systemd unable to detect running apache if invoked via "apache2ctl graceful"

[RedScourge]

[Impact]

Unattended upgrade will attempt to gracefully reload the Apache2 httpd service, but it reloads apache directly, not via systemd, so systemctl will mis-report the true status. This can cause unexpected and unnecessary server outages when e.g. upgrading from one LTS to another.

[Test Case]

root@server:/usr/local/sbin# apache2ctl stop
root@server:/usr/local/sbin# apache2ctl graceful
httpd not running, trying to start
root@server:/usr/local/sbin# systemctl status apache2.service
? apache2.service - The Apache HTTP Server
   Loaded: loaded (/lib/systemd/system/apache2.service; enabled; vendor preset: enabled)
  Drop-In: /lib/systemd/system/apache2.service.d
           +-apache2-systemd.conf
   Active: inactive (dead) since Mon 2019-06-10 01:04:13 MDT; 17s ago
  Process: 27370 ExecStop=/usr/sbin/apachectl stop (code=exited, status=0/SUCCESS)
  Process: 11462 ExecReload=/usr/sbin/apachectl graceful (code=exited, status=0/SUCCESS)
 Main PID: 1849 (code=exited, status=0/SUCCESS)

Jun 08 19:11:24 server apachectl[1408]: AH00112: Warning: DocumentRoot [/srv/apache2/www/...
Jun 08 19:11:24 server apachectl[1408]: AH00112: Warning: DocumentRoot [/srv/apache2/www/...
Jun 08 19:11:24 server apachectl[1408]: AH00112: Warning: DocumentRoot [/srv/apache2/www/...
Jun 08 19:11:24 server apachectl[1408]: AH00112: Warning: DocumentRoot [/srv/apache2/www/...
Jun 08 19:11:24 server apachectl[1408]: AH00112: Warning: DocumentRoot [/srv/apache2/www/...
Jun 08 19:11:24 server apachectl[1408]: AH00112: Warning: DocumentRoot [/srv/apache2/www/...
Jun 08 19:11:24 server systemd[1]: Started The Apache HTTP Server.
Jun 09 06:25:04 server systemd[1]: Reloading The Apache HTTP Server.
Jun 09 06:25:04 server systemd[1]: Reloaded The Apache HTTP Server.
Jun 10 01:04:13 server apachectl[27370]: httpd (no pid file) not running

Output after customizing the apache2ctl script and restarting it:

root@server:/usr/local/sbin# apache2ctl_custom graceful
Invoking 'systemctl start apache2'.
Use 'systemctl status apache2' for more info.
root@server:/usr/local/sbin# systemctl status apache2.service
? apache2.service - The Apache HTTP Server
   Loaded: loaded (/lib/systemd/system/apache2.service; enabled; vendor preset: enabled)
  Drop-In: /lib/systemd/system/apache2.service.d
           +-apache2-systemd.conf
   Active: active (running) since Mon 2019-06-10 01:04:50 MDT; 4s ago
  Process: 27370 ExecStop=/usr/sbin/apachectl stop (code=exited, status=0/SUCCESS)
  Process: 11462 ExecReload=/usr/sbin/apachectl graceful (code=exited, status=0/SUCCESS)
  Process: 27432 ExecStart=/usr/sbin/apachectl start (code=exited, status=0/SUCCESS)
 Main PID: 27444 (apache2)
    Tasks: 6 (limit: 2318)
   CGroup: /system.slice/apache2.service
           +-27444 /usr/sbin/apache2 -k start
           +-27448 /usr/sbin/apache2 -k start
           +-27449 /usr/sbin/apache2 -k start
           +-27451 /usr/sbin/apache2 -k start
           +-27454 /usr/sbin/apache2 -k start
           +-27455 /usr/sbin/apache2 -k start

Jun 10 01:04:50 server systemd[1]: Starting The Apache HTTP Server...
Jun 10 01:04:50 server systemd[1]: Started The Apache HTTP Server.

[Fix]

The fix is made somewhat obvious due to the fact that the issue was fixed for "apache2ctl start" in 18.04 LTS by replacing the regular call (which resolves to "/usr/sbin/apache2 -k start") with a check for presence of systemd, followed by an invocation of "systemctl start apache2.service" if that check is affirmative, or falling back to the old start command if that check is negative.

Attached to this bug report is a diff of the file before and after I copied the fixed invocation from the "start" subsection to the "graceful" subsection in 18.04 LTS. A fix for 16.04 LTS would require copying that block both into the "start" and "graceful" subsections.

[Regression Potential]

This alters the behavior of Debian's apache2ctl script for managing the service state. No other Apache2 code is modified. Thus things to watch for new regressions would deal with the apache service's state itself, not crashes or behavioral changes.

With a manual sysadmin-supervised service restart, if there are problems the sysadmin can intervene. The real risk of regression here would be for unsupervised or automated service updates.

[Original Report]

When starting or restarting Apache via the /usr/sbin/apache2ctl script, systemd becomes unaware of the state of Apache, causing "systemctl status apache2.service" to report "Active: inactive (dead)". Below I describe the issue and a fix; attached is the output of the diff command after I fixed it on my own 18.04 LTS system.

Despite this issue, Apache itself generally runs fine, so this is not an absolutely critical issue, however as both "systemctl status apache2.service" and "service apache2 status" become unable to detect its running state, and as some people may be scripting Apache service checks using those commands, they may experience server outages when they migrate from 14.04 LTS to 16.04 LTS or 18.04 LTS, so this could potentially have serious consequences. I suspect this caused or at least contributed to at least one outage of a web server I am responsible for, as the web file backup script was unable to restart the apache process after stopping it.

This issue affects Ubuntu 16.04 LTS when using "apache2ctl start" or "apache2ctl graceful", and Ubuntu 18.04 LTS just when using "apache2ctl graceful".

[Additional information]

lsb_release -rd :

Description: Ubuntu 18.04.2 LTS
Release: 18.04

apt-cache policy apache2:

apache2:
  Installed: 2.4.29-1ubuntu4.6
  Candidate: 2.4.29-1ubuntu4.6

[RedScourge (redscourge)]

Output showing how to reproduce the issue:

root@server:/usr/local/sbin# apache2ctl stop
root@server:/usr/local/sbin# apache2ctl graceful
httpd not running, trying to start
root@server:/usr/local/sbin# systemctl status apache2.service
? apache2.service - The Apache HTTP Server
   Loaded: loaded (/lib/systemd/system/apache2.service; enabled; vendor preset: enabled)
  Drop-In: /lib/systemd/system/apache2.service.d
           +-apache2-systemd.conf
   Active: inactive (dead) since Mon 2019-06-10 01:04:13 MDT; 17s ago
  Process: 27370 ExecStop=/usr/sbin/apachectl stop (code=exited, status=0/SUCCESS)
  Process: 11462 ExecReload=/usr/sbin/apachectl graceful (code=exited, status=0/SUCCESS)
 Main PID: 1849 (code=exited, status=0/SUCCESS)

Jun 08 19:11:24 server.wasteofcache.com apachectl[1408]: AH00112: Warning: DocumentRoot [/srv/apache2/www/russianfake
Jun 08 19:11:24 server.wasteofcache.com apachectl[1408]: AH00112: Warning: DocumentRoot [/srv/apache2/www/russianfake
Jun 08 19:11:24 server.wasteofcache.com apachectl[1408]: AH00112: Warning: DocumentRoot [/srv/apache2/www/stratmeme.c
Jun 08 19:11:24 server.wasteofcache.com apachectl[1408]: AH00112: Warning: DocumentRoot [/srv/apache2/www/stratmeme.c
Jun 08 19:11:24 server.wasteofcache.com apachectl[1408]: AH00112: Warning: DocumentRoot [/srv/apache2/www/wasteofcach
Jun 08 19:11:24 server.wasteofcache.com apachectl[1408]: AH00112: Warning: DocumentRoot [/srv/apache2/www/wasteofcach
Jun 08 19:11:24 server.wasteofcache.com systemd[1]: Started The Apache HTTP Server.
Jun 09 06:25:04 server.wasteofcache.com systemd[1]: Reloading The Apache HTTP Server.
Jun 09 06:25:04 server.wasteofcache.com systemd[1]: Reloaded The Apache HTTP Server.
Jun 10 01:04:13 server.wasteofcache.com apachectl[27370]: httpd (no pid file) not running

Output after customizing the apache2ctl script and restarting it:

root@server:/usr/local/sbin# apache2ctl_custom graceful
Invoking 'systemctl start apache2'.
Use 'systemctl status apache2' for more info.
root@server:/usr/local/sbin# systemctl status apache2.service
? apache2.service - The Apache HTTP Server
   Loaded: loaded (/lib/systemd/system/apache2.service; enabled; vendor preset: enabled)
  Drop-In: /lib/systemd/system/apache2.service.d
           +-apache2-systemd.conf
   Active: active (running) since Mon 2019-06-10 01:04:50 MDT; 4s ago
  Process: 27370 ExecStop=/usr/sbin/apachectl stop (code=exited, status=0/SUCCESS)
  Process: 11462 ExecReload=/usr/sbin/apachectl graceful (code=exited, status=0/SUCCESS)
  Process: 27432 ExecStart=/usr/sbin/apachectl start (code=exited, status=0/SUCCESS)
 Main PID: 27444 (apache2)
    Tasks: 6 (limit: 2318)
   CGroup: /system.slice/apache2.service
           +-27444 /usr/sbin/apache2 -k start
           +-27448 /usr/sbin/apache2 -k start
           +-27449 /usr/sbin/apache2 -k start
           +-27451 /usr/sbin/apache2 -k start
           +-27454 /usr/sbin/apache2 -k start
           +-27455 /usr/sbin/apache2 -k start

Jun 10 01:04:50 server.wasteofcache.com systemd[1]: Starting The Apache HTTP Server...
Jun 10 01:04:50 server.wasteofcache.com systemd[1]: Started The Apache HTTP Server.

[RedScourge (redscourge)]

(if someone could redact the domain names and file paths from the above comment that would be much appreciated)

[Ubuntu Foundations Team Bug Bot (crichton)]

The attachment "diff of /usr/sbin/apache2ctl after correcting issue" seems to be a patch. If it isn't, please remove the "patch" flag from the attachment, remove the "patch" tag, and if you are a member of the ~ubuntu-reviewers, unsubscribe the team.

[This is an automated message performed by a Launchpad user owned by ~brian-murray, for any issues please contact him.]

[Andreas Hasenack (ahasenack)]

Thanks for the detailed report. It looks like it's the same issue as https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=927302, so I'm linking them both together.

[Andreas Hasenack (ahasenack)]

Acknowledging that the bug still exists and is in our queue, but not at high priority.

[RedScourge (redscourge)]

UPDATE: I just had a server experience downtime due to this bug. An update of the apache2 package was automatically triggered on a client's 16.04 LTS server on Aug 31 via the unattended-upgrades cron script, and apparently before this update, the last time the Apache service had been restarted was done via "apachectl graceful", without an appropriately modified /usr/sbin/apachectl as described in my initial report, so the usual unattended-upgrades stop-and-start which follows an Apache update failed to perform as expected, leaving the service in the stopped state.

Therefore, it appears there is now a definite case where a user invoking "apachectl graceful" leads to downtime. To mitigate this, you can either disable/uninstall unattended-upgrades, or patch the /usr/sbin/apachectl file as described in this bug report.

I have deleted the old patch file and replaced it with an actual correct one based on the apache2ctl file on my client's server after the latest apache2 package update, now that I know how diff patch files are generated

[Bryce Harrington (bryce)]

Bumping importance on this bug, since it is causing downtime on production systems. It is already on our server-next queue and known in debian BTS.

[Bryce Harrington (bryce)]

@faisal, please don't change status on bugs without an explanation.

[Paride Legovini (paride)]

Hi RedScourge,

the attached patch is a bit dirty, probably because it has been generated by diffing apache2ctl from different package versions, and I think it has a bug: you call

  systemctl start "$APACHE_SYSTEMD_SERVICE"

in the 'restart|graceful)' case stanza, while we want 'systemctl reload' there. I'd also move the need_systemd detection logic out of the "case" to avoid duplication. For the rest I think your proposed solution is essentially correct, thanks for providing it.