It would be very useful to know how often queues reach zero or very high capacity values. Today we only have sparse snapshots of the queue length, which can change very rapidly in miliseconds
Proposed Changes
Register the queue length as histogram instead of a gauge.
This PR registers the length metric for a single queue instead of for all queues. Reasons why:
We don't want to register the same value over and over. It would make the metric less useful and be quite wasteful
Using an enum ensures that we register metrics for all queues. With the current approach it's easy to forget to add metrics for new queues: this has happened when adding lightclient messages and with peerdas.
Metrics breaking changes
All metrics previously used to track queue lengths change. Also the type label for all metrics in the processor changes capitalization from gossip_block to GossipBlock. Is that okay? Does the IntoStaticStr macro support a different capitalization?
Issue Addressed
Second attempt at
It would be very useful to know how often queues reach zero or very high capacity values. Today we only have sparse snapshots of the queue length, which can change very rapidly in miliseconds
Proposed Changes
Register the queue length as histogram instead of a gauge.
This PR registers the length metric for a single queue instead of for all queues. Reasons why:
Metrics breaking changes
All metrics previously used to track queue lengths change. Also the type label for all metrics in the processor changes capitalization from
gossip_block
toGossipBlock
. Is that okay? Does theIntoStaticStr
macro support a different capitalization?