- 14 Aug, 2018 5 commits
-
-
Danny Auble authored
-
Felip Moll authored
Bug 5453
-
Tim Wickberg authored
-
Tim Wickberg authored
Causes srun and sattach to fail on big endian platforms. Bug 5556
-
Tim Wickberg authored
Wrong author. This reverts commit 392e7cfc.
-
- 13 Aug, 2018 8 commits
-
-
Isaac Hartung authored
a job with scontrol. Bug 4964
-
Danny Auble authored
instead of an int for protocol_version. Continuation of 392e7cfc Bug 5556
-
Danny Auble authored
Bug 5556
-
Danny Auble authored
-
Tim Wickberg authored
not work when trying use sattach. Bug 5556
-
Danny Auble authored
to run the regression tests 24.* without changing that code since it links directly to the priority plugin where that function isn't defined. Bug 5558 Continuation of commit 21d2ab6e
-
Morris Jette authored
-
Brian Christiansen authored
With the change to use tres_usage_type's the stats coming back from the extern step are initialized to INFINITE64 instead of 0's. jobacctinfo_create() in _handle_stat_jobacct() initializes jobacct, but since the job->task[i]->pid == -1, no stats are gathered and the clean jobacct is returned. Related commit: fa3a8ff1 Bug 5554
-
- 12 Aug, 2018 27 commits
-
-
Tim Wickberg authored
-
Tim Wickberg authored
Most calling paths to job_signal() were sending in job_ptr->job_id alongside uid 0. This bypasses the unneeded uid validation, and forces another lookup through find_job_record() for a job_ptr we just dereferenced. So - change the less-used job_signal() to job_signal_id(), promote _job_signal to an extern taking that name over, and stop this whole unnecessary extra set of lookup calls.
-
Tim Wickberg authored
-
Tim Wickberg authored
-
Tim Wickberg authored
-
Tim Wickberg authored
Break JobId=... out from this, and only print the StepId=... bit. This allows us to use %pJ in several additional places, and follows the newer formatting used elsewhere of JobId=... StepId=... . Due to how this code works, the end result is admittedly a bit uglier, but the end result as printed to the logs is preferrable.
-
Tim Wickberg authored
-
Tim Wickberg authored
-
Tim Wickberg authored
Add xasserts() to crash development builds if the wrong struct is fed in, and fall back to checking the field and returning 'CORRUPT' if the magic value is not found for production builds rather than print garbage (and risk an out-of-bounds memory access).
-
Tim Wickberg authored
Set when step is created, xassert() when the step is being destroyed.
-
Tim Wickberg authored
Give the xassert a chance to fail instead of an invalid memory access. Also help avoid issues with new logging code if the wrong struct was passed in.
-
Tim Wickberg authored
Need to set found_other_formats to still consume the %.0s if not other format strings are included.
-
Tim Wickberg authored
Use the embedded pointer back to the job_record struct to print that followed by the step_id value.
-
Tim Wickberg authored
-
Tim Wickberg authored
Also switch to __func__ and plugin_type where applicable.
-
Tim Wickberg authored
-
Tim Wickberg authored
-
Tim Wickberg authored
-
Tim Wickberg authored
Further work to convert StepId structures will be handled later.
-
Tim Wickberg authored
-
Tim Wickberg authored
-
Tim Wickberg authored
Do this instead of looking the job_id back up immediately. Convert this to %pJ while here, and fix incorrect comments for srun_allocate_abort().
-
Tim Wickberg authored
Rather than pulling the job_id out of job_ptr just to have it call find_job_record() to find it again, just cut out the indirection.
-
Tim Wickberg authored
Switch a few fatal() log messages to fatal_abort() where appropriate.
-
Tim Wickberg authored
-
Tim Wickberg authored
Permits use of %pJ. Convert the upstream calling location _preempt_jobs() to %pJ while changing its signature as well.
-
Tim Wickberg authored
-