[REA-2374] Memory bloat when running a sysbench workload. about readyset HOT 5 CLOSED

I wonder if this is related to the fact that writes are being directed through ReadySet, rather than exclusively against the upstream

from readyset.

All queries were proxied.

from readyset.

<~accountid:6386c6e93c26ca7fa0d52b22> were you caching queries here? or was this just when running proxied queries?

from readyset.

I was unable to reproduce an OOM event with the provided instructions.

I ran the provided sysbench workload against a docker postgres upstream in 4 configurations:

1. Against the upstream directly
2. Against readyset with no caches enabled (proxying everything)
3. Against readyset with all (1672) supported caches created (cold caches)
4. Against readyset again with the caches theoretically warmed up (warm caches)

Throughout the runs, memory usage stayed stable and was fairly minimal (only
got up to 5% memory usage on my 32GB memory macbook). This makes sense given that the upstream db size is only around 3 GiB, and not all queries/rows make their way into readyset caches.

See attachments for sysbench performance results (which aren't great but that is probably due to proxying most things) as well as memory usage throughout the runs, which was stable.

sysbench_results

memory_usage

from readyset.

Thanks, Luke. I'll see if I'm able to recreate it. Its possible that we have also fixed the issue recently.

from readyset.