Server Version: Apache/2.4.6 (Unix) SVN/1.8.3
I thought I may be witnessing this bug
My question is ultimately what is reasonable for Timeout? This Apache instance is for Subversion hosting, with approximately 100 users plus any automated processes. To a lesser degree, my question of the git developers is why does git-svn not occasionally send a KeepAlive request to Apache if it knows it's still working on something?
I've been running a command (git svn fetch) to try to clone a large svn repository into a local git repo. Over the past week, I've seen the connection get reset by the server [1]. The "only" thing that has changed is that I upgraded the CollabNet Subversion Edge system. Connections being closed was not a problem prior to upgrading CollabNet Subversion Edge. After looking into it, there really aren't specific (Apache) configuration changes that appear problematic. CNSE bundles Apache. The culprit seems to be the fact that Apache was upgraded from 2.2 to 2.4
Server Settings
Server Version: Apache/2.4.6 (Unix) SVN/1.8.3
Server Built: Sep 4 2013 18:47:57
Server loaded APR Version: 1.4.8
Compiled with APR Version: 1.4.8
Server loaded APU Version: 1.5.2
Compiled with APU Version: 1.5.2
Module Magic Number: 20120211:23
Timeouts: connection: 600 keep-alive: 750 (later changed tTimout to 6000)
MPM Name: prefork
MPM Information: Max Daemons: 256 Threaded: no Forked: yes
Server Architecture: 64-bit
Server Root: /usr/local/csvn
Config File: /usr/local/csvn/data/conf/csvn_main_httpd.conf
Server Built With:
-D APR_HAS_SENDFILE
-D APR_HAS_MMAP
-D APR_HAVE_IPV6 (IPv4-mapped addresses enabled)
-D APR_USE_SYSVSEM_SERIALIZE
-D SINGLE_LISTEN_UNSERIALIZED_ACCEPT
-D APR_HAS_OTHER_CHILD
-D AP_HAVE_RELIABLE_PIPED_LOGS
-D HTTPD_ROOT="/"
-D SUEXEC_BIN="//bin/suexec"
-D DEFAULT_PIDLOG="opt/CollabNet_Subversion/data/run/httpd.pid"
-D DEFAULT_SCOREBOARD="logs/apache_runtime_status"
-D DEFAULT_ERRORLOG="logs/error_log"
-D AP_TYPES_CONFIG_FILE="opt/CollabNet_Subversion/data/conf/mime.types"
-D SERVER_CONFIG_FILE="opt/CollabNet_Subversion/data/conf/httpd.conf"
The Subversion FAQ
http://subversion.apache.org/faq.html#secure-connection-truncated suggests changing the Timeout value (after considering the layout of your repository). I've tried changing the Timeout and KeepAlive settings to no avail. [edit: changing Timeout from 60 to 600 didn't work, but setting it to 6000 appears to work as now git svn fetch has been running for a few days.]
I retried with a changed .subversion client configuration by setting servers:global:http-library=serf (I'm running the svn 1.6 client against a 1.8 server.) But that did not help.
It will fail repeatedly in the same spot, and then get to a different spot where it fails. It's almost always a binary file (.doc, .pptx) but the files are not large. The last one where it failed was 60,362 bytes.
This is what I see in the error log
[Fri Oct 11 12:48:15.693996 2013] [dav:error] [pid 17474] [client 192.168.11.14:40502] Provider encountered an error while streaming a REPORT response. [500, #0]
[Fri Oct 11 12:48:15.694045 2013] [dav:error] [pid 17474] [client 192.168.11.14:40502] A failure occurred while driving the update report editor [500, #70007]
[Fri Oct 11 12:48:15.694056 2013] [dav:error] [pid 17474] [client 192.168.11.14:40502] Error writing base64 data: The timeout specified has expired [500, #70007]
Thanks,
[1] RA layer request failed: REPORT of '/svn/platform/!svn/vcc/default': Could not read response body: connection was closed by server (
http://code.benunets.com:4443) at /home/greg/share/perl/5.14.2/Git/SVN/Ra.pm line 282
