Re: [help-gnubatch] Variable use

[Reuti]

Am 13.03.2012 um 14:23 schrieb Wechsler, Steven:

> Trouble with having the slave reset the variable is that there can be 
> multiple slaves for a single master, and they may not all start at the same 
> time.

Aha, you wrote "two jobs", so there are more.

Hence each slave should increase the its own counter by one. The next 
job_master will only start, when the maximum is reached like you intended with 
your delayed separate job idea.

$ gbch-r -c "master=0" -f S -s "slave=0" -f N -s "master=1" test.sh
$ gbch-r -c "master=1" -f N -s "slave+=1" test.sh
$ gbch-r -c "master=1" -f N -s "slave+=1" test.sh
...
$ gbch-r -c "slave=2" -f S -s "master=0" /bin/true

Please replace 2 with the desired number of overall slave instances.

BTW: If you export the variable on one machine, then all nodes should be able 
to access it just by its name.

-- Reuti


> 
> Steve
> 
> -----Original Message-----
> From: Reuti [mailto:address@hidden 
> Sent: Monday, March 12, 2012 6:03 PM
> To: Wechsler, Steven
> Cc: address@hidden
> Subject: Re: [help-gnubatch] Variable use
> 
>> 2.       Many of our jobs are tied to specific database instances, which we 
>> tend to shuffle around from host to host due to performance or space 
>> requirements. Is there a way that variables can be defined as "global" so 
>> that other jobs don't need to know which host variables are located on, or 
>> do I need to just set aside a single host that will hold all variables?
>> 3.       If I want to set up a job dependent on the successful completion of 
>> another job, how can the secondary job tell the difference between the 
>> successful variable setting from yesterday's run and today's run? For 
>> example, let's say I have two jobs, job_master and job_slave. I have 
>> job_master scheduled to run daily, and to set a variable called 
>> MASTER_COMPLETE=1 when it finishes. Job_slave is also scheduled to run 
>> daily, and to only start once it sees MASTER_COMPLETE=1. If job_ master  ran 
>> successfully yesterday, when job_slave is scheduled, it will see that 
>> MASTER_COMPLETE is still set to one, think thatjob_master finished 
>> successfully, and will start running, even though job_master has not yet 
>> run. If this is the case, then I can think of a couple of ways to get around 
>> this problem, but none of them are ideal:
> 
> Also the job_slave could reset the variable MASTER_COMPLETE.
> 
> 
>> a.       Make sure job_slave is scheduled to start after job_master. But can 
>> job_master automatically unset MASTER_COMPLETE when it starts? This solution 
>> is not ideal because it requires that any changes to job_master's scheduled 
>> time also be made to job_slave's scheduled time.
>> b.      Set up a separate job to unset MASTER_COMPLETE at some time after 
>> job_slave finishes running. This is less than ideal because it adds a whole 
>> new layer of complexity regarding all the additional jobs required to manage 
>> this.
>> BMC's Control-M deals with this problem by having each variable get a new 
>> copy at the start of a new day (And the new day can be defined to be any 
>> time of day). Variables from previous days can also be referenced.
> 
> $ gbch-r -f S -s 'MASTER_COMPLETE=0' -f N -s 'tMASTER_COMPLETE=1' job_master 
> $ gbch-r -c 'MASTER_COMPLETE=1' -f N -s MASTER_COMPLETE=0' job_slave
> 
> At start the master will reset the variable for safety reasons, and to one on 
> its normal completion.
> The slave will also reset it upon exit.
> 
> You could also implement a small finite state machine, where MASTER_COMPLETE 
> shows the actual state:
> 
> MASTER_COMPLETE=0 => idle
> MASTER_COMPLETE=1 => running master
> MASTER_COMPLETE=2 => master ended
> MASTER_COMPLETE=3 => running slave
> MASTER_COMPLETE=0 => slave ended
> 
> $ gbch-r -c 'MASTER_COMPLETE=0' -f S -s 'MASTER_COMPLETE=1' -f N -s 
> 'MASTER_COMPLETE=2' job_master $ gbch-r -c 'MASTER_COMPLETE=2' -f S -s 
> MASTER_COMPLETE=3' -f N -s 'tMASTER_COMPLETE=0' job_slave
> 
> -- Reuti
> 
>