Floyd,

Implementing a custom merge policy involves a concrete implementation
Lucene.Net.Index.MergePolicy.  The only documentation to point you to
are the javadocs as the API between Java Lucene & Lucene.Net are
identical:

http://lucene.apache.org/java/2_9_2/api/all/org/apache/lucene/index/Merg
ePolicy.html

For an implementation example take a look at
Lucene.Net.Index.LogMergePolicy.

Michael

-----Original Message-----
From: Floyd Wu [mailto:[email protected]] 
Sent: Tuesday, March 09, 2010 7:14 PM
To: [email protected]
Subject: Re: About processing handling during merge time

Hi Michael,

If I want to create a custom merge policy that act "Never do merge
during
work day". Where to start? I read something about merge policy could be
related with scheduler, but I'm not sure the relationship between these
two
objects.

many thanks.

Floyd



2010/3/9 Michael Garski <[email protected]>

> Floyd,
>
> Long-running merges are an issue with large indexes due to the CPU and
> I/O overhead required to perform the merge, but there are a few things
> you can do to reduce that time:
>
> 1) Create a custom merge policy to give you control over when merges
> occur and what segments are merged
> 2) Shard your index into smaller pieces that are searched together
using
> a MultiSearcher - this will keep your indexes smaller.
>
> Michael
>
> -----Original Message-----
> From: Floyd Wu [mailto:[email protected]]
> Sent: Monday, March 08, 2010 1:45 AM
> To: [email protected]
>  Subject: About processing handling during merge time
>
> Hi all,
>
> I have a problem that my index files is 33.4gb now, and when I insert
> new
> index through Lucene.Net few times. Lucene will execute merge index
> files
> automatically. The problem is during merge executed, writer.lock file
> will
> be created and all operations such as read/write index files were all
> stopped.
> (for a long time and there were no any status or flag to know Lucene
> were
> executeing merge process.)
>
> Are there any solutions to avoid this "long time lock" situation?
>
> Thanks in advanced.
>
> Floyd
>
>

Reply via email to