Page MenuHomePhabricator

jsub should respect .sge_request
Closed, DeclinedPublic


I have $HOME/.sge_request where I set "-l release=trusty" and other common configurations for submitted tasks. This allows me to use "jsub" on trusty without having to add "-l release=trusty" every time.

Apparently after T143282, this does not work and gets overwritten with the default "-l release=precise" instead.

It would also be nice if jsub does not overwrite the "-l h_vmem=" value in .sge_request.

Event Timeline

Restricted Application added a subscriber: Aklapper. · View Herald Transcript

If you want to use advanced SGE functionalities (which includes .sge_request), please use qsub directly instead of jsub.

If jsub cannot (or does not want to) support .sge_request, maybe it should entirely ignore .sge_request?

The current state is that some values from .sge_request are overwritten while others (those with no default values defined in jsub) go through - I find this confusing.

Jsub internally calls qsub, which is why .sge_request et al are taken into account. I'm not sure if we can make qsub ignore .sge_request.


valhallasw@tools-bastion-02:~$ sudo ls /home/*/.sge_request
valhallasw@tools-bastion-02:~$ sudo ls /data/project/*/.sge_request

i.e. you and @Merl are the only ones using .sge_request, so I think my time is better spent convincing you to use qsub than to create and support a feature in jsub to ignore .sge_request :-)

I have tried switching to qsub, but it is harder than I imagined to stop using jsub entirely, particularly for cron jobs. For example, I would have to modify my scripts to emulate the "-once" behavior of jsub.

That said, I'm fine with this being considered a low-priority task - I guess it's just another nice thing to have.

yuvipanda added a subscriber: yuvipanda.

We don't really want to add more gridengine specific things to jsub.

As for -l release=trusty, that'll become the default soon!

.jsubrc was what I was looking for - I really should have read the manual.