Re: Monit definitely not waiting for stop scripts to complete...

[Jan-Henrik Haukeland]

It should be fairly easy to change start to also become  
"synchronously". Synchronously in this context means that monit will  
wait in the same thread which started the process and poll-check for  
a pidfile to be created. This is how synchronously stop is handled  
today.

The rationale for not waiting for a process to actually start before  
moving on to the next is, I believe, that we thought if was best to  
start as soon as possible. Waiting and poll-check for a pidfile could  
take time in case of failure to start, before monit timed out and  
moved on to start the next process. With many services to check,  
there would be a potential for monit's poll cycle to take "a long  
time" in case of problems.

Today the solution is to start as soon as possible and instead spin  
of one wait-thread per started process and check if the process did  
start and if not, report an error.

Of course, this solution does not play too well with a dependency  
chain, particularly if you expect A to come up before monit starts B  
etc. Now monit just start A and B more or less simultaneously.

1) Maybe start time is not the crucial point here, but rather  
"correctness", that is, monit should wait for A to come up before  
starting B and so on?

2) Apropos correctness. This brings up another question, if you have  
a dependency chain, say, C->B->A, today if monit should fail to start  
A, it still goes on and try to start B and C. If A failed to start,  
should monit instead abort the chain and not try to start B and C  
regardless? I'm not sure.

I do not know if changing start to also become synchronously will  
have unwanted side effects, I don't think so? Martin?

In case we decide to do this change, do you have a good test case and  
would be willing to test?

What do the other comitters think? Should we change 1) and 2) above?

Best regards
--
Jan-Henrik Haukeland


On 5. sep. 2007, at 15.42, Stanisław Trytek wrote:

> From our experience monit works properly when closing services. It  
> closes services in accordance with declared dependieces and  
> services wait for another.
> Your problem may be caused by stdout, it can happen.
> The problem is with starting services. In spite of dependencies  
> monit starts them at once, and it causes mess in some environments.
> It would be great if both starting and stoping work synchronusly.  
> We made some attempts to overcome this issue.
> Stanislaw Trytek
> --
> _________________________________________
> PRZEMYSŁOWY INSTYTUT TELEKOMUNIKACJI
> TELECOMMUNICATIONS RESEARCH INSTITUTE
> ul. Poligonowa 30,  04-051 Warszawa,  Poland
> tel: +48 (22) 48 65 232
> fax: +48 (22) 48 65 494
> e-mail: address@hidden
>
>
> On Tue, 04 Sep 2007 03:21:13 +0200, Jason L. Buberel  
> <address@hidden> wrote:
>
> > My configuration is using monit 4.10-beta-1 on RHEL3. I have  
> > configured a dependency chain for squid and apache:
> >
> > squid ---depends-on--> apache
> >
> > Squid in particular can take a while to shutdown. In fact the  
> > official squid init.d script contains a section that looks like this:
> >
> > timeout=0
> > while : ; do
> >     [ -f /opt/squid/var/logs/squid.pid ] || break
> >     if [ $timeout -ge $SQUID_SHUTDOWN_TIMEOUT ]; then
> >         echo
> >         return 1
> >     fi
> >     sleep 2 && echo -n "."
> >     timeout=$((timeout+2))
> > done
> >
> > When I issue a 'monit restart apache' command, I see the following  
> > in stdout:
> >
> > 'squid' stop: /etc/init.d/squid.local
> > Stopping squid: 'apache' stop: /usr/local/bin/stop-apache.bash
> > .                                                          [  OK  ]
> >
> > Which tells me that the 'apache stop' is being executed before the  
> > 'squid stop' has actually completed. If the commands were executed  
> > serially, the output would have been:
> >
> > 'squid' stop: /etc/init.d/squid.local
> > Stopping squid: ......                     [ OK ]
> > 'apache' stop: /usr/local/bin/stop-apache.bash
> >
> > This seems similar to what is described:
> > http://lists.nongnu.org/archive/html/monit-general/2007-04/ 
> > msg00007.html
> >
> > Would it be possible to add a 'synchronous' flag to the start/stop  
> > parameters to force monit to wait for the script to exit fully:
> >
> > start synchronous "/etc/init.d/service start"
> > stop synchronous "/etc/init.d/service stop"
> >
> > Or something to that effect?
> >
> > -jason
> >
> >
> >
> >
> >
> >
> > --
> > To unsubscribe:
> > http://lists.nongnu.org/mailman/listinfo/monit-general
>
>
>
> --
> To unsubscribe:
> http://lists.nongnu.org/mailman/listinfo/monit-general