[jira] [Created] (IGNITE-10642) Cache metrics distribution mechanism should be changed from broadcast to request-response communication pattern

["Aleksey Plekhanov (JIRA)" <ji...@apache.org>]

Aleksey Plekhanov created IGNITE-10642:
------------------------------------------

             Summary: Cache metrics distribution mechanism should be changed from broadcast to request-response communication pattern
                 Key: IGNITE-10642
                 URL: https://issues.apache.org/jira/browse/IGNITE-10642
             Project: Ignite
          Issue Type: Improvement
    Affects Versions: 2.7
            Reporter: Aleksey Plekhanov


In the current implementation, all cache metrics are collected on each node for all caches and sent across the whole cluster with discovery message ({{TcpDiscoveryMetricsUpdateMessage}}) with configured frequency (MetricsUpdateFrequency, 2 seconds by default) even if no one requested them. 

This mechanism should be changed in the following ways:
* Local cache metrics should be available (if configured) on each node
* If a node needs to collect data from the cluster, it sends explicit
request over communication SPI (request should contain a limited set of caches and/or metrics)
* For performance reasons collected cluster-wide values must be cached. Previously
collected metrics should be returned without re-requesting them again if they are not too old
(configurable)
* The mechanism should be easily adaptable for other types of statistics, which probably needs to be shared between nodes in the future (IO statistics, SQL statistics, SQL execution history, etc)
* Message format should be carefully designed to minimize message size (cluster can contain thousands of caches and hundreds of nodes)
* There must be an opportunity to configure metrics in runtime



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)