We deploy cinder-api in active/backup mode, which means only one cinder-api instance processes API requests at a moment of time.
According to what Alexander found in logs it's either:
1) there was a problem with node-20 (e.g. we ran out of memory and used swap instead, etc)m which made things really slow and increased cinder-api response times to more than 60s
2) cinder-api simply can't process so many requests in time (== we reached the limit of 1 cinder-api)
Unfortunately, atop reports are missing, so we can't know for sure which one was that.
We deploy cinder-api in active/backup mode, which means only one cinder-api instance processes API requests at a moment of time.
According to what Alexander found in logs it's either:
1) there was a problem with node-20 (e.g. we ran out of memory and used swap instead, etc)m which made things really slow and increased cinder-api response times to more than 60s
2) cinder-api simply can't process so many requests in time (== we reached the limit of 1 cinder-api)
Unfortunately, atop reports are missing, so we can't know for sure which one was that.