In a recent support escalation, we received a debug zip taken by the system tenant in a serverless cluster. In the crdb_internal.cluster_contention_events.txt file we observed entries of the form:
When we dug into the logs of the debug zip we did indeed see contention on this key, but in a secondary tenants' keyspace i.e. /Tenant/123/54/2/"foo"/0. This seems to point to a bug in the interaction of contention events on keys in the secondary tenants keyspace when viewing from the system tenant. The cluster contention event keys should probably be tenant prefixed when looking at it from a point of view of the system tenant so as to make debugging easier.
In a recent support escalation, we received a debug zip taken by the system tenant in a serverless cluster. In the
crdb_internal.cluster_contention_events.txt
file we observed entries of the form:This translates to:
When we dug into the logs of the debug zip we did indeed see contention on this key, but in a secondary tenants' keyspace i.e.
/Tenant/123/54/2/"foo"/0
. This seems to point to a bug in the interaction of contention events on keys in the secondary tenants keyspace when viewing from the system tenant. The cluster contention event keys should probably be tenant prefixed when looking at it from a point of view of the system tenant so as to make debugging easier.Jira issue: CRDB-17121
Epic CRDB-20388