ignite-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Alexey Goncharuk (JIRA)" <j...@apache.org>
Subject [jira] [Created] (IGNITE-5593) Affinity change message leak on massive topology updates
Date Tue, 27 Jun 2017 08:18:00 GMT
Alexey Goncharuk created IGNITE-5593:
----------------------------------------

             Summary: Affinity change message leak on massive topology updates
                 Key: IGNITE-5593
                 URL: https://issues.apache.org/jira/browse/IGNITE-5593
             Project: Ignite
          Issue Type: Bug
          Components: cache
    Affects Versions: 1.7
            Reporter: Alexey Goncharuk
            Priority: Critical
             Fix For: 2.1


When late affinity assignment is enabled, we complete the exchange future with custom discovery
event. Since discovery topology events usually are much faster than exchange futures completion,
it is possible that a newly joined node can 'see' the affinity change messages that are related
to previous topology versions when this node even was not present in the topology.

When this message is received, an exchange future is created and this message is added to
discoEvts list. However, this future never completes on this node because init() is never
called. This means that this exchange future sits in the exchange set with the affinity change
message.

Since the number of topology changes (and, thus, messages) can be quite large, this leads
to excessive memory consumption on the starting node. I've observed ~3Gb of heap wasted on
one of the nodes when > 200 nodes were restarted.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Mime
View raw message