2 Feb 2008 04:49
Re: Exception: Couldn't read enough (EOF)
Olly Betts <olly <at> survex.com>
2008-02-02 03:49:52 GMT
2008-02-02 03:49:52 GMT
On Thu, Jan 31, 2008 at 05:39:54PM +0100, Andreas Marienborg wrote: > > On 31. jan.. 2008, at 17.28, Olly Betts wrote: > > >Are you on a multicore/multiprocessor machine? That might make this > >more likely to manifest. > > yes, 4-way system, but single proc writing (indexer-daemon) I suspect that means it's more likely that a reader can run while the base file is being rewritten. > >I think I need to write a patch and get you to test it. As a > >workaround, catching the exception and simply retrying reopen() should > >do the trick. > > I can try, but I might have a hard time reproducing it in anything but > prod, which is a bit hard to test. I guess all you can really do is apply the patch and run as normal. If the problem hasn't manifested after a reasonable period of time, then we will just have to assumed that we've nailed it. Here's the patch I've applied to SVN HEAD, which should apply to recent releases as I don't believe we've touched this code for a while: http://svn.xapian.org/trunk/xapian-core/backends/flint/flint_table.cc?r1=10081&r2=10080&view=patch&pathrev=10081 Cheers, Olly(Continue reading)
RSS Feed