lucene-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Dmitry Serebrennikov <dmit...@earthlink.net>
Subject Re: [Lucene-dev] Re: FW: Lucene 1.2 and directory write permissio ns?
Date Sat, 06 Oct 2001 00:29:22 GMT
What if we make the location of the locks a property of the index iteslf 
(meaning that it is stored in some file in the index directory)? If the 
index is trully read-only (as in a CD-ROM), we don't need locks at all 
and this property can be set to a special value indicating that index is 
read-only or that no locking should be used. Would this be the same as 
indicating that the index is process-safe?

If the index is being updated by another process, then the index is 
writable and administrator can configure locks to be wherever they want. 
This shouldn't be done by the indexing process though, or we are going 
to get into locking problems on the lock location property :).

Snyder, David wrote:

> I second that having the lock location configurable would be a "nice 
> to have"...  but this brings up the problem of the searcher process 
> being configured differently than the updating process, which will 
> defeat the usefulness of the lock.
>
>  
>
> I can imagine the sys-admin types wanting locks to go in a temp 
> directory or something like that, though.
>
>  
>
>  
>
> Dave
>
>  
>
> David Snyder
> Señor Software Engineer
>
> NetGenics, Inc.
> 1717 E. 9th St., #1700
> Cleveland, OH 44114
> (216) 861-4007
>
>     -----Original Message-----
>     From: Dmitry Serebrennikov [mailto:dmitrys@earthlink.net]
>     Sent: Friday, October 05, 2001 3:24 PM
>     To: lucene-dev@jakarta.apache.org
>     Subject: [Lucene-dev] Re: FW: Lucene 1.2 and directory write
>     permissions?
>
>     Just a though - as long as we are separating the locks, we should
>     probably make the location configurable and not require it to be a
>     subdirectory of the indexes. This will help in case the indexes
>     come burned on a CD-ROM or if the operating system does not
>     support easy linking / permission magic.
>
>     Doug Cutting wrote:
>
>>Here's one vote for putting locks in a separate directory.  Anyone dislike
>>that?
>>
>>Doug
>>
>>-----Original Message-----
>>From: Snyder, David [mailto:dsnyder@netgenics.com]
>>Sent: Friday, October 05, 2001 11:23 AM
>>To: Doug Cutting
>>Subject: RE: Lucene 1.2 and directory write permissions?
>>
>>
>>The lock file synchronization is very handy for us as we do updates in a
>>separate process from searches...  I was very pleased to see this in there!
>>
>>I think splitting out the locks into a separate directory would solve our
>>problem...  we have a reference set of data that we use for testing and our
>>sysadmin wants to make sure it doesn't accidentally get overwritten.  It's
>>still possible (using links and other permission magic) but becomes more of
>>a maintenance headache.  I definitely vote for the locks subdirectory idea.
>>It can be created when the index
>>es are initially loaded, we can go change
>>the permissions or maybe make it a link to a tmp directory or something, and
>>then our regular index files can be safe.  Do you think this is something
>>very difficult to do? (I have yet to build Lucene myself, but I would love
>>to contribute... we are actually working on some XML based loaders that may
>>be of general interest)
>>
>> Lucene has been great for us, but the way... we are indexing genetic data
>>(not the sequences themselves, but all the annotations and description stuff
>>that scientists tack on) and Lucene has been excellent...  our indexes (we
>>use many with the multisearcher) are about 13 gigs now and Lucene has hardly
>>broken a sweat.
>>
>>Thanks for your help,
>>Dave
>>
>>David Snyder
>>Señor Software Engineer
>>
>>NetGenics, Inc.
>>1717 E. 9th St., #1700
>>Cleveland, OH 44114
>>(216) 861-4007
>>
>>
>>-----Original Message-----
>>From: Doug Cutting [mailto:DCutting@grandcentral.com]
>>Sent: Friday, October 05, 2001 12:24 PM
>>To: 'Snyder, David'
>>Subject: RE: Lucene 1.2 and directory write permissions?
>>
>>
>>>From: Snyder, David [mailto:dsnyder@netgenics.com]
>>>
>>>I've been porting our application to use the 1.2 release 
>>>candidate 1 build
>>>and now have a problem opening searchers on our existing 
>>>indexes.  I get a
>>>Permission Denied exception...  our permissions are set up to 
>>>allow reading
>>>of the directory and contained files during a search, but not writing.
>>>
>>
>>Hmm.  That is a problem.  The reader now creates a lock file while it is
>>opening the index to keep a writer process from deleting files while they're
>>being opened.  When opening an index the reader must first read the list of
>>files to open, then open them.  If between reading the list and opening a
>>file that file were to disappear, then the open would fail.  This was the
>>longstanding race condition that is fixed by the lock files.  A writing
>>process will now wait for the reader to open all of the files before
>>updating things.
>>
>>Perhaps we should instead write lock files in a subdirectory of the index
>>named "locks".  You could make that directory read/write, but make the
>>parent read-only.  Alternately, we could have an flag that turns off the use
>>of lock files, for those who know that there is no other process that is
>>potentially simultaneously updating the index.  Which approach would folks
>>prefer?
>>
>>D
>>oug
>>
>


Mime
View raw message