storm
storm copied to clipboard
apache
[STORM-3759] Additional Trident Kafka Spout Metrics
The Trident Kafka Spout/Emitter does not seem to measure a couple metrics that my team has found useful over the years using the spout.
Event emit rate
Having a count of how many events are being emitted by the spout can track the rate at which the spout is consuming from the kafka topic(s). If this rate decreases, it signals performance issues to investigate. This metric has helped my team track the overall health of our Kafka spouts.
Kafka Spout Max Lag
This gauge value monitors the backlog of events. It helps determine if:
- a. the spout is not emitting fast enough to keep up with the incoming data, or
- b. the spout is current and Kafka topic delayed.
This metric can be used to distinguish between these two cases and determine the source of the slowdown. It has helped my team better understand delays. The non-Trident variant of the Kafka spout seems to have something similar.
Originally reported by jmpoholarz, imported from: Additional Trident Kafka Spout Metrics
- status: Open
- priority: Major
- resolution: Unresolved
- imported: 2025-01-24
Related PR: https://github.com/apache/storm/pull/3385
Ethan pointed out there is a builtin __emit-count metric which might be sufficient for the Event emit rate metric proposed.
Is the builtin emitted metric not sufficient for this purpose? https://github.com/apache/storm/blob/master/docs/Metrics.md#__emit-count
However, is spout/stream renaming supported for Trident topologies? For our trident storm topologies, we get metrics with names like "__emit-count-s87"; we have about 200 of them in our metric list which is too many to distinguish without proper names.
The TridentTopologyBuilder seems like it might be assigning names using only ID numbers, but I am not certain about this. https://github.com/apache/storm/blob/7bef73a6faa14558ef254efe74cbe4bfef81c2e2/storm-client/src/jvm/org/apache/storm/trident/topology/TridentTopologyBuilder.java#L142-L143