A number of times now we have needed to diagnose the source of certain operation metrics based on database. Our existing tools to discern where metrics are being incremented are not sufficient to adequately diagnose Serverless issues from the field.

One enhancement we can consider is using eBPF to trace operation metric calls. However, eBPF support needs fleshing out first.

For now, I think it would be useful to have specific debug log lines near the calls to operation metric increments that could give information about the provenance of the call (thread id and some other local info that might be helpful). These log lines could be turned on and off at runtime for brief periods to help diagnose issues in the field.