lucene-java-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ian Lea <ian....@gmail.com>
Subject Re: Slightly Off-topic: How to decide whether or not to add a document?
Date Tue, 04 Aug 2009 17:00:14 GMT
Jim


The sleep is simply

	    try { Thread.sleep(millis); }
	    catch (InterruptedException ie) { }

No threading issues that I'm aware of, despite the method living in
the Thread class.

But you're right about it possibly impacting performance, if you've
got to sleep for a reasonable amount of time for each doc, if you've
got loads of docs.  You can improve it by getting a list of possible
files + size + lastmod + whatever, sleeping, then checking them all
again i.e. only sleep once for each pass rather than once per file.

Yet another option is to forget about sleeping and check the lastmod
timestamp and only index the doc if was finished some time ago.

And yet another ... make the producer write to /a/b/c and have a
standalone non-lucene job that reads /a/b/c doing whatever checks you
like, moving files to your input directory.


That's more than enough options from me.


--
Ian.

On Tue, Aug 4, 2009 at 5:08 PM, <ohaya@cox.net> wrote:
> Ian,
>
> One question about the 4th alternative:  I was wondering how you implemented the sleep()
in Java, esp. in such a way as not to mess up any of the Lucene stuff (in case there's threading)?
>
> Right now, my indexer/inserter app doesn't explicitly do any threading stuff.
>
> Thanks,
> Jim
>
>
> ---- ohaya@cox.net wrote:
>> Hi Ian,
>>
>> Thanks for the quick response.
>>
>> I forgot to mention, but in our case, the "producers" is part of a commercial package,
so we don't have a way to get them to change anything, so I think the 1st 3 suggestions are
not feasible for us.
>>
>> I have considered something like the 4th suggestion (check file size, timeout, and
check file size again).  I'm worried that it would impact the overall index insertion process,
but that unless there's something better, that may be our best option :(...
>>
>> Thanks again,
>> Jim
>>
>>
>> ---- Ian Lea <ian.lea@gmail.com> wrote:
>> > A few suggestions:
>> >
>> > . Queue the docs once they are complete using something like JMS.
>> >
>> > . Get the document producers to write to e.g. xxx.tmp and rename to
>> > e.g. xxx.txt at the end
>> >
>> > . Get the document producers to write to a tmp folder and move to e.g.
>> > input/ when done
>> >
>> > . Find a file, store size, sleep for a while, check size and if changed, skip
>> >
>> > I've used all these at one time or another for assorted, mainly
>> > non-lucene, apps, and they are all workable.
>> >
>> >
>> > --
>> > Ian.
>> >
>> >
>> > On Tue, Aug 4, 2009 at 4:40 PM, <ohaya@cox.net> wrote:
>> > > Hi,
>> > >
>> > > I have an app to initially create a Lucene index, and to populate it with
documents.  I'm now working on that app to insert new documents into that Lucene index.
>> > >
>> > > In general, this new app, which is based loosely on the demo apps (e.g.,
IndexFiles.java), is working, i.e., I can run it with a "create" parameter, and it creates
a good/valid index from the documents, and then I can run it with an "insert" parameter, and
it inserts new documents into the index.
>> > >
>> > > [As I mentioned in an earlier thread, we only have a requirement to insert
new documents into the index, no requirements for deleting documents or updating documents
that have already been indexed).
>> > >
>> > > Ok, as I said, that works so far.
>> > >
>> > > However, in our case, the processes that are creating the documents that
we are indexing are fairly long-lived, and write fairly large documents, and I'm worried that
when an insert operation is run, some of the potential documents may still be being written
to, and we wouldn't want the indexer to insert the document into the Lucene index until the
document is "complete".
>> > >
>> > > As you know, the way that the demos such as IndexFiles work is that they
call a method called IndexDocs().  IndexDocs() then recursively walks the directory tree,
and calling the writer to add to the index.
>> > >
>> > > In this loop, IndexDocs() does a few checks (isDirectory(), canRead), and
I think that it would "pick up" (find) some documents that are still "in progress" (being
written to, and not closed) in our case.
>> > >
>> > > I was wondering if anyone here has a situation similar to this (having
to index large documents that may be "in progress/being written to"), and how you handle this
situation?
>> > >
>> > > FYI, this is on Redhat Linux (and on Windows in my test environment).
>> > >
>> > > Thanks!
>> > >
>> > > Jim
>> > >
>> > >
>> > > ---------------------------------------------------------------------
>> > > To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
>> > > For additional commands, e-mail: java-user-help@lucene.apache.org
>> > >
>> > >
>> >
>> > ---------------------------------------------------------------------
>> > To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
>> > For additional commands, e-mail: java-user-help@lucene.apache.org
>> >
>>
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
>> For additional commands, e-mail: java-user-help@lucene.apache.org
>>
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
> For additional commands, e-mail: java-user-help@lucene.apache.org
>
>

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


Mime
View raw message