Collect histogram metrics for L1/L2 sub-operations
ScottMansfield opened this issue · comments
Currently the only timing information captured is around the whole operation for each command. This means that a get that hits L1 is mixed in with the gets that miss in L1, hit L2, and set in L1. This whole series of operations takes much longer, which means that the data is a bit muddy. We should capture timing metrics for the individual operations at the orchestrator level such that L1 gets and L2 gets can be individually observed.
This will take a few extra megabytes but ¯_(ツ)_/¯
under review in #90