gannonh/sym-debug

Debug

Goals

• Find why a run is stuck, retrying, or failing.
• Correlate Linear issue identity to a Codex session quickly.
• Read the right logs in the right order to isolate root cause.

Log Sources

• Primary runtime log file: <logs-root>/log/symphony.log
  • When Symphony runs with --logs-root, it writes rotating JSON logs under this path (see apps/symphony/README.md).
  • Includes orchestrator, agent runner, and Codex app-server lifecycle logs.
• Rotated runtime logs: <logs-root>/log/symphony.log*
  • Check these when the relevant run is older than the active file.
• Stdout fallback: structured JSON log stream
  • Without --logs-root, logs stream to stdout instead of a file.

Correlation Keys

• issue_identifier: human ticket key (example: MT-625)
• issue_id: Linear UUID (stable internal ID)
• session_id: Codex thread-turn pair (<thread_id>-<turn_id>)

These fields are emitted by Symphony runtime lifecycle logs (notably in apps/symphony/src/orchestrator.rs and apps/symphony/src/codex/app_server.rs). Use them as your join keys during debugging.

Quick Triage (Stuck Run)

1. Confirm scheduler/worker symptoms for the ticket.
2. Find recent lines for the ticket (issue_identifier first).
3. Extract session_id from matching lines.
4. Trace that session_id across start, stream, completion/failure, and stall handling logs.
5. Decide class of failure: timeout/stall, app-server startup failure, turn failure, or orchestrator retry loop.

Commands

# File-log mode (`--logs-root` enabled): expand to active + rotated files.
LOG_PATHS=( ${LOG_GLOB:-log/symphony.log*} )

# 1) Narrow by ticket key (fastest entry point)
rg -n "issue_identifier=MT-625" "${LOG_PATHS[@]}"

# 2) If needed, narrow by Linear UUID
rg -n "issue_id=<linear-uuid>" "${LOG_PATHS[@]}"

# 3) Pull session IDs seen for that ticket
rg -o "session_id=[^ ;]+" "${LOG_PATHS[@]}" | sort -u

# 4) Trace one session end-to-end
rg -n "session_id=<thread>-<turn>" "${LOG_PATHS[@]}"

# 5) Focus on stuck/retry signals
rg -n "Issue stalled|scheduling retry|turn_timeout|turn_failed|Codex session failed|Codex session ended with error" "${LOG_PATHS[@]}"

# Stdout mode (startup banner shows `Logs: stdout`): use your runtime stream.
journalctl -u symphony --since "30 minutes ago" --no-pager \
  | rg -n "issue_identifier=MT-625|issue_id=<linear-uuid>|session_id=<thread>-<turn>|Issue stalled|scheduling retry|turn_timeout|turn_failed|Codex session failed|Codex session ended with error"

# Containerized deploys can use docker logs instead of journalctl.
docker logs <symphony-container> --since 30m 2>&1 \
  | rg -n "issue_identifier=MT-625|issue_id=<linear-uuid>|session_id=<thread>-<turn>|Issue stalled|scheduling retry|turn_timeout|turn_failed|Codex session failed|Codex session ended with error"

Investigation Flow

1. Locate the ticket slice:
   • Search by issue_identifier=<KEY>.
   • If noise is high, add issue_id=<UUID>.
2. Establish timeline:
   • Identify first Codex session started ... session_id=....
   • Follow with Codex session completed, ended with error, or worker exit lines.
3. Classify the problem:
   • Stall loop: Issue stalled ... restarting with backoff.
   • App-server startup: Codex session failed ....
   • Turn execution failure: turn_failed, turn_cancelled, turn_timeout, or ended with error.
   • Worker crash: Agent task exited ... reason=....
4. Validate scope:
   • Check whether failures are isolated to one issue/session or repeating across multiple tickets.
5. Capture evidence:
   • Save key log lines with timestamps, issue_identifier, issue_id, and session_id.
   • Record probable root cause and the exact failing stage.

Reading Codex Session Logs

In Symphony, Codex session diagnostics are emitted into log/symphony.log and keyed by session_id. Read them as a lifecycle:

1. Codex session started ... session_id=...
2. Session stream/lifecycle events for the same session_id
3. Terminal event:
   • Codex session completed ..., or
   • Codex session ended with error ..., or
   • Issue stalled ... restarting with backoff

For one specific session investigation, keep the trace narrow:

1. Capture one session_id for the ticket.
2. Build a timestamped slice for only that session:
   • rg -n "session_id=<thread>-<turn>" "$LOG_GLOB"
3. Mark the exact failing stage:
   • Startup failure before stream events (Codex session failed ...).
   • Turn/runtime failure after stream events (turn_* / ended with error).
   • Stall recovery (Issue stalled ... restarting with backoff).
4. Pair findings with issue_identifier and issue_id from nearby lines to confirm you are not mixing concurrent retries.

Always pair session findings with issue_identifier/issue_id to avoid mixing concurrent runs.

Notes

• Prefer rg over grep for speed on large logs.
• Check rotated logs (<logs-root>/log/symphony.log*) before concluding data is missing.
• If required context fields are missing in new log statements, align with existing structured lifecycle logging in apps/symphony/src/orchestrator.rs and apps/symphony/src/codex/app_server.rs.