kv: add observability for number of accurate- vs. approximate-stats splits #119516
Labels
A-kv
Anything in KV that doesn't belong in a more specific category.
C-enhancement
Solution expected to add code/behavior + preserve backward-compat (pg compat issues are exception)
T-kv
KV Team
Projects
As a followup to #119499, we should add some metrics around the stats estimates produced by the split.
kv.split.post-pre-split-diff-<stat>
: for eachstat
in the MVCCStats, the difference between the value of the stat at the time of the split (in splitTrigger) and before the split (in AdminSplit). Not sure if this will be useful/needed for all stats; maybe export some as metrics and log the rest. These will help keep track of drift in specific metrics and allow us to correlate that with other instabilities during investigations.kv.split.fallback-accurate-stats
: the number of splits that fall back to accurate splits computation because of a large discrepancy in the stats (batcheval: use a threshold on stats discrepancy to fall back to accurate-stats split #119503).We already log the delta that the consistency checker queue corrects. We expect these deltas to be larger after #119499.
Jira issue: CRDB-36263
Epic CRDB-34215
The text was updated successfully, but these errors were encountered: