[
https://issues.apache.org/jira/browse/IGNITE-4154?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15656992#comment-15656992
]
Semen Boikov edited comment on IGNITE-4154 at 11/11/16 12:48 PM:
-----------------------------------------------------------------
Done these optimizations as part of this issue:
Discovery history optimizations:
- remove discarded message for discovery pending messages
- remove duplicated data from TcpDiscoveryNodeAddedMessage.oldNodesDiscoData
- do not store unnecessary data in discovery EnsuredMessageHistory
- use special property for EnsuredMessageHistory size instead of
IGNITE_DISCOVERY_HISTORY_SIZE
Affinity history optimizations:
- do not store calculated primary/backup maps in history
- try save the same assignments instance for caches with similar affinity
Exchange messages optimizations:
- do not send duplicated partition state maps for caches with similar affinity
- use zip compression for data sent in exchange messages
was (Author: sboikov):
Done these optimizations as part of this issue:
Discovery history optimizations:
- remove discarded message for discovery pending messages
- remove duplicated data from TcpDiscoveryNodeAddedMessage.oldNodesDiscoData
- do not store unnecessary data in discovery EnsuredMessageHistory
- use special property for EnsuredMessageHistory size instead of
IGNITE_DISCOVERY_HISTORY_SIZE
Affinity history optimizations:
- do not store calculated primary/backup maps in history
- try save the same assignments instance for caches with similar affinity
Exchange messages optimizations:
- do not send duplicated partition state maps for caches with similar affinity
- use zip compression for data sent in exchange messages
> Optimize amount of data stored in discovery history
> ---------------------------------------------------
>
> Key: IGNITE-4154
> URL: https://issues.apache.org/jira/browse/IGNITE-4154
> Project: Ignite
> Issue Type: Task
> Components: general
> Reporter: Semen Boikov
> Assignee: Semen Boikov
> Fix For: 1.8
>
>
> Currently on large clusters with lots of caches amount of data stored in tcp
> discovery history can grow too much.
> What can be improved:
> - TcpDiscoveryNodeAddedMessage.oldNodesDiscoData contains per-node discovery
> data. For cache component essentially this is list of configured caches, and
> for joined node at the time when 'collectDiscoveryData' is called this list
> should be the same on all nodes, so it can be stored once instead of per-node
> map
> - even after message is discarded it is still stored in PendingMessages
> collection, looks like it can be safely removed after discarding
> - now system property IGNITE_DISCOVERY_HISTORY_SIZE controls both size of
> events needed for client reconnects and number of topology snapshots stored
> in GridDiscoveryManager, need have two properies
> - for clients TcpDiscoveryNodeAddFinishedMessage.clientDiscoData can be
> initialized only when message is sent to joining client
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
