Re: Error with the --progress option

[Ole Tange]

On Fri, Oct 29, 2010 at 9:50 PM, Alastair Andrew
<address@hidden> wrote:
> Hi,
>
> I've just started to use GNU parallel to distribute some of my experiments 
> across my departments Linux lab. They've recently got some quad core machines 
> which are running Ubuntu 10.04. I've placed all the commands I want to 
> execute into a text file and I'm calling parallel as follows:
>
> parallel -S .. -j+0 --progress -a cmds.txt

That ought to work fine.

> I know the -S .. option means parallel looks at the ~/.parallel/sshloginfile.
>
> My sshloginfile is populated with a list of the 30 lab machines' names
> 4/muir-01
> 4/muir-02
> 4/muir-03
> and so on.
>
> I'm always running the process from the same LAN so it doesn't need the full 
> address to resolve the machines. I've prefaced the names with 4/ indicating 
> the machines have 4 cpus for some reason parallels fails to detect this when 
> the reading the machines from the file.

Which version of GNU Parallel is installed on those machines? (run
parallel --version on the machines)

Do you get the same error if you only use one of the 30 machines?

> When I run this process without the --progress option it seems to distribute 
> fine. With the --progress option it spews forth the same errors thousands of 
> times:
> Use of uninitialized value in sprintf at /usr/local/bin/parallel line 3655.
> Sometimes the error is reported at line 3645.
> Also sometimes the error has the additional info:
> Use of uninitialized value in sprintf at /usr/local/bin/parallel line 3655, 
> <GEN0> line 38.

This sounds as if you have an undefined variable for each of the 30
hosts and that GNU Parallel uses those 30 variables every second.

Which version of GNU Parallel is this? (run parallel --version)

Do you get the same error with the newest version from git?
git clone git://git.savannah.gnu.org/parallel.git

> If I explicitly list some of the machines as arguments rather than using -S 
> .. it seems to work: parallel -S muir-02,muir-07 -j+0 --progress -a cmds.txt

If you do the 4/muir-02 as arguments for -S, which result do you get?

> Output:
> Computers / CPU cores / Max jobs to run
> 1:muir-02 / 4 / 4
> 2:muir-07 / 4 / 4
>
> Computer:jobs running/jobs completed/%of started jobs/Average seconds to 
> complete
> muir-02:4/0/50%/0.0s  muir-07:4/0/50%/0.0s
>
> Does anyone have any idea whether the error is a problem on my part or a bug 
> in parallel's progress / cpu parsing routines? If it's useful I've also 
> attached one of the /proc/cpuinfo files

Please attach an example of commands that exhibits the problem. I just
tested with a command file containing:

echo 1
echo 2
echo 3
echo 4
echo 5
echo 6
echo 7
echo 8
echo 9

and """parallel -S .. --progress -j+0 -a /tmp/commands.txt""" did not
give the error you described.


/Ole