Some customers experience a stall when the cache gets 95% full even though the cache is mostly clean and content evictable. On investigating it is apparent that the eviction has not kept up with the data being pulled into the WiredTiger cache and application threads are then forced to evict when the cache fills up to 95%. Eviction server is deemed to be the main culprit, unable to fill the eviction queues fast enough.
The objective of this ticket is to create a workload to simulate the same effect in WiredTiger standalone testing. Once we have such a workload, we would be in a better position to work on improvements to eviction under such scenarios.
- causes
-
WT-4967 Fix Jenkins perf configurations to adjust for recent wtperf scan changes
- Closed
- is related to
-
SERVER-40433 Mongodump extremely slow on 4.0.8
- Closed