Hi Liam, no, none of the indexes are corrupted. There are no errors in the log at startup.
Best, Pieter ________________________________ Van: Liam Clarke-Hutchinson <liam.cla...@adscale.co.nz> Verzonden: maandag 8 juni 2020 22:31 Aan: users@kafka.apache.org <users@kafka.apache.org> Onderwerp: Re: Kafka broker occasionally takes a very long time for log loading on startup Hi Pieter, Do the logs on the slow broker show a lot of "rebuilding corrupted indexes" messages? Cheers, Liam Clarke-Hutchinson On Tue, 9 Jun. 2020, 12:08 am Pieter Hameete, <pieter.hame...@blockbax.com> wrote: > Hello, > > we are observing Kafka brokers occasionally taking a very long time to > load logs on startup compared to usual (40 minutes versus 30-60 seconds). > This happens when restarting a broker (as part of a rolling restart > following the procedure prescribed by Confluent here > https://docs.confluent.io/current/kafka/post-deployment.html#rolling-restart). > The broker has enabled controlled shutdown which was reported to be > successful. > > We use Confluent Platform 5.5.0 (Kafka 2.5.0), and 3 brokers with a > replication factor of 3 and at least 2 in-sync replicas. Each broker uses > 1TB of AWS EBS for log storage. > > Some observations: > > * It is not always the same broker that take a long time to start up > * It does not only occur for the broker that is the active controller > * A topic partition that laods quickly normally (15ms) can take a long > time (9549ms) for the same broker only a day later. > * We experienced this on Kafka 2.4.0 as well, though it did not occur > for a few weeks after upgrading to 2.5.0 > * EBS throughput is not lower for slow startup cases compared to fast > startup. > * The logs do not show any errors on startup when this happens. > > Does anyone have an idea what could be causing this? Or any suggestions on > further tests to conduct to get closer to the cause? > > Thank you in advance and have a good day! > > -- Pieter Hameete >