Backlog by disk

Notices where this attachment appears

1. Embed this notice
   Phantasm (phnt@fluffytail.org)'s status on Saturday, 18-Jan-2025 04:24:09 JST Phantasm

   in reply to
   @7666 It's probably an IOPS quota over some time frame, many VPS providers do that. OVH being one of the worst offenders. After around 2GB, their additional disk throttle to 12,10,8,4 MB/s of sustained read or write. Slowly throttling down as you keep using the disk.
   
   What I usually see is an almost increase in disk backlog to around 5-10 seconds for 15-20 minutes and then it goes back to normal. A reliable way to trigger it is to open a few tabs of random bigger threads, or make many search queries. It's not related to an increase in network traffic/requests, dmcache deciding to write the dirty pages to the primary disk or PostgreSQL deciding to do DB maintenance. It just happens out of nowhere unprovoked.
   number of IOPS by type and disk
   Backlog by disk