Potential performance inconsistency
Opened this issue · 3 comments
Hey guys,
Just rolled that time series PR into prod so, its working and seems to be working well for the most part. We are doing around ~300ops/s. It seems that some of the operations are not consistent time though. Any suggestions on profiling this further?
You mean the 120 ms latency caused by exhgetallwithver ?
yea focusing on reads. So we are writing (to hashmaps via tairhash) across the cluster at about 240ops/s. We have a couple of apps doing very light reads, and then user driven traffic makes up the rest. I think most exhgetallwithver calls are <1ms and I think its a bit weird that sometimes its >10ms and occasionally >100ms.
Do you think all exhgetallwithver is accessing the same key? Maybe some tairhash is very big (bigkey) and exhgetallwithver is very slow. exhgetallwithver is an O(n) complexity command.