litellm/litellm-proxy-extras/litellm_proxy_extras/migration_lock.py
devin-ai-integration[bot] c66c8288c3
fix(proxy-extras): build the SpendLogs indexes in the migration job instead of in migrations (#43948)
The two SpendLogs index migrations shipped in v1.103.0 each break one table shape: the plain CREATE INDEX holds a SHARE lock on a large unpartitioned table and the CONCURRENTLY one fails with 0A000 on a partitioned parent. Both files are now inert and the indexes are built by a table-driven, shape-aware step after migrate deploy: CONCURRENTLY on a plain table, ON ONLY the parent plus per-partition CONCURRENTLY and ATTACH PARTITION on a partitioned one. The migration job builds them synchronously and exits non-zero on failure; a serving proxy that ran migrate deploy itself builds them in the background off the readiness path. A valid index of the same definition under another name is renamed and reused, an invalid one is rebuilt, and extra copies are reported with their DROP INDEX statement instead of being dropped. The migration checker rejects any CREATE INDEX on LiteLLM_SpendLogs or LiteLLM_ErrorLogs in future migrations

Co-authored-by: yassin <yassin@berri.ai>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
2026-10-01 14:12:22 -07:00

107 lines
4.6 KiB
Python

import random
import time
from collections.abc import Generator, Mapping
from contextlib import contextmanager
from dataclasses import dataclass
from typing import TYPE_CHECKING, Final
from urllib.parse import parse_qsl, urlencode, urlsplit, urlunsplit
from litellm_proxy_extras._logging import logger
from litellm_proxy_extras.prisma_toolchain import MIGRATION_LOCK_TIMEOUT_ENV_VAR, migration_lock_timeout
MIGRATION_LOCK_KEY: Final = int.from_bytes(b"llm_mig2", "big")
if TYPE_CHECKING:
import psycopg
def migration_environment(environment: Mapping[str, str]) -> Mapping[str, str]:
database_url: Final = environment.get("DATABASE_URL")
direct_url: Final = environment.get("DIRECT_URL")
if not database_url or not direct_url:
return environment
schema: Final = next((value for key, value in parse_qsl(urlsplit(database_url).query) if key == "schema"), "public")
direct: Final = urlsplit(direct_url)
parameters: Final = tuple((key, value) for key, value in parse_qsl(direct.query) if key != "schema")
return {
**environment,
"DATABASE_URL": urlunsplit(direct._replace(query=urlencode((*parameters, ("schema", schema))))),
}
@dataclass(frozen=True, slots=True)
class _LockResult:
acquired: bool
def _try_lock(connection: "psycopg.Connection[tuple[object, ...]]", key: int = MIGRATION_LOCK_KEY) -> bool:
from psycopg.rows import class_row
with connection.cursor(row_factory=class_row(_LockResult)) as cursor:
row: Final = cursor.execute("SELECT pg_try_advisory_xact_lock(%s) AS acquired", (key,)).fetchone()
return row is not None and row.acquired
@dataclass(frozen=True, slots=True)
class MigrationCoordinator:
connection: "psycopg.Connection[tuple[object, ...]]"
def check_connection(self) -> None:
self.connection.execute("SELECT 1")
def acquire_prisma_lock(self) -> None:
deadline: Final = time.monotonic() + migration_lock_timeout()
while time.monotonic() < deadline:
if _try_lock(self.connection, 72707369):
return
time.sleep(min(random.uniform(0.5, 1.5), max(0.0, deadline - time.monotonic())))
raise RuntimeError(
"Timed out waiting for Prisma's lock to recover migration history. LiteLLM startup has stopped. "
"Another migration or a pooled database session may still hold the lock. Check the database lock holder. "
"When using a transaction pooler, configure DIRECT_URL to reach the same database without the pooler."
)
@contextmanager
def migration_lock(database_url: str) -> Generator[MigrationCoordinator, None, None]:
import psycopg
wait_seconds: Final = migration_lock_timeout()
deadline: Final = time.monotonic() + wait_seconds
try:
with psycopg.connect(database_url, connect_timeout=10, autocommit=True) as connection:
coordinator: Final = MigrationCoordinator(connection)
logger.info("Waiting for the v2 migration coordinator lock (up to %ss)", wait_seconds)
while time.monotonic() < deadline:
with connection.transaction():
if _try_lock(connection):
logger.info("Acquired the v2 migration coordinator lock")
yield coordinator
coordinator.check_connection()
return
time.sleep(min(random.uniform(0.5, 1.5), max(0.0, deadline - time.monotonic())))
except psycopg.Error as exc:
raise RuntimeError(f"Lost or could not establish v2 migration coordination with the database: {exc}") from exc
raise RuntimeError(
f"Timed out waiting for another v2 migration resolver after {wait_seconds}s. "
f"Check the running migration or increase {MIGRATION_LOCK_TIMEOUT_ENV_VAR}."
)
@contextmanager
def held_migration_lock(connection: "psycopg.Connection[tuple[object, ...]]") -> Generator[bool, None, None]:
"""A session-level, non-blocking hold of the migration coordinator lock on an autocommit
connection, for DDL that cannot run inside a transaction (`CREATE INDEX CONCURRENTLY`).
Yields whether the lock was acquired; a v2 resolver or another migration job's index build
holding it yields False. Released on exit."""
from psycopg.rows import class_row
with connection.cursor(row_factory=class_row(_LockResult)) as cursor:
row: Final = cursor.execute("SELECT pg_try_advisory_lock(%s) AS acquired", (MIGRATION_LOCK_KEY,)).fetchone()
acquired: Final = row is not None and row.acquired
try:
yield acquired
finally:
if acquired:
connection.execute("SELECT pg_advisory_unlock(%s)", (MIGRATION_LOCK_KEY,))