Bug 162320 – mysql initscript lacks dependancy on net.lo

• Product: Gentoo Linux
• Status: RESOLVED NEEDINFO

[Description Peter H 2007-01-16 05:00:04 UTC]

If there is a net.* service started (other than net.lo), MySQL will fail to start with not so much as an error message.  It doesn't help that the logs are rather cryptic and appear unrelated to the actual error at hand.  Surely this should depend on the loopback connection specifically.  I'm not sure if it still needs the rest of net for the initscript.

Reproducible: Always

Steps to Reproduce:
1. Stop net.lo
2. Attempt to start mysql
3. Puzzle over logs

Actual Results:  
# /etc/init.d/mysql start
 * Starting mysql ...
 * Starting mysql (/etc/mysql/my.cnf)
 * MySQL NOT started (0)                                                                                                                                                   [ !! ]


Expected Results:  
I was expecting net.lo to be pulled in as a dependency.

I removed net.lo from the boot sequence as it was locking with udev.  I'm thinking it really belongs in 'default'?  I'm not sure exactly.  Advice there might also be appreciated.

[Comment 1 Roy Marples (RETIRED) 2007-01-16 07:24:13 UTC]

(In reply to comment #0)
> I removed net.lo from the boot sequence as it was locking with udev.  I'm
> thinking it really belongs in 'default'?  I'm not sure exactly.  Advice there
> might also be appreciated.

Well, duh, that's the error. net.lo does belong in the boot runlevel.
As udev starts before anything else, how was net.lo blocking udev?

[Comment 2 Jakub Moc (RETIRED) 2007-05-05 09:13:57 UTC]

Well, get back to us please.

[Comment 3 Peter H 2007-05-12 02:33:12 UTC]

Sorry, I had almost completely forgotten about this issue.  I had been needing to physically move the box around a lot, and it had proved quite annoying, but it's been up 68 days until today, when I rebooted it to get more info.  I'm pretty sure it's mostly that the boot process (and shutdown) are pretty messed up.

# rc-update show
           alsasound | boot
            bootmisc | boot
             checkfs | boot
           checkroot | boot
               clock | boot
         consolefont | boot default
                 gpm |      default
            hostname | boot
             hotplug |      default
             keymaps | boot
               local |      default nonetwork
          localmount | boot
             modules | boot
               mysql |      default
            netmount |      default
          ntp-client |      default
             numlock |      default
           rmnologin | boot
                sshd |      default
           syslog-ng |      default
             urandom | boot
          vixie-cron |      default

With net.lo added to boot, the boot sequence goes something as follows:
* Setting user font ... (I have to ^C here)
* ERROR: net.lo caught an interupt
* ERROR: cannot start net.eth0 as net.lo could not start
[a few other steps]
* Starting gpm ... (completely hangs here)

Without net.lo (which requires chrooting from a livecd to change, so I'm not inclined to do too much testing...):
* Setting user font ... (I have to ^C here)
* ERROR: net.eth0 caught an interrupt
[continues booting to login shell]

I can manually start all of these services, but I don't know why it's breaking on boot.  For what it's worth, it also hangs on stopping cupsd while shutting down.  I do find it interesting that sending an interrupt stops the next process, so I'm not exactly sure what it's hanging on.g

[Comment 4 Robin Johnson 2007-05-12 03:13:51 UTC]

please setup RC_BOOTLOG on your system and provide the logfiles.
the mysql init scripts do correctly specify 'use net', to indicate that they can use the network, but in many cases don't actually need it (unix sockets instead).

[Comment 5 Robin Johnson 2007-05-13 01:06:26 UTC]

to exclude a potential other bug that may have been causing this strangeness for you, please upgrade to mysql-5.0.40 and retest.

[Comment 6 Jakub Moc (RETIRED) 2008-03-20 16:35:16 UTC]

Once again, please get back to us.