Roy Dahan
Roy Dahan
The deviation is indeed big, but I don't remember what is a typical deviation we get with iotune. I wonder if there is a pattern there (e.g. IOPS are lower...
I'm even more confused than before. @syuu1228 , @yaronkaikov let's have a meeting to expedite this and let's have @avikivity and @xemul on this meeting.
We want to merge this PR as a workaround for now so scylla-cloud can start supporting i7i. I wonder if we should use at least some of the better numbers...
It could be that the peak in latency is very short and the Prometheus metrics we collect are not sensitive enough to catch it. In the HDR file, how many...
> [@juliayakovlev](https://github.com/juliayakovlev) , [@roydahan](https://github.com/roydahan) > > > 2. Keep in mind, that you are looking at the `P99` of separate 4 loaders and then compare it to the summary of...
@juliayakovlev according to @fruch we have the HDR metrics also in prometheus. Please have both graphs (HDR and the regular one we have above) and see where do we get...
So, IIUC, it seems like the latency spike is a real issue and you could even pinpoint it to specific duration "start of Scylla nodes and Bloom filter rebuilding". In...
Is it a client side only feature?
One more thing, regardless of my review above. If you would want to introduce new nemesis (which again, shouldn't be the case IMO here), the testing of this PR should...