-
Type: Bug
-
Resolution: Done
-
Priority: Critical - P2
-
None
-
Affects Version/s: 4.4.14, 5.0.8
-
Component/s: Index Maintenance, Replication, Text Search
-
Environment:MongoDB Atlas 4.4.14 M40
ReplicaSet(Primary-Secondary-Secondary)
-
ALL
Hello.
I am using the M40 cluster of MongoDB Atlas 4.4.14.
I'm inserting 1.5 million to 2 million documents into a new collection every hour, and I've confirmed that the Query Targeting: Scanned Objects / Returned Has Gone 1000 alarm occurs every hour in the secondary of the cluster.
I didn't understand this, so I did some tests
We have confirmed that the data change is caused by Atlas Search Index that exists in a different collection from the command operation collection.
I think the update should occur only for the data changes in the collection where Search Index exists (to prevent unnecessary secondary load and unnecessary alerting)
Currently, I use readPreference=secondary in the application for load balancing, but I think that the above operation could cover the colscan caused by lack of index.
In many enterprise areas using MongoDB Atlas, we believe that the readPreference=secondary option will be used for load balancing, which requires corrective action.
It took almost a month to investigate the cause of this phenomenon, so I would like you to seriously take action on the issue.
(We have also confirmed that this is the same for clusters of versions 4.4.14 and 5.0.8)
Thank you.
Have a nice day.