- 15 Sep, 2017 40 commits
-
-
Alejandro Sanchez authored
Bug 2329
-
Alejandro Sanchez authored
-
Alejandro Sanchez authored
sview Watts parsing and display was introduced in commit a503a770. There were differences between scontrol and sview, so this commit moves the way we parse and display reservation Watts to a common place and make it so sview and scontrol use these functions. The parse and display functions have been slightly refactored as compared to the original ones. Bug 2329
-
Alejandro Sanchez authored
-
Alejandro Sanchez authored
Now that _parse_resv_tres is already available in src/common/state_control, we can enable an EDIT_TEXTBOX for the SORTID_TRES and call the function there. This is the original goal of the bug. Bug 2329
-
Alejandro Sanchez authored
And make src/scontrol/create_res.c use it. Bug 2329
-
Alejandro Sanchez authored
Bug 2329
-
Alejandro Sanchez authored
Bug 2329
-
Alejandro Sanchez authored
Bug 2329
-
Alejandro Sanchez authored
Testing with valgrind I found some cases where memory leaked or program got SIGSEV. This was due to bad memory tracking of these resv_desc_msg_t members. Some example requests that now end up with the expected behavior and no memory errors nor leaked memory: scontrol create reservationname=test start=now end=now+1day users=alex ... - nodecnt=2 - nodecnt=badvalue - nodecnt=2 tres=node=1 - nodecnt=2 tres=node=badvalue - corecnt=1 - corecnt=badvalue - corecnt=1 tres=cpu=2 - corecnt=1 tres=cpu=badvalue sview reservation creation/update. Ideally we should use slurm_free_resv_desc_msg() instead of tracking each resv_desc_msg_t member separately, but this function unconditionally xfrees each member, even if they have a value but memory was not allocated, thus receiving a SIGABRT. Bug 2329
-
Alejandro Sanchez authored
resv_msg_ptr->core_cnt memory needed to be freed before returning, since it was previously allocated by xrealloc(). Analogous to commit 8063a08cbf2f.
-
Alejandro Sanchez authored
Bug 2329
-
Alejandro Sanchez authored
_parse_resv_core_cnt has been moved to src/common/state_control.[ch]. Furthermore, it has been split in two functions: - _parse_resv_core_cnt - _is_corecnt_supported Bug 2329
-
Alejandro Sanchez authored
Bug 2329
-
Alejandro Sanchez authored
Bug was introduced in 22c9e25c. Stack trace: Thread 1 (Thread 0x7ff64228c300 (LWP 6652)): ../sysdeps/unix/sysv/linux/raise.c:58 fmt=fmt@entry=0x7ff6409bb000 "*** Error in `%s': %s: 0x%s ***\n") at ../sysdeps/posix/libc_fatal.c:175 ptr=<optimized out>, str=0x7ff6409bb110 "double free or corruption (out)", action=3) at malloc.c:5048 have_lock=<optimized out>) at malloc.c:3904 malloc.c:2984 new_text=0x561a4f8b4d10 "asfasdf", column=21) at ../../../../slurm/src/sview/resv_info.c:331 event=0x561a4f8a8f10, resv_msg=0x561a4f8b1410) at ../../../../slurm/src/sview/resv_info.c:515 Bug 2329
-
Alejandro Sanchez authored
resv_msg_ptr->node_cnt memory needed to be freed before returning, since it was previously allocated by xrealloc().
-
Alejandro Sanchez authored
The scontrol and sview clients print errors in different ways (i.e. error() vs global_edit_error_msg respectively). By adding this err_msg parameter to the common functions we decouple the message itself from the way we print it. Bug 2329
-
Alejandro Sanchez authored
And make it so src/scontrol/create_res.c makes use of it. Also set exit_code = 1 when there's an error in the calling function instead of the function itself. Bug 2329
-
Alejandro Sanchez authored
-
Alejandro Sanchez authored
Different clients provide similar functionality such as scontrol and sview. These new files are intended to hold common functions that can be used by these clients. This will help to avoid behavior divergence across them and stop duplicating code. src/common/Makefile.am libcommon_la_SOURCES have been modified accordingly, and a following autogen.sh execution should be committed right after to reflect the changes in the generated src/common/Makefile.in file. Bug 2329
-
Tim Wickberg authored
-
Tim Wickberg authored
-
Tim Wickberg authored
-
Tim Wickberg authored
Letting the thread_id go out of context means that _safe_signal_while_allocating should be detached or we leak the return value.
-
Tim Wickberg authored
Since the thread_id is never referenced elsewhere, don't bother storing it in the struct.
-
Tim Wickberg authored
Letting the thread_id go out of scope leaks the thread's return value. So this should be a detached thread instead.
-
Tim Wickberg authored
-
Tim Wickberg authored
-
Tim Wickberg authored
-
Tim Wickberg authored
Remove references to deprecated/obsolete/removed MPI stacks, and generally tidy things up.
-
Tim Wickberg authored
-
Tim Wickberg authored
Remove plugins for: - lam - mpich1_p4 - mpich1_shmem - mvapich All modern MPI stacks should be talking PMI2 / PMIx, these are all obsolete and should have been retired long ago.
-
Tim Wickberg authored
-
Tim Wickberg authored
-
Tim Wickberg authored
Rather than creating a detached thread and using pthread_kill to stop it (which could lead to data loss), make an attached thread, signal it as appropriate then wait on proper shutdown with pthread_join.
-
Tim Wickberg authored
-
Tim Wickberg authored
-
Tim Wickberg authored
-
Tim Wickberg authored
-
Tim Wickberg authored
-