On Thu, Feb 11, 2010 at 1:33 PM, Julian Foad <[email protected]> wrote: > I tried a dry-run merge, using a tr...@908933 client, and it hangs with > Serf, but completes in a few seconds with Neon. (When I say it hangs, an > identical test except using a trunk build from 2010-02-04 hung for 25 > minutes before I interrupted it.) > > The test case here uses a 1....@902803 WC, clean except for a local mod > to the 'STATUS' file. The "##" lines are comments I typed while I was > waiting, to remind me when I issued the command. > > [[[ > $ svn merge --dry-run ^/subversion/branches/1.6.x-r891672/@902803 > --- Merging r891676 through r902803 into '.': > U subversion/tests/cmdline/externals_tests.py > ## Started ~10:49 ... > ^C/home/julianfoad/src/subversion-a/subversion/svn/util.c:900: (apr_err=4) > /home/julianfoad/src/subversion-a/subversion/libsvn_client/merge.c:9460: > (apr_err=4) > /home/julianfoad/src/subversion-a/subversion/libsvn_client/merge.c:8082: > (apr_err=4) > /home/julianfoad/src/subversion-a/subversion/libsvn_client/merge.c:7753: > (apr_err=4) > /home/julianfoad/src/subversion-a/subversion/libsvn_client/merge.c:4863: > (apr_err=4) > /home/julianfoad/src/subversion-a/subversion/libsvn_ra_serf/update.c:2248: > (apr_err=4) > svn: Error retrieving REPORT (4): Interrupted system call > > $ svn merge --dry-run ^/subversion/branches/1.6.x-r891672/@902803 > --config-option=servers:global:http-library=neon > --- Merging r891676 through r902803 into '.': > U subversion/tests/cmdline/externals_tests.py > U subversion/libsvn_client/commit_util.c > > $ svn merge --dry-run ^/subversion/branches/1.6.x-r891672/@902803 > --config-option=servers:global:http-library=neon > --- Merging r891676 through r902803 into '.': > U subversion/tests/cmdline/externals_tests.py > U subversion/libsvn_client/commit_util.c > > $ svn merge --dry-run ^/subversion/branches/1.6.x-r891672/@902803 > --config-option=servers:global:http-library=serf > --- Merging r891676 through r902803 into '.': > U subversion/tests/cmdline/externals_tests.py > ## Started 11:16 ... > ]]] > > Here's the stack trace if I interrupt it in GDB after about 10 seconds: > [[[ > --- Merging r891676 through r902803 into '.': > U subversion/tests/cmdline/externals_tests.py > ^C > Program received signal SIGINT, Interrupt. > 0x0040e422 in __kernel_vsyscall () > (gdb) bt > #0 0x0040e422 in __kernel_vsyscall () > #1 0x004dc156 in epoll_wait () from /lib/tls/i686/cmov/libc.so.6 > #2 0x0061d3ad in apr_pollset_poll () from /usr/lib/libapr-1.so.0 > #3 0x00d149e8 in serf_context_run () from /usr/lib/libserf-0.so.0 > #4 0x00af2b9d in finish_report (report_baton=0x966f148, pool=0x966ef30) > at > /home/julianfoad/src/subversion-b/subversion/libsvn_ra_serf/update.c:2237 > #5 0x002eff90 in drive_merge_report_editor ( > target_abspath=0x81b5c70 "/home/julianfoad/src/subversion-1.6.x", > url1=0x80ad3d0 > "https://svn.apache.org/repos/asf/subversion/branches/1.6.x";, > revision1=891675, > url2=0x80ad448 > "https://svn.apache.org/repos/asf/subversion/branches/1.6.x-r891672";, > revision2=902803, children_with_mergeinfo=0x9123318, > depth=svn_depth_infinity, > notify_b=0xbfffede4, callbacks=0x31be60, merge_b=0xbfffed74, > pool=0x966ef30) > at /home/julianfoad/src/subversion-b/subversion/libsvn_client/merge.c:4863 > #6 0x002f5b76 in do_directory_merge ( > url1=0x80ad3d0 > "https://svn.apache.org/repos/asf/subversion/branches/1.6.x";, > revision1=891675, > url2=0x80ad448 > "https://svn.apache.org/repos/asf/subversion/branches/1.6.x-r891672";, > revision2=902803, target_abspath=0x81b5c70 > "/home/julianfoad/src/subversion-1.6.x", > depth=svn_depth_infinity, squelch_mergeinfo_notifications=0, > notify_b=0xbfffede4, > merge_b=0xbfffed74, pool=0x9123270) > at /home/julianfoad/src/subversion-b/subversion/libsvn_client/merge.c:7746 > #7 0x002f66cf in do_merge (merge_sources=0x80ac9f0, > target_abspath=0x81b5c70 "/home/julianfoad/src/subversion-1.6.x", > sources_ancestral=1, > sources_related=1, same_repos=1, ignore_ancestry=0, force=0, dry_run=1, > record_only=0, > reintegrate_merge=0, squelch_mergeinfo_notifications=0, > depth=svn_depth_infinity, > merge_options=0x0, use_sleep=0xbfffef30, ctx=0x808fb60, pool=0x808f200) > at /home/julianfoad/src/subversion-b/subversion/libsvn_client/merge.c:8079 > #8 0x002fa646 in svn_client_merge_peg3 ( > source=0x81b5968 > "https://svn.apache.org/repos/asf/subversion/branches/1.6.x-r891672";, > ranges_to_merge=0x81b5b80, peg_revision=0xbfffefe0, > target_wcpath=0x8073548 "", > depth=svn_depth_unknown, ignore_ancestry=0, force=0, record_only=0, > dry_run=1, > merge_options=0x0, ctx=0x808fb60, pool=0x808f200) > at /home/julianfoad/src/subversion-b/subversion/libsvn_client/merge.c:9451 > #9 0x0805b0b2 in svn_cl__merge (os=0x808f3a8, baton=0xbffff248, > pool=0x808f200) > at /home/julianfoad/src/subversion-b/subversion/svn/merge-cmd.c:322 > #10 0x0805a53e in main (argc=5, argv=0xbffff394) > at /home/julianfoad/src/subversion-b/subversion/svn/main.c:2249 > ]]] > > If in GDB I step through the big loop in RA-serf's finish_report(), > pausing in each iteration, then it works [1]. If, however, I break at > the beginning of the loop, set a breakpoint at the end of the loop, and > let that loop run through at full speed, then it hangs in > serf_context_run() in some iteration of the loop. [2] > > If I insert debug prints [3], I get: > > [[[ > ## ra-serf finish_report() loop: > ## done=0, fetches=0, propfinds=0, t=-694967296 > ## done=0, num_conns=2, cur_conn=1 > ## done=0, fetches=0, propfinds=0, t=-694967296 > ## done=0, num_conns=2, cur_conn=1 > ## done=0, fetches=0, propfinds=0, t=-694967296 > ## done=0, num_conns=2, cur_conn=1 > ## done=0, fetches=0, propfinds=0, t=-694967296 > ## done=0, num_conns=2, cur_conn=1 > ## done=0, fetches=2, propfinds=0, t=-694967296 > ## done=1, num_conns=2, cur_conn=1 > ## done=1, fetches=2, propfinds=0, t=-694967296 > ## done=1, num_conns=2, cur_conn=1 > [... hundreds of repeats ...] > ## done=1, fetches=2, propfinds=0, t=-694967296 > ## done=1, num_conns=2, cur_conn=1 > ## done=1, fetches=2, propfinds=0, t=-694967296 > --- Merging r891676 through r902803 into '.': > U subversion/tests/cmdline/externals_tests.py > ## done=1, num_conns=2, cur_conn=1 > ## done=1, fetches=1, propfinds=0, t=-694967296 > [... hangs here ...] > ]]] > > The timeout looks wrong, for a start. Could that be the problem? > > My Serf is version 0.3.0. > [..]
Can you try with serf branch 0.3.x? I fixed a connection cleanup bug some time ago that caused serf to hang in apr_pollset_poll. http://serf.googlecode.com/svn/branches/0.3.x/ Lieven
