mirror of
https://github.com/fabro-sh/fabro.git
synced 2026-09-09 22:33:37 +00:00
test(cli): reduce logs snapshot surface area
Replace the brittle raw NDJSON snapshots in logs tests with direct assertions on the stable contract: the command succeeds, all events belong to the requested run, and the expected lifecycle events appear in order. This keeps coverage on logs behavior while avoiding churn from unrelated run.created payload details like default model configuration.
This commit is contained in:
Bryan Helmkamp
parent
3df499c45f
commit
d30d8d01d3
1 changed files with 78 additions and 96 deletions
|
|
@ -1,7 +1,45 @@
|
|||
use fabro_test::{fabro_snapshot, test_context};
|
||||
use serde_json::Value;
|
||||
|
||||
use super::support::{setup_completed_dry_run, setup_detached_dry_run};
|
||||
|
||||
fn parse_ndjson(stdout: &[u8]) -> Vec<Value> {
|
||||
String::from_utf8(stdout.to_vec())
|
||||
.expect("stdout should be valid UTF-8")
|
||||
.lines()
|
||||
.filter(|line| !line.trim().is_empty())
|
||||
.map(|line| serde_json::from_str::<Value>(line).expect("logs output should be valid NDJSON"))
|
||||
.collect()
|
||||
}
|
||||
|
||||
fn assert_event_sequence_contains(events: &[Value], expected: &[&str]) {
|
||||
let event_names: Vec<&str> = events
|
||||
.iter()
|
||||
.filter_map(|event| event["event"].as_str())
|
||||
.collect();
|
||||
|
||||
let mut cursor = 0;
|
||||
for expected_name in expected {
|
||||
let Some(found_at) = event_names[cursor..]
|
||||
.iter()
|
||||
.position(|name| name == expected_name) else {
|
||||
panic!("missing event {expected_name} in sequence: {event_names:?}");
|
||||
};
|
||||
cursor += found_at + 1;
|
||||
}
|
||||
}
|
||||
|
||||
fn assert_events_belong_to_run(events: &[Value], run_id: &str) {
|
||||
assert!(!events.is_empty(), "expected at least one log event");
|
||||
for event in events {
|
||||
assert_eq!(
|
||||
event["run_id"].as_str(),
|
||||
Some(run_id),
|
||||
"event should belong to requested run: {event}"
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn help() {
|
||||
let context = test_context!();
|
||||
|
|
@ -39,57 +77,29 @@ fn help() {
|
|||
fn logs_completed_run_outputs_raw_ndjson() {
|
||||
let context = test_context!();
|
||||
let run = setup_completed_dry_run(&context);
|
||||
let mut filters = context.filters();
|
||||
filters.push((
|
||||
r"\d{4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2}(?:\.\d+)?Z".to_string(),
|
||||
"[TIMESTAMP]".to_string(),
|
||||
));
|
||||
filters.push((
|
||||
r#""duration_ms":\s*\d+"#.to_string(),
|
||||
r#""duration_ms": [DURATION_MS]"#.to_string(),
|
||||
));
|
||||
filters.push((
|
||||
r#""id":"[0-9a-f-]+""#.to_string(),
|
||||
r#""id":"[EVENT_ID]""#.to_string(),
|
||||
));
|
||||
filters.push((
|
||||
r#""run_dir":"(?:\[DRY_RUN_DIR\]|\[STORAGE_DIR\]/runs/REDACTED)""#.to_string(),
|
||||
r#""run_dir":"[RUN_DIR]""#.to_string(),
|
||||
));
|
||||
let mut cmd = context.command();
|
||||
cmd.args(["logs", &run.run_id]);
|
||||
let output = cmd.output().expect("command should execute");
|
||||
assert!(
|
||||
output.status.success(),
|
||||
"logs should succeed\nstdout:\n{}\nstderr:\n{}",
|
||||
String::from_utf8_lossy(&output.stdout),
|
||||
String::from_utf8_lossy(&output.stderr)
|
||||
);
|
||||
|
||||
fabro_snapshot!(filters, cmd, @r#"
|
||||
success: true
|
||||
exit_code: 0
|
||||
----- stdout -----
|
||||
{"event":"run.created","id":"[EVENT_ID]","properties":{"graph":{"attrs":{"goal":{"String":"Run tests and report results"},"rankdir":{"String":"LR"}},"edges":[{"attrs":{},"from":"start","to":"run_tests"},{"attrs":{},"from":"run_tests","to":"report"},{"attrs":{},"from":"report","to":"exit"}],"name":"Simple","nodes":{"exit":{"attrs":{"label":{"String":"Exit"},"shape":{"String":"Msquare"}},"id":"exit"},"report":{"attrs":{"label":{"String":"Report"},"prompt":{"String":"Summarize the test results"}},"id":"report"},"run_tests":{"attrs":{"label":{"String":"Run Tests"},"prompt":{"String":"Run the test suite and report results"}},"id":"run_tests"},"start":{"attrs":{"label":{"String":"Start"},"shape":{"String":"Mdiamond"}},"id":"start"}}},"host_repo_path":"[TEMP_DIR]","labels":{"fabro_test_case":"[TEST_CASE]","fabro_test_run":"[TEST_RUN]"},"run_dir":"[RUN_DIR]","settings":{"auto_approve":true,"dry_run":true,"fabro":{"root":"fabro/"},"features":{"retros":false,"session_sandboxes":false},"goal":"Run tests and report results","hooks":[{"blocking":true,"command":"cargo fmt","event":"post_tool_use","matcher":"write_file|edit_file|apply_patch","name":"cargo-fmt","sandbox":null,"timeout_ms":null}],"labels":{"fabro_test_case":"[TEST_CASE]","fabro_test_run":"[TEST_RUN]"},"llm":{"fallbacks":null,"model":"gpt-5.4","provider":"openai"},"mode":"standalone","no_retro":true,"pull_request":{"auto_merge":false,"draft":false,"enabled":true,"merge_strategy":"squash"},"sandbox":{"daytona":{"auto_stop_interval":30,"labels":{"repo":"fabro-sh/fabro"},"network":null,"skip_clone":false,"snapshot":{"cpu":4,"disk":20,"dockerfile":"FROM ubuntu:24.04/n/nRUN apt-get update && apt-get install -y --no-install-recommends curl git ca-certificates build-essential pkg-config libssl-dev unzip python3 && rm -rf /var/lib/apt/lists/*/n/n# GitHub CLI/nRUN curl -fsSL https://cli.github.com/packages/githubcli-archive-keyring.gpg | dd of=/usr/share/keyrings/githubcli-archive-keyring.gpg && echo \"deb [arch=$(dpkg --print-architecture) signed-by=/usr/share/keyrings/githubcli-archive-keyring.gpg] https://cli.github.com/packages stable main\" | tee /etc/apt/sources.list.d/github-cli.list > /dev/null && apt-get update && apt-get install -y --no-install-recommends gh && rm -rf /var/lib/apt/lists/*/n/n# Rust/nRUN curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y/nENV PATH=\"/root/.cargo/bin:${PATH}\"/nRUN cargo install cargo-nextest --locked/nENV CARGO_INCREMENTAL=0/n/n# Bun/nRUN curl -fsSL https://bun.sh/install | bash/nENV PATH=\"/root/.bun/bin:${PATH}\"/n/nWORKDIR /root/n","memory":8,"name":"fabro-v6"}},"devcontainer":null,"env":null,"local":null,"preserve":null,"provider":"local"},"storage_dir":"[STORAGE_DIR]","version":1},"workflow_slug":"simple","workflow_source":"digraph Simple {/n graph [goal=\"Run tests and report results\"]/n rankdir=LR/n/n start [shape=Mdiamond, label=\"Start\"]/n exit [shape=Msquare, label=\"Exit\"]/n/n run_tests [label=\"Run Tests\", prompt=\"Run the test suite and report results\"]/n report [label=\"Report\", prompt=\"Summarize the test results\"]/n/n start -> run_tests -> report -> exit/n}/n","working_directory":"[TEMP_DIR]"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"run.submitted","id":"[EVENT_ID]","properties":{},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"run.starting","id":"[EVENT_ID]","properties":{"reason":"sandbox_initializing"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"sandbox.initializing","id":"[EVENT_ID]","properties":{"provider":"local"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"sandbox.ready","id":"[EVENT_ID]","properties":{"duration_ms": [DURATION_MS],"provider":"local"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"sandbox.initialized","id":"[EVENT_ID]","properties":{"provider":"local","working_directory":"[TEMP_DIR]"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"run.started","id":"[EVENT_ID]","properties":{"goal":"Run tests and report results","name":"Simple"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"run.running","id":"[EVENT_ID]","properties":{},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"stage.started","id":"[EVENT_ID]","node_id":"start","node_label":"Start","properties":{"attempt":1,"handler_type":"start","index":0,"max_attempts":1},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"stage.completed","id":"[EVENT_ID]","node_id":"start","node_label":"Start","properties":{"attempt":1,"context_values":{"current.preamble":"Goal: Run tests and report results/n","current_node":"start","graph.goal":"Run tests and report results","graph.rankdir":"LR","internal.fidelity":"compact","internal.node_visit_count":1,"internal.run_id":"[ULID]","internal.thread_id":null},"duration_ms": [DURATION_MS],"index":0,"max_attempts":1,"node_visits":{"start":1},"notes":"[Simulated] start","status":"success"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"edge.selected","id":"[EVENT_ID]","properties":{"from_node":"start","is_jump":false,"reason":"unconditional","stage_status":"success","to_node":"run_tests"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"checkpoint.completed","id":"[EVENT_ID]","node_id":"start","node_label":"start","properties":{"completed_nodes":["start"],"context_values":{"current.preamble":"Goal: Run tests and report results/n","current_node":"start","failure_class":"","failure_signature":"","graph.goal":"Run tests and report results","graph.rankdir":"LR","internal.fidelity":"compact","internal.node_visit_count":1,"internal.retry_count.start":0,"internal.run_id":"[ULID]","internal.thread_id":null,"outcome":"success"},"current_node":"start","next_node_id":"run_tests","node_outcomes":{"start":{"notes":"[Simulated] start","status":"success","usage":null}},"node_visits":{"start":1},"status":"success"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"stage.started","id":"[EVENT_ID]","node_id":"run_tests","node_label":"Run Tests","properties":{"attempt":1,"handler_type":"agent","index":1,"max_attempts":1},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"stage.completed","id":"[EVENT_ID]","node_id":"run_tests","node_label":"Run Tests","properties":{"attempt":1,"context_updates":{"last_response":"[Simulated] Response for stage: run_tests","last_stage":"run_tests","response.run_tests":"[Simulated] Response for stage: run_tests"},"context_values":{"current.preamble":"Goal: Run tests and report results/n","current_node":"run_tests","failure_class":"","failure_signature":"","graph.goal":"Run tests and report results","graph.rankdir":"LR","internal.fidelity":"compact","internal.node_visit_count":1,"internal.retry_count.start":0,"internal.run_id":"[ULID]","internal.thread_id":"start","outcome":"success","thread.start.current_node":"run_tests"},"duration_ms": [DURATION_MS],"index":1,"max_attempts":1,"node_visits":{"run_tests":1,"start":1},"notes":"[Simulated] run_tests","response":"[Simulated] Response for stage: run_tests","status":"success"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"edge.selected","id":"[EVENT_ID]","properties":{"from_node":"run_tests","is_jump":false,"reason":"unconditional","stage_status":"success","to_node":"report"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"checkpoint.completed","id":"[EVENT_ID]","node_id":"run_tests","node_label":"run_tests","properties":{"completed_nodes":["start","run_tests"],"context_values":{"current.preamble":"Goal: Run tests and report results/n","current_node":"run_tests","failure_class":"","failure_signature":"","graph.goal":"Run tests and report results","graph.rankdir":"LR","internal.fidelity":"compact","internal.node_visit_count":1,"internal.retry_count.run_tests":0,"internal.retry_count.start":0,"internal.run_id":"[ULID]","internal.thread_id":"start","last_response":"[Simulated] Response for stage: run_tests","last_stage":"run_tests","outcome":"success","response.run_tests":"[Simulated] Response for stage: run_tests","thread.start.current_node":"run_tests"},"current_node":"run_tests","next_node_id":"report","node_outcomes":{"run_tests":{"context_updates":{"last_response":"[Simulated] Response for stage: run_tests","last_stage":"run_tests","response.run_tests":"[Simulated] Response for stage: run_tests"},"notes":"[Simulated] run_tests","status":"success","usage":null},"start":{"notes":"[Simulated] start","status":"success","usage":null}},"node_visits":{"run_tests":1,"start":1},"status":"success"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"stage.started","id":"[EVENT_ID]","node_id":"report","node_label":"Report","properties":{"attempt":1,"handler_type":"agent","index":2,"max_attempts":1},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"stage.completed","id":"[EVENT_ID]","node_id":"report","node_label":"Report","properties":{"attempt":1,"context_updates":{"last_response":"[Simulated] Response for stage: report","last_stage":"report","response.report":"[Simulated] Response for stage: report"},"context_values":{"current.preamble":"Goal: Run tests and report results/n/n## Completed stages/n- **run_tests**: success/n","current_node":"report","failure_class":"","failure_signature":"","graph.goal":"Run tests and report results","graph.rankdir":"LR","internal.fidelity":"compact","internal.node_visit_count":1,"internal.retry_count.run_tests":0,"internal.retry_count.start":0,"internal.run_id":"[ULID]","internal.thread_id":"run_tests","last_response":"[Simulated] Response for stage: run_tests","last_stage":"run_tests","outcome":"success","response.run_tests":"[Simulated] Response for stage: run_tests","thread.run_tests.current_node":"report","thread.start.current_node":"run_tests"},"duration_ms": [DURATION_MS],"index":2,"max_attempts":1,"node_visits":{"report":1,"run_tests":1,"start":1},"notes":"[Simulated] report","response":"[Simulated] Response for stage: report","status":"success"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"edge.selected","id":"[EVENT_ID]","properties":{"from_node":"report","is_jump":false,"reason":"unconditional","stage_status":"success","to_node":"exit"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"checkpoint.completed","id":"[EVENT_ID]","node_id":"report","node_label":"report","properties":{"completed_nodes":["start","run_tests","report"],"context_values":{"current.preamble":"Goal: Run tests and report results/n/n## Completed stages/n- **run_tests**: success/n","current_node":"report","failure_class":"","failure_signature":"","graph.goal":"Run tests and report results","graph.rankdir":"LR","internal.fidelity":"compact","internal.node_visit_count":1,"internal.retry_count.report":0,"internal.retry_count.run_tests":0,"internal.retry_count.start":0,"internal.run_id":"[ULID]","internal.thread_id":"run_tests","last_response":"[Simulated] Response for stage: report","last_stage":"report","outcome":"success","response.report":"[Simulated] Response for stage: report","response.run_tests":"[Simulated] Response for stage: run_tests","thread.run_tests.current_node":"report","thread.start.current_node":"run_tests"},"current_node":"report","next_node_id":"exit","node_outcomes":{"report":{"context_updates":{"last_response":"[Simulated] Response for stage: report","last_stage":"report","response.report":"[Simulated] Response for stage: report"},"notes":"[Simulated] report","status":"success","usage":null},"run_tests":{"context_updates":{"last_response":"[Simulated] Response for stage: run_tests","last_stage":"run_tests","response.run_tests":"[Simulated] Response for stage: run_tests"},"notes":"[Simulated] run_tests","status":"success","usage":null},"start":{"notes":"[Simulated] start","status":"success","usage":null}},"node_visits":{"report":1,"run_tests":1,"start":1},"status":"success"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"stage.started","id":"[EVENT_ID]","node_id":"exit","node_label":"Exit","properties":{"attempt":1,"handler_type":"exit","index":3,"max_attempts":1},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"stage.completed","id":"[EVENT_ID]","node_id":"exit","node_label":"Exit","properties":{"attempt":1,"duration_ms": [DURATION_MS],"index":3,"max_attempts":1,"status":"success"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"run.completed","id":"[EVENT_ID]","properties":{"artifact_count":0,"duration_ms": [DURATION_MS],"reason":"completed","status":"success"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"sandbox.cleanup.started","id":"[EVENT_ID]","properties":{"provider":"local"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"sandbox.cleanup.completed","id":"[EVENT_ID]","properties":{"duration_ms": [DURATION_MS],"provider":"local"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
----- stderr -----
|
||||
"#);
|
||||
let events = parse_ndjson(&output.stdout);
|
||||
assert_events_belong_to_run(&events, &run.run_id);
|
||||
assert_event_sequence_contains(
|
||||
&events,
|
||||
&[
|
||||
"run.created",
|
||||
"run.running",
|
||||
"stage.started",
|
||||
"stage.completed",
|
||||
"run.completed",
|
||||
"sandbox.cleanup.completed",
|
||||
],
|
||||
);
|
||||
}
|
||||
|
||||
#[test]
|
||||
|
|
@ -204,55 +214,27 @@ fn logs_pretty_formats_small_run() {
|
|||
fn logs_follow_detached_run_streams_until_completion() {
|
||||
let context = test_context!();
|
||||
let run = setup_detached_dry_run(&context);
|
||||
let mut filters = context.filters();
|
||||
filters.push((
|
||||
r"\d{4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2}(?:\.\d+)?Z".to_string(),
|
||||
"[TIMESTAMP]".to_string(),
|
||||
));
|
||||
filters.push((
|
||||
r#""duration_ms":\s*\d+"#.to_string(),
|
||||
r#""duration_ms": [DURATION_MS]"#.to_string(),
|
||||
));
|
||||
filters.push((
|
||||
r#""id":"[0-9a-f-]+""#.to_string(),
|
||||
r#""id":"[EVENT_ID]""#.to_string(),
|
||||
));
|
||||
filters.push((
|
||||
r#""run_dir":"(?:\[DRY_RUN_DIR\]|\[STORAGE_DIR\]/runs/REDACTED)""#.to_string(),
|
||||
r#""run_dir":"[RUN_DIR]""#.to_string(),
|
||||
));
|
||||
let mut cmd = context.command();
|
||||
cmd.args(["logs", "--follow", &run.run_id]);
|
||||
let output = cmd.output().expect("command should execute");
|
||||
assert!(
|
||||
output.status.success(),
|
||||
"logs --follow should succeed\nstdout:\n{}\nstderr:\n{}",
|
||||
String::from_utf8_lossy(&output.stdout),
|
||||
String::from_utf8_lossy(&output.stderr)
|
||||
);
|
||||
|
||||
fabro_snapshot!(filters, cmd, @r#"
|
||||
success: true
|
||||
exit_code: 0
|
||||
----- stdout -----
|
||||
{"event":"run.created","id":"[EVENT_ID]","properties":{"graph":{"attrs":{"goal":{"String":"Run tests and report results"},"rankdir":{"String":"LR"}},"edges":[{"attrs":{},"from":"start","to":"run_tests"},{"attrs":{},"from":"run_tests","to":"report"},{"attrs":{},"from":"report","to":"exit"}],"name":"Simple","nodes":{"exit":{"attrs":{"label":{"String":"Exit"},"shape":{"String":"Msquare"}},"id":"exit"},"report":{"attrs":{"label":{"String":"Report"},"prompt":{"String":"Summarize the test results"}},"id":"report"},"run_tests":{"attrs":{"label":{"String":"Run Tests"},"prompt":{"String":"Run the test suite and report results"}},"id":"run_tests"},"start":{"attrs":{"label":{"String":"Start"},"shape":{"String":"Mdiamond"}},"id":"start"}}},"host_repo_path":"[TEMP_DIR]","labels":{"fabro_test_case":"[TEST_CASE]","fabro_test_run":"[TEST_RUN]"},"run_dir":"[RUN_DIR]","settings":{"auto_approve":true,"dry_run":true,"fabro":{"root":"fabro/"},"features":{"retros":false,"session_sandboxes":false},"goal":"Run tests and report results","hooks":[{"blocking":true,"command":"cargo fmt","event":"post_tool_use","matcher":"write_file|edit_file|apply_patch","name":"cargo-fmt","sandbox":null,"timeout_ms":null}],"labels":{"fabro_test_case":"[TEST_CASE]","fabro_test_run":"[TEST_RUN]"},"llm":{"fallbacks":null,"model":"gpt-5.4","provider":"openai"},"mode":"standalone","no_retro":true,"pull_request":{"auto_merge":false,"draft":false,"enabled":true,"merge_strategy":"squash"},"sandbox":{"daytona":{"auto_stop_interval":30,"labels":{"repo":"fabro-sh/fabro"},"network":null,"skip_clone":false,"snapshot":{"cpu":4,"disk":20,"dockerfile":"FROM ubuntu:24.04/n/nRUN apt-get update && apt-get install -y --no-install-recommends curl git ca-certificates build-essential pkg-config libssl-dev unzip python3 && rm -rf /var/lib/apt/lists/*/n/n# GitHub CLI/nRUN curl -fsSL https://cli.github.com/packages/githubcli-archive-keyring.gpg | dd of=/usr/share/keyrings/githubcli-archive-keyring.gpg && echo \"deb [arch=$(dpkg --print-architecture) signed-by=/usr/share/keyrings/githubcli-archive-keyring.gpg] https://cli.github.com/packages stable main\" | tee /etc/apt/sources.list.d/github-cli.list > /dev/null && apt-get update && apt-get install -y --no-install-recommends gh && rm -rf /var/lib/apt/lists/*/n/n# Rust/nRUN curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y/nENV PATH=\"/root/.cargo/bin:${PATH}\"/nRUN cargo install cargo-nextest --locked/nENV CARGO_INCREMENTAL=0/n/n# Bun/nRUN curl -fsSL https://bun.sh/install | bash/nENV PATH=\"/root/.bun/bin:${PATH}\"/n/nWORKDIR /root/n","memory":8,"name":"fabro-v6"}},"devcontainer":null,"env":null,"local":null,"preserve":null,"provider":"local"},"storage_dir":"[STORAGE_DIR]","version":1},"workflow_slug":"simple","workflow_source":"digraph Simple {/n graph [goal=\"Run tests and report results\"]/n rankdir=LR/n/n start [shape=Mdiamond, label=\"Start\"]/n exit [shape=Msquare, label=\"Exit\"]/n/n run_tests [label=\"Run Tests\", prompt=\"Run the test suite and report results\"]/n report [label=\"Report\", prompt=\"Summarize the test results\"]/n/n start -> run_tests -> report -> exit/n}/n","working_directory":"[TEMP_DIR]"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"run.submitted","id":"[EVENT_ID]","properties":{},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"run.starting","id":"[EVENT_ID]","properties":{"reason":"sandbox_initializing"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"sandbox.initializing","id":"[EVENT_ID]","properties":{"provider":"local"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"sandbox.ready","id":"[EVENT_ID]","properties":{"duration_ms": [DURATION_MS],"provider":"local"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"sandbox.initialized","id":"[EVENT_ID]","properties":{"provider":"local","working_directory":"[TEMP_DIR]"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"run.started","id":"[EVENT_ID]","properties":{"goal":"Run tests and report results","name":"Simple"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"run.running","id":"[EVENT_ID]","properties":{},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"stage.started","id":"[EVENT_ID]","node_id":"start","node_label":"Start","properties":{"attempt":1,"handler_type":"start","index":0,"max_attempts":1},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"stage.completed","id":"[EVENT_ID]","node_id":"start","node_label":"Start","properties":{"attempt":1,"context_values":{"current.preamble":"Goal: Run tests and report results/n","current_node":"start","graph.goal":"Run tests and report results","graph.rankdir":"LR","internal.fidelity":"compact","internal.node_visit_count":1,"internal.run_id":"[ULID]","internal.thread_id":null},"duration_ms": [DURATION_MS],"index":0,"max_attempts":1,"node_visits":{"start":1},"notes":"[Simulated] start","status":"success"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"edge.selected","id":"[EVENT_ID]","properties":{"from_node":"start","is_jump":false,"reason":"unconditional","stage_status":"success","to_node":"run_tests"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"checkpoint.completed","id":"[EVENT_ID]","node_id":"start","node_label":"start","properties":{"completed_nodes":["start"],"context_values":{"current.preamble":"Goal: Run tests and report results/n","current_node":"start","failure_class":"","failure_signature":"","graph.goal":"Run tests and report results","graph.rankdir":"LR","internal.fidelity":"compact","internal.node_visit_count":1,"internal.retry_count.start":0,"internal.run_id":"[ULID]","internal.thread_id":null,"outcome":"success"},"current_node":"start","next_node_id":"run_tests","node_outcomes":{"start":{"notes":"[Simulated] start","status":"success","usage":null}},"node_visits":{"start":1},"status":"success"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"stage.started","id":"[EVENT_ID]","node_id":"run_tests","node_label":"Run Tests","properties":{"attempt":1,"handler_type":"agent","index":1,"max_attempts":1},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"stage.completed","id":"[EVENT_ID]","node_id":"run_tests","node_label":"Run Tests","properties":{"attempt":1,"context_updates":{"last_response":"[Simulated] Response for stage: run_tests","last_stage":"run_tests","response.run_tests":"[Simulated] Response for stage: run_tests"},"context_values":{"current.preamble":"Goal: Run tests and report results/n","current_node":"run_tests","failure_class":"","failure_signature":"","graph.goal":"Run tests and report results","graph.rankdir":"LR","internal.fidelity":"compact","internal.node_visit_count":1,"internal.retry_count.start":0,"internal.run_id":"[ULID]","internal.thread_id":"start","outcome":"success","thread.start.current_node":"run_tests"},"duration_ms": [DURATION_MS],"index":1,"max_attempts":1,"node_visits":{"run_tests":1,"start":1},"notes":"[Simulated] run_tests","response":"[Simulated] Response for stage: run_tests","status":"success"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"edge.selected","id":"[EVENT_ID]","properties":{"from_node":"run_tests","is_jump":false,"reason":"unconditional","stage_status":"success","to_node":"report"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"checkpoint.completed","id":"[EVENT_ID]","node_id":"run_tests","node_label":"run_tests","properties":{"completed_nodes":["start","run_tests"],"context_values":{"current.preamble":"Goal: Run tests and report results/n","current_node":"run_tests","failure_class":"","failure_signature":"","graph.goal":"Run tests and report results","graph.rankdir":"LR","internal.fidelity":"compact","internal.node_visit_count":1,"internal.retry_count.run_tests":0,"internal.retry_count.start":0,"internal.run_id":"[ULID]","internal.thread_id":"start","last_response":"[Simulated] Response for stage: run_tests","last_stage":"run_tests","outcome":"success","response.run_tests":"[Simulated] Response for stage: run_tests","thread.start.current_node":"run_tests"},"current_node":"run_tests","next_node_id":"report","node_outcomes":{"run_tests":{"context_updates":{"last_response":"[Simulated] Response for stage: run_tests","last_stage":"run_tests","response.run_tests":"[Simulated] Response for stage: run_tests"},"notes":"[Simulated] run_tests","status":"success","usage":null},"start":{"notes":"[Simulated] start","status":"success","usage":null}},"node_visits":{"run_tests":1,"start":1},"status":"success"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"stage.started","id":"[EVENT_ID]","node_id":"report","node_label":"Report","properties":{"attempt":1,"handler_type":"agent","index":2,"max_attempts":1},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"stage.completed","id":"[EVENT_ID]","node_id":"report","node_label":"Report","properties":{"attempt":1,"context_updates":{"last_response":"[Simulated] Response for stage: report","last_stage":"report","response.report":"[Simulated] Response for stage: report"},"context_values":{"current.preamble":"Goal: Run tests and report results/n/n## Completed stages/n- **run_tests**: success/n","current_node":"report","failure_class":"","failure_signature":"","graph.goal":"Run tests and report results","graph.rankdir":"LR","internal.fidelity":"compact","internal.node_visit_count":1,"internal.retry_count.run_tests":0,"internal.retry_count.start":0,"internal.run_id":"[ULID]","internal.thread_id":"run_tests","last_response":"[Simulated] Response for stage: run_tests","last_stage":"run_tests","outcome":"success","response.run_tests":"[Simulated] Response for stage: run_tests","thread.run_tests.current_node":"report","thread.start.current_node":"run_tests"},"duration_ms": [DURATION_MS],"index":2,"max_attempts":1,"node_visits":{"report":1,"run_tests":1,"start":1},"notes":"[Simulated] report","response":"[Simulated] Response for stage: report","status":"success"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"edge.selected","id":"[EVENT_ID]","properties":{"from_node":"report","is_jump":false,"reason":"unconditional","stage_status":"success","to_node":"exit"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"checkpoint.completed","id":"[EVENT_ID]","node_id":"report","node_label":"report","properties":{"completed_nodes":["start","run_tests","report"],"context_values":{"current.preamble":"Goal: Run tests and report results/n/n## Completed stages/n- **run_tests**: success/n","current_node":"report","failure_class":"","failure_signature":"","graph.goal":"Run tests and report results","graph.rankdir":"LR","internal.fidelity":"compact","internal.node_visit_count":1,"internal.retry_count.report":0,"internal.retry_count.run_tests":0,"internal.retry_count.start":0,"internal.run_id":"[ULID]","internal.thread_id":"run_tests","last_response":"[Simulated] Response for stage: report","last_stage":"report","outcome":"success","response.report":"[Simulated] Response for stage: report","response.run_tests":"[Simulated] Response for stage: run_tests","thread.run_tests.current_node":"report","thread.start.current_node":"run_tests"},"current_node":"report","next_node_id":"exit","node_outcomes":{"report":{"context_updates":{"last_response":"[Simulated] Response for stage: report","last_stage":"report","response.report":"[Simulated] Response for stage: report"},"notes":"[Simulated] report","status":"success","usage":null},"run_tests":{"context_updates":{"last_response":"[Simulated] Response for stage: run_tests","last_stage":"run_tests","response.run_tests":"[Simulated] Response for stage: run_tests"},"notes":"[Simulated] run_tests","status":"success","usage":null},"start":{"notes":"[Simulated] start","status":"success","usage":null}},"node_visits":{"report":1,"run_tests":1,"start":1},"status":"success"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"stage.started","id":"[EVENT_ID]","node_id":"exit","node_label":"Exit","properties":{"attempt":1,"handler_type":"exit","index":3,"max_attempts":1},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"stage.completed","id":"[EVENT_ID]","node_id":"exit","node_label":"Exit","properties":{"attempt":1,"duration_ms": [DURATION_MS],"index":3,"max_attempts":1,"status":"success"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"run.completed","id":"[EVENT_ID]","properties":{"artifact_count":0,"duration_ms": [DURATION_MS],"reason":"completed","status":"success"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"sandbox.cleanup.started","id":"[EVENT_ID]","properties":{"provider":"local"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
{"event":"sandbox.cleanup.completed","id":"[EVENT_ID]","properties":{"duration_ms": [DURATION_MS],"provider":"local"},"run_id":"[ULID]","ts":"[TIMESTAMP]"}
|
||||
----- stderr -----
|
||||
"#);
|
||||
let events = parse_ndjson(&output.stdout);
|
||||
assert_events_belong_to_run(&events, &run.run_id);
|
||||
assert_event_sequence_contains(
|
||||
&events,
|
||||
&[
|
||||
"run.created",
|
||||
"run.running",
|
||||
"stage.started",
|
||||
"stage.completed",
|
||||
"run.completed",
|
||||
"sandbox.cleanup.completed",
|
||||
],
|
||||
);
|
||||
}
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue