MODEL_ID: Final = "global.moonshotai.kimi-k3" _CONVERSE_MODEL: Final = f"bedrock/converse/{_MODEL_ID}" _STREAM_TARGET: Final = f"/model/{_MODEL_ID}/converse-stream" diff --git a/tests/integration/providers/test_bedrock_kimi_k3_cache_point_wire.py b/tests/integration/providers/test_bedrock_kimi_k3_cache_point_wire.py index 7d978bc0561..854da6fce0b 100644 --- a/tests/integration/providers/test_bedrock_kimi_k3_cache_point_wire.py +++ b/tests/integration/providers/test_bedrock_kimi_k3_cache_point_wire.py @@ -31,6 +31,8 @@ from pydantic import JsonValue, TypeAdapter from litellm.proxy.common_utils.encrypt_decrypt_utils import decrypt_if_encrypted_with +pytestmark: Final = pytest.mark.shares_worker_state + _KIMI_US: Final = "us.moonshotai.kimi-k3" _KIMI_GLOBAL: Final = "global.moonshotai.kimi-k3" _KIMI_BASE: Final = "moonshotai.kimi-k3" diff --git a/tests/integration/providers/test_bedrock_mantle_claude_chat_chaos.py b/tests/integration/providers/test_bedrock_mantle_claude_chat_chaos.py index 9c7903497e3..bcbf6d0f6f3 100644 --- a/tests/integration/providers/test_bedrock_mantle_claude_chat_chaos.py +++ b/tests/integration/providers/test_bedrock_mantle_claude_chat_chaos.py @@ -21,6 +21,8 @@ from integration._support.process import owned_proxy_process from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + _BACKEND: Final = "anthropic.claude-haiku-4-5" _API_KEY: Final = "synthetic-mantle-bearer" _CONFIG_MODEL: Final = "bedrock-mantle-claude-chat-chaos" diff --git a/tests/integration/providers/test_bedrock_mantle_output_limit_wire.py b/tests/integration/providers/test_bedrock_mantle_output_limit_wire.py index 1e599103a27..89ad226761e 100644 --- a/tests/integration/providers/test_bedrock_mantle_output_limit_wire.py +++ b/tests/integration/providers/test_bedrock_mantle_output_limit_wire.py @@ -27,6 +27,8 @@ from pydantic import JsonValue, TypeAdapter from litellm.proxy.common_utils.encrypt_decrypt_utils import decrypt_if_encrypted_with +pytestmark: Final = pytest.mark.shares_worker_state + _API_KEY: Final = "synthetic-mantle-bearer" _SIGNING_KEY: Final = os.environ.get("LITELLM_SALT_KEY", "sk-integration-salt") _GPT56: Final = "openai.gpt-5.6-terra" diff --git a/tests/integration/providers/test_cache_control_tool_call_marks_owned_proxy.py b/tests/integration/providers/test_cache_control_tool_call_marks_owned_proxy.py index 02f9e42ff31..1ab6d9fd54f 100644 --- a/tests/integration/providers/test_cache_control_tool_call_marks_owned_proxy.py +++ b/tests/integration/providers/test_cache_control_tool_call_marks_owned_proxy.py @@ -39,7 +39,7 @@ from integration.providers._cache_control_marks_support import ( ) from pydantic import JsonValue -pytestmark: Final = pytest.mark.timeout(2 * graceful_stop_seconds() + 120) +pytestmark: Final = [pytest.mark.timeout(2 * graceful_stop_seconds() + 120), pytest.mark.shares_worker_state] _STARTED_WORKER: Final = re.compile(r"Started server process \[(\d+)\]") _SURFACES: Final = ("chat", "chat-stream", "chat-unmarked", "messages", "responses") diff --git a/tests/integration/providers/test_cache_control_tool_call_marks_wire.py b/tests/integration/providers/test_cache_control_tool_call_marks_wire.py index 47d28860018..6389d90597e 100644 --- a/tests/integration/providers/test_cache_control_tool_call_marks_wire.py +++ b/tests/integration/providers/test_cache_control_tool_call_marks_wire.py @@ -61,6 +61,8 @@ from pydantic import JsonValue, TypeAdapter from litellm.utils import get_prompt_cache_min_tokens +pytestmark: Final = pytest.mark.shares_worker_state + _OWNED_PROXY_TIMEOUT: Final = 2 * graceful_stop_seconds() + 120 _JSON_OBJECT: Final = TypeAdapter(dict[str, JsonValue]) _LANE_CONFIG: Final = Path(__file__).resolve().parents[1] / "proxy_config.yaml" diff --git a/tests/integration/providers/test_chatgpt_responses_caller_stream_chaos.py b/tests/integration/providers/test_chatgpt_responses_caller_stream_chaos.py index d2787d49bd0..eed0e3c3641 100644 --- a/tests/integration/providers/test_chatgpt_responses_caller_stream_chaos.py +++ b/tests/integration/providers/test_chatgpt_responses_caller_stream_chaos.py @@ -23,7 +23,7 @@ from integration._support.wire import Reply, Request, wire_server from integration._support.ports import allocate_port from pydantic import JsonValue -pytestmark: Final = pytest.mark.timeout(240) +pytestmark: Final = [pytest.mark.timeout(240), pytest.mark.shares_worker_state] _MODEL: Final = "chatgpt/gpt-5.5" _CONFIG_MODEL: Final = "chatgpt-caller-stream-chaos" diff --git a/tests/integration/providers/test_chatgpt_responses_caller_stream_wire.py b/tests/integration/providers/test_chatgpt_responses_caller_stream_wire.py index d4ae7e0b4ae..57eb30e28fe 100644 --- a/tests/integration/providers/test_chatgpt_responses_caller_stream_wire.py +++ b/tests/integration/providers/test_chatgpt_responses_caller_stream_wire.py @@ -26,7 +26,7 @@ from openai.types.responses import ( ) from pydantic import JsonValue -pytestmark: Final = pytest.mark.timeout(240) +pytestmark: Final = [pytest.mark.timeout(240), pytest.mark.shares_worker_state] _MODEL: Final = "chatgpt/gpt-5.5" _NO_CACHE: Final[Mapping[str, JsonValue]] = {"cache": {"no-cache": True}} diff --git a/tests/integration/providers/test_coralbricks_provider_chaos.py b/tests/integration/providers/test_coralbricks_provider_chaos.py index dc4014fdb75..9ec37da52a9 100644 --- a/tests/integration/providers/test_coralbricks_provider_chaos.py +++ b/tests/integration/providers/test_coralbricks_provider_chaos.py @@ -36,6 +36,8 @@ from integration.providers._coralbricks import ( ) from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + Endpoint = Literal["chat", "messages", "responses"] _STARTED_WORKER: Final = re.compile(r"Started server process \[(\d+)\]") diff --git a/tests/integration/providers/test_coralbricks_provider_env_wire.py b/tests/integration/providers/test_coralbricks_provider_env_wire.py index 6d8cc766b62..23a68dcd42c 100644 --- a/tests/integration/providers/test_coralbricks_provider_env_wire.py +++ b/tests/integration/providers/test_coralbricks_provider_env_wire.py @@ -27,6 +27,8 @@ from integration.providers._coralbricks import ( ) from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + ENV_KEY: Final = "cb_env_integration_key" CONFIG_MODEL: Final = "coralbricks-env-config" diff --git a/tests/integration/providers/test_databricks_decisions_wire.py b/tests/integration/providers/test_databricks_decisions_wire.py index d3b17044207..d96d6ab1be1 100644 --- a/tests/integration/providers/test_databricks_decisions_wire.py +++ b/tests/integration/providers/test_databricks_decisions_wire.py @@ -15,6 +15,8 @@ from integration._support.upstream import ScenarioHandle, delete_scenario, regis from integration.cost_calculation.cost_tracking_case import JsonResponse from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + _MODEL: Final = "ai_decide" _ROUTE: Final = "/api/2.0/ai-functions/ai-decide" _ENDPOINT: Final = "databricks-openjev-qwen35-4b" diff --git a/tests/integration/providers/test_decision_model_guardrail.py b/tests/integration/providers/test_decision_model_guardrail.py index 7f584810df3..67e6f0918d1 100644 --- a/tests/integration/providers/test_decision_model_guardrail.py +++ b/tests/integration/providers/test_decision_model_guardrail.py @@ -23,6 +23,8 @@ from pydantic import JsonValue from tests.integration.cost_calculation.cost_tracking_case import JsonResponse, RoutedResponse +pytestmark: Final = pytest.mark.shares_worker_state + _PROBABILITY_ABOVE: Final = 0.99 _PROBABILITY_BELOW: Final = 0.01 _THRESHOLD: Final = 0.5 diff --git a/tests/integration/providers/test_decisions_box_deployment_flag_chaos.py b/tests/integration/providers/test_decisions_box_deployment_flag_chaos.py index d01cbde915e..a7a7c58ceed 100644 --- a/tests/integration/providers/test_decisions_box_deployment_flag_chaos.py +++ b/tests/integration/providers/test_decisions_box_deployment_flag_chaos.py @@ -21,6 +21,8 @@ from integration._support.wire import Reply, Request, Wire, wire_server from integration._support.ports import allocate_port from pydantic import JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + _API_KEY: Final = "synthetic-decisions-key" _JSON_OBJECT: Final = TypeAdapter(dict[str, JsonValue]) _BOX: Final[dict[str, JsonValue]] = {"top": 0.1, "bottom": 0.4, "left": 0.2, "right": 0.6} diff --git a/tests/integration/providers/test_decisions_chaos.py b/tests/integration/providers/test_decisions_chaos.py index 7bbbb09f638..86edc02bc57 100644 --- a/tests/integration/providers/test_decisions_chaos.py +++ b/tests/integration/providers/test_decisions_chaos.py @@ -21,6 +21,8 @@ from integration._support.wire import Reply, Request, Wire, wire_server from integration._support.ports import allocate_port from pydantic import JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + _BACKEND: Final = "pplx-decider-v1-27b" _CONFIG_MODEL: Final = "decisions-chaos" _API_KEY: Final = "synthetic-decisions-key" diff --git a/tests/integration/providers/test_device_code_login_guard_boot.py b/tests/integration/providers/test_device_code_login_guard_boot.py index 34ab7f1597f..8aba6841cf9 100644 --- a/tests/integration/providers/test_device_code_login_guard_boot.py +++ b/tests/integration/providers/test_device_code_login_guard_boot.py @@ -25,7 +25,7 @@ from integration._support.process import OwnedProxy, owned_proxy_process from integration._support.wire import Reply, Request, wire_server from pydantic import JsonValue, TypeAdapter -pytestmark: Final = pytest.mark.timeout(900) +pytestmark: Final = [pytest.mark.timeout(900), pytest.mark.shares_worker_state] _STARTED_WORKER: Final = re.compile(r"Started server process \[(\d+)\]") _DROPPED: Final = re.compile(r"original model: (\S+), ignoring and continuing") diff --git a/tests/integration/providers/test_device_code_login_guard_wire.py b/tests/integration/providers/test_device_code_login_guard_wire.py index c88bd0b9273..4f01d79d43b 100644 --- a/tests/integration/providers/test_device_code_login_guard_wire.py +++ b/tests/integration/providers/test_device_code_login_guard_wire.py @@ -29,7 +29,7 @@ from integration._support.wire import Request from litellm.llms.chatgpt.authenticator import DEVICE_CODE_COOLDOWN_SECONDS from pydantic import JsonValue -pytestmark: Final = pytest.mark.timeout(300) +pytestmark: Final = [pytest.mark.timeout(300), pytest.mark.shares_worker_state] Provider: TypeAlias = Literal["chatgpt", "copilot"] Endpoint: TypeAlias = Literal["chat", "responses", "messages"] diff --git a/tests/integration/providers/test_fal_ai_passthrough_wire.py b/tests/integration/providers/test_fal_ai_passthrough_wire.py index b80cc601b3b..b97bd1f2c1c 100644 --- a/tests/integration/providers/test_fal_ai_passthrough_wire.py +++ b/tests/integration/providers/test_fal_ai_passthrough_wire.py @@ -8,6 +8,8 @@ from integration._support.database import read_rows from integration._support.process import owned_proxy from integration._support.wire import Reply, Request, wire_server +pytestmark: Final = pytest.mark.shares_worker_state + _MODEL: Final = "fal-ai/trellis-2" _REQUEST_BODY: Final = {"image_url": "https://example.com/in.png", "resolution": 1536} _UPSTREAM_BODY: Final = { diff --git a/tests/integration/providers/test_gemini_embedding_file_block_chaos.py b/tests/integration/providers/test_gemini_embedding_file_block_chaos.py index 17483147107..2be02565f76 100644 --- a/tests/integration/providers/test_gemini_embedding_file_block_chaos.py +++ b/tests/integration/providers/test_gemini_embedding_file_block_chaos.py @@ -19,6 +19,8 @@ from integration._support.process import group_members, owned_proxy_process from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + BACKEND: Final = "gemini-embedding-001" TARGET: Final = f"/models/{BACKEND}:batchEmbedContents" END_OFFSETS: Final = {"fast": "3s", "slow": "5s", "malformed": "3s", "dropped": "9s"} diff --git a/tests/integration/providers/test_gemini_long_decode_chaos.py b/tests/integration/providers/test_gemini_long_decode_chaos.py index 7f86a01d69a..096e30ff814 100644 --- a/tests/integration/providers/test_gemini_long_decode_chaos.py +++ b/tests/integration/providers/test_gemini_long_decode_chaos.py @@ -55,6 +55,8 @@ from integration.providers._gemini_long_decode_support import ( ) from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + Continuation = Callable[[Request, str], Reply] _MARKER: Final = re.compile(r"marker-([0-9a-f]{32})") diff --git a/tests/integration/providers/test_gemini_long_decode_wire.py b/tests/integration/providers/test_gemini_long_decode_wire.py index eccfb8e3486..1270b864cb7 100644 --- a/tests/integration/providers/test_gemini_long_decode_wire.py +++ b/tests/integration/providers/test_gemini_long_decode_wire.py @@ -88,6 +88,8 @@ from integration.providers._gemini_long_decode_support import ( ) from pydantic import BaseModel, JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + Client = Literal["sdk_sync", "sdk_async", "httpx"] Failure = Literal["400", "500", "drop", "notjson", "errorframe"] Dropped = Literal["first", "continued"] diff --git a/tests/integration/providers/test_gemini_thinking_replay_wire.py b/tests/integration/providers/test_gemini_thinking_replay_wire.py index c0fbe0c4fa3..4f965f079ec 100644 --- a/tests/integration/providers/test_gemini_thinking_replay_wire.py +++ b/tests/integration/providers/test_gemini_thinking_replay_wire.py @@ -26,6 +26,8 @@ from pydantic import JsonValue, TypeAdapter from litellm.proxy.common_utils.encrypt_decrypt_utils import decrypt_if_encrypted_with +pytestmark: Final = pytest.mark.shares_worker_state + _BACKEND: Final = "gemini-2.5-flash" _API_KEY: Final = "synthetic-gemini-key" _PROJECT: Final = "scripted-project" diff --git a/tests/integration/providers/test_gemini_tool_result_part_reference_chaos.py b/tests/integration/providers/test_gemini_tool_result_part_reference_chaos.py index 0194f138689..ead83ed66aa 100644 --- a/tests/integration/providers/test_gemini_tool_result_part_reference_chaos.py +++ b/tests/integration/providers/test_gemini_tool_result_part_reference_chaos.py @@ -21,6 +21,8 @@ from integration._support.process import graceful_stop_seconds, owned_proxy_proc from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + _BACKEND: Final = "gemini-2.5-flash" _API_KEY: Final = "synthetic-gemini-key" _CONFIG_MODEL: Final = "gemini-tool-result-part-reference-chaos" diff --git a/tests/integration/providers/test_gemini_tts_voice_mapping_owned_proxy.py b/tests/integration/providers/test_gemini_tts_voice_mapping_owned_proxy.py index 3a5a9bf5d06..08ac73cd8e7 100644 --- a/tests/integration/providers/test_gemini_tts_voice_mapping_owned_proxy.py +++ b/tests/integration/providers/test_gemini_tts_voice_mapping_owned_proxy.py @@ -58,7 +58,7 @@ from integration.providers._gemini_tts_voice_support import ( ) from pydantic import JsonValue -pytestmark: Final = pytest.mark.timeout(2 * graceful_stop_seconds() + 120) +pytestmark: Final = [pytest.mark.timeout(2 * graceful_stop_seconds() + 120), pytest.mark.shares_worker_state] _STARTED_WORKER: Final = re.compile(r"Started server process \[(\d+)\]") BURST: Final = 12 diff --git a/tests/integration/providers/test_inferenceapis_context_window_fallbacks_wire.py b/tests/integration/providers/test_inferenceapis_context_window_fallbacks_wire.py index 04869ebb77f..435b4c08c29 100644 --- a/tests/integration/providers/test_inferenceapis_context_window_fallbacks_wire.py +++ b/tests/integration/providers/test_inferenceapis_context_window_fallbacks_wire.py @@ -21,7 +21,7 @@ from integration._support.process import OwnedProxy, graceful_stop_seconds, owne from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import JsonValue, TypeAdapter -pytestmark = pytest.mark.timeout(int(2 * graceful_stop_seconds() + 120)) +pytestmark: Final = [pytest.mark.timeout(int(2 * graceful_stop_seconds() + 120)), pytest.mark.shares_worker_state] _BACKEND: Final = "deepseek-ai/DeepSeek-V4-Flash" _VENDOR_KEY: Final = "synthetic-inferenceapis-key" diff --git a/tests/integration/providers/test_malformed_tools_chaos.py b/tests/integration/providers/test_malformed_tools_chaos.py index 0ab50a5afbe..5a8ced9d4bb 100644 --- a/tests/integration/providers/test_malformed_tools_chaos.py +++ b/tests/integration/providers/test_malformed_tools_chaos.py @@ -25,6 +25,8 @@ from integration._support.wire import Reply, Request, Wire, wire_server from integration._support.worker_readiness import wait_for_worker_readiness from pydantic import JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + _OPENAI_BACKEND: Final = "gpt-4o-mini" _CLAUDE_BACKEND: Final = "claude-haiku-4-5" _GEMINI_BACKEND: Final = "gemini-2.5-flash" diff --git a/tests/integration/providers/test_ollama_prompt_tools_wire.py b/tests/integration/providers/test_ollama_prompt_tools_wire.py index 59a2ee58a44..428af5dcfe6 100644 --- a/tests/integration/providers/test_ollama_prompt_tools_wire.py +++ b/tests/integration/providers/test_ollama_prompt_tools_wire.py @@ -21,6 +21,8 @@ from integration._support.process import graceful_stop_seconds, owned_proxy from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + _BACKEND: Final = "llama3-prompt-tools" _API_KEY: Final = "synthetic-ollama-key" _INSTRUCTION: Final = ( diff --git a/tests/integration/providers/test_openai_dialect_prompt_cache_breakpoint_owned_proxy.py b/tests/integration/providers/test_openai_dialect_prompt_cache_breakpoint_owned_proxy.py index 00f2c430ea5..3df73d3b767 100644 --- a/tests/integration/providers/test_openai_dialect_prompt_cache_breakpoint_owned_proxy.py +++ b/tests/integration/providers/test_openai_dialect_prompt_cache_breakpoint_owned_proxy.py @@ -51,6 +51,8 @@ from integration.providers._mantle_gpt_prompt_cache_support import ( ) from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + _STARTED_WORKER: Final = re.compile(r"Started server process \[(\d+)\]") _FOUNDRY_BASE: Final = "http://prompt-cache-breakpoint-audit.services.ai.azure.com" _FOUNDRY_HOST: Final = urlsplit(_FOUNDRY_BASE).netloc diff --git a/tests/integration/providers/test_openai_realtime_transcription_url.py b/tests/integration/providers/test_openai_realtime_transcription_url.py index 81c450d8f70..eda8819beb5 100644 --- a/tests/integration/providers/test_openai_realtime_transcription_url.py +++ b/tests/integration/providers/test_openai_realtime_transcription_url.py @@ -36,7 +36,7 @@ from tests.integration.cost_calculation.cost_tracking_case import RealtimeRespon RecordProperty = Callable[[str, object], None] -pytestmark: Final = pytest.mark.timeout(180) +pytestmark: Final = [pytest.mark.timeout(180), pytest.mark.shares_worker_state] TRANSCRIBE_MODEL: Final = "gpt-live-transcribe" SECOND_TRANSCRIBE_MODEL: Final = "gpt-live-transcribe-second" diff --git a/tests/integration/providers/test_openai_responses_tool_search_wire.py b/tests/integration/providers/test_openai_responses_tool_search_wire.py index 0e5c539e3f9..39a955be13b 100644 --- a/tests/integration/providers/test_openai_responses_tool_search_wire.py +++ b/tests/integration/providers/test_openai_responses_tool_search_wire.py @@ -51,7 +51,7 @@ from integration.providers.test_responses_bridge_tool_search_wire import ( ) from pydantic import JsonValue -pytestmark = pytest.mark.timeout(2 * graceful_stop_seconds() + 120) +pytestmark: Final = [pytest.mark.timeout(2 * graceful_stop_seconds() + 120), pytest.mark.shares_worker_state] _OPENAI_HOST: Final = "api.openai.com" _OPENAI_API_BASE: Final = f"https://{_OPENAI_HOST}/v1" diff --git a/tests/integration/providers/test_openai_responses_websocket_wire.py b/tests/integration/providers/test_openai_responses_websocket_wire.py index 4f9a119db2f..267e175fe19 100644 --- a/tests/integration/providers/test_openai_responses_websocket_wire.py +++ b/tests/integration/providers/test_openai_responses_websocket_wire.py @@ -22,7 +22,7 @@ from integration._support.responses_ws_peer import ( from integration._support.tls import write_self_signed_cert from pydantic import JsonValue -pytestmark: Final = pytest.mark.timeout(2 * graceful_stop_seconds() + 120) +pytestmark: Final = [pytest.mark.timeout(2 * graceful_stop_seconds() + 120), pytest.mark.shares_worker_state] PROXY_OPENAI_KEY: Final = "sk-proxy-environment-openai-key" PROXY_AZURE_KEY: Final = "proxy-environment-azure-key" diff --git a/tests/integration/providers/test_realtime_unguarded_voice_turns.py b/tests/integration/providers/test_realtime_unguarded_voice_turns.py index c7489685273..c033e390076 100644 --- a/tests/integration/providers/test_realtime_unguarded_voice_turns.py +++ b/tests/integration/providers/test_realtime_unguarded_voice_turns.py @@ -54,7 +54,7 @@ from tests.integration._support.process import ( from tests.integration._support.upstream import ScenarioHandle, delete_scenario, register_scenario from tests.integration.cost_calculation.cost_tracking_case import RealtimeResponse -pytestmark: Final = pytest.mark.timeout(2 * graceful_stop_seconds() + 120) +pytestmark: Final = [pytest.mark.timeout(2 * graceful_stop_seconds() + 120), pytest.mark.shares_worker_state] WORKERS: Final = 2 BURST: Final = 20 diff --git a/tests/integration/providers/test_responses_bridge_message_cache_control_wire_chaos.py b/tests/integration/providers/test_responses_bridge_message_cache_control_wire_chaos.py index 3de825ae4a6..d8fb0f8f993 100644 --- a/tests/integration/providers/test_responses_bridge_message_cache_control_wire_chaos.py +++ b/tests/integration/providers/test_responses_bridge_message_cache_control_wire_chaos.py @@ -49,7 +49,7 @@ from integration.providers._cache_control_marks_support import ( ) from pydantic import JsonValue, TypeAdapter -pytestmark: Final = pytest.mark.timeout(2 * graceful_stop_seconds() + 120) +pytestmark: Final = [pytest.mark.timeout(2 * graceful_stop_seconds() + 120), pytest.mark.shares_worker_state] _JSON_OBJECT: Final = TypeAdapter(dict[str, JsonValue]) _OWNED: Final = "bridge-mark-owned" diff --git a/tests/integration/providers/test_responses_bridge_prompt_cache_breakpoint_chaos.py b/tests/integration/providers/test_responses_bridge_prompt_cache_breakpoint_chaos.py index 5acf3b10e04..6fa310c7cf5 100644 --- a/tests/integration/providers/test_responses_bridge_prompt_cache_breakpoint_chaos.py +++ b/tests/integration/providers/test_responses_bridge_prompt_cache_breakpoint_chaos.py @@ -30,6 +30,8 @@ from integration.providers._responses_bridge_prompt_cache_breakpoint import ( _send_call, ) +pytestmark: Final = pytest.mark.shares_worker_state + _CONFIG_MODEL: Final = "responses-bridge-cache-breakpoint-chaos" _API_KEY: Final = "synthetic-responses-bridge-key" diff --git a/tests/integration/providers/test_responses_bridge_prompt_cache_breakpoint_wire_chaos.py b/tests/integration/providers/test_responses_bridge_prompt_cache_breakpoint_wire_chaos.py index 1bc7853437b..f49d7c59710 100644 --- a/tests/integration/providers/test_responses_bridge_prompt_cache_breakpoint_wire_chaos.py +++ b/tests/integration/providers/test_responses_bridge_prompt_cache_breakpoint_wire_chaos.py @@ -21,7 +21,7 @@ from integration._support.wire import Reply, Request, Wire, wire_server from integration._support.ports import allocate_port from pydantic import JsonValue -pytestmark: Final = pytest.mark.timeout(2 * graceful_stop_seconds() + 120) +pytestmark: Final = [pytest.mark.timeout(2 * graceful_stop_seconds() + 120), pytest.mark.shares_worker_state] _GLOBAL_UNSET: Final = "bridge-breakpoint-global-unset" _GLOBAL_FALSE: Final = "bridge-breakpoint-global-false" diff --git a/tests/integration/providers/test_responses_bridge_tool_search_chaos.py b/tests/integration/providers/test_responses_bridge_tool_search_chaos.py index 6beccefe58f..d4a5e6bf0db 100644 --- a/tests/integration/providers/test_responses_bridge_tool_search_chaos.py +++ b/tests/integration/providers/test_responses_bridge_tool_search_chaos.py @@ -30,6 +30,8 @@ from integration._support.wire import Reply, Request, Wire, wire_server from integration._support.ports import allocate_port from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + _CLAUDE: Final = cc.OPUS _BEDROCK_CLAUDE: Final = "anthropic.claude-sonnet-5-5" _MISTRAL: Final = "mistral-large-latest" diff --git a/tests/integration/providers/test_rust_anthropic_safeguards_beta_wire.py b/tests/integration/providers/test_rust_anthropic_safeguards_beta_wire.py index 75c5efc3aa0..c13d4da7dbb 100644 --- a/tests/integration/providers/test_rust_anthropic_safeguards_beta_wire.py +++ b/tests/integration/providers/test_rust_anthropic_safeguards_beta_wire.py @@ -61,7 +61,7 @@ from integration.providers._anthropic_safeguards_beta_support import ( ) from pydantic import JsonValue -pytestmark: Final = pytest.mark.xdist_group("anthropic_safeguards_beta") +pytestmark: Final = [pytest.mark.xdist_group("anthropic_safeguards_beta"), pytest.mark.shares_worker_state] _CELL_TIMEOUT: Final = int(2 * graceful_stop_seconds() + 120) _BURST: Final = 24 diff --git a/tests/integration/providers/test_sagemaker_embedding_chaos.py b/tests/integration/providers/test_sagemaker_embedding_chaos.py index 8e984c81ff9..ca04aace346 100644 --- a/tests/integration/providers/test_sagemaker_embedding_chaos.py +++ b/tests/integration/providers/test_sagemaker_embedding_chaos.py @@ -31,7 +31,7 @@ from integration._support.wire import Reply, Request, wire_server from integration._support.ports import allocate_port from pydantic import JsonValue -pytestmark = pytest.mark.timeout(300) +pytestmark: Final = [pytest.mark.timeout(300), pytest.mark.shares_worker_state] _MODEL: Final = "sm-openai-chaos" _ALIASES: Final = ( diff --git a/tests/integration/providers/test_sagemaker_embedding_wire.py b/tests/integration/providers/test_sagemaker_embedding_wire.py index 0ed38b1a768..74f3c8c8a00 100644 --- a/tests/integration/providers/test_sagemaker_embedding_wire.py +++ b/tests/integration/providers/test_sagemaker_embedding_wire.py @@ -37,7 +37,7 @@ from integration._support.sagemaker_embedding import ( from integration._support.wire import Request, Wire, wire_server from pydantic import JsonValue -pytestmark = pytest.mark.timeout(240) +pytestmark: Final = [pytest.mark.timeout(240), pytest.mark.shares_worker_state] _NO_CACHE: Final[dict[str, JsonValue]] = {"no-cache": True} _SPEND_ROW: Final = 'SELECT status, call_type, model_group FROM "LiteLLM_SpendLogs" WHERE litellm_call_id=%s' diff --git a/tests/integration/providers/test_strands_decider_agentcore_wire.py b/tests/integration/providers/test_strands_decider_agentcore_wire.py index 9b7a9dbbfd0..98ddcca2bae 100644 --- a/tests/integration/providers/test_strands_decider_agentcore_wire.py +++ b/tests/integration/providers/test_strands_decider_agentcore_wire.py @@ -166,7 +166,7 @@ _SCRUBBED_NAMES: Final = frozenset( _load_yaml: Final[Callable[[str], object]] = yaml.safe_load _JSON: Final[TypeAdapter[JsonValue]] = TypeAdapter(JsonValue) _OWNED_PROXY_CELL_SECONDS: Final = 2 * max(30.0, float(os.environ.get("INTEGRATION_PROXY_READY_SECONDS") or 70)) + 120 -pytestmark: Final = pytest.mark.timeout(_OWNED_PROXY_CELL_SECONDS) +pytestmark: Final = [pytest.mark.timeout(_OWNED_PROXY_CELL_SECONDS), pytest.mark.shares_worker_state] _LINEAR_FACTOR: Final = 20 _CLASSIFIER_SAMPLES: Final = 5 diff --git a/tests/integration/providers/test_vertex_anthropic_structured_output_wire.py b/tests/integration/providers/test_vertex_anthropic_structured_output_wire.py index bba79049283..cb295815ae2 100644 --- a/tests/integration/providers/test_vertex_anthropic_structured_output_wire.py +++ b/tests/integration/providers/test_vertex_anthropic_structured_output_wire.py @@ -31,6 +31,8 @@ from integration._support.wire import Reply, Request, Wire, wire_server from integration._support.ports import allocate_port from pydantic import JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + _JSON_TOOL: Final = "json_tool_call" _USER_TOOL: Final = "lookup" _STRUCTURED_OUTPUTS_BETA: Final = "structured-outputs-2025-11-13" diff --git a/tests/integration/providers/test_vertex_passthrough_wire.py b/tests/integration/providers/test_vertex_passthrough_wire.py index 06c7e8e03ae..98eb0547d42 100644 --- a/tests/integration/providers/test_vertex_passthrough_wire.py +++ b/tests/integration/providers/test_vertex_passthrough_wire.py @@ -41,7 +41,7 @@ from integration.providers._tool_search_support import ( ) from pydantic import JsonValue, TypeAdapter -pytestmark = pytest.mark.timeout(2 * graceful_stop_seconds() + 120) +pytestmark: Final = [pytest.mark.timeout(2 * graceful_stop_seconds() + 120), pytest.mark.shares_worker_state] _PROJECT: Final = "scripted-project" _TOKEN: Final = "scripted-vertex-token" diff --git a/tests/integration/providers/test_voyage_mongodb_host_wire.py b/tests/integration/providers/test_voyage_mongodb_host_wire.py index 62860c52568..4a1f79a609a 100644 --- a/tests/integration/providers/test_voyage_mongodb_host_wire.py +++ b/tests/integration/providers/test_voyage_mongodb_host_wire.py @@ -62,7 +62,7 @@ _JSON_OBJECTS: Final = TypeAdapter(list[dict[str, JsonValue]]) _WORKER_PIDS: Final = TypeAdapter(list[int]) _load_yaml: Final[Callable[[str], object]] = yaml.safe_load _OWNED_PROXY_CELL_SECONDS: Final = 2 * max(30.0, float(os.environ.get("INTEGRATION_PROXY_READY_SECONDS") or 70)) + 120 -pytestmark: Final = pytest.mark.timeout(_OWNED_PROXY_CELL_SECONDS) +pytestmark: Final = [pytest.mark.timeout(_OWNED_PROXY_CELL_SECONDS), pytest.mark.shares_worker_state] def _deployment(name: str, model: str, mode: str, **litellm_params: JsonValue) -> dict[str, JsonValue]: diff --git a/tests/integration/routing/test_complexity_router_databricks_classifier.py b/tests/integration/routing/test_complexity_router_databricks_classifier.py index af4b76e888f..73a665cee7b 100644 --- a/tests/integration/routing/test_complexity_router_databricks_classifier.py +++ b/tests/integration/routing/test_complexity_router_databricks_classifier.py @@ -30,6 +30,8 @@ from pydantic import JsonValue, TypeAdapter from litellm.proxy.common_utils.encrypt_decrypt_utils import decrypt_if_encrypted_with from litellm.router_strategy.complexity_router.config import DEFAULT_JEV_INSTRUCTIONS +pytestmark: Final = pytest.mark.shares_worker_state + _MODEL: Final = "ai_decide" _ROUTE: Final = "/api/2.0/ai-functions/ai-decide" _API_KEY: Final = "synthetic-databricks-key" diff --git a/tests/integration/routing/test_deployment_mode_routing_wire.py b/tests/integration/routing/test_deployment_mode_routing_wire.py index 65e64883b18..fd7091d9e05 100644 --- a/tests/integration/routing/test_deployment_mode_routing_wire.py +++ b/tests/integration/routing/test_deployment_mode_routing_wire.py @@ -26,6 +26,8 @@ from integration._support.responses_vendor import same_response from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + _API_KEY: Final = "synthetic-routing-bearer" _MANTLE_GPT: Final = "bedrock_mantle/openai.gpt-5.6-terra" _MANTLE_OSS: Final = "bedrock_mantle/openai.gpt-oss-120b" diff --git a/tests/integration/routing/test_fallback_hop_foreign_encrypted_reasoning.py b/tests/integration/routing/test_fallback_hop_foreign_encrypted_reasoning.py index f30c13993d4..e6ec3444369 100644 --- a/tests/integration/routing/test_fallback_hop_foreign_encrypted_reasoning.py +++ b/tests/integration/routing/test_fallback_hop_foreign_encrypted_reasoning.py @@ -50,6 +50,8 @@ from integration._support.ports import allocate_port from integration.cost_calculation.cost_tracking_case import JsonResponse, RoutedResponse, SseResponse, StoredResponse from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + ORDER_ONE_BLOB: Final = "gAAAAA-minted-by-order-1" ORDER_TWO_BLOB: Final = "gAAAAA-minted-by-order-2" SUMMARY_TEXT: Final = "multiply 17 by 23" diff --git a/tests/integration/routing/test_observed_routing.py b/tests/integration/routing/test_observed_routing.py index d9ac0652a0d..673aa54fb78 100644 --- a/tests/integration/routing/test_observed_routing.py +++ b/tests/integration/routing/test_observed_routing.py @@ -13,6 +13,8 @@ from integration._support.process import owned_proxy from integration._support.wire import Reply, Request, wire_server from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + @pytest.mark.covers( "other.routing.retries.several_attempts_reach_success_without_hidden_retries", diff --git a/tests/integration/routing/test_payment_required_cooldown.py b/tests/integration/routing/test_payment_required_cooldown.py index 12190efee26..bc5cf1a5913 100644 --- a/tests/integration/routing/test_payment_required_cooldown.py +++ b/tests/integration/routing/test_payment_required_cooldown.py @@ -43,7 +43,7 @@ from integration.routing.test_payment_required_mapping import ( from pydantic import JsonValue from redis import Redis -pytestmark: Final = pytest.mark.timeout(300) +pytestmark: Final = [pytest.mark.timeout(300), pytest.mark.shares_worker_state] COOLDOWN_SECONDS: Final = 300 RELOAD_SECONDS: Final = 2 diff --git a/tests/integration/routing/test_pdf_document_pre_call_checks_owned_proxy.py b/tests/integration/routing/test_pdf_document_pre_call_checks_owned_proxy.py index c9a4e8e8881..a5e93258087 100644 --- a/tests/integration/routing/test_pdf_document_pre_call_checks_owned_proxy.py +++ b/tests/integration/routing/test_pdf_document_pre_call_checks_owned_proxy.py @@ -29,6 +29,8 @@ from integration.providers._cache_control_marks_support import anthropic_peer, m from pydantic import JsonValue, TypeAdapter from redis import Redis +pytestmark: Final = pytest.mark.shares_worker_state + _WINDOW: Final = "window-claude" _ITPM: Final = "itpm-claude" _BUDGET: Final = "budget-claude" diff --git a/tests/integration/routing/test_prompt_cache_pin_ttl.py b/tests/integration/routing/test_prompt_cache_pin_ttl.py index 1ca92505191..0a758d51b7e 100644 --- a/tests/integration/routing/test_prompt_cache_pin_ttl.py +++ b/tests/integration/routing/test_prompt_cache_pin_ttl.py @@ -59,7 +59,7 @@ from integration.routing._prompt_cache_pin_support import ( from pydantic import JsonValue from redis import Redis -pytestmark: Final = pytest.mark.timeout(2 * graceful_stop_seconds() + 300) +pytestmark: Final = [pytest.mark.timeout(2 * graceful_stop_seconds() + 300), pytest.mark.shares_worker_state] _CLAUDE: Final = "pin-claude" _SLOW: Final = "pin-slow" diff --git a/tests/integration/routing/test_prompt_cache_pin_ttl_chaos.py b/tests/integration/routing/test_prompt_cache_pin_ttl_chaos.py index 4cdda116d23..af418ee9eae 100644 --- a/tests/integration/routing/test_prompt_cache_pin_ttl_chaos.py +++ b/tests/integration/routing/test_prompt_cache_pin_ttl_chaos.py @@ -36,7 +36,7 @@ from integration.routing._prompt_cache_pin_support import ( from pydantic import JsonValue from redis import Redis -pytestmark: Final = pytest.mark.timeout(2 * graceful_stop_seconds() + 600) +pytestmark: Final = [pytest.mark.timeout(2 * graceful_stop_seconds() + 600), pytest.mark.shares_worker_state] _GROUP: Final = "pin-chaos" _LEFT: Final = "pin-chaos-left" diff --git a/tests/integration/routing/test_ptu_shares.py b/tests/integration/routing/test_ptu_shares.py index 9855394c83c..9f57d31a3a0 100644 --- a/tests/integration/routing/test_ptu_shares.py +++ b/tests/integration/routing/test_ptu_shares.py @@ -45,7 +45,7 @@ from redis import Redis if TYPE_CHECKING: from litellm.llms.azure.ptu_capacity import PTUCapacity -pytestmark: Final = pytest.mark.timeout(2 * graceful_stop_seconds() + 120) +pytestmark: Final = [pytest.mark.timeout(2 * graceful_stop_seconds() + 120), pytest.mark.shares_worker_state] _SHARED: Final = "ptu-shared" _MIXED: Final = "ptu-mixed" diff --git a/tests/integration/routing/test_ptu_shares_regex_timing.py b/tests/integration/routing/test_ptu_shares_regex_timing.py index 19d5261dcb5..66c1fbad7cb 100644 --- a/tests/integration/routing/test_ptu_shares_regex_timing.py +++ b/tests/integration/routing/test_ptu_shares_regex_timing.py @@ -18,7 +18,7 @@ from integration._support.timing import assert_within_factor_of_control, timed from integration._support.wire import Reply, Request, wire_server from pydantic import JsonValue, TypeAdapter -pytestmark: Final = pytest.mark.timeout(2 * graceful_stop_seconds() + 240) +pytestmark: Final = [pytest.mark.timeout(2 * graceful_stop_seconds() + 240), pytest.mark.shares_worker_state] _WORST_GROUP: Final = "ptu-regex-worst" _CONTROL_GROUP: Final = "ptu-regex-unshared-worst" diff --git a/tests/integration/routing/test_vllm_passthrough_stream_usage_settings.py b/tests/integration/routing/test_vllm_passthrough_stream_usage_settings.py index 02e70a82c1c..25ccb6202ed 100644 --- a/tests/integration/routing/test_vllm_passthrough_stream_usage_settings.py +++ b/tests/integration/routing/test_vllm_passthrough_stream_usage_settings.py @@ -23,7 +23,7 @@ from integration._support.process import OwnedProxy, graceful_stop_seconds, owne from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import JsonValue -pytestmark: Final = pytest.mark.timeout(2 * graceful_stop_seconds() + 120) +pytestmark: Final = [pytest.mark.timeout(2 * graceful_stop_seconds() + 120), pytest.mark.shares_worker_state] _PROXY_CONFIG: Final = Path(__file__).resolve().parents[1] / "proxy_config.yaml" _SERVED_MODEL: Final = "Qwen/Qwen2.5-0.5B-Instruct" diff --git a/tests/integration/security/_sinks.py b/tests/integration/security/_sinks.py index 906384e8760..79aa61dfc90 100644 --- a/tests/integration/security/_sinks.py +++ b/tests/integration/security/_sinks.py @@ -62,7 +62,7 @@ from typing import Final import yaml from integration._support.client import Gateway, Scenario, eventually, gateway_from_environment from integration._support.database import read_rows -from integration._support.process import OwnedProxy, owned_proxy_process +from integration._support.process import OwnedProxy, OwnedState, owned_proxy_process from integration._support.wire import Reply, Request, Wire, wire_server from integration.security._canary import Canary, canary @@ -240,7 +240,9 @@ def canary_rig( "GENERIC_LOGGER_HEADERS": f"Authorization=Bearer {token}", **(environment or {}), } - with owned_proxy_process(gateway, root, overrides, config=config) as owned: + with owned_proxy_process( + gateway, root, overrides, config=config, state=OwnedState.SHARED_WITH_WORKER + ) as owned: admin = Gateway( owned.gateway.client, overrides.get("LITELLM_MASTER_KEY", owned.gateway.key), owned.gateway.upstream_url ) diff --git a/tests/integration/security/test_fine_tuning_managed_ids.py b/tests/integration/security/test_fine_tuning_managed_ids.py index 59daa8cb820..8f179e5a25f 100644 --- a/tests/integration/security/test_fine_tuning_managed_ids.py +++ b/tests/integration/security/test_fine_tuning_managed_ids.py @@ -19,6 +19,8 @@ from integration._support.wire import Reply, Request, Wire, wire_server from openai import OpenAI from pydantic import BaseModel, ConfigDict, JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + MANAGED_MODEL: Final = "gpt-4.1-managed-ids" DEPLOYMENT_KEY: Final = "sk-fine-tuning-managed-deployment" TRAINING_LINE: Final = b'{"messages": [{"role": "user", "content": "hi"}, {"role": "assistant", "content": "hello"}]}\n' diff --git a/tests/integration/spend/test_background_interaction_settlement.py b/tests/integration/spend/test_background_interaction_settlement.py index 4408bd147f9..ca18ba8a96a 100644 --- a/tests/integration/spend/test_background_interaction_settlement.py +++ b/tests/integration/spend/test_background_interaction_settlement.py @@ -42,7 +42,7 @@ from pydantic import JsonValue from litellm.proxy.spend_tracking.budget_reservation import DEFAULT_MAX_OUTPUT_TOKENS_FALLBACK -pytestmark: Final = pytest.mark.timeout(900) +pytestmark: Final = [pytest.mark.timeout(900), pytest.mark.shares_worker_state] _MODEL: Final = "gemini/gemini-3.8-flash" _INPUT_TOKENS: Final = 300 diff --git a/tests/integration/spend/test_daily_activity_routes.py b/tests/integration/spend/test_daily_activity_routes.py index 73dc95dc47b..520c8ccc6d8 100644 --- a/tests/integration/spend/test_daily_activity_routes.py +++ b/tests/integration/spend/test_daily_activity_routes.py @@ -37,6 +37,8 @@ from litellm.types.proxy.management_endpoints.common_daily_activity import ( SpendAnalyticsPaginatedResponse, ) +pytestmark: Final = pytest.mark.shares_worker_state + def _delete_organization(gateway: Gateway, organization_id: str) -> None: response: Final = gateway.request("DELETE", "/organization/delete", {"organization_ids": [organization_id]}) diff --git a/tests/integration/spend/test_daily_tag_spend_retention.py b/tests/integration/spend/test_daily_tag_spend_retention.py index 0bdb981c878..2ccd11332a1 100644 --- a/tests/integration/spend/test_daily_tag_spend_retention.py +++ b/tests/integration/spend/test_daily_tag_spend_retention.py @@ -14,6 +14,8 @@ from tests.integration._support.client import Gateway, eventually, string_value from tests.integration._support.database import read_rows from tests.integration._support.process import owned_proxy, owned_proxy_process +pytestmark: Final = pytest.mark.shares_worker_state + CLEANUP_EVERY_MINUTE: Final = "* * * * *" RETENTION_SETTING: Final = "maximum_daily_tag_spend_retention_period" _MAPPING: Final = TypeAdapter(dict[str, JsonValue]) diff --git a/tests/integration/spend/test_disable_end_user_cost_tracking.py b/tests/integration/spend/test_disable_end_user_cost_tracking.py index d0233378d1e..0db72a06703 100644 --- a/tests/integration/spend/test_disable_end_user_cost_tracking.py +++ b/tests/integration/spend/test_disable_end_user_cost_tracking.py @@ -22,7 +22,7 @@ from integration._support.upstream import JsonResponse, SseResponse, delete_scen from integration._support.wire import Reply, Request, wire_server from pydantic import JsonValue, TypeAdapter -pytestmark: Final = pytest.mark.timeout(2 * graceful_stop_seconds() + 120) +pytestmark: Final = [pytest.mark.timeout(2 * graceful_stop_seconds() + 120), pytest.mark.shares_worker_state] Surface = Literal["chat", "chat-stream", "messages", "messages-stream", "responses", "responses-stream"] diff --git a/tests/integration/spend/test_end_user_spend_without_proxy_user.py b/tests/integration/spend/test_end_user_spend_without_proxy_user.py index 16210438e82..68beedd087d 100644 --- a/tests/integration/spend/test_end_user_spend_without_proxy_user.py +++ b/tests/integration/spend/test_end_user_spend_without_proxy_user.py @@ -13,6 +13,8 @@ from integration._support.process import owned_proxy from integration._support.upstream import JsonResponse, SseResponse, delete_scenario, register_scenario from pydantic import BaseModel, JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + _JSON_OBJECT: Final = TypeAdapter(dict[str, JsonValue]) diff --git a/tests/integration/spend/test_expired_ui_session_failure_rows.py b/tests/integration/spend/test_expired_ui_session_failure_rows.py index 0acdc86709e..7bd41eff8e9 100644 --- a/tests/integration/spend/test_expired_ui_session_failure_rows.py +++ b/tests/integration/spend/test_expired_ui_session_failure_rows.py @@ -11,7 +11,7 @@ from integration._support.process import OwnedProxy, graceful_stop_seconds, owne from litellm.proxy._types import hash_token -pytestmark: Final = pytest.mark.timeout(2 * graceful_stop_seconds() + 240) +pytestmark: Final = [pytest.mark.timeout(2 * graceful_stop_seconds() + 240), pytest.mark.shares_worker_state] SESSION_SECONDS: Final = 3 DASHBOARD_INFO_ROUTES: Final = ( diff --git a/tests/integration/spend/test_key_budget_lockout.py b/tests/integration/spend/test_key_budget_lockout.py index 56b4db89bcb..74470b80543 100644 --- a/tests/integration/spend/test_key_budget_lockout.py +++ b/tests/integration/spend/test_key_budget_lockout.py @@ -15,6 +15,8 @@ from openai.types.chat import ChatCompletionChunk from pydantic import BaseModel, JsonValue, TypeAdapter from redis import Redis +pytestmark: Final = pytest.mark.shares_worker_state + _CHAT_BODY: Final = TypeAdapter(dict[str, JsonValue]) _CHANNEL: Final = "litellm_proxy.auth_cache_invalidation" _EXPECTED_MODEL_DISCOVERY: Final = ("GET", "/v1/models") diff --git a/tests/integration/spend/test_messages_stream_usage_cost.py b/tests/integration/spend/test_messages_stream_usage_cost.py index 692dd01e0b0..e28ab121de4 100644 --- a/tests/integration/spend/test_messages_stream_usage_cost.py +++ b/tests/integration/spend/test_messages_stream_usage_cost.py @@ -12,6 +12,8 @@ from integration._support.upstream import _aws_event_frame from integration._support.wire import Reply, Request, wire_server from pydantic import JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + BEDROCK_MODEL: Final = "anthropic.claude-haiku-4-5-20251001-v1:0" INPUT_TOKENS: Final = 30 CACHE_READ_TOKENS: Final = 900 diff --git a/tests/integration/spend/test_normalized_error_long_message.py b/tests/integration/spend/test_normalized_error_long_message.py index 407b40d36c3..a364327290f 100644 --- a/tests/integration/spend/test_normalized_error_long_message.py +++ b/tests/integration/spend/test_normalized_error_long_message.py @@ -22,6 +22,8 @@ from integration._support.timing import assert_not_delayed_by, assert_within_fac from integration._support.wire import Reply, Request, wire_server from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + CRAFTED_MODEL: Final = ("exceeded " * 32_000)[:288_000] HOSTILE_5KB_MODEL: Final = ("exceeded budget " * 400)[:5_000] BENIGN_288KB_MODEL: Final = "x" * len(CRAFTED_MODEL) diff --git a/tests/integration/spend/test_passthrough_budget_reservation.py b/tests/integration/spend/test_passthrough_budget_reservation.py index f072333c616..4940e6f61c6 100644 --- a/tests/integration/spend/test_passthrough_budget_reservation.py +++ b/tests/integration/spend/test_passthrough_budget_reservation.py @@ -12,6 +12,8 @@ from integration._support.upstream import delete_scenario, register_scenario from integration.cost_calculation.cost_tracking_case import JsonResponse from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + INPUT_COST_PER_TOKEN: Final = 0.000001 OUTPUT_COST_PER_TOKEN: Final = 0.001 PROMPT_TOKENS: Final = 10 diff --git a/tests/integration/spend/test_passthrough_request_tags.py b/tests/integration/spend/test_passthrough_request_tags.py index f3c0b66b20d..980cb90ff79 100644 --- a/tests/integration/spend/test_passthrough_request_tags.py +++ b/tests/integration/spend/test_passthrough_request_tags.py @@ -15,6 +15,8 @@ from integration._support.process import owned_proxy from integration._support.wire import Reply, Request, wire_server from pydantic import TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + def _chat_reply(marker: str) -> dict[str, JsonValue]: return { diff --git a/tests/integration/spend/test_ptu_share_attribution.py b/tests/integration/spend/test_ptu_share_attribution.py index 2b2405c27de..4a2c7fd2417 100644 --- a/tests/integration/spend/test_ptu_share_attribution.py +++ b/tests/integration/spend/test_ptu_share_attribution.py @@ -32,7 +32,7 @@ from litellm.constants import PTU_SENTINEL_API_KEY if TYPE_CHECKING: from litellm.llms.azure.ptu_capacity import PTUCapacity -pytestmark: Final = pytest.mark.timeout(4 * graceful_stop_seconds() + 240) +pytestmark: Final = [pytest.mark.timeout(4 * graceful_stop_seconds() + 240), pytest.mark.shares_worker_state] _GROUP: Final = "ptu-ledger" _SHARED_DEPLOYMENT: Final = "ledger-shared-gpt41" diff --git a/tests/integration/spend/test_requester_ip_address_logging.py b/tests/integration/spend/test_requester_ip_address_logging.py index 23e3a56db10..74af0afb59c 100644 --- a/tests/integration/spend/test_requester_ip_address_logging.py +++ b/tests/integration/spend/test_requester_ip_address_logging.py @@ -24,6 +24,8 @@ from integration._support.process import owned_proxy from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + FORWARDED_IP: Final = "203.0.113.74" PROXY_HOP: Final = "198.51.100.23" REAL_IP: Final = "203.0.113.75" diff --git a/tests/integration/spend/test_reset_budget_leader_election.py b/tests/integration/spend/test_reset_budget_leader_election.py index f9c165a0be0..58c018c4181 100644 --- a/tests/integration/spend/test_reset_budget_leader_election.py +++ b/tests/integration/spend/test_reset_budget_leader_election.py @@ -11,6 +11,8 @@ from integration._support.database import read_rows from integration._support.process import owned_proxy from redis import Redis +pytestmark: Final = pytest.mark.shares_worker_state + RESET_LEASE_KEY: Final = "cronjob_lock:reset_budget_job" PEER_POD_LEASE: Final = json.dumps("integration-peer-pod-holding-the-reset-lease") FAST_RESET_TICK: Final = MappingProxyType( diff --git a/tests/integration/spend/test_responses_websocket_turn_rows.py b/tests/integration/spend/test_responses_websocket_turn_rows.py index 59b5925c6eb..4d72c5b17cf 100644 --- a/tests/integration/spend/test_responses_websocket_turn_rows.py +++ b/tests/integration/spend/test_responses_websocket_turn_rows.py @@ -22,7 +22,7 @@ from integration._support.responses_ws_peer import PROVIDER_MODEL, USAGE, comple from integration._support.tls import write_self_signed_cert from pydantic import JsonValue -pytestmark: Final = pytest.mark.timeout(2 * graceful_stop_seconds() + 180) +pytestmark: Final = [pytest.mark.timeout(2 * graceful_stop_seconds() + 180), pytest.mark.shares_worker_state] INPUT_RATE: Final = 0.001 OUTPUT_RATE: Final = 0.002 diff --git a/tests/integration/spend/test_spend_log_caller_session.py b/tests/integration/spend/test_spend_log_caller_session.py index 213583d21a1..756e02a81ee 100644 --- a/tests/integration/spend/test_spend_log_caller_session.py +++ b/tests/integration/spend/test_spend_log_caller_session.py @@ -38,7 +38,10 @@ from integration._support.process import OwnedProxy, graceful_stop_seconds, owne from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import JsonValue -pytestmark: Final = pytest.mark.timeout(2 * readiness_seconds(2) + 2 * graceful_stop_seconds() + 240) +pytestmark: Final = [ + pytest.mark.timeout(2 * readiness_seconds(2) + 2 * graceful_stop_seconds() + 240), + pytest.mark.shares_worker_state, +] MARKER: Final = re.compile(rb"callersess-[0-9a-f]{32}") CALLER_SESSION_KEY: Final = "litellm_caller_session_id" diff --git a/tests/integration/spend/test_spend_log_tool_payload_content.py b/tests/integration/spend/test_spend_log_tool_payload_content.py index 672c0a38407..9055f6f06c4 100644 --- a/tests/integration/spend/test_spend_log_tool_payload_content.py +++ b/tests/integration/spend/test_spend_log_tool_payload_content.py @@ -22,6 +22,8 @@ from pydantic import JsonValue, TypeAdapter from litellm.constants import LITELLM_TRUNCATED_PAYLOAD_FIELD, LITELLM_TRUNCATION_DB_SAFEGUARD_NOTE from litellm.responses.utils import ResponsesAPIRequestUtils +pytestmark: Final = pytest.mark.shares_worker_state + JSON_OBJECT: Final = TypeAdapter(dict[str, JsonValue]) REDACTED: Final = "REDACTED_BY_LITELM" TOOL_INPUT: Final = {"key": "order-123", "sort_key": "created_at"} diff --git a/tests/integration/spend/test_spend_log_write_batching.py b/tests/integration/spend/test_spend_log_write_batching.py index 386f99b1d07..0db4942ceb1 100644 --- a/tests/integration/spend/test_spend_log_write_batching.py +++ b/tests/integration/spend/test_spend_log_write_batching.py @@ -9,6 +9,8 @@ from integration._support.database import read_rows from integration._support.process import owned_proxy from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + WRITE_STATEMENT_MAX_BYTES: Final = 200_000 MESSAGES_PER_REQUEST: Final = 60 MESSAGE_CHARACTERS: Final = 2_000 diff --git a/tests/integration/spend/test_spend_logs_metadata_fields.py b/tests/integration/spend/test_spend_logs_metadata_fields.py index 9941edf0f26..cf5b123d1d5 100644 --- a/tests/integration/spend/test_spend_logs_metadata_fields.py +++ b/tests/integration/spend/test_spend_logs_metadata_fields.py @@ -18,6 +18,8 @@ from integration.spend.test_batch_completion_accounting import batch_input_file, from integration.streaming.test_stream_contracts import text_stream from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + CACHED_PROMPT_TOKENS: Final = 4 diff --git a/tests/integration/spend/test_team_member_budget_alerts.py b/tests/integration/spend/test_team_member_budget_alerts.py index f12bcb9748a..be4bbb17e63 100644 --- a/tests/integration/spend/test_team_member_budget_alerts.py +++ b/tests/integration/spend/test_team_member_budget_alerts.py @@ -9,6 +9,8 @@ from integration._support.database import read_rows from integration._support.mail import smtp_sink from integration._support.process import owned_proxy +pytestmark: Final = pytest.mark.shares_worker_state + MEMBER_BUDGET: Final = 0.10 CALL_COST: Final = 20 * 0.001 + 20 * 0.002 diff --git a/tests/integration/spend/test_team_member_spend_flush.py b/tests/integration/spend/test_team_member_spend_flush.py index 2433731f733..e4b897e087e 100644 --- a/tests/integration/spend/test_team_member_spend_flush.py +++ b/tests/integration/spend/test_team_member_spend_flush.py @@ -19,6 +19,8 @@ from integration._support.database import read_rows from integration._support.process import owned_proxy from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + SINGLE_CONNECTION_CONFIG: Final = """ model_list: [] general_settings: diff --git a/tests/integration/spend/test_user_budget_on_team_keys.py b/tests/integration/spend/test_user_budget_on_team_keys.py index f8e0eea3454..f80859555f0 100644 --- a/tests/integration/spend/test_user_budget_on_team_keys.py +++ b/tests/integration/spend/test_user_budget_on_team_keys.py @@ -10,6 +10,8 @@ from integration._support.client import Gateway, eventually from integration._support.database import read_rows from integration._support.process import owned_proxy +pytestmark: Final = pytest.mark.shares_worker_state + @pytest.mark.covers("spend.user_budget.opted_in_team_key_is_denied_once_owner_budget_is_exhausted") def test_team_key_is_denied_before_provider_once_owner_personal_budget_is_exhausted_when_opted_in( diff --git a/tests/integration/spend/test_vertex_image_regional_uplift.py b/tests/integration/spend/test_vertex_image_regional_uplift.py index 9aedb81c3e0..6900a6d489a 100644 --- a/tests/integration/spend/test_vertex_image_regional_uplift.py +++ b/tests/integration/spend/test_vertex_image_regional_uplift.py @@ -28,6 +28,8 @@ from integration._support.vertex import service_account_json from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + _IMAGE_BACKEND: Final = "gemini-scripted-image" _CHAT_BACKEND: Final = "gemini-scripted-chat" _PROJECT: Final = "scripted-project" diff --git a/tests/integration/streaming/test_chat_stream_guardrail_error_events.py b/tests/integration/streaming/test_chat_stream_guardrail_error_events.py index 021ca68ce8e..a8ebe510b8e 100644 --- a/tests/integration/streaming/test_chat_stream_guardrail_error_events.py +++ b/tests/integration/streaming/test_chat_stream_guardrail_error_events.py @@ -23,6 +23,7 @@ from litellm.proxy import client as proxy_client pytestmark: Final = [ pytest.mark.timeout(2 * graceful_stop_seconds() + 120), pytest.mark.xdist_group("chat-stream-guardrail-error-events"), + pytest.mark.shares_worker_state, ] _MODEL: Final = "stream-error-events" diff --git a/tests/integration/streaming/test_faked_responses_stream_chaos.py b/tests/integration/streaming/test_faked_responses_stream_chaos.py index fca99c9d525..612451c237e 100644 --- a/tests/integration/streaming/test_faked_responses_stream_chaos.py +++ b/tests/integration/streaming/test_faked_responses_stream_chaos.py @@ -41,7 +41,7 @@ from integration.streaming.faked_responses_case import ( ) from pydantic import JsonValue -pytestmark: Final = pytest.mark.timeout(2 * graceful_stop_seconds() + 120) +pytestmark: Final = [pytest.mark.timeout(2 * graceful_stop_seconds() + 120), pytest.mark.shares_worker_state] _RESPONSES_MODEL: Final = "faked-stream-chaos" _BRIDGE_MODEL: Final = "faked-stream-chaos-bridge" diff --git a/tests/integration/streaming/test_google_genai_stream_usage_settings_wire.py b/tests/integration/streaming/test_google_genai_stream_usage_settings_wire.py index 796e706c569..1cb778723fd 100644 --- a/tests/integration/streaming/test_google_genai_stream_usage_settings_wire.py +++ b/tests/integration/streaming/test_google_genai_stream_usage_settings_wire.py @@ -18,6 +18,8 @@ from integration._support.vertex import service_account_json from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + _JSON_OBJECT: Final = TypeAdapter(dict[str, JsonValue]) _PROJECT: Final = "scripted-project" _ADAPTER_MODEL: Final = "settings-adapter-openai" diff --git a/tests/integration/streaming/test_interception_converted_stream_options.py b/tests/integration/streaming/test_interception_converted_stream_options.py index 83a3af06355..0ca62d2f014 100644 --- a/tests/integration/streaming/test_interception_converted_stream_options.py +++ b/tests/integration/streaming/test_interception_converted_stream_options.py @@ -36,7 +36,7 @@ from openai.types.responses import ResponseCompletedEvent, ResponseTextDeltaEven from pydantic import JsonValue _CELL_SECONDS: Final = 2 * graceful_stop_seconds() + 120 -pytestmark: Final = pytest.mark.timeout(_CELL_SECONDS) +pytestmark: Final = [pytest.mark.timeout(_CELL_SECONDS), pytest.mark.shares_worker_state] _PROXY_CONFIG: Final = Path(__file__).resolve().parents[1] / "proxy_config.yaml" _NO_CACHE: Final[Mapping[str, JsonValue]] = MappingProxyType({"cache": {"no-cache": True}}) diff --git a/tests/integration/streaming/test_noma_v2_streaming_settings.py b/tests/integration/streaming/test_noma_v2_streaming_settings.py index 6354a2c9135..877978732f8 100644 --- a/tests/integration/streaming/test_noma_v2_streaming_settings.py +++ b/tests/integration/streaming/test_noma_v2_streaming_settings.py @@ -21,6 +21,8 @@ from integration._support.process import OwnedProxy, WorkerHealthcheck, owned_pr from integration._support.wire import Reply, Request, Wire, wire_server from pydantic import JsonValue +pytestmark: Final = pytest.mark.shares_worker_state + Endpoint: TypeAlias = Literal["chat", "messages", "responses"] Spelling: TypeAlias = Literal["noma_v2", "noma_use_v2", "stored", "custom"] Variant: TypeAlias = Literal[ diff --git a/tests/integration/streaming/test_responses_bridge_stream_chaos.py b/tests/integration/streaming/test_responses_bridge_stream_chaos.py index 754ee5d3981..839a19b64a8 100644 --- a/tests/integration/streaming/test_responses_bridge_stream_chaos.py +++ b/tests/integration/streaming/test_responses_bridge_stream_chaos.py @@ -37,7 +37,7 @@ from integration._support.wire import Reply, Request, wire_server from integration._support.ports import allocate_port from pydantic import JsonValue -pytestmark: Final = pytest.mark.timeout(2 * graceful_stop_seconds() + 120) +pytestmark: Final = [pytest.mark.timeout(2 * graceful_stop_seconds() + 120), pytest.mark.shares_worker_state] _MODEL: Final = "bridged-stream-chaos" _MARKER: Final = re.compile(r"marker-([0-9a-f]{32})") diff --git a/tests/integration/streaming/test_responses_bridge_stream_errors.py b/tests/integration/streaming/test_responses_bridge_stream_errors.py index a3efbd233b9..8a5655be0f8 100644 --- a/tests/integration/streaming/test_responses_bridge_stream_errors.py +++ b/tests/integration/streaming/test_responses_bridge_stream_errors.py @@ -50,7 +50,7 @@ from pydantic import JsonValue _PRIMARY: Final = "bridged-primary" _SPARE: Final = "bridged-spare" -pytestmark: Final = pytest.mark.timeout(2 * graceful_stop_seconds() + 120) +pytestmark: Final = [pytest.mark.timeout(2 * graceful_stop_seconds() + 120), pytest.mark.shares_worker_state] def assert_failed_once(wire: Wire, call_id: str, model: str, attempts: int = 1) -> tuple[Request, ...]: diff --git a/tests/integration/streaming/test_stream_contracts.py b/tests/integration/streaming/test_stream_contracts.py index 3f8243033d7..d7537a78db8 100644 --- a/tests/integration/streaming/test_stream_contracts.py +++ b/tests/integration/streaming/test_stream_contracts.py @@ -18,6 +18,8 @@ from integration._support.wire import Reply, Request, wire_server from openai import OpenAI from pydantic import JsonValue, TypeAdapter +pytestmark: Final = pytest.mark.shares_worker_state + JSON_OBJECT: Final = TypeAdapter(dict[str, JsonValue]) diff --git a/tests/integration/translation/chat_completions/reasoning/test_chat_completions_reasoning_gemini.py b/tests/integration/translation/chat_completions/reasoning/test_chat_completions_reasoning_gemini.py index 194e6972453..efd649f9383 100644 --- a/tests/integration/translation/chat_completions/reasoning/test_chat_completions_reasoning_gemini.py +++ b/tests/integration/translation/chat_completions/reasoning/test_chat_completions_reasoning_gemini.py @@ -41,7 +41,7 @@ from pydantic import JsonValue __all__ = ("thoughts_off",) -pytestmark = pytest.mark.timeout(2 * graceful_stop_seconds() + 120) +pytestmark = [pytest.mark.timeout(2 * graceful_stop_seconds() + 120), pytest.mark.shares_worker_state] GEMINI_3_8_FLASH_THINKING_SHOWN_THOUGHT: Final = "**My Inspection Strategy: Validation First**\n\nOkay, here's how I'm approaching this project review. The instruction specifically mentions considering zero, negative, and non-integer quantities. To efficiently address that, I'm going to start with `shop/validation.py`. My reasoning is simple: the validation logic is the first line of defense. Understanding how the system *expects* to handle these edge cases will inform everything else. It's the logical place to begin my investigation to understand the code. I'll use the `read_file` tool to examine that file. I am **not** changing any files. I will call the `chat_completions` client for this project review. I am now proceeding with reading `shop/validation.py`.\n\n\n" diff --git a/tests/integration/translation/responses/cross_model/test_responses_cross_model.py b/tests/integration/translation/responses/cross_model/test_responses_cross_model.py index cb6f973d2df..1439ee3fed1 100644 --- a/tests/integration/translation/responses/cross_model/test_responses_cross_model.py +++ b/tests/integration/translation/responses/cross_model/test_responses_cross_model.py @@ -21,6 +21,8 @@ from integration.translation.responses.bases.azure_ai import ( from integration.translation.responses.bases.openai_responses import GPT_5_6_SOL_TEST_CASE, GPT_6_1_SOL_TEST_CASE from integration.translation.runner import assert_translation +pytestmark: Final = pytest.mark.shares_worker_state + GPT_6_1_SOL_TWO_MESSAGES_BEFORE_TOOLS_TEST_CASE: Final = two_messages_before_tools(GPT_6_1_SOL_TEST_CASE) AZURE_CLAUDE_SONNET_5_5_CROSS_MODEL_TEST_CASE: Final = replace( AZURE_CLAUDE_SONNET_5_5_TEST_CASE, diff --git a/tests/unit/integration_support/test_process.py b/tests/unit/integration_support/test_process.py index 7741de93a80..ea9c9fcc854 100644 --- a/tests/unit/integration_support/test_process.py +++ b/tests/unit/integration_support/test_process.py @@ -8,11 +8,11 @@ import subprocess import sys import threading import time -from collections.abc import Iterator +from collections.abc import Iterator, Mapping from contextlib import contextmanager from functools import reduce from pathlib import Path -from types import ModuleType +from types import MappingProxyType, ModuleType from typing import Final from unittest.mock import MagicMock @@ -25,13 +25,20 @@ from pydantic import TypeAdapter from litellm.proxy.types_utils.utils import get_instance_fn from tests.integration._support.client import Gateway, eventually from tests.integration._support.process import ( + OWNED_STATE_VARIABLE, OwnedProxy, + OwnedState, + PrivateState, WorkerHealthcheck, WorkerKill, _is_gone, + _proxy_environment, + boots_private_state, healthcheck_seconds, kill_stray_process, + owned_environment, owned_proxy_process, + private_configuration, replacement_pid, stop_process_group, worker_configuration, @@ -385,6 +392,98 @@ def test_owned_runtime_configuration_preserves_custom_worker_startup_hooks( ) +WORKER_STATE: Final = MappingProxyType( + {"DATABASE_URL": "postgresql://admin@127.0.0.1/worker", "REDIS_HOST": "127.0.0.1", "REDIS_PORT": "6379"} +) +PRIVATE_DATABASE: Final = "postgresql://admin@127.0.0.1/owned_private" + + +@contextmanager +def _private_state(directory: Path, inherited: Mapping[str, str]) -> Iterator[PrivateState]: + yield PrivateState( + PRIVATE_DATABASE, + MappingProxyType({"DATABASE_URL": PRIVATE_DATABASE, "REDIS_HOST": "127.0.0.2", "REDIS_PORT": "6390"}), + ) + + +@pytest.mark.parametrize( + ("state", "overrides", "remove_environment", "marked", "private"), + ( + (None, {}, (), False, True), + (None, {"DATABASE_URL": "postgresql://admin@127.0.0.1/scratch"}, (), False, False), + (None, {}, ("DATABASE_URL",), False, False), + (OwnedState.SHARED_WITH_WORKER, {}, (), False, False), + (None, {}, (), True, False), + (OwnedState.PRIVATE, {}, (), True, True), + ), + ids=("default", "explicit-database", "stripped-database", "shared", "marked-module", "private-in-marked-module"), +) +def test_an_owned_proxy_boots_private_state_unless_it_shares_or_brings_its_own_database( + state: OwnedState | None, + overrides: Mapping[str, str], + remove_environment: tuple[str, ...], + marked: bool, + private: bool, +) -> None: + marker: Final = {OWNED_STATE_VARIABLE: OwnedState.SHARED_WITH_WORKER.value} if marked else {} + inherited: Final = {**WORKER_STATE, **marker} + assert boots_private_state(state, overrides, remove_environment, inherited) is private + + +def test_a_default_owned_proxy_runs_on_its_own_database_and_redis(tmp_path: Path) -> None: + with ( + httpx.Client(base_url="http://127.0.0.1:1", trust_env=False) as client, + owned_environment(tmp_path, {}, (), None, WORKER_STATE, _private_state) as owned, + ): + environment: Final = _proxy_environment( + Gateway(client, "owned-test-key", "http://127.0.0.1:1"), owned.overrides, (), WORKER_STATE + ) + assert environment["DATABASE_URL"] == PRIVATE_DATABASE != WORKER_STATE["DATABASE_URL"] + assert (environment["REDIS_HOST"], environment["REDIS_PORT"]) == ("127.0.0.2", "6390") + assert owned.database_url == PRIVATE_DATABASE + + +@pytest.mark.parametrize( + ("overrides", "remove_environment", "redis"), + ( + ({}, ("REDIS_HOST", "REDIS_PORT"), (None, None)), + ({"REDIS_HOST": "redis.caller", "REDIS_PORT": "7000"}, (), ("redis.caller", "7000")), + ), + ids=("stripped", "caller-owned"), +) +def test_a_private_owned_proxy_leaves_redis_to_a_caller_that_chose_it( + tmp_path: Path, + overrides: Mapping[str, str], + remove_environment: tuple[str, ...], + redis: tuple[str | None, str | None], +) -> None: + with ( + httpx.Client(base_url="http://127.0.0.1:1", trust_env=False) as client, + owned_environment(tmp_path, overrides, remove_environment, None, WORKER_STATE, _private_state) as owned, + ): + environment: Final = _proxy_environment( + Gateway(client, "owned-test-key", "http://127.0.0.1:1"), owned.overrides, remove_environment, WORKER_STATE + ) + assert environment["DATABASE_URL"] == PRIVATE_DATABASE + assert (environment.get("REDIS_HOST"), environment.get("REDIS_PORT")) == redis + + +@pytest.mark.parametrize( + ("router_settings", "retries"), + (({"allowed_fails": 3}, 0), ({"num_retries": 2}, 2), (None, 0)), + ids=("inherits-worker-default", "keeps-explicit", "no-router-settings"), +) +def test_a_private_owned_proxy_keeps_the_worker_retry_default_unless_its_config_sets_one( + tmp_path: Path, router_settings: dict[str, int] | None, retries: int +) -> None: + source: Final = tmp_path / "source.yaml" + router: Final = {"router_settings": router_settings} if router_settings else {} + source.write_text(yaml.safe_dump({"model_list": [], **router})) + configured: Final = yaml.safe_load(private_configuration(source, tmp_path / "proxy").read_text()) + assert configured["router_settings"]["num_retries"] == retries + assert configured["model_list"] == [] + + _SUPERVISOR: Final = """import multiprocessing import time