[
https://issues.apache.org/jira/browse/KUDU-1539?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Todd Lipcon updated KUDU-1539:
------------------------------
Component/s: master
> Read/Write operations using new schema on recently altered tables may fail
> --------------------------------------------------------------------------
>
> Key: KUDU-1539
> URL: https://issues.apache.org/jira/browse/KUDU-1539
> Project: Kudu
> Issue Type: Bug
> Components: master
> Reporter: Dan Burkert
>
> Currently the master will transition from state ALTERING to RUNNING when a
> single replica from every tablet heartbeats with the new schema. This can
> cause clients who issue the alter and wait for the alter to be done (the
> master to transitioning the table back to RUNNING) to subsequently fail reads
> and writes, since an arbitrary replica (even the leader) is not guaranteed to
> have committed the schema alteration (only a majority is necessary).
> We could make the situation better by only considering the leader heartbeat
> when determining whether to go from ALTERING to RUNNING, but it doesn't close
> the hole of reading/writing from a follower.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
