bloom filter request hits

[tarzanek]

Please make sure that this is a feature request.

System information

• Scylla version (you are using): 2025.1
• Are you willing to contribute it (Yes/No): No

Describe the feature and the current behavior/state.

right now we see number of requests hitting Scylla, but if Scylla doesn't reply with any data (e.g. with bloom filters) we don't see it
we do see cache hits and cache misses, so we can kinda assume the query was refused by bloom filter, but at the same time there are other CQL queries from driver which will skew this assumption, so if we can exactly know how much is served by bloom filters (resp. how effective they are)
we'd like to see how many of requests

Who will benefit with this feature?

scoping, sizing, benchmarking, use case testing
tuning corner cases where they have lots of misses (you can then adjust fp chance based on this)

Any Other info.

I have no clue if Scylla has such metric, most likely not, 2025.3.1 list of metrics with "bloom" in them:

# HELP scylla_database_clustering_filter_count Counts bloom filter invocations.
# HELP scylla_database_clustering_filter_fast_path_count Counts number of times bloom filtering short cut to include all sstables when only one full range was specified.
# HELP scylla_database_clustering_filter_sstables_checked Counts sstables checked after applying the bloom filter. High value indicates that bloom filter is not very efficient.
# HELP scylla_database_clustering_filter_surviving_sstables Counts sstables that survived the clustering key filtering. High value indicates that bloom filter is not very efficient and still have to access a lot of sstables to get data.
# HELP scylla_sstables_bloom_filter_memory_size Bloom filter memory usage in bytes.
# TYPE scylla_sstables_bloom_filter_memory_size gauge

so not sure if above (or any other without bloom in its name) can be used

[tarzanek]

I know I can check nodetool cfstats, but I'd like to see it for whole cluster, per node at least (or by table too? though that might be too much)
Can we perhaps reuse what cfstats print ?

[mykaul]

I don't see what benefit will the above metrics give being in some dashboard. What will you be doing with them? Tweak the BF parameters?

[tarzanek]

yes, tweaking BF and also seeing how effective client queries are

[mykaul]

yes, tweaking BF and also seeing how effective client queries are

I'm not sure it's a good enough reason to add more panels in our overcrowded dashboards.

[mykaul]

(there's also an implicit assumption here that data fetch patterns don't change over time, etc.)

[tarzanek]

Aerospike has this, so that's main motivation, to compare to aerospike and see if we can handle similar workload, it'd be good to have these