Getting Started

What to Expect in the First Hour

Why some pages are blank at first, when data lights up, and what initial interpretation to trust.

Most pages in MonPG come alive within sixty seconds. A few take longer, and that's by design. Here's a rough timeline so you know what to trust during the first hour.

Timeline

  • Within 30 seconds — first snapshot arrives. The dashboard's metric tiles populate, the server status flips to online, the header dropdown shows your server with a green dot.
  • By 1 minute — Queries page has rows. The numbers aren't lies but they aren't rates either; pg_stat_statements counters are cumulative since server start (or last reset). Sorting by total_time tells you what queries have done the most work overall, not what's slow right now.
  • Around 90 seconds — rate-based charts (queries per second, tuples per second) flip from flat zero to actual values. They need at least two snapshots to compute a rate.
  • ~5 minutes — time-series charts on Dashboard, Connections, and Backends start drawing curves. They need a handful of points to be readable.
  • ~10 minutes — health checkup runs against pg_stat_bgwriter, the WAL archiver state, replication slots, and autovacuum freshness. Results land on the Alerts page.
  • ~30 minutes — Query Advisor starts surfacing regressions if any queries have changed their mean time.
  • ~1 hour — enough data for "last hour vs. previous hour" comparisons to actually mean something.

Pages that stay quiet longer

Some pages need supporting extensions or configuration before they fill in:

  • Query Advisor needs slow queries to capture. If nothing in your workload runs longer than a few hundred milliseconds, the page legitimately stays empty. Install auto_explain in shared_preload_libraries and set auto_explain.log_min_duration to capture plans for queries above your threshold. Without it, Query Advisor is heuristic-only.
  • Index Advisor needs queries whose plans would benefit from a different index. Without hypopg the suggestions are pattern-match heuristics ("this column is in a WHERE clause and has no index"). With hypopg installed, we run actual EXPLAIN simulations against hypothetical indexes — real "would this help" answers, not guesses.
  • VACUUM Advisor needs pgstattuple to compute real bloat percentages. Without it we estimate from pg_stat_user_tables, which works but is noisier.
  • Log Insights needs log collection wired up. On managed PG that's a separate IAM step (CloudWatch role on RDS, Diagnostic Settings to Event Hub on Azure, service account on Cloud SQL). The setup pages under Settings → Servers walk through it.

Things that look alarming but aren't

  • "WAL archiver failed" warnings in the first cycle — usually means the monitor role doesn't have pg_read_server_files grant. The collector shrugs and continues. Affects exactly one System-page widget; everything else works.
  • "permission denied for view pg_buffercache" — you didn't install pg_buffercache. Not an error, just a missing optional extension. Either install it (it's safe — pure DDL) or ignore.
  • "calculated statement deltas: count=0" on the very first cycle — normal. There's nothing to delta against yet.