Reuti <re...@staff.uni-marburg.de> writes: > Hi, > >> Am 23.12.2016 um 11:58 schrieb Ondrej Valousek <ondrej.valou...@s3group.com>: >> >> Hi List, >> >> Question, I run: >> - Qrsh (to get a shell on remote compute node) >> - “/bin/yes > /dev/null &” (run some bogus detached process there) >> - Control-D – I lose the shell connection, but the job keeps running >> in the ‘r’ state >> - If I run ‘qdel’ to terminate my ‘/bin/yes’ job, the job remains in >> ‘dr’ state forever. I would expect qdel kills the shell as well as >> the spawned /bin/yes job – but this does not happen, why? > > do you have this setting in SGE's configuration: > > $qconf -sconf > … > execd_params ENABLE_ADDGRP_KILL=TRUE > > -- Reuti
That's not actually relevant. It just waits until background tasks finish -- try with sleep, for instance. Offhand, I don't know why, and whether it should be considered a bug. It may depend on whether you use the builtin qrsh support. >> Also, does SoGE have something like ‘qps’ to see processes >> associated with the job id? No. Under Linux, I think the "pdc" program that's intended as a test (not installed) may provide that sort of information, but I can't remember what output it provides, and there isn't a way of interrogating execd for it. If you're going to debug that sort of thing, you probably want a session on the node anyhow. [You may need similar information to attach a debugger, and there's an example of getting that for the padb with openmpi under https://arc.liv.ac.uk/SGE/tools.html] _______________________________________________ SGE-discuss mailing list SGE-discuss@liv.ac.uk https://arc.liv.ac.uk/mailman/listinfo/sge-discuss
