Re: 4.0 showstopper?

[Martin Pala]

Jan-Henrik Haukeland wrote:

> Martin Pala <address@hidden> writes:
>
>  
>
> > during "lastminute" tests i saw following problem:
> >
> > [CEST Sep 17 14:51:46] AssertException: at socket.c:333
> > aborting..
> >    
>
> Oops! 
I identified the problem - it is caused by race condition between 
methods execution:

1.) monit detected that the process is not running and main monit thread 
forks the start method (separate process, which will inherite all 
filedescriptors)

2.) monit main thread creates new thread which waits for the service to 
start - in the case that the service will not start (timeout will 
occure), this thread posts timeout event, which causes alert and 
continues by smtp server connection and sending the message. The 
fildescriptor of socket opened to smtp server is shared between all 
threads (wait_start and main)

3.) while wait_start is waiting for service to start, monit main thread 
executed another validate cycle and detected that the process is not 
running (independently of wait_start thread) - as usual, new process 
inherites all open filedescriptors (including smtp server socket fd)


Solution:

- secure runtime filedescriptors from inheritance by forked process
- synchronize main and wait_start thread to not check the service which 
is in wait_start stage. This is standalone problem - monit can try to 
start the service in paralel without realy waiting for service to start. 
In addition it allows the fd race condition in the case of alert message.

In theory this race condition can affect all filedescriptors - for 
example in the case that there is connection to monit httpd server 
(filedescriptor of the client connection will be inherited the same way 
as described above => the consequences will be probably the same - 
assert exception on socket write will shutdown monit).//

>  
>
> > I'll look on it ...
> >    
>
> Super! We'll wait until tomorrow then so we can add the "-i"
> replacement doc as well to the release. I can add this doc, if it is
> okay with you.
>
>  
Thanks :)


Martin