[jboss-jira] [JBoss JIRA] (JGRP-1858) TP: update the timestamp of an entry in logical_addr_cache when it is removed

[Bela Ban (JIRA)]

     [ https://issues.jboss.org/browse/JGRP-1858?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Bela Ban resolved JGRP-1858.
----------------------------

    Resolution: Done


> TP: update the timestamp of an entry in logical_addr_cache when it is removed
> -----------------------------------------------------------------------------
>
>                 Key: JGRP-1858
>                 URL: https://issues.jboss.org/browse/JGRP-1858
>             Project: JGroups
>          Issue Type: Bug
>      Security Level: Public(Everyone can see) 
>            Reporter: Bela Ban
>            Assignee: Bela Ban
>             Fix For: 3.4.5, 3.5
>
>
> {{TP.logical_addr_cache}} maintains mappings between UUIDs and IP addresses. When an address is removed from it, it is not really removed but only marked as 'removable'.
> When the cache exceeds its capacity, all removable values are removed.
> However, we also have a cache reaper, which periodically (every {{logical_addr_cache_expiration}} ms) removes all values marked as removable. The default is 2 minutes. This means that the reaper runs every 2 minutes and drops removable entries that are older than 2 minutes.
> If we have an entry for member P which was added 2 hours ago, and now member P leaves, then that entry is marked as removable. If the reaper happens to kick in just after P left, P's entry will be removed *immediately* !
> SOLUTION:
> * Update the timestamp of an entry when it is marked from normal to removable (don't mark it twice !)
> * Allow for 0 as a valid reaper interval (disabling reaping)
> * Introduce a {{reaper_interval}} (when the reaper runs) and {{logical_addr_cache_expiration}} (already exists)
> ** The former has to be smaller than the latter



--
This message was sent by Atlassian JIRA
(v6.2.6#6264)