Solr connection/socket timeouts

["Ramkumar R. Aiyengar" <an...@gmail.com>]

Currently HttpShardHandlerFactory and DUH2 set socket and connection
timeouts to unlimited by default. Should there be some finite default
instead? With the current defaults, a machine crash for example will lead
to requests in progress hang and occupy threads till the process is shut
down, which doesn't seem desirable. These are all configurable, we could
set some default (60s for conn, 600s for socket?), let people know and have
them up the limits if they really need it..

Thoughts?

Re: Solr connection/socket timeouts

["Ramkumar R. Aiyengar" <an...@gmail.com>]

I agree we need conservative defaults. I thought 10 mins is reasonable.
Obviously there will be longer requests, but that's where I would have
thought notice in CHANGES.txt would hopefully be good enough, you are
already looking at expert cases in such cases. If some other number is more
reasonable, that's fine, just that infinity isn't reasonable and will
eventually bite people :-) Will raise a JIRA..
On 5 Aug 2014 15:19, "Mark Miller" <ma...@gmail.com> wrote:

> Part of the issue is picking the right numbers I think. Some of our calls
> can legit take a *long* time.
>
> I think timeouts become a little more important with SolrCloud over
> ‘legacy mode’ Solr.
>
> Anyway, IMO, at some point we should introduce large defaults. It’s been
> on my mind before. A good distributed system requires timeouts for proper
> long term “chugging along”.
>
> --
> Mark Miller
> about.me/markrmiller
>
> On August 5, 2014 at 7:46:36 AM, Ramkumar R. Aiyengar (
> andyetitmoves@gmail.com) wrote:
>
> Currently HttpShardHandlerFactory and DUH2 set socket and connection
> timeouts to unlimited by default. Should there be some finite default
> instead? With the current defaults, a machine crash for example will lead
> to requests in progress hang and occupy threads till the process is shut
> down, which doesn't seem desirable. These are all configurable, we could
> set some default (60s for conn, 600s for socket?), let people know and have
> them up the limits if they really need it..
>
> Thoughts?
>
>

Re: Solr connection/socket timeouts

[Mark Miller <ma...@gmail.com>]

Part of the issue is picking the right numbers I think. Some of our calls can legit take a *long* time. 

I think timeouts become a little more important with SolrCloud over ‘legacy mode’ Solr.

Anyway, IMO, at some point we should introduce large defaults. It’s been on my mind before. A good distributed system requires timeouts for proper long term “chugging along”.

-- 
Mark Miller
about.me/markrmiller

On August 5, 2014 at 7:46:36 AM, Ramkumar R. Aiyengar (andyetitmoves@gmail.com) wrote:

Currently HttpShardHandlerFactory and DUH2 set socket and connection timeouts to unlimited by default. Should there be some finite default instead? With the current defaults, a machine crash for example will lead to requests in progress hang and occupy threads till the process is shut down, which doesn't seem desirable. These are all configurable, we could set some default (60s for conn, 600s for socket?), let people know and have them up the limits if they really need it..

Thoughts?