My report 2022-01-26

[Bob Proulx]

I am typing this in ten minutes before the meeting.  If it looks
hectic and cluttered it is only because it is that way.

As reported previously have git running on the new vcs2.  Starting
this week I have been working on getting subversion up and running.

* Three of the six components of subversion are working.

    FAIL: svn-http
    FAIL: svn-https
    PASS: svn-rsync
    PASS: svn-ssh
    FAIL: svn-svn
    PASS: svn-web

** svn over ssh is working.

svn+ssh leverages the same libnss-mysql module that is used for
git+ssh and was part of the earlier work.  And it will be the same for
hg+ssh and bzr+ssh too.

** ViewVC web interface is online

This uses Nginx on the frontend and FastCGI on the backend.  I had
some hassle with systemd.  The traditional method uses spawn-fcgi to
launch multiwatch which uses socket activation to monitor and respawn
multiple fcgiwrap daemons.  We normally run 10 fcgiwrap daemons to
handle the load on the cgi-bin services such as ViewVC.  The shipped
systemd service unit for this is terrible.  I ended up replacing the
packaged one entirely.  It was a non-trivial effort to get to this
trivial result.  It's working now however.

    root@vcs2:/etc# ps -ef |grep -e multiwatch -e fcgiwrap
    www-data 30930     1  0  2021 ?        00:00:02 /usr/bin/multiwatch -f 10 
-- /usr/sbin/fcgiwrap -f
    www-data 30933 30930  0  2021 ?        00:28:32 /usr/sbin/fcgiwrap -f
    www-data 30934 30930  0  2021 ?        00:27:48 /usr/sbin/fcgiwrap -f
    www-data 30935 30930  0  2021 ?        00:27:30 /usr/sbin/fcgiwrap -f
    www-data 30936 30930  0  2021 ?        00:27:52 /usr/sbin/fcgiwrap -f
    www-data 30938 30930  0  2021 ?        00:28:31 /usr/sbin/fcgiwrap -f
    www-data 30939 30930  0  2021 ?        00:27:46 /usr/sbin/fcgiwrap -f
    www-data 30940 30930  0  2021 ?        00:28:40 /usr/sbin/fcgiwrap -f
    www-data 30941 30930  0  2021 ?        00:28:40 /usr/sbin/fcgiwrap -f
    www-data 30942 30930  0  2021 ?        00:27:50 /usr/sbin/fcgiwrap -f
    www-data 30943 30930  0  2021 ?        00:27:33 /usr/sbin/fcgiwrap -f
    root     31212   938  0 10:53 pts/3    00:00:00 grep -e multiwatch -e 
fcgiwrap

** Down to two cron jobs on vcs0 that still need to be migrated to vcs2

This is the main problematic one that will need more work to do.

    # Create project areas via web ui checkbox and database.
    35,05 * * * *   root    nice -n 11 sv_groups --cron --only-svn --only-git 
--only-hg --only-bzr

And then this one is giving problems.  It's hanging.  And reporting
stale lock problems.  Plus its just ugly.  It needs cleanup.

    2,32 * * * *      root    
/usr/src/infra-old-vcs/infra/git/refresh-project-list.sh && 
/usr/src/infra-old-vcs/infra/git/sv_cgit.pl 2>&1 | grep -v 'No such file or 
directory'

* Missing /etc/logrotate.d/rsyslog file?

For whatever reason on the new VMs the /etc/logrotate.d/rsyslog file
was found to be missing.  This caused the /var/log directory to grow
without pruning for those files.  Here is an easy way to get it back.

    apt-get -o DPkg::Options::=--force-confnew -o 
DPkg::Options::=--force-confmiss install --reinstall rsyslog

I don't like it but one of the prominent members of the community
successfully argued that if a conffile is missing that it is an
explicit configuration and that dpkg should not replace a conffile if
one is missing.  Therefore some years ago dpkg was changed to require
a force option if that case is detected in order to replace the
conffile.  Personally I think it is silly to consider a missing
conffile as a configuration.  If one wants that then one can zero out
the file.

In any case it means I now always specify the following option when
doing these things in order to get conffiles that have been removed.

    -o DPkg::Options::=--force-confmiss

I couldn't pull this one out of etckeeper's git because from square
one the file wasn't there.

And it's meeting time.

Bob