lucene-java-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Uwe Schindler" <>
Subject RE: Effectiveness MMapDirectory on NFS Mounted indexes
Date Wed, 05 Nov 2014 10:55:04 GMT

In general, storing an index on NFS mounts is a really bad idea, because Lucene Commits don't
work correctly with NFS (this is an issue since the early beginning and is not fixable). If
you use NFS, you need to use SimpleFSLockFactory for locking (because NativeFSLockFactory
does not work) and you may also need to use a different IndexFileDeleter.

MMapDirectory does not behave different with NFS, but the risk of cruashing your JVM is very
high. If the connection to the NFS server drops, the OS kernel will unmap the mapped file
and all accesses to it will cause a segmentation fault (crush) of your Java VM. Otherwise
there is no real difference between the directory implementations. Writing and locking the
index is the main problem (see above).


Uwe Schindler
H.-H.-Meier-Allee 63, D-28213 Bremen

> -----Original Message-----
> From: Buddhavarapu, Suresh []
> Sent: Wednesday, November 05, 2014 7:00 AM
> To:
> Subject: Effectiveness MMapDirectory on NFS Mounted indexes
> Hi,
> I'm using Lucene 2.9.3 on a 64 bit machine. Many a times we are observing
> that the systems gets into to thrashing mode during merges.
> We are experimenting with using MMapDirectory. Our index is stored on
> NFS/CIFS mounted file shares.
> My question, is this MMapDirectory useful in such scenarios, I mean  for
> indexes on a network file share? Any experiences or thoughts on this?
> I cannot really migrate to latest lucene as of now.
> Thanks,
> Suresh

To unsubscribe, e-mail:
For additional commands, e-mail:

View raw message