Skip to content

PerfMonComps: Lock component data maps for each slot separately

This should achieve better scalability, although the original implementation was not really a bottleneck in practice. At some point we should perform a detailed profiling study of this service (especially for the full monitoring) to see if there are any major performance issues (such as cache invalidation between threads, lock contention in different workflows etc.). In production, we only use the fast monitoring functionality, so these changes shouldn't make any difference there.

Merge request reports