- Make it in a separate goroutine, so it doesn't slow down regular intern() calls.
- Do not lock internStringMap.mutableLock during the cleanup routine, since now
it is called from a single goroutine and reads only the readonly part of the internStringMap.
This should prevent from locking regular intern() calls for new strings during cleanups.
- Add jitter to the cleanup interval in order to prevent from synchornous increase in resource usage
during cleanups.
- Run the cleanup twice per -internStringCacheExpireDuration . This should save 30% CPU time spent
on cleanup comparing to the previous code, which was running the cleanup 3 times per -internStringCacheExpireDuration .
This should reduce GC overhead when tens of millions of strings are interned (for example, during stream deduplication
of millions of active time series).
The limit has been increased from 300 bytes to 500 bytes according to the collected production stats.
This allows reducing CPU usage without significant increase of RAM usage in most practical cases.
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3692
Allow users fine-tuning the maximum string length for interning via -internStringMaxLen command-line flag.
This may be used for fine-tuning RAM vs CPU usage for certain workloads.
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3692
Previously only up to 100K results were cached.
This could result in sub-optimal performance when more than 100K unique strings were actually used.
For example, when the relabeling rule was applied to a million of unique Graphite metric names
like in the https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3466
This commit should reduce the long-term CPU usage for https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3466
after all the unique Graphite metrics are registered in the FastStringMatcher.Transform() cache.
It is expected that the number of unique strings, which are passed to FastStringMatcher.Match(),
FastStringTransformer.Transform() and to InternString() during the last 5 minutes,
is limited, so the function results fit memory. Otherwise OOM crash can occur.
This should be the case for typical production workloads.
