[
https://issues.apache.org/jira/browse/SOLR-3666?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16279001#comment-16279001
]
Shawn Heisey commented on SOLR-3666:
------------------------------------
This problem still comes up on the mailing list, five years after the issue was
created.
Fixing this problem would require that when an import is started, that the
handler name is written somewhere in zookeeper under the znode for the
collection, along with the host/port/context of the core where the import was
started. That way, when a status request comes in, Solr will know where the
import is actually happening, and can direct the status command to the same
place.
One potential problem I thought of: What should Solr do if the
host/port/context that has been written into ZK stops responding? If we remove
the info we wrote to ZK when that happens, but the problem is transitory in
nature and the import didn't actually stop, then further status requests after
the transitory problem fixes itself might not go to the core that is actually
running the import, so the user may see an idle status for an import that
actually still is running.
Another wrinkle: If there is no info in ZK for the handler name, should a
"status" command write the info to ZK, or should that be reserved to commands
that start imports?
Something else to think about: Should the info pointing further requests to the
same core have an expiration? Some imports can take *days*, and there might
even be some that take longer. If there is an expiration after which the znode
is removed, it would be absolutely mandatory for the time interval to be
configurable, in several time units.
> DataImportHandler status command in SolrCloud does not work properly
> ---------------------------------------------------------------------
>
> Key: SOLR-3666
> URL: https://issues.apache.org/jira/browse/SOLR-3666
> Project: Solr
> Issue Type: Bug
> Components: contrib - DataImportHandler, SolrCloud
> Affects Versions: 4.0-ALPHA
> Reporter: Sauvik Sarkar
>
> The dataimport?command=status command does not work correctly when invoked on
> the node not running the DIH in a SolrCloud configuration.
> The expectation is that no matter which node is importing any other node
> should be able get the import status information.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
