[
https://issues.apache.org/jira/browse/KUDU-1997?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15998943#comment-15998943
]
Todd Lipcon commented on KUDU-1997:
-----------------------------------
I'm not quite following what this JIRA is proposing. As the description states,
edge-triggered actions are part of the design of the master, and are explicitly
meant to be that way for simplicity and scalability. On failover or restart,
the new leader master should be re-triggering, but only at that point, rather
than having to rely on some periodic background scan
> Catalog manager tasks are edge-triggered
> ----------------------------------------
>
> Key: KUDU-1997
> URL: https://issues.apache.org/jira/browse/KUDU-1997
> Project: Kudu
> Issue Type: Bug
> Components: master
> Affects Versions: 1.3.1
> Reporter: Mike Percy
>
> Catalog manager tasks are edge-triggered (based on a response to a tablet
> report) instead of level-triggered (based on periodic state polling).
> The upshot of this (original) design is that when there is a master failover,
> ongoing tasks will stop being retried.
> Examples of these tasks include things like adding a new replica, deleting an
> evicted replica, etc. See SendDeleteReplicaRequest() and
> SendAddServerRequest() in the code base for examples.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
