Github user revans2 commented on a diff in the pull request:
https://github.com/apache/storm/pull/2840#discussion_r218485913
--- Diff:
storm-server/src/main/java/org/apache/storm/daemon/nimbus/Nimbus.java ---
@@ -4767,4 +4797,205 @@ public IScheduler getForcedScheduler() {
}
+ private static class ClusterSummaryMetrics implements MetricSet {
+ private static final String SUMMARY = "summary";
+ private final Map<String, com.codahale.metrics.Metric> metrics =
new HashMap<>();
+
+ public com.codahale.metrics.Metric put(String key,
com.codahale.metrics.Metric value) {
+ return metrics.put(MetricRegistry.name(SUMMARY, key), value);
+ }
+
+ @Override
+ public Map<String, com.codahale.metrics.Metric> getMetrics() {
+ return metrics;
+ }
+ }
+
+ private class ClusterSummaryMetricSet implements Runnable {
+ private static final int CACHING_WINDOW = 5;
+
+ private final ClusterSummaryMetrics clusterSummaryMetrics = new
ClusterSummaryMetrics();
+
+ private final Function<String, Histogram> registerHistogram =
(name) -> {
+ //This histogram reflects the data distribution across only
one ClusterSummary, i.e.,
+ // data distribution across all entities of a type (e.g., data
from all nimbus/topologies) at one moment.
+ // Hence we use half of the CACHING_WINDOW time to ensure it
retains only data from the most recent update
+ final Histogram histogram = new Histogram(new
SlidingTimeWindowReservoir(CACHING_WINDOW / 2, TimeUnit.SECONDS));
+ clusterSummaryMetrics.put(name, histogram);
+ return histogram;
+ };
+ private volatile boolean active = false;
+
+ //NImbus metrics distribution
+ private final Histogram nimbusUptime =
registerHistogram.apply("nimbuses:uptime-secs");
+
+ //Supervisor metrics distribution
+ private final Histogram supervisorsUptime =
registerHistogram.apply("supervisors:uptime-secs");
+ private final Histogram supervisorsNumWorkers =
registerHistogram.apply("supervisors:num-workers");
+ private final Histogram supervisorsNumUsedWorkers =
registerHistogram.apply("supervisors:num-used-workers");
+ private final Histogram supervisorsUsedMem =
registerHistogram.apply("supervisors:used-mem");
+ private final Histogram supervisorsUsedCpu =
registerHistogram.apply("supervisors:used-cpu");
+ private final Histogram supervisorsFragmentedMem =
registerHistogram.apply("supervisors:fragmented-mem");
+ private final Histogram supervisorsFragmentedCpu =
registerHistogram.apply("supervisors:fragmented-cpu");
+
+ //Topology metrics distribution
+ private final Histogram topologiesNumTasks =
registerHistogram.apply("topologies:num-tasks");
+ private final Histogram topologiesNumExecutors =
registerHistogram.apply("topologies:num-executors");
+ private final Histogram topologiesNumWorker =
registerHistogram.apply("topologies:num-workers");
+ private final Histogram topologiesUptime =
registerHistogram.apply("topologies:uptime-secs");
+ private final Histogram topologiesReplicationCount =
registerHistogram.apply("topologies:replication-count");
+ private final Histogram topologiesRequestedMemOnHeap =
registerHistogram.apply("topologies:requested-mem-on-heap");
+ private final Histogram topologiesRequestedMemOffHeap =
registerHistogram.apply("topologies:requested-mem-off-heap");
+ private final Histogram topologiesRequestedCpu =
registerHistogram.apply("topologies:requested-cpu");
+ private final Histogram topologiesAssignedMemOnHeap =
registerHistogram.apply("topologies:assigned-mem-on-heap");
+ private final Histogram topologiesAssignedMemOffHeap =
registerHistogram.apply("topologies:assigned-mem-off-heap");
+ private final Histogram topologiesAssignedCpu =
registerHistogram.apply("topologies:assigned-cpu");
--- End diff --
I am sorry, but I am really having a hard time of seeing a histogram of
these being valuable, but perhaps I am wrong. I would much rather see
dimensions/tags being used, but since those are not available yet, I am okay
with this.
---
