GNU Parallel Bug Reports Re: GNU Parallel reports Deep recursion on subroutine "main::get_job_with_sshlogin"

[Ole Tange]

On Thu, Jan 20, 2011 at 5:32 PM, Ole Tange <address@hidden> wrote:
> 2011/1/20  <address@hidden>:
>>> 2011/1/20  <address@hidden>:
:
>>>> One more comment - if I kill parallel, programs at computers
>>>> where it was spreading overs ssh keep running. Which
>>>> might cause troubles if you need to restart computation
>>>> as the computers you want to use are already computing
>>>> the previous, but now killed, computation.
>>>
>>> Please show an example that shows this behaviour.
>>
>> The provided example is exhibiting the behaviour. Only
>> with sleep 10 it is not as obvious. But with sleep 1000
>> the process at target machines are just waiting for the sleep to finish.
>> Although the main command is closed.
>
> I can reproduce this, too: https://savannah.gnu.org/bugs/index.php?32193
>
> The workaround for this is to kill GNU Parallel with CTRL-C.

The cause of this is ssh: Signals (such as kill) are only passed
through ssh if ssh is started with a terminal. It is possible to force
ssh to grab a terminal (using -tt), however, this will cause the
output to be changed (\n -> \r\n) which is unacceptable if the output
of the computation is important.

So the solution is to have ssh fixed, which apparently takes a looong time:

https://bugzilla.mindrot.org/show_bug.cgi?id=396

After that has been fixed GNU Parallel will probably work without any
changes. If it does not work, re-open this bug:

https://savannah.gnu.org/bugs/?32193

/Ole