Alexey Kukushkin created IGNITE-16448:
-----------------------------------------
Summary: Optimize unacknowledged messages queue for FULL_SYNC
caches
Key: IGNITE-16448
URL: https://issues.apache.org/jira/browse/IGNITE-16448
Project: Ignite
Issue Type: Improvement
Affects Versions: 2.12
Reporter: Alexey Kukushkin
Attachments: apache-ignite-msgReqs-heap-usage.png
Apache Ignite queues all unacknowledged messages in the
{{GridNioRecoveryDescriptor#msgReqs}} queue. Size of this queue is effectively
controlled by the {{TcpCommunicationSpi.ackSendThreshold}} property, which is
32 by default.
Queuing PUTs of large entries results in significant heap memory usage.
Please check if we really need to queue PUTs into FULL_SYNC caches. It seems to
me we should not since a failure to write to a backup would be sent back to the
client right away without any retries. If that is the case then please optimize
the heap usage by not queueing FULL_SYNC cache update messages.
See the heap usage example attached: !apache-ignite-msgReqs-heap-usage.png!
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
