This is a separate issue for the third item in #65954.
sqlNodeCPUOverloadIndicator is the implementation of the cpuOverloadIndicator interface for a single-tenant SQL node in a multi-tenant cluster. This has to rely on the periodic load information from the cpu scheduler and will therefore be tuned towards indicating overload at higher overload points (otherwise we could fluctuate into underloaded territory due to restricting admission, and not be work conserving). Such tuning towards more overload, and therefore more queueing inside the scheduler, is somewhat acceptable since a SQL node is not multi-tenant.
This issue covers the work to implement it, since isOverloaded() currently just returns false.
Epic CRDB-13762
Jira issue: CRDB-14838
This is a separate issue for the third item in #65954.
sqlNodeCPUOverloadIndicatoris the implementation of thecpuOverloadIndicatorinterface for a single-tenant SQL node in a multi-tenant cluster. This has to rely on the periodic load information from the cpu scheduler and will therefore be tuned towards indicating overload at higher overload points (otherwise we could fluctuate into underloaded territory due to restricting admission, and not be work conserving). Such tuning towards more overload, and therefore more queueing inside the scheduler, is somewhat acceptable since a SQL node is not multi-tenant.This issue covers the work to implement it, since
isOverloaded()currently just returnsfalse.Epic CRDB-13762
Jira issue: CRDB-14838