Vladimir Steshin created IGNITE-26708:
-----------------------------------------
Summary: Revise TcpDiscoveryMetricsUpdateMessage
Key: IGNITE-26708
URL: https://issues.apache.org/jira/browse/IGNITE-26708
Project: Ignite
Issue Type: Improvement
Reporter: Vladimir Steshin
We have _TcpDiscoveryMetricsUpdateMessage_ and related
{_}EVT_NODE_METRICS_UPDATED{_}. And also similar _ClusterMetricsUpdateMessage_
in {_}Communication{_}. Both of them send the metrics. If discovery is not of
_TPC_ type (ZooKeeper for instance), then the metrics are sent through
{_}Communication{_}:
{code:java}
ClusterProcessor:
sndMetrics = !(ctx.config().getDiscoverySpi() instanceof TcpDiscoverySpi);
if (sndMetrics) {
metricsProvider = ctx.discovery().createMetricsProvider();
long updateFreq = ctx.config().getMetricsUpdateFrequency();
ctx.timeout().addTimeoutObject(new MetricsUpdateTimeoutObject(updateFreq));
} {code}
Why not sending the metrics via _Communication_ only? Pros: fewer discovery
messages; no duplication of functionality, messages and serialization; more
lightweight discovery.
No need of delivery guarantee here.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
