Re: Question on the dependency of processes

[Martin Pala]

Yiwen Jiang wrote:
> Hi there,
>
> I sent a couple of emails to the general mailing list under the same 
> subject line, but got busy in some other things. I finally got some time 
> to look at monit again. And I am in the process of building a test case 
> as requested, and will keep you posted on the results.
>
> However, I noticed that in the 'Dependencies' section of the monit 
> manual, there was an example of process start/stop/crashed scenario. In 
> the example, there was not a mentioning of multiple process crash 
> scenarioes. In fact, in the multiple crash/exit scenario, the start up 
> sequence is different from a clean start scenario. I am starting to 
> wonder if what I'm using monit for is indeed its design intent.
>
> Using the following server startup as an example:
>
> WEB-SERVER -> APPLICATION-SERVER -> DATABASE -> FILESYSTEM
>       (a)               (b)             (c)          (d)
>
> If b does not run, monit will "first stop a then start b and finally 
> start a again".  Cool.
>
> The behaviour changes quite a bit if both a and b are found to be not 
> running by monit:
> Because of the way monit is implemented, a will be started first, and 
> then b. This is in conflicting with the 'If no servers are running' 
> behaviour, where as b is started before a.

No.

In the case that both processes are stopped, two cases are possible 
depending on the rank of service during testing cycle ('a' before 'b' or 
vice versa):

1.) monit tested 'a' before 'b':
It will first try to START 'b' and then RESTART 'a'

2.) monit tested 'b' before 'a':
It will first try to STOP 'a', then RESTART 'b', then START 'a'


> While servers should be designed to accommodate the disappearing of its 
> own communication parties, there are applications that do not behave 
> like this. Instead, they count on process dependencies framework (i.e. 
> restarting dependant processes) to ensure the proper re-establishment of 
> communication once the most depending process is restarted.
>
> Is monit designed to handle cases like this? Or is there a pre-requisite 
> where if process a depends on process b, and if b is down (crashed, 
> killed, etc), a will remain up and running?

In the case that 'b' crashed, monit will stop 'a' ... see above


Martin