GitHub user yatriks added a comment to the discussion: Memory leak and 
confusing logs

@git-hulk yes, the second test was with the `max_compaction_bytes` as well.

I left the load running for a few more days, and had another memory-related oom 
kill (with both kvrocks and docker reported memory going beyond the limit when 
kill happens).

I think I've realized one other thing, since I'm trying to use 'realistic' load 
testing, I set up very large batches of inserts, but when the batches run out 
after ~4 days or so, the select loads run unimpeded, hammering the server and 
potentially triggering the large memory usage.

I'm now going to run two tests:
1. intensive select workload only (large sunions), see if that causes the 
memory spike to happen any faster / at the time of a compaction, and after that,
2. try disabling auto compactions and seeing if that changes anything.

Thanks for your guidance. I was surprised to see this last test also run into a 
memory spike, and operationally I'm sure this kind of workload would only 
happen in bursts at first but I would like to continue investigating why it 
happens at all. Appreciate any additional thoughts and insights :pray: 

GitHub link: 
https://github.com/apache/kvrocks/discussions/2984#discussioncomment-13273848

----
This is an automatically sent email for [email protected].
To unsubscribe, please send an email to: [email protected]

Reply via email to