-
Notifications
You must be signed in to change notification settings - Fork 1.2k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
bloom-filter: default value for components_memory_reclaim_threshold is too strict #18607
Comments
…fault Incremented the components_memory_reclaim_threshold config's default value to 0.8 as the previous value was too strict and caused unnecessary eviction in otherwise healthy clusters. Fixes scylladb#18607 Signed-off-by: Lakshmi Narayanan Sreethar <[email protected]>
According to what @avikivity wrote before the OOM is "imminent" when it reaches 0.5. BF amount can contribute but this is not the root cause - it's just a trigger together with many other possible triggers for the OOM. IMO we should not enable this feature by default just like we don't enable compaction I/O bandwidth throttling by default. Instead we should invest in the actual fixing of the problem:
fyi @denesb |
components_memory_reclaim_threshold
controls when we start evicting bloom filters from memory. It defaults to 0.1, which comes down to 10% of the shard's memory. It looks like this is too strict and causes unnecessary bloom filter eviction in otherwise healthy clusters. One cluster which was impacted by this, had a BF/memory ratio of 0.6. The eviction mechanism was introduced to avert OOM disasters, not to enforce lean bloom filters and so the default should be adjusted accordingly and increased to as high as 0.8 or even maybe 0.9. The idea is that by default it should only kick in when OOM is iminent, it should not intervene on clusters which have high BF memory consumption but are otherwise fine.The text was updated successfully, but these errors were encountered: