abij opened a new pull request, #11623:
URL: https://github.com/apache/nifi/pull/11623
### Summary
The Prometheus flow metrics endpoint (`/nifi-api/flow/metrics/prometheus`)
exposes `cluster_connected_node_count` with a `connected_nodes` label
containing a human-readable ratio string, e.g. `connected_nodes="3 / 4"`.
Because `ClusterMetricsRegistry` is a long-lived singleton on
`StandardNiFiServiceFacade` that is never cleared, and this label's value
changes every time cluster membership changes, every distinct ratio the cluster
has ever reported over the process's lifetime remains permanently exposed as a
separate, frozen time series alongside the current one — e.g. a fully-connected
4-node cluster can still show a stale `connected_nodes="3 / 4"` series from an
earlier transient disconnect, in addition to the correct `connected_nodes="4 /
4"` series. This makes the metric unreliable for dashboards/alerts without
manually filtering out label combinations that can't be distinguished from
current ones.
This is the same class of bug fixed twice before in this file for other
registries (NIFI-8272, NIFI-11899), both times by no longer using a
churning/derived value as a metric label.
### What this PR does
Removes `connected_nodes` from the gauge's label set entirely, since it's
purely a cosmetic rendering of `connectedNodeCount`/`totalNodeCount`, both of
which are already exposed numerically (`cluster_connected_node_count`'s own
value, and the separate `cluster_total_node_count` gauge). This brings
`CONNECTED_NODE_COUNT` in line with its sibling gauges in the same registry
(`cluster_is_clustered`, `cluster_is_connected_to_cluster`,
`cluster_total_node_count`), none of which carry a churning label.
- `ClusterMetricsRegistry`: `.labelNames("instance", "connected_nodes")` →
`.labelNames("instance")`
- `PrometheusMetricsUtil#createClusterMetrics`: drop the now-unused
`connectedNodes` parameter
- `StandardNiFiServiceFacade#populateFlowMetrics`: drop the now-unused
`connectedNodesLabel` string assembly
- `TestFlowResource`: update the corresponding test fixture call
No information is lost — clients relying on the ratio can compute it from
the two existing numeric gauges instead. This is technically a breaking change
to the metric's label surface for anyone rendering the `connected_nodes` label
text directly (e.g. a Grafana table column), documented here for visibility.
### Tests
- `./mvnw -pl nifi-framework-bundle/nifi-framework/nifi-web/nifi-web-api -am
install` — build succeeds
- `./mvnw -pl nifi-framework-bundle/nifi-framework/nifi-web/nifi-web-api
test -Dtest=TestFlowResource` — 22/22 tests pass
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
