lucene-java-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Christopher Kolstad" <chri...@ovitas.no>
Subject Re: Best practice for updating an index when reindexing is not an option
Date Fri, 11 Jul 2008 11:34:58 GMT
>
> The only way to make this work with svn is if you can have svn perform a
> switch without doing any removal, then restart your IndexSearcher, then do a
> normal svn switch to remove the now unused files.  Does svn have an option
> to "switch but don't remove any removed files"?  Because IndexSearcher holds
> these files open, svn will not be able to remove the now unused files until
> IndexSearcher switches.

Exactly, I'll hear with the svn-user list to see if anyone has had any
experiece with a similar problem, but it looks like the svn switch command
automatically performs an svn up instantly after having switched to the new
URL. (I.e. it will remove the file)

Or ... for every update to your index you could plant 2 tags, the first one
> reflecting only added files, and the 2nd one reflecting removed files.
>  Then, with your IndexSearcher still running, do an svn switch to the first
> tag, then restart the IndexSearcher, then svn switch to the 2nd one.  I
> think that'd work?

That sounds like a rather good idea, SVN does have diff between two
different tags, so it should be possible to quickly figure out which files
are added and which are removed (or modified (in the case of the index I
guess all the files should be new)).

This should only be necessary on Windows ... UNIX platforms let you remove a
> file even when it's held open by a process ... the file's bytes still exist
> on disk (just without a filename pointing to them) and are only deleted when
> the last open file handle on that file is closed -- "delete on last close".


That's what I thought, but it feels good to get confirmation. We've been
trying to move our customer to linux for a long time. Hopefully it will
happen this fall and we won't have to do it this way anymore.

Appreciate all the help Mike, we got an OK from the customer to wait until
the fall and hopefully a move to linux. So I'll leave it be for now, though
not perfect, at least it works the way it did before I started to attempt
the improvement ;)

Christopher (Chris also works ;) )


On Fri, Jul 11, 2008 at 11:47 AM, Michael McCandless <
lucene@mikemccandless.com> wrote:

> OK, got it.
>
> The only way to make this work with svn is if you can have svn perform a
> switch without doing any removal, then restart your IndexSearcher, then do a
> normal svn switch to remove the now unused files.  Does svn have an option
> to "switch but don't remove any removed files"?  Because IndexSearcher holds
> these files open, svn will not be able to remove the now unused files until
> IndexSearcher switches.
>
> Or ... for every update to your index you could plant 2 tags, the first one
> reflecting only added files, and the 2nd one reflecting removed files.
>  Then, with your IndexSearcher still running, do an svn switch to the first
> tag, then restart the IndexSearcher, then svn switch to the 2nd one.  I
> think that'd work?
>
> This should only be necessary on Windows ... UNIX platforms let you remove
> a file even when it's held open by a process ... the file's bytes still
> exist on disk (just without a filename pointing to them) and are only
> deleted when the last open file handle on that file is closed -- "delete on
> last close".
>
>
> Mike
>
> Christopher Kolstad wrote:
>
>  Hi.
>>
>> First, thanks for the reply.
>>
>> Why does SubversionUpdate require shutting down the IndexSearcher?  What
>>
>>> goes wrong?
>>>
>>>
>> SubversionUpdate requires shutting down the IndexSearcher in our current
>> implementation because the old index files are deleted in the tag we're
>> switching to. Sorry, just realised that my last mail didn't state that we
>> don't in fact to an "svn up", but rather an "svn switch". Thus, when we
>> try
>> to perform the update SubversionUpdate fails due to file lock issues when
>> trying to update (deleting the old lucene files) the lucene index
>> directory
>> (The relevant code for the update action is quoted below).
>>
>> You might want to switch instead to rsync.
>>
>>>
>>>
>> I'm hoping I won't have to, firstly because I'm more familiar with
>> subversion, secondly because that would require me to configure rsync for
>> windows, and I'm still not sure if that will help anything with the file
>> lock issues we're trying to avoid.
>>
>> A Lucene index is fundamentally write once, so, syncing changes over
>> should
>>
>>> simply be copying over new files and removing now-deleted files.  You
>>> won't
>>> be able to remove files held open by the IndexSearcher, but, once the
>>> IndexSearcher restarts you'd then be able to delete those files on the
>>> next
>>> sync.
>>>
>>
>>
>> So I should be able to run the switch and then restart the IndexSearcher,
>> instead of turning off the IndexSearcher, run the switch, turn on the
>> IndexSearcher. I'd see how that would work with a linux box, having a bit
>> more trouble seeing how I will get it to work with a windows box (and my
>> live server is unfortunately a Windows 2003 box), Subversion keeps running
>> into file lock issues when I switch from one tag to the other if I try to
>> keep the search active. With Lucene 2.1 it even ran into file lock issues
>> after I'd disabled the search and was performing the switch. Now, when
>> we're
>> using the Lucene 2.3.2 jar the lock issues has mostly gone (once in 3
>> months, instead of every switch/update).
>>
>> Current code:
>>
>> disableSearch(request); //Sets the SearchActive boolean to false
>>
>>>
>>> Search searcher =
>>> (Search)ctx.getAttribute(FelleskatalogenStartupServlet.SEARCH);
>>>     if (searcher != null) {
>>>       searcher.clear();
>>>    }
>>>
>>
>>          String latestTag =
>>
>>> SubversionUtil.getInstance().getLatestTag(getTagUrl(request));
>>>
>>>           SubversionUtil.getInstance().runSwitch(getRoot(request),
>>> getTagUrl(request) + "/" + latestTag);
>>>
>>>           if (log.isDebugEnabled()) {
>>>               log.debug("Index set to " + getRoot(request) + "/lucene");
>>>           }
>>>
>>>           ctx.setAttribute(FelleskatalogenStartupServlet.SEARCH, new
>>> Search(getRoot(request) + "/lucene"));
>>>           ctx.setAttribute(FelleskatalogenStartupServlet.SEARCHACTIVE,
>>> new Boolean(true));
>>>
>>>
>>
>> BR,
>>
>> Christopher
>>
>> On Thu, Jul 10, 2008 at 4:27 PM, Michael McCandless <
>> lucene@mikemccandless.com> wrote:
>>
>>
>>
>>
>>> Why does SubversionUpdate require shutting down the IndexSearcher?  What
>>> goes wrong?
>>>
>>> You might want to switch instead to rsync.
>>>
>>> A Lucene index is fundamentally write once, so, syncing changes over
>>> should
>>> simply be copying over new files and removing now-deleted files.  You
>>> won't
>>> be able to remove files held open by the IndexSearcher, but, once the
>>> IndexSearcher restarts you'd then be able to delete those files on the
>>> next
>>> sync.
>>>
>>> Mike
>>>
>>>
>>> Christopher Kolstad wrote:
>>>
>>> Hi.
>>>
>>>>
>>>> Currently using Lucene 2.3.2 in a tomcat webapp. We have an action
>>>> configured that performs reindexing on our staging server. However, our
>>>> live
>>>> server can not reindex since it does not have the necessary dtd files to
>>>> process the xml.
>>>>
>>>> To update the index on the live server we perform a subversion update on
>>>> the
>>>> lucene index directory.
>>>> Unfortunately this makes it necessary to stop the IndexSearcher while
>>>> the
>>>> SubversionUpdate is doing its thing.
>>>>
>>>> Presently we've had a requirement from our customer to not disable
>>>> search.
>>>>
>>>> So my idea was to copy the index directory to another directory and then
>>>> switch the IndexSearcher from the original index directory to the
>>>> temporary
>>>> directory.
>>>> Then perform the Subversion update, and when done, switch the
>>>> IndexSearcher
>>>> back to the original (now, updated) index directory.
>>>>
>>>> Does anyone have any other suggestions on how to update the index
>>>> directory
>>>> from subversion without having to disable the IndexSearcher?
>>>>
>>>> BR
>>>> Christopher
>>>>
>>>> --
>>>> Regards,
>>>> Christopher Kolstad
>>>> =============================
>>>> |100 little bugs in the code, debug one, |
>>>> |recompile, 101 little bugs in the code |
>>>> =============================
>>>>
>>>> E-mail: chriswk@ifi.uio.no (University)
>>>> christopher.kolstad@gmail.com (Home)
>>>> chriswk@ovitas.no (Job)
>>>>
>>>>
>>>
>>> ---------------------------------------------------------------------
>>> To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
>>> For additional commands, e-mail: java-user-help@lucene.apache.org
>>>
>>>
>>>
>>
>> --
>> Regards,
>> Christopher Kolstad
>> =============================
>> |100 little bugs in the code, debug one, |
>> |recompile, 101 little bugs in the code |
>> =============================
>>
>> E-mail: chriswk@ifi.uio.no (University)
>> christopher.kolstad@gmail.com (Home)
>> chriswk@ovitas.no (Job)
>>
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
> For additional commands, e-mail: java-user-help@lucene.apache.org
>
>


-- 
Regards,
Christopher Kolstad
=============================
|100 little bugs in the code, debug one, |
|recompile, 101 little bugs in the code |
=============================

E-mail: chriswk@ifi.uio.no (University)
christopher.kolstad@gmail.com (Home)
chriswk@ovitas.no (Job)

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message