We recently saw an incident where followers were overloaded due to intent resolution triggered by the GC queue. @sumeerbhola pointed out:
Looking at mvccGCQueue.process , it uses gcAdmissionHeader and that uses kv.gc.admission_priority which is BulkNormalPri. So these should be getting classified as elastic. So we need to look at that code again to see why this did not behave as expected.
Describe the problem
We recently saw an incident where followers were overloaded due to intent resolution triggered by the GC queue. @sumeerbhola pointed out:
Looking at mvccGCQueue.process , it uses gcAdmissionHeader and that uses kv.gc.admission_priority which is BulkNormalPri. So these should be getting classified as elastic. So we need to look at that code again to see why this did not behave as expected.
Tracking issue to investigate this.
Jira issue: CRDB-44746
Epic CRDB-43580