[
https://issues.apache.org/jira/browse/SOLR-7344?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14585242#comment-14585242
]
Yonik Seeley commented on SOLR-7344:
------------------------------------
bq. Yeah, but you are being obtuse about that. Where do you see the issue?
The only thing I can figure is that I had an underlying assumption that you
didn't. I assumed that at least *some* queues would be (essentially)
unbounded (like our 10K today) to solve all of the distributed deadlock
problems. But I didn't see a description of which ones would be unbounded.
Which ones you *need* to make unbounded also obviously affect the other goal -
limit concurrent requests to make better use of resources. Too many unknowns
make it hard to figure out where the problems are.
The reasoning behind thinking that at least one queue would need to be
unbounded is that the streaming API already has the ability to make unlimited
recursive requests. So it would seem that we either need at least one
unbounded queue, or barring that, an unbounded number of limited queues (i.e. a
request-depth number could be incremented by 1 for each level and a queue for
that level created on demand).
> Allow Jetty thread pool limits while still avoiding distributed deadlock.
> -------------------------------------------------------------------------
>
> Key: SOLR-7344
> URL: https://issues.apache.org/jira/browse/SOLR-7344
> Project: Solr
> Issue Type: Improvement
> Components: SolrCloud
> Reporter: Mark Miller
> Attachments: SOLR-7344.patch
>
>
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
