lucene-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "David Smiley (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (SOLR-9824) Documents indexed in bulk are replicated using too many HTTP requests
Date Fri, 09 Dec 2016 06:01:58 GMT

    [ https://issues.apache.org/jira/browse/SOLR-9824?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15734440#comment-15734440
] 

David Smiley commented on SOLR-9824:
------------------------------------

I like the idea of removing lastDoc; it's hard to work with.

I'm not sure why we even need a longLastingThreads flag.  Can't we just ensure that blockUntilFinished()
triggers the runners to interrupt their poll's always?  You're doing this now in the patch
(albeit only when longLastingThreads==true) but why not simply always?  And why are Future's
needed to do the cancel() when we can interrupt the Threads directly? We could expose them
easily by having the Runner store it's current thread when run() is called.  If we didn't
need a "longLastingThreads" boolean then the client could set the poll time independently,
perhaps defaulting to '0' but might set it to be very long if it intends to call blockUntilFinished().
 Arguably, blockUntilFinished() might log a warning if the poll time is zero because it would
amount to misconfiguration.

It may be safer to ensure that interrupt() *only* affects the queue.poll calls and not anything
else; but I'm unsure if anything else internal to writing the document to the stream would
interrupt/cancel part way to warrant caring.  Do you know?  It's do-able but would require
some extra boolean volatile state variables like doStop and currentlyPolling.

I'm confused about something: the first line of sendUpdateStream() is {{while(!queue.isEmpty)}}
but why even do that given that given we poll the queue?  i.e. why not {{while(true)}}?  Or
perhaps why even loop at all given the caller has a similar loop.

> Documents indexed in bulk are replicated using too many HTTP requests
> ---------------------------------------------------------------------
>
>                 Key: SOLR-9824
>                 URL: https://issues.apache.org/jira/browse/SOLR-9824
>             Project: Solr
>          Issue Type: Improvement
>      Security Level: Public(Default Security Level. Issues are Public) 
>          Components: SolrCloud
>    Affects Versions: 6.3
>            Reporter: David Smiley
>         Attachments: SOLR-9824.patch, SOLR-9824.patch, SOLR-9824.patch
>
>
> This takes awhile to explain; bear with me. While working on bulk indexing small documents,
I looked at the logs of my SolrCloud nodes.  I noticed that shards would see an /update log
message every ~6ms which is *way* too much.  These are requests from one shard (that isn't
a leader/replica for these docs but the recipient from my client) to the target shard leader
(no additional replicas).  One might ask why I'm not sending docs to the right shard in the
first place; I have a reason but it's besides the point -- there's a real Solr perf problem
here and this probably applies equally to replicationFactor>1 situations too.  I could
turn off the logs but that would hide useful stuff, and it's disconcerting to me that so many
short-lived HTTP requests are happening, somehow at the bequest of DistributedUpdateProcessor.
 After lots of analysis and debugging and hair pulling, I finally figured it out.  
> In SOLR-7333 ([~tpot]) introduced an optimization called {{UpdateRequest.isLastDocInBatch()}}
in which ConcurrentUpdateSolrClient will poll with a '0' timeout to the internal queue, so
that it can close the connection without it hanging around any longer than needed.  This part
makes sense to me.  Currently the only spot that has the smarts to set this flag is {{JavaBinUpdateRequestCodec.unmarshal.readOuterMostDocIterator()}}
at the last document.  So if a shard received docs in a javabin stream (but not other formats)
one would expect the _last_ document to have this flag.  There's even a test.  Docs without
this flag get the default poll time; for javabin it's 25ms.  Okay.
> I _suspect_ that if someone used CloudSolrClient or HttpSolrClient to send javabin data
in a batch, the intended efficiencies of SOLR-7333 would apply.  I didn't try. In my case,
I'm using ConcurrentUpdateSolrClient (and BTW DistributedUpdateProcessor uses CUSC too). 
CUSC uses the RequestWriter (defaulting to javabin) to send each document separately without
any leading marker or trailing marker.  For the XML format by comparison, there is a leading
and trailing marker (<stream> ... </stream>).  Since there's no outer container
for the javabin unmarshalling to detect the last document, it marks _every_ document as {{req.lastDocInBatch()}}!
 Ouch!



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org


Mime
View raw message