You are viewing a plain text version of this content. The canonical link for it is here.
Posted to solr-user@lucene.apache.org by Kaan Erdener <ke...@rackspace.com> on 2007/03/06 20:49:13 UTC
improve performance after commit
hello,
I'm looking for some tips / suggestions around reducing the query
time for Solr after I've post'ed a commit request. My Lucene index
contains around 2,000,000 documents, and I have a job that
periodically removes artibrary documents from Lucene and replaces
them with fresh copies from a database. Whenever that cycle occurs, I
send a commit to Solr to expose the updates. The problem is that
immediately after the commit, a Solr query that previously took
5-20ms now takes 20-25 seconds. Ouch.
I know that commit can be expensive, although I don't know by how
much, or what I might do to mitigate the expense. I haven't much doc
around this topic. I've also tried different cache settings
(basically using high values for cache and auto-warm sizes) but that
doesn't seem to make much of a difference.
I'll keep investigating on my own, but if anyone has any suggestions
or additional info, I would greatly appreciate it.
thanks,
Kaan
Confidentiality Notice: This e-mail message (including any attached or
embedded documents) is intended for the exclusive and confidential use of the
individual or entity to which this message is addressed, and unless otherwise
expressly indicated, is confidential and privileged information of Rackspace
Managed Hosting. Any dissemination, distribution or copying of the enclosed
material is prohibited. If you receive this transmission in error, please
notify us immediately by e-mail at abuse@rackspace.com, and delete the
original message. Your cooperation is appreciated.
Re: improve performance after commit
Posted by Kaan Erdener <ke...@rackspace.com>.
On Mar 7, 2007, at 11:34 AM, Chris Hostetter wrote:
> : back in just now. Here's an example trying to warm using a sort on
> : field name "subject". I tried query of
> : "allMessageContent:trying;subject+asc" as well as
> : "allMessageContent:trying;subject" (without "+asc") - either way
>
> when expressing params in XML (either as init params for a request
> handler, or in a QuerySenderListener the params don't need to be URL
> escaped ... they just need to be XML escaped, try something like...
>
>
> <listener event="newSearcher" class="solr.QuerySenderListener">
> <arr name="queries">
> <lst>
> <str name="q">allMessageContent:test; subject asc</str>
> <str name="start">0</str>
> <str name="rows">10</str>
> </lst>
> </arr>
> </listener>
>
> -Hoss
Thanks to you and Ryan for that suggestion, that was indeed the
problem. Using a warming query of "allMessageContent:trying;subject
asc" (without my hand-escaped whitespace) worked great.
In the end, this is what I've got in my solrconfig.xml, and the
overall query performance is now consistently fast, even after
post'ing a commit message.
<listener event="newSearcher" class="solr.QuerySenderListener">
<arr name="queries">
<lst> <str name="q">text:trying;date asc</str> <str
name="start">0</str> <str name="rows">50</str> </lst>
<lst> <str name="q">text:trying;refId asc</str> <str
name="start">0</str> <str name="rows">50</str> </lst>
<lst> <str name="q">text:trying;subject asc</str> <str
name="start">0</str> <str name="rows">50</str> </lst>
<lst> <str name="q">text:trying;name asc</str> <str
name="start">0</str> <str name="rows">50</str> </lst>
</arr>
</listener>
<listener event="firstSearcher" class="solr.QuerySenderListener">
<arr name="queries">
<lst> <str name="q">text:trying;date asc</str> <str
name="start">0</str> <str name="rows">50</str> </lst>
<lst> <str name="q">text:trying;refId asc</str> <str
name="start">0</str> <str name="rows">50</str> </lst>
<lst> <str name="q">text:trying;subject asc</str> <str
name="start">0</str> <str name="rows">50</str> </lst>
<lst> <str name="q">text:trying;name asc</str> <str
name="start">0</str> <str name="rows">50</str> </lst>
</arr>
</listener>
Thanks again,
Kaan
Confidentiality Notice: This e-mail message (including any attached or
embedded documents) is intended for the exclusive and confidential use of the
individual or entity to which this message is addressed, and unless otherwise
expressly indicated, is confidential and privileged information of Rackspace
Managed Hosting. Any dissemination, distribution or copying of the enclosed
material is prohibited. If you receive this transmission in error, please
notify us immediately by e-mail at abuse@rackspace.com, and delete the
original message. Your cooperation is appreciated.
Re: improve performance after commit
Posted by Chris Hostetter <ho...@fucit.org>.
: back in just now. Here's an example trying to warm using a sort on
: field name "subject". I tried query of
: "allMessageContent:trying;subject+asc" as well as
: "allMessageContent:trying;subject" (without "+asc") - either way
when expressing params in XML (either as init params for a request
handler, or in a QuerySenderListener the params don't need to be URL
escaped ... they just need to be XML escaped, try something like...
<listener event="newSearcher" class="solr.QuerySenderListener">
<arr name="queries">
<lst>
<str name="q">allMessageContent:test; subject asc</str>
<str name="start">0</str>
<str name="rows">10</str>
</lst>
</arr>
</listener>
-Hoss
Re: improve performance after commit
Posted by Ryan McKinley <ry...@gmail.com>.
>
> <str name="q">allMessageContent:test;subject+asc</str>
>
there should be a space between subject and asc,
try: http://host/select?q=allMessageContent:test;subject%20asc
+ is supposed to become a space, but it looks like it is staying "+"
Re: improve performance after commit
Posted by Kaan Erdener <ke...@rackspace.com>.
On Mar 6, 2007, at 9:50 PM, Yonik Seeley wrote:
> On 3/6/07, Kaan Erdener <ke...@rackspace.com> wrote:
>> From what I can see in the logs, these are both invoked after the
>> commit. However, the query times after a commit are still slow
>> (around 20 seconds).
>
> Your warming script didn't do any sorts.
> Why don't you also show the part of the log with the slow query...
> that would make it much easier for people to help.
>
> -Yonik
Right, I had some initially, but Solr threw exceptions. I put them
back in just now. Here's an example trying to warm using a sort on
field name "subject". I tried query of
"allMessageContent:trying;subject+asc" as well as
"allMessageContent:trying;subject" (without "+asc") - either way
throws an exception. Both are shown below. The generated exception
isn't clear (not to me, anyway), and I didn't find any examples of
this elsewhere for reference. What's the correct way to specify a
sort and direction when setting up a listener?
thanks,
Kaan
<lst> <str name="q">allMessageContent:test;subject+asc</str> <str
name="start">0</str> <str name="rows">50</str> </lst>
Mar 6, 2007 11:47:27 PM org.apache.solr.core.SolrCore execute
INFO: rows=50&start=0&q=allMessageContent:test;subject%2Basc 0 4
Mar 6, 2007 11:47:27 PM org.apache.solr.core.QuerySenderListener
newSearcher
INFO: QuerySenderListener done.
Mar 6, 2007 11:47:27 PM org.apache.solr.core.SolrException log
SEVERE: java.lang.ArrayIndexOutOfBoundsException: 1
at org.apache.solr.search.QueryParsing.parseSort
(QueryParsing.java:189)
at
org.apache.solr.request.StandardRequestHandler.handleRequest
(StandardRequestHandler.java:115)
at org.apache.solr.core.SolrCore.execute(SolrCore.java:595)
at org.apache.solr.core.QuerySenderListener.newSearcher
(QuerySenderListener.java:51)
at org.apache.solr.core.SolrCore$3.call(SolrCore.java:451)
at java.util.concurrent.FutureTask$Sync.innerRun
(FutureTask.java:269)
at java.util.concurrent.FutureTask.run(FutureTask.java:123)
at java.util.concurrent.ThreadPoolExecutor$Worker.runTask
(ThreadPoolExecutor.java:650)
at java.util.concurrent.ThreadPoolExecutor$Worker.run
(ThreadPoolExecutor.java:675)
at java.lang.Thread.run(Thread.java:595)
<lst> <str name="q">allMessageContent:test;subject+asc</str> <str
name="start">0</str> <str name="rows">50</str> </lst>
Mar 6, 2007 11:51:57 PM org.apache.solr.core.SolrCore execute
INFO: rows=50&start=0&q=allMessageContent:test;subject 0 2
Mar 6, 2007 11:51:57 PM org.apache.solr.core.QuerySenderListener
newSearcher
INFO: QuerySenderListener done.
Mar 6, 2007 11:51:57 PM org.apache.solr.core.SolrException log
SEVERE: java.lang.ArrayIndexOutOfBoundsException: 1
at org.apache.solr.search.QueryParsing.parseSort
(QueryParsing.java:189)
at
org.apache.solr.request.StandardRequestHandler.handleRequest
(StandardRequestHandler.java:115)
at org.apache.solr.core.SolrCore.execute(SolrCore.java:595)
at org.apache.solr.core.QuerySenderListener.newSearcher
(QuerySenderListener.java:51)
at org.apache.solr.core.SolrCore$3.call(SolrCore.java:451)
at java.util.concurrent.FutureTask$Sync.innerRun
(FutureTask.java:269)
at java.util.concurrent.FutureTask.run(FutureTask.java:123)
at java.util.concurrent.ThreadPoolExecutor$Worker.runTask
(ThreadPoolExecutor.java:650)
at java.util.concurrent.ThreadPoolExecutor$Worker.run
(ThreadPoolExecutor.java:675)
at java.lang.Thread.run(Thread.java:595)
Confidentiality Notice: This e-mail message (including any attached or
embedded documents) is intended for the exclusive and confidential use of the
individual or entity to which this message is addressed, and unless otherwise
expressly indicated, is confidential and privileged information of Rackspace
Managed Hosting. Any dissemination, distribution or copying of the enclosed
material is prohibited. If you receive this transmission in error, please
notify us immediately by e-mail at abuse@rackspace.com, and delete the
original message. Your cooperation is appreciated.
Re: improve performance after commit
Posted by Yonik Seeley <yo...@apache.org>.
On 3/6/07, Kaan Erdener <ke...@rackspace.com> wrote:
> From what I can see in the logs, these are both invoked after the
> commit. However, the query times after a commit are still slow
> (around 20 seconds).
Your warming script didn't do any sorts.
Why don't you also show the part of the log with the slow query...
that would make it much easier for people to help.
-Yonik
Re: improve performance after commit
Posted by Kaan Erdener <ke...@rackspace.com>.
On Mar 6, 2007, at 1:55 PM, Yonik Seeley wrote:
> On 3/6/07, Kaan Erdener <ke...@rackspace.com> wrote:
>> I'm looking for some tips / suggestions around reducing the query
>> time for Solr after I've post'ed a commit request. My Lucene index
>> contains around 2,000,000 documents, and I have a job that
>> periodically removes artibrary documents from Lucene and replaces
>> them with fresh copies from a database. Whenever that cycle occurs, I
>> send a commit to Solr to expose the updates. The problem is that
>> immediately after the commit, a Solr query that previously took
>> 5-20ms now takes 20-25 seconds. Ouch.
>
> If this is a normal query (no faceting) then most likely the time
> is spent
> populating a lucene FieldCache entry used for sorting results.
> Put a static warming entry in solrconfig.xml that queries for a small
> number of documents and sorts that query by all the fields you
> commonly sort by.
>
> -Yonik
I'm not exactly sure this is what you meant, but I did some more
research and it looks close. I added the following to my solrconfig.xml:
<listener event="newSearcher" class="solr.QuerySenderListener">
<arr name="queries">
<lst> <str name="q">allMessageContent:test</str> <str
name="start">0</str> <str name="rows">10</str> </lst>
</arr>
</listener>
and also:
<listener event="firstSearcher" class="solr.QuerySenderListener">
<arr name="queries">
<lst> <str name="q">allMessageContent:trying</str> <str
name="start">0</str> <str name="rows">10</str> </lst>
</arr>
</listener>
From what I can see in the logs, these are both invoked after the
commit. However, the query times after a commit are still slow
(around 20 seconds). I'm guessing I didn't set up the warming
correctly? I had some sorting parameters in there, but the syntax was
wrong, produced errors on startup, so I took them out for now.
Mar 6, 2007 4:51:52 PM org.apache.solr.update.DirectUpdateHandler2
commit
INFO: end_commit_flush
Mar 6, 2007 4:51:52 PM org.apache.solr.search.SolrIndexSearcher warm
INFO: autowarming Searcher@4cd580 main from Searcher@7be8c2 main
documentCache
{lookups=10,hits=0,hitratio=0.00,inserts=20,evictions=0,size=20,cumulati
ve_lookups=120,cumulative_hits=68,cumulative_hitratio=0.56,cumulative_in
serts=52,cumulative_evictions=0}
Mar 6, 2007 4:51:52 PM org.apache.solr.search.SolrIndexSearcher warm
INFO: autowarming result for Searcher@4cd580 main
documentCache
{lookups=0,hits=0,hitratio=0.00,inserts=0,evictions=0,size=0,cumulative_
lookups=120,cumulative_hits=68,cumulative_hitratio=0.56,cumulative_inser
ts=52,cumulative_evictions=0}
Mar 6, 2007 4:51:52 PM org.apache.solr.core.QuerySenderListener
newSearcher
INFO: QuerySenderListener sending requests to Searcher@4cd580 main
Mar 6, 2007 4:51:52 PM org.apache.solr.core.SolrCore execute
INFO: rows=10&start=0&q=allMessageContent:trying 0 410
Mar 6, 2007 4:51:52 PM org.apache.solr.core.QuerySenderListener
newSearcher
Confidentiality Notice: This e-mail message (including any attached or
embedded documents) is intended for the exclusive and confidential use of the
individual or entity to which this message is addressed, and unless otherwise
expressly indicated, is confidential and privileged information of Rackspace
Managed Hosting. Any dissemination, distribution or copying of the enclosed
material is prohibited. If you receive this transmission in error, please
notify us immediately by e-mail at abuse@rackspace.com, and delete the
original message. Your cooperation is appreciated.
Re: improve performance after commit
Posted by Yonik Seeley <yo...@apache.org>.
On 3/6/07, Kaan Erdener <ke...@rackspace.com> wrote:
> I'm looking for some tips / suggestions around reducing the query
> time for Solr after I've post'ed a commit request. My Lucene index
> contains around 2,000,000 documents, and I have a job that
> periodically removes artibrary documents from Lucene and replaces
> them with fresh copies from a database. Whenever that cycle occurs, I
> send a commit to Solr to expose the updates. The problem is that
> immediately after the commit, a Solr query that previously took
> 5-20ms now takes 20-25 seconds. Ouch.
If this is a normal query (no faceting) then most likely the time is spent
populating a lucene FieldCache entry used for sorting results.
Put a static warming entry in solrconfig.xml that queries for a small
number of documents and sorts that query by all the fields you
commonly sort by.
-Yonik