SmartStore Eviction Failure

[dersonje2]

Hello,

I am experiencing a periodic issue with smartstore where a bucket will try to be evicted then proceeds to fail and does that cycle thousands of times. The indexer IO is fine, the bucket is warm, we have enough cache sizing, and I have not been able to correlate any cache logs with when these failures begin on multiple indexer nodes in the cluster (~33% of indexers).

2 questions:
* What is an urgent mode eviction?
* What can cause warm buckets to be unable to be evicted when they rolled to warm ~a full day earlier?

[richgalloway]

A warm bucket will not be evicted if it is too new on the premise that new data is more likely to be searched than old data.  "new" is defined by hotlist_recency_secs  and hotlist_bloom_filter_recency_hours  in indexes.conf.

Urgent mode eviction comes into play when there are not enough files eligible for normal eviction.  In urgent mode, the hotlist_recency_secs and hotlist_bloom_filter_recency_hours settings are ignored.