Logs

• The user mentions a specific error string, service name, or log level
• You need to correlate a symptom with a specific moment in time
• A trace span failed and you need the structured log output around it
• You’re looking for a pattern (repeated error, frequency change) before narrowing to specific records

Logs are your best tool for verbatim evidence. They’re your worst tool for “what’s the overall picture” — reach for query-aggregations instead.

Tool guidance (query-logs)

The query-logs tool returns paginated log entries with timestamp, level, message, and core attributes.

• start_date is required; ISO format with timezone (e.g. 2026-01-24T15:25:19.000-08:00 or 2026-01-24T15:25:19Z).
• end_date defaults to now.
• query is a filter expression like message="error" AND level=error. Empty string returns all logs in range. See query syntax below.
• limit defaults to 20, max 50. Do not request more.
• direction is ASC or DESC. Defaults to DESC (newest first).

Session-scoped queries. To pull logs from a specific session, use secure_session_id=<id> in the query. Set start_date to the session’s created_at and end_date to created_at + 3 hours. Narrower windows will miss tail events.

Typical patterns

1. Error pattern search — query="level=error AND service_name=<svc>", 24h window, aggregate first with query-aggregations if volume is high.
2. Specific error message — escape quotes in the query: query="message=\"Cannot use 'in' operator*\"". Wildcards work.
3. Logs around a trace — pull the trace first, grab the timestamp and service, then query="trace_id=<id>" with a 1-minute window centered on the event.
4. Service-wide health snapshot — query-aggregations with product_type="logs" and group_by="level" for a time window, then drill into high-count levels with query-logs.

Interpreting results

• Don’t paraphrase — cite the exact log line. “The database is slow” is not actionable; {"ts": "2026-04-20T14:23:11Z", "level": "error", "message": "pg_pool: connection timeout after 30s"} is.
• If you see repeated errors, note the exact cadence — a burst suggests an incident onset; steady-state suggests a persistent bug.
• Log level drift often correlates with deploys or flag flips — cross-reference with query-flag-evaluations when a regression appears.

Common mistakes

• Asking for 100+ entries. Max is 50. Use query-aggregations for aggregates.
• Forgetting to specify start_date. It’s required.
• Wildcarding too broadly (message="*") — returns everything, wastes token budget.
• Assuming attribute names. Call get-keys(product_type="Logs") if you’re unsure which field to filter on.