User:Cook Me Plox/Performance housekeeping

From Meta Wiki
Jump to navigation Jump to search

Things to do every couple months:

  • Check Cloudflare uncached requests, or nginx BigQuery logs, for any large clumps of uncached requests, either due to new APIs, overzealous bots, bad client code, etc
  • Check nginx BigQuery logs for very long-running requests
  • Check mysql slow log
  • Turn on database query logging for a few hours (incl. some GE update if possible) to profile the percentage of DB queries from various sources. Look for opportunities to batch or cache (although redis is not really much better than mysql for latency)
  • Look at total job queue run time per day, and long-running parses (like from bad DPLs)
  • Test parse preview time against known benchmarks from the past
  • Look at CloudSQL total utilization graph