Re: svn commit: r1067687 - in /subversion/trunk: ./ subversion/libsvn_fs_util/caching.c subversion/svnserve/main.c

Stefan Sperling <stsp_at_elego.de> writes:

> On Mon, Feb 07, 2011 at 11:51:11AM +0000, Philip Martin wrote:
>> stefan2_at_apache.org writes:
>>
>> > Author: stefan2
>> > Date: Sun Feb 6 15:51:15 2011
>> > New Revision: 1067687
>> >
>> > URL: http://svn.apache.org/viewvc?rev=1067687&view=rev
>> > Log:
>> > Merged latest caching bug fixes from performance branch:
>> > revisions 1029232, 1032333, 1033040, 1033057 and 1033294
>> > (support for "no threads", typos, error leaks)
>>
>> This commit* appears to cause a huge increase in Apache memory use when
>> running the regression tests, enough to make the machine unuseable and
>> prevent the tests running.
>>
>> [*]Strictly it's r1067712 since r1067687 doesn't build, but I think
>> r1067687 is the one that causes the problem.
>
> My buildbot seems to be affected by this, too.
>>From the web gui I can tell that it's currently running neon tests.
> And it's responding to pings... but I cannot ssh into it.
> ssh can establish a TCP connection but then it hangs.

It appears to be this part that causes the problem:

Index: ../src/subversion/libsvn_fs_util/caching.c
===================================================================
--- ../src/subversion/libsvn_fs_util/caching.c (revision 1067686)
+++ ../src/subversion/libsvn_fs_util/caching.c (revision 1067687)
@@ -38,8 +38,12 @@
     16, /* up to 16 files kept open */
     TRUE, /* cache fulltexts */
     FALSE, /* don't cache text deltas */
+
+#ifdef APR_HAS_THREADS
     FALSE /* assume multi-threaded operation */
- };
+#else
+ TRUE /* single-threaded is the only supported mode of operation */
+};

-- 
Philip