Re: duplicate processes

[Charlie Brady <charlieb-supervision@budge.apana.org.au>]

On Tue, 27 Sep 2005, Jussi Ramo wrote:

>> What evidence do you have that the processes are started directly by 
> init? Remember that a process will be inherited by init if its direct 
>> parent dies.
>
> No evidence. Just looked at the parent process.

I'm pretty sure that's misleading.

> So you suggest that the "runsv ndb_mgmd" dies and the ndb_mgmd is 
> inherited by init. Then "runsv ndb_mgmd" is respawned by runsvdir (?) 
> and that starts another ndb_mgmd. This makes sense to me but now the 
> question is why runsv first dies once for certain processes.

No, I didn't mean to suggest that runsv dies. I expect that ndb_mgmd has 
forked, and the parent died. Perhaps it was designed to do that, to 
"daemonise" the child. You will have to find some way to prevent that from 
happening.

>>> So duplicate process will be generated and system becomes unstable. 
> Both of those processes (the one started by init and the one started by 
> runsv) react on sv command.
>> React in what ways?
>
> Do not know if this brings any extra information but if I have first the 
> following ndb_mgmd processes (one of them badly as child of init) :
>
> root      1950  1945  0 07:04 ?        00:00:00 runsv ndb_mgmd
> ais       1963     1  0 07:04 ?        00:00:00 /opt/SGC/bin/ndb_mgmd -f 
> /opt/SGC/etc/ndbconfig.ini
> ais       2276  1950  2 07:05 ?        00:00:00 /opt/SGC/bin/ndb_mgmd -f 
> /opt/SGC/etc/ndbconfig.ini
>
> Then I do like:
>
> blade_0_7:~ # /opt/SGC/bin/sv down /var/services/ndb_mgmd/
>
> and the other "right" ndb_mgmd disappears:
>
> root      1950  1945  0 07:04 ?        00:00:00 runsv ndb_mgmd
> ais       1963     1  0 07:04 ?        00:00:00 /opt/SGC/bin/ndb_mgmd -f 
> /opt/SGC/etc/ndbconfig.ini

OK.

> I then kill the "wrong" ndb_mgmd
>
> blade_0_7:~ # kill -9 1963
> root      1950  1945  0 07:04 ?        00:00:00 runsv ndb_mgmd

OK, but you shouldn't be using -9 unless there is no alternative.

> And when ndb_mgmd is put "up" there are again those two processes:
>
> blade_0_7:~ # /opt/SGC/bin/sv up /var/services/ndb_mgmd/
>
> root      1950  1945  0 07:04 ?        00:00:00 runsv ndb_mgmd
> ais       2805     1  0 07:08 ?        00:00:00 /opt/SGC/bin/ndb_mgmd -f 
> /opt/SGC/etc/ndbconfig.ini
> ais       2837  1950  4 07:08 ?        00:00:00 /opt/SGC/bin/ndb_mgmd -f 
> /opt/SGC/etc/ndbconfig.ini

OK. Assuming that you have pids allocated sequentially, look how many 
intervening processes there are between the two invocations.

>>> The other ndb_mgmd is restarted frequently by runsv because of the 
> same process is started directly by init for some reason.
>> Again, what evidence do you have that there is any process started 
> directly by init? Even if so, why would runsv restart the process it is 
> managing? I expect that the process runsv is monitoring is exiting, and that 
> is why runsv is starting a new process.
>
> Right. The process runsv is monitoring is exiting because the port it tries 
> to use is reserved by the extra process (now hopefully correct phrasing:) 
> whose parent is init.

Not quite. The process runsv is monitoring exits when told to by runsv. 
You showed us that. The other ndb_mgmd process however doesn't exit, and 
runsv will therefore not be able to start a new process which is able to 
keep running.

Does ndb_mgmd create a pid file? If so, what is its content?