-
Type: Task
-
Resolution: Incomplete
-
Priority: Major - P3
-
None
-
Affects Version/s: None
-
Component/s: None
-
None
In WT-5124, we clarified that the WT lookaside score cannot be used as a reliable cache pressure indicator because its value can sky-rocket while 1% of the cache is full.
I have heard tell that the lookaside table inserts metric also is an unreliable indicator of cache pressure because inserts can occur long before cache pressure rises high, and the number of inserts can decrease (become infrequent) while cache pressure remains high. I had originally thought to use frequency of inserts as a cache pressure indicator.
1. Could someone confirm that the lookaside table insert metric is not a satisfactory indicator of WT cache pressure?
2. Additionally, could someone explore whether there are any reliable metrics MongoDB could use?
milkie, if no metric can be found, the only alternative I can think of is for MongoDB to maintain a statically sized time window between stable_timestamp and oldest_timestamp, to support sharded transaction PIT reads. It would not react to cache pressure changes. Set a default window size, and allow users to modify the value via a run-time adjustable server parameter.
- is depended on by
-
SERVER-46435 Alternatives to using the lookaside-score for snapshot (history) window control [v4.2 only]
- Closed
- related to
-
WT-5124 Investigate why the WT lookaside score can be high when 1% of the cache is in use and no evictions occur
- Closed