ops-telemetry-query

Connect to the Redshift Serverless telemetry database with psql using a REDSHIFT_URL connection string, then query the telemetry.broker_heartbeats, telemetry.request_evaluations, and telemetry.request_completions views. Always prefix tables with the telemetry schema and use LIMIT on exploratory queries.

Filter telemetry.request_evaluations on outcome = 'Skipped' and group by skip_code to get counts and percentages over a time window. The example queries file includes a ready-made top-skip-reasons query for the last 24 hours.

You need the Redshift endpoint and the readonly user password, either from network_secrets.toml under [environments.<env>.telemetry] or provided directly as REDSHIFT_ENDPOINT and REDSHIFT_PASSWORD environment variables. These are combined into a postgres://readonly connection string.

zsh treats ! as history expansion inside double quotes, so != in psql -c strings gets mangled. Use <> instead of != in SQL, or run queries through a heredoc which is immune to shell expansion.

No. Redshift lacks LATERAL joins and DISTINCT ON, requires UNION ALL with ORDER BY to be wrapped in a subquery, and allows only one MEDIAN per SELECT. Use GETDATE() instead of NOW() and JSON_PARSE for SUPER columns.

No. Telemetry is opt-in, so the data only represents brokers that have enabled reporting. Do not assume the brokers visible in telemetry are the only ones active on the network.