Exploring Endpoint Execution Logs

Every endpoint run emits one execution log entry to PostHog’s log_entries store. This skill reads those entries for a specific endpoint to answer “what happened when it ran?”. It is the log-level counterpart to diagnosing-endpoint-performance (which reasons about cache/materialisation strategy from config and query_log).

When to use this skill

• “Why is my endpoint failing / erroring?”
• “Show me the logs / recent runs for endpoint X”
• “Did the last run hit cache? How many rows did it return?”
• “What happened the last time endpoint Y ran?”

If the question is “this endpoint is slow, what should I change?”, use diagnosing-endpoint-performance. If it’s project-wide (“what can I clean up?”), use auditing-endpoints.

What an execution log entry looks like

Each run produces exactly one entry. The level is INFO on success and ERROR on failure, and the message carries the extra data as searchable key=value tokens:

Endpoint executed · path=materialized cache=hit duration_ms=142 rows=1024 version=3
Endpoint execution failed · path=inline error=ResolutionError version=3

Token meanings:

Each run gets a distinct instance_id, so logs group one-per-execution in the viewer.

Available tools

Filtering

endpoint-logs exposes the standard log filters:

• level — comma-separated, e.g. ERROR to see only failed runs, or INFO,ERROR for all.
• search — case-insensitive substring over the message. Because the extra data is in key=value tokens, you can search cache=miss, path=inline, error=ResolutionError, or a specific version=3.
• after / before — ISO timestamps to bound the time range.
• instance_id — pin a single execution.
• limit — 1–500 (default 50).

Workflow

1. Identify the endpoint by name. If given a URL, parse it from /api/projects/{team_id}/endpoints/{name}/run.

2. Start broad: endpoint-logs for the endpoint with a recent time range. Skim levels and tokens.

3. Narrow to the symptom:

   • Failures → level=ERROR; read the error= token and path= to see where it broke.
   • Cache concerns → search=cache=miss to see how often runs miss cache.
   • Wrong results → compare rows= across runs, and version= to spot a regression after a version bump.

4. For counts/trends across many runs (e.g. error rate over a week), drop to execute-sql against log_entries:

   sqlCopy

   SELECT toDate(timestamp) AS day, upper(level) AS level, count() AS runs
   FROM log_entries
   WHERE log_source =
   

Example interaction

User: "weekly_signups started erroring this morning"

Agent steps:
- endpoint-logs weekly_signups, level=ERROR, after=<this morning>
  → several "Endpoint execution failed · path=inline error=ResolutionError version=5"
- endpoint-get weekly_signups → current version is v5 (bumped today)
- endpoint-logs weekly_signups, level=INFO, before=<this morning>
  → prior runs: "path=inline cache=hit ... version=4" succeeded

- "v5 (created this morning) is failing with a ResolutionError on the inline path — it can't
   resolve a table or field reference. v4 ran fine. This looks like a bad query in the new
   version. Want me to pull the v5 query (endpoint-versions) so we can fix it, or roll back to v4?"

Important notes

• One entry per run. Don’t expect step-by-step traces — endpoints log a single completion line. The detail lives in the tokens, not in multiple lines.
• log_source_id is the endpoint UUID, not the name. For execute-sql, fetch it via endpoint-get first.
• Logs are retained ~90 days (the log_entries TTL). Older runs won’t appear.
• Execution logs ≠ query performance. endpoint-logs tells you what happened and why a run failed; for “should I materialise / bump cache TTL?” use diagnosing-endpoint-performance, which reasons over config and query_log cost metrics.
• Best-effort emission. A log line is emitted after each run but never blocks it — if a run succeeded for the caller but no log shows, the emit was dropped, not the query.