Maciej Moscicki
Maciej Moscicki
Enable 2 kafka producer metrics: - metadata-age - so we can monitor stale cluster metadata - record-queue-time-max - how much time do batches spend in send buffers
Test that reproduces the bug: https://github.com/allegro/hermes/commit/b9773ce4bdb0c7ddf5b1edb630e5bd658ec897e9 ``` org.opentest4j.AssertionFailedError: expected: "1.20" but was: "x" ```
Test dependency https://github.com/allegro/embedded-elasticsearch is no longer maintained. We should migrate to [testcontainers](https://www.testcontainers.org/modules/elasticsearch/) instead.
Bump kafka clients in order to make use of [Strictly Uniform Sticky Partitioner](https://cwiki.apache.org/confluence/display/KAFKA/KIP-794%3A+Strictly+Uniform+Sticky+Partitioner). Rewrite / remove tests which depend on previous partitioner behaviour i.e. round robin.
1. Split fail fast producer config to local / remote 2. Expose kafka send_total metric which can be used to calculate true record send rate (with retries)
Remove dropwizard (and graphite) support from hermes and hermes-client. Metrics are now only reported using micrometer and are exposed in prometheus format. The only place where dropwizard is currently used...