-
Type:
Improvement
-
Resolution: Fixed
-
Priority:
Major - P3
-
Affects Version/s: None
-
Component/s: None
-
None
-
Workload Resilience
-
Fully Compatible
-
200
-
None
-
None
-
None
-
None
-
None
-
None
-
None
WiredTiger will conscript application threads into eviction once a workload exceeds the dirty fill threshold. If the workload continues to add pressure under these circumstances we often see eviction stalls, which present to the user as total unavailability (all user threads are held hostage until the storage engine can comfortably evict its cache). The answer here is to slow down writes in response to untenable cache pressure, possibly rejecting the writes if we exceed a maximum queue depth.