From cf0696c52a394afb81c064f2a8c101dfb33a9d86 Mon Sep 17 00:00:00 2001
From: Mehdi Dogguy <mehdi@debian.org>
Date: Mon, 8 Sep 2014 21:33:39 +0200
Subject: [PATCH] Imported Upstream version 2.1.8
---
BUILD.NOTES | 8 +-
META | 4 +-
NEWS | 39 +++-
contribs/lua/proctrack.lua | 81 ++++---
doc/html/accounting.shtml | 2 +-
doc/html/team.shtml | 4 +-
doc/man/man1/sacct.1 | 8 +
doc/man/man1/sacctmgr.1 | 3 +
doc/man/man1/salloc.1 | 4 +-
doc/man/man1/sbatch.1 | 4 +-
doc/man/man1/scontrol.1 | 19 +-
doc/man/man1/srun.1 | 4 +-
doc/man/man5/slurm.conf.5 | 3 +-
slurm.spec | 6 +-
slurm/slurm.h.in | 9 +-
src/api/job_info.c | 5 +-
src/common/plugstack.c | 8 +-
src/common/proc_args.c | 27 ++-
src/common/slurm_resource_info.c | 33 ++-
src/common/slurm_resource_info.h | 5 +-
src/common/slurmdbd_defs.c | 5 +
src/database/mysql_common.h | 1 +
.../mysql/accounting_storage_mysql.c | 68 ++++--
.../mysql/mysql_jobacct_process.c | 15 +-
src/plugins/proctrack/lua/proctrack_lua.c | 200 +++++++++++++----
src/plugins/sched/backfill/backfill.c | 58 ++++-
src/plugins/sched/wiki2/job_modify.c | 64 +++++-
.../select/bluegene/plugin/bg_block_info.c | 8 +-
.../select/bluegene/plugin/bg_job_place.c | 210 +++++++-----------
.../select/bluegene/plugin/bg_job_run.c | 6 +-
.../bluegene/plugin/bg_record_functions.c | 28 +++
.../bluegene/plugin/bg_record_functions.h | 1 +
.../select/bluegene/plugin/dynamic_block.c | 10 +-
.../select/bluegene/plugin/state_test.c | 10 +-
src/plugins/select/cons_res/dist_tasks.c | 43 +++-
src/plugins/select/cons_res/job_test.c | 86 +++++--
src/plugins/select/cons_res/select_cons_res.c | 6 +-
src/plugins/select/linear/select_linear.c | 16 +-
src/sacct/options.c | 12 +-
src/sacctmgr/account_functions.c | 7 +-
src/sacctmgr/archive_functions.c | 6 +-
src/sacctmgr/cluster_functions.c | 3 +-
src/sacctmgr/file_functions.c | 6 +-
src/sacctmgr/qos_functions.c | 3 +-
src/sacctmgr/user_functions.c | 16 +-
src/salloc/opt.c | 17 +-
src/salloc/salloc.c | 21 +-
src/sbatch/opt.c | 17 +-
src/sbatch/sbatch.c | 21 +-
src/slurmctld/job_mgr.c | 119 ++++------
src/slurmctld/job_scheduler.c | 19 +-
src/slurmctld/node_mgr.c | 26 ++-
src/slurmctld/node_scheduler.c | 53 +++--
src/squeue/sort.c | 3 +
src/srun/opt.c | 46 +++-
src/srun/opt.h | 6 +-
src/srun/srun.c | 5 +-
src/sview/block_info.c | 7 +
testsuite/expect/test7.7.prog.c | 2 +-
59 files changed, 1038 insertions(+), 488 deletions(-)
diff --git a/BUILD.NOTES b/BUILD.NOTES
index 20ef2e186..46285a0d5 100644
--- a/BUILD.NOTES
+++ b/BUILD.NOTES
@@ -63,7 +63,12 @@ Linux cluster (See BlueGene and AIX specific notes below for some differences).
Use SourceForge admin tool to add new release, including changelog.
BlueGene build notes:
-3. Use the rpm make target to create the new RPMs. This requires a .rpmmacros
+0. If on a bgp system and you want sview export these variables
+ export CFLAGS="-I/opt/gnome/lib/gtk-2.0/include -I/opt/gnome/lib/glib-2.0/include $CFLAGS"
+ export LIBS="-L/usr/X11R6/lib64 $LIBS"
+ export CMD_LDFLAGS='-L/usr/X11R6/lib64'
+ export PKG_CONFIG_PATH="/opt/gnome/lib64/pkgconfig/:$PKG_CONFIG_PATH"
+1. Use the rpm make target to create the new RPMs. This requires a .rpmmacros
(.rpmrc for newer versions of rpmbuild) file containing:
%_prefix /usr
%_slurm_sysconfdir /etc/slurm
@@ -89,6 +94,7 @@ To build and run on AIX:
then making the buildfarm directory the first one in my PATH.
Also, make certain that the "proctrack" rpm is installed.
1. export OBJECT_MODE=32
+ export PKG_CONFIG="/usr/bin/pkg-config"
2. Build with:
./configure --enable-debug --prefix=/opt/freeware \
--sysconfdir=/opt/freeware/etc/slurm \
diff --git a/META b/META
index b9f60e141..3ef94ea78 100644
--- a/META
+++ b/META
@@ -3,9 +3,9 @@
Api_revision: 0
Major: 2
Meta: 1
- Micro: 7
+ Micro: 8
Minor: 1
Name: slurm
Release: 1
Release_tags: dist
- Version: 2.1.7
+ Version: 2.1.8
diff --git a/NEWS b/NEWS
index 85f380304..85c97fe69 100644
--- a/NEWS
+++ b/NEWS
@@ -3,6 +3,43 @@ documents those changes that are of interest to users and admins.
* Changes in SLURM 2.1.8
========================
+ -- Update BUILD_NOTES for AIX and bgp systems on how to get sview to
+ build correctly.
+ -- Update man page for scontrol when nodes are in the "MIXED" state.
+ -- Better error messages for sacctmgr.
+ -- Fix bug in allocation of CPUs with select/cons_res and --cpus-per-task
+ option.
+ -- Fix bug in dependency support for afterok and afternotok options to insure
+ that the job's exit status gets checked for dependent jobs prior to puring
+ completed job records.
+ -- Fix bug in sched/backfill that could set an incorrect expected start time
+ for a job.
+ -- BLUEGENE - Fix for systems that have midplanes defined in the database
+ that don't exist.
+ -- Accounting, fixed bug where if removing an object a rollback wasn't
+ possible.
+ -- Fix possible scontrol stack corruption when listing jobs with very a long
+ job or working directory name (over 511 characters).
+ -- Insure that SPANK environment variables set by salloc or sbatch get
+ propagated to the Prolog on all nodes by setting SLURM_SPANK_* environment
+ variables for srun's use.
+ -- In sched/wiki2 - Add support for the MODIFYJOB command to alter a job's
+ comment field
+ -- When a cluster first registers with the SlurmDBD only send nodes in an
+ non-usable state. Before all nodes were sent.
+ -- Alter sacct to be able to query jobs by association id.
+ -- Edit documentation for scontrol stating ExitCode as something not alterable.
+ -- Update documentation about ReturnToService and silently rebooting nodes.
+ -- When combining --ntasks-per-node and --exclusive in an allocation request
+ the correct thing, giving the allocation the entire node but only
+ ntasks-per-node, happens.
+ -- Fix accounting transaction logs when deleting associations to put the
+ ids instead of the lfts which could change over time.
+ -- Fix support for salloc, sbatch and srun's --hint option to avoid allocating
+ a job more sockets per node or more cores per socket than desired. Also
+ when --hint=compute_bound or --hint=memory_bound then avoid allocating more
+ than one task per hyperthread (a change in behavior, but almost certainly
+ a preferable mode of operation).
* Changes in SLURM 2.1.7
========================
@@ -5000,4 +5037,4 @@ documents those changes that are of interest to users and admins.
-- Change directory to /tmp in slurmd if daemonizing.
-- Logfiles are reopened on reconfigure.
-$Id: NEWS 20049 2010-04-16 00:34:17Z da $
+$Id: NEWS 20218 2010-05-07 21:29:56Z jette $
diff --git a/contribs/lua/proctrack.lua b/contribs/lua/proctrack.lua
index 61e6b45f9..29b23701b 100644
--- a/contribs/lua/proctrack.lua
+++ b/contribs/lua/proctrack.lua
@@ -14,6 +14,8 @@ require "posix"
--
--########################################################################--
+local use_release_agent = false
+
function slurm_container_create (job)
local id = cpuset_id_create (job)
local cpu_list = cpumap:convert_ids (job.CPUs)
@@ -32,7 +34,7 @@ end
function slurm_container_signal (id, signo)
log_verbose ("slurm_container_signal(%d, %d)\n", id, signo)
cpuset_kill (id, signo)
- return 0
+ return slurm.SUCCESS
end
function slurm_container_destroy (id)
@@ -49,7 +51,7 @@ function slurm_container_find (pid)
return id
end
end
- return -1
+ return slurm.FAILURE
end
function slurm_container_has_pid (id, pid)
@@ -69,7 +71,7 @@ function slurm_container_wait (id)
posix.sleep (s)
s = (2*s <= 30) and 2*s or 30 -- Wait a max of 30s
end
- return 0
+ return slurm.SUCCESS
end
function slurm_container_get_pids (id)
@@ -135,6 +137,9 @@ function cpuset_create (name, cpus)
posix.umask (mask)
cpuset_set_f (path, "cpus", cpus)
cpuset_set_f (path, "mems")
+ if (use_release_agent == true) then
+ cpuset_set_f (path, "notify_on_release", 1)
+ end
return true
end
@@ -159,9 +164,15 @@ function cpuset_kill (name, signo)
if (not cpuset_exists (name)) then return end
local path = string.format ("%s/%s/tasks", cpuset_dir, name)
- for pid in io.lines (cpuset_dir .. "/" .. name .. "/tasks") do
- log_debug ("Sending signal %d to pid %d", signo, pid)
- posix.kill (pid, signo)
+
+ local path_fh = io.open(path)
+ if path_fh then
+ while true do
+ local pid = path_fh:read()
+ if pid == nil then break end
+ log_debug ("Sending signal %d to pid %d", signo, pid)
+ posix.kill (pid, signo)
+ end
end
end
@@ -200,7 +211,7 @@ function cpuset_id_create (job)
end
-- Add the lower 16 bits of the stepid:
- id = id truncate_to_n_bits (job.stepid, 16)
+ id = id + truncate_to_n_bits (job.stepid, 16)
-- Must truncate result to 32bits until SLURM's job container
-- id is no longer represented by uint32_t :
@@ -222,12 +233,33 @@ function cpuset_has_pid (id, process_id)
return false
end
+function pid_is_thread (process_id)
+ local pid_status_path = string.format ("/proc/%d/status",process_id)
+ local pid_status_fh = io.open(pid_status_path)
+
+ if pid_status_fh then
+ while true do
+ local pid_status_line=pid_status_fh:read()
+ if pid_status_line == nil then break end
+ if string.match(pid_status_line,'^Tgid:%s+' .. process_id .. '$') then return false end
+ end
+ end
+ return true
+end
+
function cpuset_pids (id)
local pids = {}
if (cpuset_exists (id)) then
local path = string.format ("%s/%s/tasks", cpuset_dir, id)
- for task in io.lines (path) do
- table.insert (pids, task)
+ local path_fh = io.open(path)
+ if path_fh then
+ while true do
+ local task=path_fh:read()
+ if task == nil then break end
+ if not ( pid_is_thread(task) ) then
+ table.insert (pids, task)
+ end
+ end
end
end
return pids
@@ -317,33 +349,18 @@ function cpumap_create ()
return cpu_map
end
-function v_log_msg (l, fmt, ...)
- slurm.log (l, string.format (fmt, ...))
-end
-
-function log_msg (fmt, ...)
- v_log_msg (0, fmt, ...)
-end
-
-function log_verbose (fmt, ...)
- v_log_msg (1, fmt, ...)
-end
-
-function log_debug (fmt, ...)
- v_log_msg (2, fmt, ...)
-end
-
-function log_err (fmt, ...)
- slurm.error (string.format (fmt, ...))
-end
-
-
--########################################################################--
--
-- Initialization code:
--
--########################################################################--
+log_msg = slurm.log_info
+log_verbose = slurm.log_verbose
+log_debug = slurm.log_debug
+log_err = slurm.error
+
+
cpuset_dir = get_cpuset_dir ()
if cpuset_dir == nil then
print "cpuset must be mounted"
@@ -355,8 +372,8 @@ root_cpuset.mems = cpuset_read (cpuset_dir, "mems")
cpumap = cpumap_create ()
-slurm.log (string.format ("initialized: root cpuset = %s\n", cpuset_dir))
+log_msg ("initialized: root cpuset = %s\n", cpuset_dir)
-return 0
+return slurm.SUCCESS
-- vi: filetype=lua ts=4 sw=4 expandtab
diff --git a/doc/html/accounting.shtml b/doc/html/accounting.shtml
index f4952fcb5..6f4804bb3 100644
--- a/doc/html/accounting.shtml
+++ b/doc/html/accounting.shtml
@@ -230,7 +230,7 @@ referenced or recorded.</p>
<ul>
<li><b>AccountingStorageEnforce</b>:
This option contains a comma separated list of options you may want to
- enforce. The valid options are
+ enforce. The valid options are any comma separated combination of
<ul>
<li>associations - This will prevent users from running jobs if
their <i>association</i> is not in the database. This option will
diff --git a/doc/html/team.shtml b/doc/html/team.shtml
index 0a6d9a507..b87e9fdf4 100644
--- a/doc/html/team.shtml
+++ b/doc/html/team.shtml
@@ -11,7 +11,6 @@ Linux NetworX and many other contributors.
<ul>
<li>Morris Jette (LLNL, Project leader)</li>
<li>Danny Auble (LLNL)</li>
-<li>Chris Holmes (HP)</li>
<li>Don Lipari (LLNL)</li>
</ul>
@@ -23,6 +22,7 @@ Linux NetworX and many other contributors.
<li>Ernest Artiaga (Barcelona Supercomputer Center, Spain)</li>
<li>Susanne Balle (HP)</li>
<li>Anton Blanchard (Samba)</li>
+<li>Janne Blomqvist (Aalto University, Finland)</li>
<li>David Bremer (LLNL)</li>
<li>Hongjia Cao (National University of Defense Techonogy, China)</li>
<li>Daniel Christians (HP)</li>
@@ -35,9 +35,11 @@ Linux NetworX and many other contributors.
<li>Kent Engström (National Supercomputer Centre, Sweden)</li>
<li>Jim Garlick (LLNL)</li>
<li>Didier Gazen (Laboratoire d'Aerologie, France)</li>
+<li>Yiannis Georgiou (Bull)</li>
<li>Mark Grondona (LLNL)</li>
<li>Takao Hatazaki (HP, Japan)</li>
<li>Matthieu Hautreux (CEA, France)</li>
+<li>Chris Holmes (HP)</li>
<li>David Höppner</li>
<li>Nathan Huff (North Dakota State University)</li>
<li>David Jackson (Adaptive Computing)</li>
diff --git a/doc/man/man1/sacct.1 b/doc/man/man1/sacct.1
index 9d8f219b6..fd452fc79 100644
--- a/doc/man/man1/sacct.1
+++ b/doc/man/man1/sacct.1
@@ -352,6 +352,14 @@ wckey names.
Space characters are not allowed in the \f2wckey_list\fP. Default is
all wckeys\&.
+.TP
+\f3\-x \fP\f2associd_list\fP\f3,\fP \f3\-\-associations\fP\f3=\fP\f2assoc_list\fP
+Displays the statistics only for the jobs running under the
+association ids specified by the \f2assoc_list\fP operand, which is a
+comma\-separated list of association ids.
+Space characters are not allowed in the \f2assoc_list\fP. Default is
+all associations\&.
+
.TP
\f3\-X \fP\f3,\fP \f3\-\-allocations\fP
Only show cumulative statistics for each job, not the intermediate steps.
diff --git a/doc/man/man1/sacctmgr.1 b/doc/man/man1/sacctmgr.1
index b649d091c..7431f49c8 100644
--- a/doc/man/man1/sacctmgr.1
+++ b/doc/man/man1/sacctmgr.1
@@ -173,6 +173,9 @@ The hierarchy may have an arbitrary depth.
\fIassociation\fP
The entity used to group information consisting of four parameters:
\fIaccount\fR, \fIcluster\fR, \fIpartition (optional)\fR, and \fIuser\fR.
+Used only with the \fIlist\fR or \fIshow\fR command. Add, modify, and
+delete should be done to a user, account or cluster entity. This will
+in\-turn update the underlying associations.
.TP
\fIcluster\fP
diff --git a/doc/man/man1/salloc.1 b/doc/man/man1/salloc.1
index f0dcb8c97..beb5d4b2d 100644
--- a/doc/man/man1/salloc.1
+++ b/doc/man/man1/salloc.1
@@ -365,11 +365,11 @@ Bind tasks according to application hints
.TP
.B compute_bound
Select settings for compute bound applications:
-use all cores in each socket
+use all cores in each socket, one thread per core
.TP
.B memory_bound
Select settings for memory bound applications:
-use only one core in each socket
+use only one core in each socket, one thread per core
.TP
.B [no]multithread
[don't] use extra threads with in-core multi-threading
diff --git a/doc/man/man1/sbatch.1 b/doc/man/man1/sbatch.1
index 65925e843..53e286a59 100644
--- a/doc/man/man1/sbatch.1
+++ b/doc/man/man1/sbatch.1
@@ -380,11 +380,11 @@ Bind tasks according to application hints
.TP
.B compute_bound
Select settings for compute bound applications:
-use all cores in each socket
+use all cores in each socket, one thread per core
.TP
.B memory_bound
Select settings for memory bound applications:
-use only one core in each socket
+use only one core in each socket, one thread per core
.TP
.B [no]multithread
[don't] use extra threads with in-core multi-threading
diff --git a/doc/man/man1/scontrol.1 b/doc/man/man1/scontrol.1
index 626466e3f..553c1c911 100644
--- a/doc/man/man1/scontrol.1
+++ b/doc/man/man1/scontrol.1
@@ -374,12 +374,6 @@ Set the job's list of excluded node. Multiple node names may be
specified using simple node range expressions (e.g. "lx[10\-20]").
Value may be cleared with blank data value, "ExcNodeList=".
.TP
-\fIExitCode\fP=<exit>:<sig>
-Exit status reported for the job by the wait() function.
-The first number is the exit code, typically as set by the exit() function.
-The second number of the signal that caused the process to terminate if
-it was terminated by a signal.
-.TP
\fIFeatures\fP=<features>
Set the job's required node features.
The list of features may include multiple feature names separated
@@ -546,6 +540,12 @@ The time the job is expected to terminate based on the job's time
limit. When the job ends sooner, this field will be updated with the
actual end time.
.TP
+\fIExitCode\fP=<exit>:<sig>
+Exit status reported for the job by the wait() function.
+The first number is the exit code, typically as set by the exit() function.
+The second number of the signal that caused the process to terminate if
+it was terminated by a signal.
+.TP
\fIJobState\fP
The current state of the job.
.TP
@@ -602,8 +602,11 @@ Use quotes to enclose a reason having more than one word.
.TP
\fIState\fP=<state>
Identify the state to be assigned to the node. Possible values are "NoResp",
-"ALLOC", "ALLOCATED", "DOWN", "DRAIN", "FAIL", "FAILING", "IDLE", "MAINT",
-"POWER_DOWN", "POWER_UP", or "RESUME".
+"ALLOC", "ALLOCATED", "DOWN", "DRAIN", "FAIL", "FAILING", "IDLE",
+"MIXED", "MAINT", "POWER_DOWN", "POWER_UP", or "RESUME".
+If a node is in a "MIXED" state it usually means the node is in
+multiple states. For instance if only part of the node is "ALLOCATED"
+and the rest of the node is "IDLE" the state will be "MIXED".
If you want to remove a node from service, you typically want to set
it's state to "DRAIN".
"FAILING" is similar to "DRAIN" except that some applications will
diff --git a/doc/man/man1/srun.1 b/doc/man/man1/srun.1
index 364c54043..8fd76ecc5 100644
--- a/doc/man/man1/srun.1
+++ b/doc/man/man1/srun.1
@@ -404,11 +404,11 @@ Bind tasks according to application hints
.TP
.B compute_bound
Select settings for compute bound applications:
-use all cores in each socket
+use all cores in each socket, one thread per core
.TP
.B memory_bound
Select settings for memory bound applications:
-use only one core in each socket
+use only one core in each socket, one thread per core
.TP
.B [no]multithread
[don't] use extra threads with in-core multi-threading
diff --git a/doc/man/man5/slurm.conf.5 b/doc/man/man5/slurm.conf.5
index e6f25ed1f..0dbeeec9a 100644
--- a/doc/man/man5/slurm.conf.5
+++ b/doc/man/man5/slurm.conf.5
@@ -1183,7 +1183,8 @@ and resumes communications).
A DOWN node will become available for use upon registration with a
valid configuration only if it was set DOWN due to being non\-responsive.
If the node was set DOWN for any other reason (low memory, prolog failure,
-epilog failure, etc.), its state will not automatically be changed.
+epilog failure, silently rebooting, etc.), its state will not automatically
+be changed.
.TP
\fB2\fR
A DOWN node will become available for use upon registration with a
diff --git a/slurm.spec b/slurm.spec
index abcaec43d..d03fd7ce5 100644
--- a/slurm.spec
+++ b/slurm.spec
@@ -83,14 +83,14 @@
%endif
Name: slurm
-Version: 2.1.7
+Version: 2.1.8
Release: 1%{?dist}
Summary: Simple Linux Utility for Resource Management
License: GPL
Group: System Environment/Base
-Source: slurm-2.1.7.tar.bz2
+Source: slurm-2.1.8.tar.bz2
BuildRoot: %{_tmppath}/%{name}-%{version}-%{release}
URL: https://computing.llnl.gov/linux/slurm/
@@ -352,7 +352,7 @@ Gives the ability for SLURM to use Berkeley Lab Checkpoint/Restart
#############################################################################
%prep
-%setup -n slurm-2.1.7
+%setup -n slurm-2.1.8
%build
%configure --program-prefix=%{?_program_prefix:%{_program_prefix}} \
diff --git a/slurm/slurm.h.in b/slurm/slurm.h.in
index f72214ce4..3a62b6a35 100644
--- a/slurm/slurm.h.in
+++ b/slurm/slurm.h.in
@@ -623,12 +623,9 @@ typedef struct job_descriptor { /* For submit, allocate, and update requests */
* default=0 */
uint32_t max_nodes; /* maximum number of nodes usable by job,
* default=0 */
- uint16_t min_sockets; /* minimum number of sockets per node required
- * by job, default=0 */
- uint16_t min_cores; /* minimum number of cores per cpu required
- * by job, default=0 */
- uint16_t min_threads; /* minimum number of threads per core required
- * by job, default=0 */
+ uint16_t min_sockets; /* number of sockets per node required by job */
+ uint16_t min_cores; /* number of cores per cpu required by job */
+ uint16_t min_threads; /* number of threads per core required by job */
uint16_t cpus_per_task; /* number of processors required for each task */
uint16_t ntasks_per_node;/* number of tasks to invoke on each node */
uint16_t ntasks_per_socket;/* number of tasks to invoke on each socket */
diff --git a/src/api/job_info.c b/src/api/job_info.c
index a68522cdb..9dd924ce8 100644
--- a/src/api/job_info.c
+++ b/src/api/job_info.c
@@ -581,7 +581,8 @@ line13:
xstrcat(out, " ");
else
xstrcat(out, "\n ");
- sprintf(tmp_line, "Command=%s", job_ptr->command);
+ snprintf(tmp_line, sizeof(tmp_line), "Command=%s",
+ job_ptr->command);
xstrcat(out, tmp_line);
}
if (one_liner)
@@ -590,7 +591,7 @@ line13:
xstrcat(out, "\n ");
/****** Line 17 ******/
- sprintf(tmp_line, "WorkDir=%s", job_ptr->work_dir);
+ snprintf(tmp_line, sizeof(tmp_line), "WorkDir=%s", job_ptr->work_dir);
xstrcat(out, tmp_line);
#ifdef HAVE_BG
diff --git a/src/common/plugstack.c b/src/common/plugstack.c
index b4a53dbcb..f748867dd 100644
--- a/src/common/plugstack.c
+++ b/src/common/plugstack.c
@@ -1297,7 +1297,7 @@ static int _opt_find(struct spank_plugin_opt *p,
static struct spank_plugin_opt *_find_remote_option_by_name(const char
*str)
{
- struct spank_plugin_opt *opt;
+ struct spank_plugin_opt *opt = NULL;
struct opt_find_args args;
char buf[256];
char *name;
@@ -1318,8 +1318,10 @@ static struct spank_plugin_opt *_find_remote_option_by_name(const char
args.optname = buf;
args.plugin_name = name;
- opt = list_find_first(option_cache, (ListFindF) _opt_find, &args);
-
+ if (option_cache) {
+ opt = list_find_first(option_cache, (ListFindF) _opt_find,
+ &args);
+ }
if (opt == NULL) {
error("warning: plugin \"%s\" option \"%s\" not found.",
name, buf);
diff --git a/src/common/proc_args.c b/src/common/proc_args.c
index d68811a5c..252f68e39 100644
--- a/src/common/proc_args.c
+++ b/src/common/proc_args.c
@@ -483,6 +483,7 @@ bool verify_socket_core_thread_count(const char *arg, int *min_sockets,
{
bool tmp_val,ret_val;
int i,j;
+ int max_sockets, max_cores, max_threads;
const char *cur_ptr = arg;
char buf[3][48]; /* each can hold INT64_MAX - INT64_MAX */
buf[0][0] = '\0';
@@ -519,13 +520,21 @@ bool verify_socket_core_thread_count(const char *arg, int *min_sockets,
ret_val = true;
tmp_val = get_resource_arg_range(&buf[0][0], "first arg of -B",
- min_sockets, NULL, true);
+ min_sockets, &max_sockets, true);
+ if ((*min_sockets == 1) && (max_sockets == INT_MAX))
+ *min_sockets = NO_VAL; /* Use full range of values */
ret_val = ret_val && tmp_val;
+
tmp_val = get_resource_arg_range(&buf[1][0], "second arg of -B",
- min_cores, NULL, true);
+ min_cores, &max_cores, true);
+ if ((*min_cores == 1) && (max_cores == INT_MAX))
+ *min_cores = NO_VAL; /* Use full range of values */
ret_val = ret_val && tmp_val;
+
tmp_val = get_resource_arg_range(&buf[2][0], "third arg of -B",
- min_threads, NULL, true);
+ min_threads, &max_threads, true);
+ if ((*min_threads == 1) && (max_threads == INT_MAX))
+ *min_threads = NO_VAL; /* Use full range of values */
ret_val = ret_val && tmp_val;
return ret_val;
@@ -566,20 +575,20 @@ bool verify_hint(const char *arg, int *min_sockets, int *min_cores,
" help show this help message\n");
return 1;
} else if (strcasecmp(tok, "compute_bound") == 0) {
- *min_sockets = 1;
- *min_cores = 1;
+ *min_sockets = NO_VAL;
+ *min_cores = NO_VAL;
+ *min_threads = 1;
*cpu_bind_type |= CPU_BIND_TO_CORES;
} else if (strcasecmp(tok, "memory_bound") == 0) {
- *min_cores = 1;
+ *min_cores = 1;
+ *min_threads = 1;
*cpu_bind_type |= CPU_BIND_TO_CORES;
} else if (strcasecmp(tok, "multithread") == 0) {
- *min_threads = 1;
+ *min_threads = NO_VAL;
*cpu_bind_type |= CPU_BIND_TO_THREADS;
} else if (strcasecmp(tok, "nomultithread") == 0) {
*min_threads = 1;
*cpu_bind_type |= CPU_BIND_TO_THREADS;
- if (*ntasks_per_core == NO_VAL)
- *ntasks_per_core = 1;
} else {
error("unrecognized --hint argument \"%s\", "
"see --hint=help", tok);
diff --git a/src/common/slurm_resource_info.c b/src/common/slurm_resource_info.c
index 6896fe0c1..c6c543f8f 100644
--- a/src/common/slurm_resource_info.c
+++ b/src/common/slurm_resource_info.c
@@ -91,8 +91,9 @@ static int _isvalue(char *arg) {
* given this number given the number of cpus_per_task and
* maximum sockets, cores, threads. Note that the value of
* cpus is the lowest-level logical processor (LLLP).
- * IN min_sockets - Job requested min sockets
- * IN min_cores - Job requested min cores
+ * IN socket_cnt - Job requested socket count
+ * IN core_cnt - Job requested core count
+ * IN threads_cnt - Job requested thread count
* IN cpus_per_task - Job requested cpus per task
* IN ntaskspernode - number of tasks per node
* IN ntaskspersocket- number of tasks per socket
@@ -106,8 +107,9 @@ static int _isvalue(char *arg) {
*
* Note: currently only used in the select/linear plugin.
*/
-int slurm_get_avail_procs(const uint16_t min_sockets,
- const uint16_t min_cores,
+int slurm_get_avail_procs(const uint16_t socket_cnt,
+ const uint16_t core_cnt,
+ const uint16_t thread_cnt,
uint16_t cpus_per_task,
const uint16_t ntaskspernode,
const uint16_t ntaskspersocket,
@@ -124,9 +126,18 @@ int slurm_get_avail_procs(const uint16_t min_sockets,
uint16_t avail_cpus = 0, max_cpus = 0;
uint16_t allocated_cpus = 0, allocated_cores = 0, allocated_sockets = 0;
uint16_t max_avail_cpus = 0xffff; /* for alloc_* accounting */
+ uint16_t min_sockets = 1, max_sockets = 0xffff;
+ uint16_t min_cores = 1, max_cores = 0xffff;
+ uint16_t min_threads = 1, max_threads = 0xffff;
int i;
/* pick defaults for any unspecified items */
+ if (socket_cnt != (uint16_t) NO_VAL)
+ min_sockets = max_sockets = socket_cnt;
+ if (core_cnt != (uint16_t) NO_VAL)
+ min_cores = max_cores = core_cnt;
+ if (thread_cnt != (uint16_t) NO_VAL)
+ min_threads = max_threads = thread_cnt;
if (cpus_per_task <= 0)
cpus_per_task = 1;
if (*threads <= 0)
@@ -141,8 +152,8 @@ int slurm_get_avail_procs(const uint16_t min_sockets,
allocated_sockets++;
}
#if(DEBUG)
- info("get_avail_procs %u %s MIN User_ sockets %u cores %u",
- job_id, name, min_sockets, min_cores);
+ info("get_avail_procs %u %s User_ sockets %u cores %u threads %u",
+ job_id, name, socket_cnt, core_cnt, thread_cnt);
info("get_avail_procs %u %s HW_ sockets %u cores %u threads %u",
job_id, name, *sockets, *cores, *threads);
info("get_avail_procs %u %s Ntask node %u sockets %u core %u",
@@ -200,6 +211,11 @@ int slurm_get_avail_procs(const uint16_t min_sockets,
}
}
+ /*** honor socket/core/thread maximums ***/
+ *sockets = MIN(*sockets, max_sockets);
+ *cores = MIN(*cores, max_cores);
+ *threads = MIN(*threads, max_threads);
+
if (min_sockets > *sockets) {
*cpus = 0;
} else {
@@ -251,6 +267,11 @@ int slurm_get_avail_procs(const uint16_t min_sockets,
if (min_sockets > *sockets)
*cpus = 0;
+ /*** honor socket/core/thread maximums ***/
+ *sockets = MIN(*sockets, max_sockets);
+ *cores = MIN(*cores, max_cores);
+ *threads = MIN(*threads, max_threads);
+
/*** compute an overall maximum cpu count honoring ntasks* ***/
max_cpus = *threads;
if (ntaskspercore > 0) {
diff --git a/src/common/slurm_resource_info.h b/src/common/slurm_resource_info.h
index c1a0743ff..b9f6a4fdb 100644
--- a/src/common/slurm_resource_info.h
+++ b/src/common/slurm_resource_info.h
@@ -50,8 +50,9 @@
# endif /* HAVE_INTTYPES_H */
#endif
-int slurm_get_avail_procs(const uint16_t minsockets,
- const uint16_t mincores,
+int slurm_get_avail_procs(const uint16_t socket_cnt,
+ const uint16_t core_cnt,
+ const uint16_t thread_cnt,
const uint16_t cpuspertask,
const uint16_t ntaskspernode,
const uint16_t ntaskspersocket,
diff --git a/src/common/slurmdbd_defs.c b/src/common/slurmdbd_defs.c
index 50ee94fe3..fa0285a81 100644
--- a/src/common/slurmdbd_defs.c
+++ b/src/common/slurmdbd_defs.c
@@ -1360,6 +1360,11 @@ static int _send_fini_msg(void)
Buf buffer;
dbd_fini_msg_t req;
+ /* If the connection is already gone, we don't need to send a
+ fini. */
+ if(_fd_writeable(slurmdbd_fd) == -1)
+ return SLURM_SUCCESS;
+
buffer = init_buf(1024);
pack16((uint16_t) DBD_FINI, buffer);
req.commit = 0;
diff --git a/src/database/mysql_common.h b/src/database/mysql_common.h
index ea416df26..c0bebd485 100644
--- a/src/database/mysql_common.h
+++ b/src/database/mysql_common.h
@@ -60,6 +60,7 @@
#include <mysqld_error.h>
typedef struct {
+ char *auto_incr_query;
MYSQL *db_conn;
bool rollback;
List update_list;
diff --git a/src/plugins/accounting_storage/mysql/accounting_storage_mysql.c b/src/plugins/accounting_storage/mysql/accounting_storage_mysql.c
index d5be23ee5..52eac1bcb 100644
--- a/src/plugins/accounting_storage/mysql/accounting_storage_mysql.c
+++ b/src/plugins/accounting_storage/mysql/accounting_storage_mysql.c
@@ -2363,7 +2363,7 @@ static int _remove_common(mysql_conn_t *mysql_conn,
MYSQL_ROW row;
time_t day_old = now - DELETE_SEC_BACK;
bool has_jobs = false;
- char *tmp_name_char = _fix_double_quotes(name_char);
+ char *tmp_name_char = NULL;
/* If we have jobs associated with this we do not want to
* really delete it for accounting purposes. This is for
@@ -2385,9 +2385,13 @@ static int _remove_common(mysql_conn_t *mysql_conn,
/* we want to remove completely all that is less than a day old */
if(!has_jobs && table != assoc_table) {
query = xstrdup_printf("delete from %s where creation_time>%d "
- "&& (%s);"
- "alter table %s AUTO_INCREMENT=0;",
- table, day_old, name_char, table);
+ "&& (%s);",
+ table, day_old, name_char);
+ /* Make sure the next id we get doesn't create holes
+ * in the ids. */
+ xstrfmtcat(mysql_conn->auto_incr_query,
+ "alter table %s AUTO_INCREMENT=0;",
+ table);
}
if(table != assoc_table)
@@ -2396,6 +2400,16 @@ static int _remove_common(mysql_conn_t *mysql_conn,
"where deleted=0 && (%s);",
table, now, name_char);
+ /* If we are removing assocs use the assoc_char since the
+ name_char has lft between statements that can change over
+ time. The assoc_char has the actual ids of the assocs
+ which never change.
+ */
+ if(type == DBD_REMOVE_ASSOCS && assoc_char)
+ tmp_name_char = _fix_double_quotes(assoc_char);
+ else
+ tmp_name_char = _fix_double_quotes(name_char);
+
xstrfmtcat(query,
"insert into %s (timestamp, action, name, actor) "
"values (%d, %d, \"%s\", \"%s\");",
@@ -2598,12 +2612,13 @@ just_update:
"fairshare=1, max_jobs=NULL, "
"max_nodes_per_job=NULL, "
"max_wall_duration_per_job=NULL, "
- "max_cpu_mins_per_job=NULL "
- "where (%s);"
- "alter table %s AUTO_INCREMENT=0;",
+ "max_cpu_mins_per_job=NULL where (%s);",
assoc_table, now,
- loc_assoc_char,
- assoc_table);
+ loc_assoc_char);
+ /* Make sure the next id we get doesn't create holes in the ids. */
+ xstrfmtcat(mysql_conn->auto_incr_query,
+ "alter table %s AUTO_INCREMENT=0;",
+ assoc_table);
if(table != assoc_table)
xfree(loc_assoc_char);
@@ -2615,6 +2630,7 @@ just_update:
if(mysql_conn->rollback) {
mysql_db_rollback(mysql_conn->db_conn);
}
+ xfree(mysql_conn->auto_incr_query);
list_flush(mysql_conn->update_list);
}
@@ -3665,7 +3681,8 @@ extern void *acct_storage_p_get_connection(bool make_agent, int conn_num,
if(!mysql_db_info)
init();
- debug2("acct_storage_p_get_connection: request new connection");
+ debug2("acct_storage_p_get_connection: request new connection %d",
+ rollback);
mysql_conn->rollback = rollback;
mysql_conn->conn = conn_num;
@@ -3690,6 +3707,7 @@ extern int acct_storage_p_close_connection(mysql_conn_t **mysql_conn)
acct_storage_p_commit((*mysql_conn), 0);
mysql_close_db_connection(&(*mysql_conn)->db_conn);
+ xfree((*mysql_conn)->auto_incr_query);
list_destroy((*mysql_conn)->update_list);
xfree((*mysql_conn));
@@ -3708,8 +3726,32 @@ extern int acct_storage_p_commit(mysql_conn_t *mysql_conn, bool commit)
if(mysql_db_rollback(mysql_conn->db_conn))
error("rollback failed");
} else {
- if(mysql_db_commit(mysql_conn->db_conn))
- error("commit failed");
+ int rc = SLURM_SUCCESS;
+ /* Since any use of altering a tables
+ AUTO_INCREMENT will make it so you can't
+ rollback, save it until right at the end.
+ for now we also want to check if
+ update_list exists since I didn't want to
+ alter the code all sorts since it is
+ different in 2.2.
+ */
+ if(mysql_conn->auto_incr_query
+ && list_count(mysql_conn->update_list)) {
+ debug3("%d(%d) query\n%s",
+ mysql_conn->conn, __LINE__,
+ mysql_conn->auto_incr_query);
+ rc = mysql_db_query(
+ mysql_conn->db_conn,
+ mysql_conn->auto_incr_query);
+ }
+
+ if(rc != SLURM_SUCCESS) {
+ if(mysql_db_rollback(mysql_conn->db_conn))
+ error("rollback failed");
+ } else {
+ if(mysql_db_commit(mysql_conn->db_conn))
+ error("commit failed");
+ }
}
}
@@ -3833,7 +3875,7 @@ extern int acct_storage_p_commit(mysql_conn_t *mysql_conn, bool commit)
if(get_qos_count)
_set_qos_cnt(mysql_conn->db_conn);
}
-
+ xfree(mysql_conn->auto_incr_query);
list_flush(mysql_conn->update_list);
return SLURM_SUCCESS;
diff --git a/src/plugins/accounting_storage/mysql/mysql_jobacct_process.c b/src/plugins/accounting_storage/mysql/mysql_jobacct_process.c
index af64285b2..0280748e0 100644
--- a/src/plugins/accounting_storage/mysql/mysql_jobacct_process.c
+++ b/src/plugins/accounting_storage/mysql/mysql_jobacct_process.c
@@ -551,26 +551,21 @@ extern int setup_job_cond_limits(mysql_conn_t *mysql_conn,
if(!job_cond)
return 0;
- /* THIS ASSOCID CHECK ALWAYS NEEDS TO BE FIRST!!!!!!! */
if(job_cond->associd_list && list_count(job_cond->associd_list)) {
set = 0;
- xstrfmtcat(*extra, ", %s as t3 where (", assoc_table);
+ if(*extra)
+ xstrcat(*extra, " && (");
+ else
+ xstrcat(*extra, " where (");
itr = list_iterator_create(job_cond->associd_list);
while((object = list_next(itr))) {
if(set)
xstrcat(*extra, " || ");
- xstrfmtcat(*extra, "t3.id=%s", object);
+ xstrfmtcat(*extra, "t1.associd='%s'", object);
set = 1;
}
list_iterator_destroy(itr);
xstrcat(*extra, ")");
- table_level="t3";
- /* just incase the association is gone */
- if(set)
- xstrcat(*extra, " || ");
- xstrfmtcat(*extra, "t3.id is null) && "
- "(t2.lft between t3.lft and t3.rgt "
- "|| t2.lft is null)");
}
if(job_cond->acct_list && list_count(job_cond->acct_list)) {
diff --git a/src/plugins/proctrack/lua/proctrack_lua.c b/src/plugins/proctrack/lua/proctrack_lua.c
index 00d1dad41..b6a529eda 100644
--- a/src/plugins/proctrack/lua/proctrack_lua.c
+++ b/src/plugins/proctrack/lua/proctrack_lua.c
@@ -52,6 +52,7 @@
#include <stdlib.h>
#include <unistd.h>
#include <dlfcn.h>
+#include <pthread.h>
#include <slurm/slurm.h>
#include <slurm/slurm_errno.h>
@@ -61,6 +62,7 @@
#include <lualib.h>
#include "src/common/log.h"
+#include "src/common/macros.h"
#include "src/slurmd/slurmstepd/slurmstepd_job.h"
@@ -71,6 +73,14 @@ const uint32_t plugin_version = 90;
static const char lua_script_path[] = DEFAULT_SCRIPT_DIR "/proctrack.lua";
static lua_State *L = NULL;
+/*
+ * Mutex for protecting multi-threaded access to this plugin.
+ * (Only 1 thread at a time should be in here)
+ */
+#ifdef WITH_PTHREADS
+static pthread_mutex_t lua_lock = PTHREAD_MUTEX_INITIALIZER;
+#endif
+
/*
* Lua interface to SLURM log facility:
*/
@@ -133,6 +143,29 @@ static int lua_register_slurm_output_functions ()
*/
lua_newtable (L);
luaL_register (L, NULL, slurm_functions);
+
+ /*
+ * Create more user-friendly lua versions of SLURM log functions.
+ */
+ luaL_loadstring (L, "slurm.error (string.format(unpack({...})))");
+ lua_setfield (L, -2, "log_error");
+ luaL_loadstring (L, "slurm.log (0, string.format(unpack({...})))");
+ lua_setfield (L, -2, "log_info");
+ luaL_loadstring (L, "slurm.log (1, string.format(unpack({...})))");
+ lua_setfield (L, -2, "log_verbose");
+ luaL_loadstring (L, "slurm.log (2, string.format(unpack({...})))");
+ lua_setfield (L, -2, "log_debug");
+ luaL_loadstring (L, "slurm.log (3, string.format(unpack({...})))");
+ lua_setfield (L, -2, "log_debug2");
+
+ /*
+ * slurm.SUCCESS and slurm.FAILURE
+ */
+ lua_pushnumber (L, -1);
+ lua_setfield (L, -2, "FAILURE");
+ lua_pushnumber (L, 0);
+ lua_setfield (L, -2, "SUCCESS");
+
lua_setglobal (L, "slurm");
return 0;
}
@@ -181,6 +214,11 @@ static int check_lua_script_functions ()
return (rc);
}
+/*
+ * NOTE: The init callback should never be called multiple times,
+ * let alone called from multiple threads. Therefore, locking
+ * is unecessary here.
+ */
int init (void)
{
int rc = SLURM_SUCCESS;
@@ -275,173 +313,237 @@ static int lua_job_table_create (slurmd_job_t *job)
int slurm_container_create (slurmd_job_t *job)
{
+ int rc = SLURM_ERROR;
double id;
+
+ slurm_mutex_lock (&lua_lock);
+
/*
* All lua script functions should have been verified during
* initialization:
*/
lua_getglobal (L, "slurm_container_create");
if (lua_isnil (L, -1))
- return SLURM_FAILURE;
+ goto out;
lua_job_table_create (job);
- if (lua_pcall (L, 1, 1, 0) != 0)
- return error ("proctrack/lua: %s: slurm_container_create: %s",
- lua_script_path, lua_tostring (L, -1));
+ if (lua_pcall (L, 1, 1, 0) != 0) {
+ error ("proctrack/lua: %s: slurm_container_create: %s",
+ lua_script_path, lua_tostring (L, -1));
+ goto out;
+ }
/*
* Get the container id off the stack:
*/
if (lua_isnil (L, -1)) {
+ error ("proctrack/lua: slurm_container_create did not return id");
lua_pop (L, -1);
- return (-1);
+ goto out;
}
id = lua_tonumber (L, -1);
job->cont_id = id;
info ("job->cont_id = %u (%.0f) \n", job->cont_id, id);
lua_pop (L, -1);
- return (0);
+
+ rc = SLURM_SUCCESS;
+out:
+ slurm_mutex_unlock (&lua_lock);
+ return rc;
}
int slurm_container_add (slurmd_job_t *job, pid_t pid)
{
- int rc;
+ int rc = SLURM_ERROR;
+
+ slurm_mutex_lock (&lua_lock);
lua_getglobal (L, "slurm_container_add");
if (lua_isnil (L, -1))
- return SLURM_FAILURE;
+ goto out;
lua_job_table_create (job);
lua_pushnumber (L, job->cont_id);
lua_pushnumber (L, pid);
- if (lua_pcall (L, 3, 1, 0) != 0)
- return error ("running lua function 'slurm_container_add': %s",
- lua_tostring (L, -1));
+ if (lua_pcall (L, 3, 1, 0) != 0) {
+ error ("running lua function 'slurm_container_add': %s",
+ lua_tostring (L, -1));
+ goto out;
+ }
rc = lua_tonumber (L, -1);
lua_pop (L, -1);
+out:
+ slurm_mutex_unlock (&lua_lock);
return (rc);
}
int slurm_container_signal (uint32_t id, int sig)
{
- int rc;
+ int rc = SLURM_ERROR;
+
+ slurm_mutex_lock (&lua_lock);
+
lua_getglobal (L, "slurm_container_signal");
if (lua_isnil (L, -1))
- return SLURM_FAILURE;
+ goto out;
lua_pushnumber (L, id);
lua_pushnumber (L, sig);
- if (lua_pcall (L, 2, 1, 0) != 0)
- return error ("running lua function 'slurm_container_signal': %s",
- lua_tostring (L, -1));
+ if (lua_pcall (L, 2, 1, 0) != 0) {
+ error ("running lua function 'slurm_container_signal': %s",
+ lua_tostring (L, -1));
+ goto out;
+ }
rc = lua_tonumber (L, -1);
lua_pop (L, -1);
+out:
+ slurm_mutex_unlock (&lua_lock);
return (rc);
}
int slurm_container_destroy (uint32_t id)
{
- int rc;
+ int rc = SLURM_ERROR;
+
+ slurm_mutex_lock (&lua_lock);
+
lua_getglobal (L, "slurm_container_destroy");
if (lua_isnil (L, -1))
- return SLURM_FAILURE;
+ goto out;
lua_pushnumber (L, id);
- if (lua_pcall (L, 1, 1, 0) != 0)
- return error ("running lua function 'slurm_container_destroy': %s",
- lua_tostring (L, -1));
+ if (lua_pcall (L, 1, 1, 0) != 0) {
+ error ("running lua function 'slurm_container_destroy': %s",
+ lua_tostring (L, -1));
+ goto out;
+ }
rc = lua_tonumber (L, -1);
lua_pop (L, -1);
+
+out:
+ slurm_mutex_unlock (&lua_lock);
return (rc);
}
uint32_t slurm_container_find (pid_t pid)
{
- uint32_t id;
+ uint32_t id = (uint32_t) SLURM_ERROR;
+
+ slurm_mutex_lock (&lua_lock);
+
lua_getglobal (L, "slurm_container_find");
if (lua_isnil (L, -1))
- return SLURM_FAILURE;
+ goto out;
lua_pushnumber (L, pid);
- if (lua_pcall (L, 1, 1, 0) != 0)
- return error ("running lua function 'slurm_container_find': %s",
- lua_tostring (L, -1));
+ if (lua_pcall (L, 1, 1, 0) != 0) {
+ error ("running lua function 'slurm_container_find': %s",
+ lua_tostring (L, -1));
+ goto out;
+ }
id = (uint32_t) lua_tonumber (L, -1);
lua_pop (L, -1);
+
+out:
+ slurm_mutex_lock (&lua_lock);
return (id);
}
bool slurm_container_has_pid (uint32_t id, pid_t pid)
{
- int rc;
+ int rc = 0;
+
+ slurm_mutex_lock (&lua_lock);
+
lua_getglobal (L, "slurm_container_has_pid");
if (lua_isnil (L, -1))
- return SLURM_FAILURE;
+ goto out;
lua_pushnumber (L, id);
lua_pushnumber (L, pid);
- if (lua_pcall (L, 2, 1, 0) != 0)
- return error ("running lua function 'slurm_container_has_pid': %s",
- lua_tostring (L, -1));
+ if (lua_pcall (L, 2, 1, 0) != 0) {
+ error ("running lua function 'slurm_container_has_pid': %s",
+ lua_tostring (L, -1));
+ goto out;
+ }
rc = lua_toboolean (L, -1);
lua_pop (L, -1);
+
+out:
+ slurm_mutex_unlock (&lua_lock);
return (rc == 1);
}
int slurm_container_wait (uint32_t id)
{
- int rc;
+ int rc = SLURM_ERROR;
+
+ slurm_mutex_lock (&lua_lock);
+
lua_getglobal (L, "slurm_container_wait");
if (lua_isnil (L, -1))
- return SLURM_FAILURE;
+ goto out;
lua_pushnumber (L, id);
- if (lua_pcall (L, 1, 1, 0) != 0)
- return error ("running lua function 'slurm_container_wait': %s",
- lua_tostring (L, -1));
+ if (lua_pcall (L, 1, 1, 0) != 0) {
+ error ("running lua function 'slurm_container_wait': %s",
+ lua_tostring (L, -1));
+ goto out;
+ }
rc = lua_tonumber (L, -1);
lua_pop (L, -1);
+out:
+ slurm_mutex_unlock (&lua_lock);
return (rc);
}
int slurm_container_get_pids (uint32_t cont_id, pid_t **pids, int *npids)
{
+ int rc = SLURM_ERROR;
int i = 0;
int t = 0;
pid_t *p;
+ *npids = 0;
+
+ slurm_mutex_lock (&lua_lock);
+
lua_getglobal (L, "slurm_container_get_pids");
if (lua_isnil (L, -1))
- return SLURM_FAILURE;
+ goto out;
lua_pushnumber (L, cont_id);
- if (lua_pcall (L, 1, 1, 0) != 0)
- return (error ("%s: %s: %s",
- "proctrack/lua",
- __func__,
- lua_tostring (L, -1)));
+ if (lua_pcall (L, 1, 1, 0) != 0) {
+ error ("%s: %s: %s",
+ "proctrack/lua",
+ __func__,
+ lua_tostring (L, -1));
+ goto out;
+ }
/*
* list of PIDs should be returned in a table from the lua
* script. If a table wasn't returned then generate an error:
*/
- if (!lua_istable(L, -1))
- return (error ("%s: %s: function should return a table",
- "proctrack/lua",
- __func__));
+ if (!lua_istable(L, -1)) {
+ error ("%s: %s: function should return a table",
+ "proctrack/lua",
+ __func__);
+ goto out;
+ }
/*
* Save absolute position of table in lua stack
@@ -468,6 +570,10 @@ int slurm_container_get_pids (uint32_t cont_id, pid_t **pids, int *npids)
lua_pop (L, 1);
*pids = p;
- return SLURM_SUCCESS;
+
+ rc = SLURM_SUCCESS;
+out:
+ slurm_mutex_unlock (&lua_lock);
+ return rc;
}
diff --git a/src/plugins/sched/backfill/backfill.c b/src/plugins/sched/backfill/backfill.c
index 8c194fa6e..af28f7821 100644
--- a/src/plugins/sched/backfill/backfill.c
+++ b/src/plugins/sched/backfill/backfill.c
@@ -121,6 +121,9 @@ static bool _job_is_completing(void);
static bool _more_work(void);
static int _num_feature_count(struct job_record *job_ptr);
static int _start_job(struct job_record *job_ptr, bitstr_t *avail_bitmap);
+static bool _test_resv_overlap(node_space_map_t *node_space,
+ bitstr_t *use_bitmap, uint32_t start_time,
+ uint32_t end_reserve);
static int _try_sched(struct job_record *job_ptr, bitstr_t **avail_bitmap,
uint32_t min_nodes, uint32_t max_nodes,
uint32_t req_nodes);
@@ -421,7 +424,11 @@ static void _attempt_backfill(void)
if (part_ptr == NULL) {
part_ptr = find_part_record(job_ptr->partition);
- xassert(part_ptr);
+ if (part_ptr == NULL) {
+ error("Could not find partition %s for job %u",
+ job_ptr->partition, job_ptr->job_id);
+ continue;
+ }
job_ptr->part_ptr = part_ptr;
error("partition pointer reset for job %u, part %s",
job_ptr->job_id, job_ptr->partition);
@@ -536,14 +543,20 @@ static void _attempt_backfill(void)
}
job_ptr->start_time = MAX(job_ptr->start_time, start_res);
+ last_job_update = now;
if (job_ptr->start_time <= now) {
int rc = _start_job(job_ptr, resv_bitmap);
- if (rc == ESLURM_ACCOUNTING_POLICY)
+ if (rc == ESLURM_ACCOUNTING_POLICY) {
+ /* Unknown future start time, just skip job */
continue;
- else if (rc != SLURM_SUCCESS)
+ } else if (rc != SLURM_SUCCESS) {
/* Planned to start job, but something bad
* happended. */
break;
+ } else {
+ /* Started this job, move to next one */
+ continue;
+ }
}
if (job_ptr->start_time > (now + BACKFILL_WINDOW)) {
/* Starts too far in the future to worry about */
@@ -555,10 +568,19 @@ static void _attempt_backfill(void)
break;
}
+ end_reserve = job_ptr->start_time + (time_limit * 60);
+ if (_test_resv_overlap(node_space, avail_bitmap,
+ job_ptr->start_time, end_reserve)) {
+ /* This job overlaps with an existing reservation for
+ * job to be backfill scheduled, which the sched
+ * plugin does not know about. Try again later. */
+ later_start = job_ptr->start_time;
+ goto TRY_LATER;
+ }
+
/*
* Add reservation to scheduling table
*/
- end_reserve = job_ptr->start_time + (time_limit * 60);
bit_not(avail_bitmap);
_add_reservation(job_ptr->start_time, end_reserve,
avail_bitmap, node_space, &node_space_recs);
@@ -721,3 +743,31 @@ static void _add_reservation(uint32_t start_time, uint32_t end_reserve,
break;
}
}
+
+/*
+ * Determine if the resource specification for a new job overlaps with a
+ * reservation that the backfill scheduler has made for a job to be
+ * started in the future.
+ * IN use_bitmap - nodes to be allocated
+ * IN start_time - start time of job
+ * IN end_reserve - end time of job
+ */
+static bool _test_resv_overlap(node_space_map_t *node_space,
+ bitstr_t *use_bitmap, uint32_t start_time,
+ uint32_t end_reserve)
+{
+ bool overlap = false;
+ int j;
+
+ for (j=0; ; ) {
+ if ((node_space[j].end_time > start_time) &&
+ (node_space[j].begin_time < end_reserve) &&
+ (!bit_super_set(use_bitmap, node_space[j].avail_bitmap))) {
+ overlap = true;
+ break;
+ }
+ if ((j = node_space[j].next) == 0)
+ break;
+ }
+ return overlap;
+}
diff --git a/src/plugins/sched/wiki2/job_modify.c b/src/plugins/sched/wiki2/job_modify.c
index 6f77e3606..fd1d01df3 100644
--- a/src/plugins/sched/wiki2/job_modify.c
+++ b/src/plugins/sched/wiki2/job_modify.c
@@ -62,7 +62,8 @@ static int _job_modify(uint32_t jobid, char *bank_ptr,
char *depend_ptr, char *new_hostlist,
uint32_t new_node_cnt, char *part_name_ptr,
uint32_t new_time_limit, char *name_ptr,
- char *start_ptr, char *feature_ptr, char *env_ptr)
+ char *start_ptr, char *feature_ptr, char *env_ptr,
+ char *comment_ptr)
{
struct job_record *job_ptr;
time_t now = time(NULL);
@@ -78,6 +79,13 @@ static int _job_modify(uint32_t jobid, char *bank_ptr,
return ESLURM_DISABLED;
}
+ if (comment_ptr) {
+ info("wiki: change job %u comment %s", jobid, comment_ptr);
+ xfree(job_ptr->comment);
+ job_ptr->comment = xstrdup(comment_ptr);
+ last_job_update = now;
+ }
+
if (depend_ptr) {
int rc = update_job_dependency(job_ptr, depend_ptr);
if (rc == SLURM_SUCCESS) {
@@ -90,7 +98,6 @@ static int _job_modify(uint32_t jobid, char *bank_ptr,
}
}
-
if (env_ptr) {
bool have_equal = false;
char old_sep[1];
@@ -393,16 +400,24 @@ host_fini: if (rc) {
}
/* Modify a job:
- * CMD=MODIFYJOB ARG=<jobid> PARTITION=<name> NODES=<number>
- * DEPEND=afterany:<jobid> TIMELIMT=<seconds> BANK=<name>
- * JOBNAME=<name> MINSTARTTIME=<uts> RFEATURES=<features>
- * VARIABLELIST=<env_vars>
+ * CMD=MODIFYJOB ARG=<jobid>
+ * [BANK=<name>]
+ * [COMMENT=<whatever>;]
+ * [DEPEND=afterany:<jobid>]
+ * [JOBNAME=<name>]
+ * [MINSTARTTIME=<uts>]
+ * [NODES=<number>]
+ * [PARTITION=<name>]
+ * [RFEATURES=<features>]
+ * [TIMELIMT=<seconds>]
+ * [VARIABLELIST=<env_vars>]
+ *
* RET 0 on success, -1 on failure */
extern int job_modify_wiki(char *cmd_ptr, int *err_code, char **err_msg)
{
char *arg_ptr, *bank_ptr, *depend_ptr, *nodes_ptr, *start_ptr;
char *host_ptr, *name_ptr, *part_ptr, *time_ptr, *tmp_char;
- char *feature_ptr, *env_ptr;
+ char *comment_ptr, *feature_ptr, *env_ptr;
int i, slurm_rc;
uint32_t jobid, new_node_cnt = 0, new_time_limit = 0;
static char reply_msg[128];
@@ -429,6 +444,7 @@ extern int job_modify_wiki(char *cmd_ptr, int *err_code, char **err_msg)
return -1;
}
bank_ptr = strstr(cmd_ptr, "BANK=");
+ comment_ptr = strstr(cmd_ptr, "COMMENT=");
depend_ptr = strstr(cmd_ptr, "DEPEND=");
host_ptr = strstr(cmd_ptr, "HOSTLIST=");
name_ptr = strstr(cmd_ptr, "JOBNAME=");
@@ -443,6 +459,32 @@ extern int job_modify_wiki(char *cmd_ptr, int *err_code, char **err_msg)
bank_ptr += 5;
null_term(bank_ptr);
}
+ if (comment_ptr) {
+ comment_ptr[7] = ':';
+ comment_ptr += 8;
+ if (comment_ptr[0] == '\"') {
+ comment_ptr++;
+ for (i=0; ; i++) {
+ if (comment_ptr[i] == '\0')
+ break;
+ if (comment_ptr[i] == '\"') {
+ comment_ptr[i] = '\0';
+ break;
+ }
+ }
+ } else if (comment_ptr[0] == '\'') {
+ comment_ptr++;
+ for (i=0; ; i++) {
+ if (comment_ptr[i] == '\0')
+ break;
+ if (comment_ptr[i] == '\'') {
+ comment_ptr[i] = '\0';
+ break;
+ }
+ }
+ } else
+ null_term(comment_ptr);
+ }
if (depend_ptr) {
depend_ptr[6] = ':';
depend_ptr += 7;
@@ -510,10 +552,10 @@ extern int job_modify_wiki(char *cmd_ptr, int *err_code, char **err_msg)
null_term(env_ptr);
}
- /* Look for any un-parsed "=" */
+ /* Look for any un-parsed "=" ignoring anything after VARIABLELIST
+ * which is expected to contain "=" in its value*/
tmp_char = strchr(cmd_ptr, '=');
- if (( env_ptr && (tmp_char < env_ptr)) ||
- (!env_ptr && tmp_char)) {
+ if (tmp_char && (!env_ptr || (env_ptr > tmp_char))) {
tmp_char[0] = '\0';
while (tmp_char[-1] && (!isspace(tmp_char[-1])))
tmp_char--;
@@ -523,7 +565,7 @@ extern int job_modify_wiki(char *cmd_ptr, int *err_code, char **err_msg)
lock_slurmctld(job_write_lock);
slurm_rc = _job_modify(jobid, bank_ptr, depend_ptr, host_ptr,
new_node_cnt, part_ptr, new_time_limit, name_ptr,
- start_ptr, feature_ptr, env_ptr);
+ start_ptr, feature_ptr, env_ptr, comment_ptr);
unlock_slurmctld(job_write_lock);
if (slurm_rc != SLURM_SUCCESS) {
*err_code = -700;
diff --git a/src/plugins/select/bluegene/plugin/bg_block_info.c b/src/plugins/select/bluegene/plugin/bg_block_info.c
index cce430540..f0d296fdc 100644
--- a/src/plugins/select/bluegene/plugin/bg_block_info.c
+++ b/src/plugins/select/bluegene/plugin/bg_block_info.c
@@ -1,7 +1,7 @@
/*****************************************************************************\
* bg_block_info.c - bluegene block information from the db2 database.
*
- * $Id: bg_block_info.c 19270 2010-01-19 19:46:45Z da $
+ * $Id: bg_block_info.c 20097 2010-04-20 16:42:16Z da $
*****************************************************************************
* Copyright (C) 2004-2006 The Regents of the University of California.
* Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
@@ -184,6 +184,12 @@ extern int block_ready(struct job_record *job_ptr)
} else if ((bg_record->user_uid == job_ptr->user_id)
&& (bg_record->state
== RM_PARTITION_READY)) {
+ /* Clear the state just incase we
+ missed it somehow.
+ */
+ job_ptr->job_state &= (~JOB_CONFIGURING);
+ last_job_update = time(NULL);
+
rc = 1;
} else if (bg_record->user_uid != job_ptr->user_id)
rc = 0;
diff --git a/src/plugins/select/bluegene/plugin/bg_job_place.c b/src/plugins/select/bluegene/plugin/bg_job_place.c
index acd5c679d..e8139e4a0 100644
--- a/src/plugins/select/bluegene/plugin/bg_job_place.c
+++ b/src/plugins/select/bluegene/plugin/bg_job_place.c
@@ -59,15 +59,7 @@ _STMT_START { \
pthread_mutex_t create_dynamic_mutex = PTHREAD_MUTEX_INITIALIZER;
-/* This list is for the test_job_list function because we will be
- * adding and removing blocks off the bg_lists->job_running and don't want
- * to ruin that list in submit_job it should = bg_lists->job_running
- * otherwise it should be a copy of that list.
- */
-List job_block_test_list = NULL;
-
static void _rotate_geo(uint16_t *req_geometry, int rot_cnt);
-static int _bg_record_sort_aval_inc(bg_record_t* rec_a, bg_record_t* rec_b);
static int _get_user_groups(uint32_t user_id, uint32_t group_id,
gid_t *groups, int max_groups, int *ngroups);
static int _test_image_perms(char *image_name, List image_list,
@@ -126,62 +118,6 @@ static void _rotate_geo(uint16_t *req_geometry, int rot_cnt)
}
}
-/*
- * Comparator used for sorting blocks smallest to largest
- *
- * returns: -1: rec_a < rec_b 0: rec_a == rec_b 1: rec_a > rec_b
- *
- */
-static int _bg_record_sort_aval_inc(bg_record_t* rec_a, bg_record_t* rec_b)
-{
- if((rec_a->job_running == BLOCK_ERROR_STATE)
- && (rec_b->job_running != BLOCK_ERROR_STATE))
- return 1;
- else if((rec_a->job_running != BLOCK_ERROR_STATE)
- && (rec_b->job_running == BLOCK_ERROR_STATE))
- return -1;
- else if(!rec_a->job_ptr && rec_b->job_ptr)
- return 1;
- else if(rec_a->job_ptr && !rec_b->job_ptr)
- return -1;
- else if(rec_a->job_ptr && rec_b->job_ptr) {
- if(rec_a->job_ptr->start_time > rec_b->job_ptr->start_time)
- return 1;
- else if(rec_a->job_ptr->start_time < rec_b->job_ptr->start_time)
- return -1;
- }
-
- return bg_record_cmpf_inc(rec_a, rec_b);
-}
-
-/*
- * Comparator used for sorting blocks smallest to largest
- *
- * returns: -1: rec_a > rec_b 0: rec_a == rec_b 1: rec_a < rec_b
- *
- */
-static int _bg_record_sort_aval_dec(bg_record_t* rec_a, bg_record_t* rec_b)
-{
- if((rec_a->job_running == BLOCK_ERROR_STATE)
- && (rec_b->job_running != BLOCK_ERROR_STATE))
- return -1;
- else if((rec_a->job_running != BLOCK_ERROR_STATE)
- && (rec_b->job_running == BLOCK_ERROR_STATE))
- return 1;
- else if(!rec_a->job_ptr && rec_b->job_ptr)
- return -1;
- else if(rec_a->job_ptr && !rec_b->job_ptr)
- return 1;
- else if(rec_a->job_ptr && rec_b->job_ptr) {
- if(rec_a->job_ptr->start_time > rec_b->job_ptr->start_time)
- return -1;
- else if(rec_a->job_ptr->start_time < rec_b->job_ptr->start_time)
- return 1;
- }
-
- return bg_record_cmpf_inc(rec_a, rec_b);
-}
-
/*
* Get a list of groups associated with a specific user_id
* Return 0 on success, -1 on failure
@@ -360,6 +296,13 @@ static bg_record_t *_find_matching_block(List block_list,
/* If test_only we want to fall through to tell the
scheduler that it is runnable just not right now.
*/
+
+ /* The job running could be reset so set it back up
+ here if there is a job_ptr
+ */
+ if(bg_record->job_ptr)
+ bg_record->job_running = bg_record->job_ptr->job_id;
+
debug3("%s job_running = %d",
bg_record->bg_block_id, bg_record->job_running);
/*block is messed up some how (BLOCK_ERROR_STATE)
@@ -724,18 +667,18 @@ static int _dynamically_request(List block_list, int *blocks_added,
list_of_lists = list_create(NULL);
if(user_req_nodes)
- list_append(list_of_lists, job_block_test_list);
+ list_append(list_of_lists, bg_lists->job_running);
else {
list_append(list_of_lists, block_list);
- if(job_block_test_list == bg_lists->job_running &&
- list_count(block_list) != list_count(bg_lists->booted)) {
+ if(list_count(block_list) != list_count(bg_lists->booted)) {
list_append(list_of_lists, bg_lists->booted);
if(list_count(bg_lists->booted)
- != list_count(job_block_test_list))
- list_append(list_of_lists, job_block_test_list);
+ != list_count(bg_lists->job_running))
+ list_append(list_of_lists,
+ bg_lists->job_running);
} else if(list_count(block_list)
- != list_count(job_block_test_list)) {
- list_append(list_of_lists, job_block_test_list);
+ != list_count(bg_lists->job_running)) {
+ list_append(list_of_lists, bg_lists->job_running);
}
}
itr = list_iterator_create(list_of_lists);
@@ -765,20 +708,14 @@ static int _dynamically_request(List block_list, int *blocks_added,
list_append(block_list, bg_record);
(*blocks_added) = 1;
} else {
- if(job_block_test_list
- == bg_lists->job_running) {
- if(configure_block(bg_record)
- == SLURM_ERROR) {
- destroy_bg_record(
- bg_record);
- error("_dynamically_"
- "request: "
- "unable to "
- "configure "
- "block");
- rc = SLURM_ERROR;
- break;
- }
+ if(configure_block(bg_record)
+ == SLURM_ERROR) {
+ destroy_bg_record(bg_record);
+ error("_dynamically_request: "
+ "unable to configure "
+ "block");
+ rc = SLURM_ERROR;
+ break;
}
list_append(block_list, bg_record);
print_bg_record(bg_record);
@@ -793,7 +730,7 @@ static int _dynamically_request(List block_list, int *blocks_added,
continue;
}
list_sort(block_list,
- (ListCmpF)_bg_record_sort_aval_dec);
+ (ListCmpF)bg_record_sort_aval_inc);
rc = SLURM_SUCCESS;
break;
@@ -1061,15 +998,40 @@ static int _find_best_block_match(List block_list,
if(is_test) {
List new_blocks = NULL;
- List job_list = NULL;
+ List job_list = list_create(NULL);
+ ListIterator itr = NULL;
debug("trying with empty machine");
- slurm_mutex_lock(&block_state_mutex);
- if(job_block_test_list == bg_lists->job_running)
- job_list = copy_bg_list(job_block_test_list);
- else
- job_list = job_block_test_list;
- slurm_mutex_unlock(&block_state_mutex);
- list_sort(job_list, (ListCmpF)_bg_record_sort_aval_inc);
+
+ /* Here we need to make sure the blocks in the
+ job list are those in the block list so go
+ through and grab them and add them to a
+ separate list.
+ */
+ itr = list_iterator_create(block_list);
+ while((bg_record = list_next(itr))) {
+ if(bg_record->job_running != NO_JOB_RUNNING)
+ list_append(job_list, bg_record);
+ /* Since the error blocks are at the
+ end we only really need to look at
+ the first one to make sure it will
+ work, so don't add more than one to
+ the job list.
+ We do need to check for at least
+ one error block because that lets
+ us know not to hold up the entire
+ machine for a job that won't run
+ until the error is removed which
+ could be a very long time.
+ */
+ if(bg_record->job_running == BLOCK_ERROR_STATE)
+ break;
+ }
+ list_iterator_destroy(itr);
+
+ /* Block list is already in the correct order,
+ earliest avaliable first,
+ so the job list will also be. No need to
+ sort. */
while(1) {
bool track_down_nodes = true;
/* this gets altered in
@@ -1077,9 +1039,8 @@ static int _find_best_block_match(List block_list,
for(i=0; i<BA_SYSTEM_DIMENSIONS; i++)
request.geometry[i] = req_geometry[i];
- bg_record = list_pop(job_list);
- if(bg_record) {
- if(bg_record->job_ptr)
+ if((bg_record = list_pop(job_list))) {
+ if(bg_record->job_ptr) {
debug2("taking off %d(%s) "
"started at %d "
"ends at %d",
@@ -1089,8 +1050,21 @@ static int _find_best_block_match(List block_list,
start_time,
bg_record->job_ptr->
end_time);
- else if(bg_record->job_running
- == BLOCK_ERROR_STATE)
+ /* Mark the block as
+ not running a job,
+ this should
+ correspond to the
+ pointer in the
+ block_list. We
+ only look at the
+ job_running var so
+ don't remove the
+ job_ptr.
+ */
+ bg_record->job_running =
+ NO_JOB_RUNNING;
+ } else if(bg_record->job_running
+ == BLOCK_ERROR_STATE)
debug2("taking off (%s) "
"which is in an error "
"state",
@@ -1106,7 +1080,6 @@ static int _find_best_block_match(List block_list,
if(!(new_blocks = create_dynamic_block(
block_list, &request, job_list,
track_down_nodes))) {
- destroy_bg_record(bg_record);
if(errno == ESLURM_INTERCONNECT_FAILURE
|| !list_count(job_list)) {
char *nodes;
@@ -1133,10 +1106,9 @@ static int _find_best_block_match(List block_list,
if(!(*found_bg_record)) {
error("got an empty list back");
list_destroy(new_blocks);
- if(bg_record) {
- destroy_bg_record(bg_record);
+ if(bg_record)
continue;
- } else {
+ else {
rc = SLURM_ERROR;
break;
}
@@ -1145,36 +1117,16 @@ static int _find_best_block_match(List block_list,
(*found_bg_record)->bitmap);
if(bg_record) {
+ (*found_bg_record)->job_running =
+ bg_record->job_running;
(*found_bg_record)->job_ptr
= bg_record->job_ptr;
- destroy_bg_record(bg_record);
}
-
- if(job_block_test_list
- != bg_lists->job_running) {
- list_append(block_list,
- (*found_bg_record));
- while((bg_record =
- list_pop(new_blocks))) {
- if(block_exist_in_list(
- block_list,
- bg_record))
- destroy_bg_record(
- bg_record);
- else {
- list_append(block_list,
- bg_record);
-// print_bg_record(bg_record);
- }
- }
- }
-
list_destroy(new_blocks);
break;
}
- if(job_block_test_list == bg_lists->job_running)
- list_destroy(job_list);
+ list_destroy(job_list);
goto end_it;
} else {
@@ -1474,8 +1426,6 @@ extern int submit_job(struct job_record *job_ptr, bitstr_t *slurm_block_bitmap,
if(bg_conf->layout_mode == LAYOUT_DYNAMIC)
slurm_mutex_lock(&create_dynamic_mutex);
- job_block_test_list = bg_lists->job_running;
-
select_g_select_jobinfo_get(job_ptr->select_jobinfo,
SELECT_JOBDATA_CONN_TYPE, &conn_type);
if(conn_type == SELECT_NAV) {
@@ -1503,7 +1453,7 @@ extern int submit_job(struct job_record *job_ptr, bitstr_t *slurm_block_bitmap,
select_g_select_jobinfo_sprint(job_ptr->select_jobinfo,
buf, sizeof(buf),
SELECT_PRINT_MIXED);
- debug("bluegene:submit_job: %u mode=%d %s nodes=%u-%u-%u",
+ debug("bluegene:submit_job: %u mode=%d %s nodes=%u-%u-%u",
job_ptr->job_id, local_mode, buf,
min_nodes, req_nodes, max_nodes);
select_g_select_jobinfo_sprint(job_ptr->select_jobinfo,
@@ -1540,7 +1490,7 @@ extern int submit_job(struct job_record *job_ptr, bitstr_t *slurm_block_bitmap,
/* just remove the preemptable jobs now since we are treating
this as a run now deal */
preempt:
- list_sort(block_list, (ListCmpF)_bg_record_sort_aval_dec);
+ list_sort(block_list, (ListCmpF)bg_record_sort_aval_inc);
rc = _find_best_block_match(block_list, &blocks_added,
job_ptr, slurm_block_bitmap, min_nodes,
diff --git a/src/plugins/select/bluegene/plugin/bg_job_run.c b/src/plugins/select/bluegene/plugin/bg_job_run.c
index d69e4addd..839f27178 100644
--- a/src/plugins/select/bluegene/plugin/bg_job_run.c
+++ b/src/plugins/select/bluegene/plugin/bg_job_run.c
@@ -2,7 +2,7 @@
* bg_job_run.c - blue gene job execution (e.g. initiation and termination)
* functions.
*
- * $Id: bg_job_run.c 20011 2010-04-13 19:10:21Z da $
+ * $Id: bg_job_run.c 20097 2010-04-20 16:42:16Z da $
*****************************************************************************
* Copyright (C) 2004-2006 The Regents of the University of California.
* Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
@@ -547,6 +547,10 @@ static void _sync_agent(bg_update_t *bg_update_ptr)
list_push(bg_lists->booted, bg_record);
if(bg_record->state == RM_PARTITION_READY) {
+ if(bg_record->job_ptr) {
+ bg_record->job_ptr->job_state &= (~JOB_CONFIGURING);
+ last_job_update = time(NULL);
+ }
if(bg_record->user_uid != bg_update_ptr->job_ptr->user_id) {
int set_user_rc = SLURM_SUCCESS;
diff --git a/src/plugins/select/bluegene/plugin/bg_record_functions.c b/src/plugins/select/bluegene/plugin/bg_record_functions.c
index fe79d6e5f..dea4d8248 100644
--- a/src/plugins/select/bluegene/plugin/bg_record_functions.c
+++ b/src/plugins/select/bluegene/plugin/bg_record_functions.c
@@ -522,6 +522,34 @@ extern int bg_record_cmpf_inc(bg_record_t* rec_a, bg_record_t* rec_b)
return 0;
}
+/*
+ * Comparator used for sorting blocks from earliest avaliable to lastest
+ *
+ * returns: -1: rec_a < rec_b 0: rec_a == rec_b 1: rec_a > rec_b
+ *
+ */
+extern int bg_record_sort_aval_inc(bg_record_t* rec_a, bg_record_t* rec_b)
+{
+ if((rec_a->job_running == BLOCK_ERROR_STATE)
+ && (rec_b->job_running != BLOCK_ERROR_STATE))
+ return 1;
+ else if((rec_a->job_running != BLOCK_ERROR_STATE)
+ && (rec_b->job_running == BLOCK_ERROR_STATE))
+ return -1;
+ else if(!rec_a->job_ptr && rec_b->job_ptr)
+ return -1;
+ else if(rec_a->job_ptr && !rec_b->job_ptr)
+ return 1;
+ else if(rec_a->job_ptr && rec_b->job_ptr) {
+ if(rec_a->job_ptr->end_time > rec_b->job_ptr->end_time)
+ return 1;
+ else if(rec_a->job_ptr->end_time < rec_b->job_ptr->end_time)
+ return -1;
+ }
+
+ return bg_record_cmpf_inc(rec_a, rec_b);
+}
+
/* if looking at the main list this should have some nice
* block_state_mutex locks around it.
*/
diff --git a/src/plugins/select/bluegene/plugin/bg_record_functions.h b/src/plugins/select/bluegene/plugin/bg_record_functions.h
index 26ae7daea..45e2b2a9d 100644
--- a/src/plugins/select/bluegene/plugin/bg_record_functions.h
+++ b/src/plugins/select/bluegene/plugin/bg_record_functions.h
@@ -122,6 +122,7 @@ extern void process_nodes(bg_record_t *bg_reord, bool startup);
extern List copy_bg_list(List in_list);
extern void copy_bg_record(bg_record_t *fir_record, bg_record_t *sec_record);
extern int bg_record_cmpf_inc(bg_record_t *rec_a, bg_record_t *rec_b);
+extern int bg_record_sort_aval_inc(bg_record_t* rec_a, bg_record_t* rec_b);
/* return bg_record from a bg_list */
extern bg_record_t *find_bg_record_in_list(List my_list, char *bg_block_id);
diff --git a/src/plugins/select/bluegene/plugin/dynamic_block.c b/src/plugins/select/bluegene/plugin/dynamic_block.c
index bca33d8e5..ff6c65e0a 100644
--- a/src/plugins/select/bluegene/plugin/dynamic_block.c
+++ b/src/plugins/select/bluegene/plugin/dynamic_block.c
@@ -179,6 +179,9 @@ extern List create_dynamic_block(List block_list,
break;
}
+ /* Sort the list so the small blocks are in the order
+ * of ionodes. */
+ list_sort(block_list, (ListCmpF)bg_record_cmpf_inc);
request->conn_type = SELECT_SMALL;
new_blocks = list_create(destroy_bg_record);
/* check only blocks that are free and small */
@@ -209,6 +212,8 @@ extern List create_dynamic_block(List block_list,
== SLURM_SUCCESS)
goto finished;
+ /* Re-sort the list back to the original order. */
+ list_sort(block_list, (ListCmpF)bg_record_sort_aval_inc);
list_destroy(new_blocks);
new_blocks = NULL;
debug2("small block not able to be placed inside others");
@@ -612,8 +617,8 @@ static int _breakup_blocks(List block_list, List new_blocks,
int cnodes = request->procs / bg_conf->cpu_ratio;
int curr_bp_bit = -1;
- debug2("proc count = %d cnodes = %d size = %d",
- request->procs, cnodes, request->size);
+ debug2("cpu_count= %d cnodes=%d o_free=%d o_small=%d",
+ request->procs, cnodes, only_free, only_small);
switch(cnodes) {
case 16:
@@ -651,6 +656,7 @@ static int _breakup_blocks(List block_list, List new_blocks,
/* check for free blocks on the first and second time */
if(only_free && (bg_record->state != RM_PARTITION_FREE))
continue;
+
/* check small blocks first */
if(only_small && (bg_record->node_cnt > bg_conf->bp_node_cnt))
continue;
diff --git a/src/plugins/select/bluegene/plugin/state_test.c b/src/plugins/select/bluegene/plugin/state_test.c
index 03c3595bc..2c81188f9 100644
--- a/src/plugins/select/bluegene/plugin/state_test.c
+++ b/src/plugins/select/bluegene/plugin/state_test.c
@@ -2,7 +2,7 @@
* state_test.c - Test state of Bluegene base partitions and switches.
* DRAIN nodes in SLURM that are not usable.
*
- * $Id: state_test.c 19998 2010-04-12 19:18:46Z da $
+ * $Id: state_test.c 20152 2010-04-27 18:01:39Z da $
*****************************************************************************
* Copyright (C) 2004-2007 The Regents of the University of California.
* Copyright (C) 2008-2009 Lawrence Livermore National Security.
@@ -321,7 +321,13 @@ static int _test_down_nodecards(rm_BP_t *bp_ptr, bool slurmctld_locked)
goto clean_up;
}
- node_name = _get_bp_node_name(bp_ptr);
+ /* The node_name will only be NULL if this system doesn't
+ really have the node.
+ */
+ if(!(node_name = _get_bp_node_name(bp_ptr))) {
+ rc = SLURM_ERROR;
+ goto clean_up;
+ }
if((rc = bridge_get_data(ncard_list, RM_NodeCardListSize, &num))
!= STATUS_OK) {
diff --git a/src/plugins/select/cons_res/dist_tasks.c b/src/plugins/select/cons_res/dist_tasks.c
index f83e37f23..9714117c8 100644
--- a/src/plugins/select/cons_res/dist_tasks.c
+++ b/src/plugins/select/cons_res/dist_tasks.c
@@ -71,7 +71,7 @@
static int _compute_c_b_task_dist(struct job_record *job_ptr)
{
bool over_subscribe = false;
- uint32_t n, i, tid, maxtasks;
+ uint32_t n, i, tid, maxtasks, l;
uint16_t *avail_cpus;
job_resources_t *job_res = job_ptr->job_resrcs;
if (!job_res || !job_res->cpus) {
@@ -83,6 +83,9 @@ static int _compute_c_b_task_dist(struct job_record *job_ptr)
avail_cpus = job_res->cpus;
job_res->cpus = xmalloc(job_res->nhosts * sizeof(uint16_t));
+ if (job_ptr->details->cpus_per_task > 1)
+ maxtasks = maxtasks / job_ptr->details->cpus_per_task;
+
for (tid = 0, i = 0; (tid < maxtasks); i++) { /* cycle counter */
bool space_remaining = false;
if (over_subscribe) {
@@ -96,10 +99,12 @@ static int _compute_c_b_task_dist(struct job_record *job_ptr)
for (n = 0; ((n < job_res->nhosts) && (tid < maxtasks)); n++) {
if ((i < avail_cpus[n]) || over_subscribe) {
tid++;
- if (job_res->cpus[n] < avail_cpus[n])
- job_res->cpus[n]++;
- if ((i + 1) < avail_cpus[n])
- space_remaining = true;
+ for (l = 0; l < job_ptr->details->cpus_per_task; l++) {
+ if (job_res->cpus[n] < avail_cpus[n])
+ job_res->cpus[n]++;
+ if ((i + 1) < avail_cpus[n])
+ space_remaining = true;
+ }
}
}
if (!space_remaining) {
@@ -192,9 +197,17 @@ static void _block_sync_core_bitmap(struct job_record *job_ptr,
alloc_cores = true;
#endif
- if (job_ptr->details && job_ptr->details->mc_ptr &&
- job_ptr->details->mc_ptr->ntasks_per_core)
- ntasks_per_core = job_ptr->details->mc_ptr->ntasks_per_core;
+ if (job_ptr->details && job_ptr->details->mc_ptr) {
+ if (job_ptr->details->mc_ptr->ntasks_per_core) {
+ ntasks_per_core = job_ptr->details->mc_ptr->
+ ntasks_per_core;
+ }
+ if ((job_ptr->details->mc_ptr->min_threads !=
+ (uint16_t) NO_VAL) &&
+ (job_ptr->details->mc_ptr->min_threads < ntasks_per_core)) {
+ ntasks_per_core = job_ptr->details->mc_ptr->min_threads;
+ }
+ }
size = bit_size(job_res->node_bitmap);
csize = bit_size(job_res->core_bitmap);
@@ -276,9 +289,17 @@ static void _cyclic_sync_core_bitmap(struct job_record *job_ptr,
alloc_cores = true;
#endif
core_map = job_res->core_bitmap;
- if (job_ptr->details && job_ptr->details->mc_ptr &&
- job_ptr->details->mc_ptr->ntasks_per_core)
- ntasks_per_core = job_ptr->details->mc_ptr->ntasks_per_core;
+ if (job_ptr->details && job_ptr->details->mc_ptr) {
+ if (job_ptr->details->mc_ptr->ntasks_per_core) {
+ ntasks_per_core = job_ptr->details->mc_ptr->
+ ntasks_per_core;
+ }
+ if ((job_ptr->details->mc_ptr->min_threads !=
+ (uint16_t) NO_VAL) &&
+ (job_ptr->details->mc_ptr->min_threads < ntasks_per_core)) {
+ ntasks_per_core = job_ptr->details->mc_ptr->min_threads;
+ }
+ }
sock_size = select_node_record[0].sockets;
sock_start = xmalloc(sock_size * sizeof(uint32_t));
diff --git a/src/plugins/select/cons_res/job_test.c b/src/plugins/select/cons_res/job_test.c
index 596577adf..af31c8f1c 100644
--- a/src/plugins/select/cons_res/job_test.c
+++ b/src/plugins/select/cons_res/job_test.c
@@ -138,12 +138,21 @@ uint16_t _allocate_sockets(struct job_record *job_ptr, bitstr_t *core_map,
uint16_t i, c, sockets = select_node_record[node_i].sockets;
uint16_t cores_per_socket = select_node_record[node_i].cores;
uint16_t threads_per_core = select_node_record[node_i].vpus;
- uint16_t min_cores = 0, min_sockets = 0, ntasks_per_socket = 0;
+ uint16_t min_cores = 1, min_sockets = 1, ntasks_per_socket = 0;
+ uint16_t max_cores = 0xffff, max_sockets = 0xffff, max_threads = 0xffff;
uint16_t ntasks_per_core = 0xffff;
if (job_ptr->details && job_ptr->details->mc_ptr) {
- min_cores = job_ptr->details->mc_ptr->min_cores;
- min_sockets = job_ptr->details->mc_ptr->min_sockets;
+ if (job_ptr->details->mc_ptr->min_cores != (uint16_t) NO_VAL) {
+ min_cores = job_ptr->details->mc_ptr->min_cores;
+ max_cores = job_ptr->details->mc_ptr->min_cores;
+ }
+ if (job_ptr->details->mc_ptr->min_sockets != (uint16_t) NO_VAL){
+ min_sockets = job_ptr->details->mc_ptr->min_sockets;
+ max_sockets = job_ptr->details->mc_ptr->min_sockets;
+ }
+ if (job_ptr->details->mc_ptr->min_threads != (uint16_t) NO_VAL)
+ max_threads = job_ptr->details->mc_ptr->min_threads;
if (job_ptr->details->mc_ptr->ntasks_per_core) {
ntasks_per_core = job_ptr->details->mc_ptr->
ntasks_per_core;
@@ -249,10 +258,22 @@ uint16_t _allocate_sockets(struct job_record *job_ptr, bitstr_t *core_map,
goto fini;
}
+ /* check max_cores and max_sockets */
c = 0;
for (i = 0; i < sockets; i++) {
+ if (free_cores[i] > max_cores) {
+ /* remove extra cores from this socket */
+ uint16_t tmp = free_cores[i] - max_cores;
+ free_core_count -= tmp;
+ free_cores[i] -= tmp;
+ }
if (free_cores[i] > 0)
c++;
+ if (free_cores[i] && c > max_sockets) {
+ /* remove extra sockets from use */
+ free_core_count -= free_cores[i];
+ free_cores[i] = 0;
+ }
}
if (free_core_count < 1) {
/* no available resources on this node */
@@ -260,7 +281,6 @@ uint16_t _allocate_sockets(struct job_record *job_ptr, bitstr_t *core_map,
goto fini;
}
-
/* Step 3: Compute task-related data:
* ntasks_per_socket, ntasks_per_node and cpus_per_task
* to determine the number of tasks to run on this node
@@ -270,7 +290,7 @@ uint16_t _allocate_sockets(struct job_record *job_ptr, bitstr_t *core_map,
*/
avail_cpus = 0;
num_tasks = 0;
- threads_per_core = MIN(threads_per_core, ntasks_per_core);
+ threads_per_core = MIN(threads_per_core, max_threads);
for (i = 0; i < sockets; i++) {
uint16_t tmp = free_cores[i] * threads_per_core;
avail_cpus += tmp;
@@ -279,7 +299,11 @@ uint16_t _allocate_sockets(struct job_record *job_ptr, bitstr_t *core_map,
else
num_tasks += tmp;
}
- if (job_ptr->details->ntasks_per_node)
+
+ /* If job requested exclusive rights to the node don't do the
+ min here since it will make it so we don't allocate the
+ entire node. */
+ if (job_ptr->details->ntasks_per_node && job_ptr->details->shared)
num_tasks = MIN(num_tasks, job_ptr->details->ntasks_per_node);
if (cpus_per_task < 2) {
@@ -371,12 +395,21 @@ uint16_t _allocate_cores(struct job_record *job_ptr, bitstr_t *core_map,
uint16_t i, c, sockets = select_node_record[node_i].sockets;
uint16_t cores_per_socket = select_node_record[node_i].cores;
uint16_t threads_per_core = select_node_record[node_i].vpus;
- uint16_t min_cores = 0, min_sockets = 0;
+ uint16_t min_cores = 1, min_sockets = 1;
+ uint16_t max_cores = 0xffff, max_sockets = 0xffff, max_threads = 0xffff;
uint16_t ntasks_per_core = 0xffff;
if (!cpu_type && job_ptr->details && job_ptr->details->mc_ptr) {
- min_cores = job_ptr->details->mc_ptr->min_cores;
- min_sockets = job_ptr->details->mc_ptr->min_sockets;
+ if (job_ptr->details->mc_ptr->min_cores != (uint16_t) NO_VAL) {
+ min_cores = job_ptr->details->mc_ptr->min_cores;
+ max_cores = job_ptr->details->mc_ptr->min_cores;
+ }
+ if (job_ptr->details->mc_ptr->min_sockets != (uint16_t) NO_VAL){
+ min_sockets = job_ptr->details->mc_ptr->min_sockets;
+ max_sockets = job_ptr->details->mc_ptr->min_sockets;
+ }
+ if (job_ptr->details->mc_ptr->min_threads != (uint16_t) NO_VAL)
+ max_threads = job_ptr->details->mc_ptr->min_threads;
if (job_ptr->details->mc_ptr->ntasks_per_core) {
ntasks_per_core = job_ptr->details->mc_ptr->
ntasks_per_core;
@@ -463,10 +496,22 @@ uint16_t _allocate_cores(struct job_record *job_ptr, bitstr_t *core_map,
goto fini;
}
+ /* Step 2b: check max_cores per socket and max_sockets per node */
c = 0;
for (i = 0; i < sockets; i++) {
+ if (free_cores[i] > max_cores) {
+ /* remove extra cores from this socket */
+ uint16_t tmp = free_cores[i] - max_cores;
+ free_core_count -= tmp;
+ free_cores[i] -= tmp;
+ }
if (free_cores[i] > 0)
c++;
+ if (free_cores[i] && (c > max_sockets)) {
+ /* remove extra sockets from use */
+ free_core_count -= free_cores[i];
+ free_cores[i] = 0;
+ }
}
if (free_core_count < 1) {
/* no available resources on this node */
@@ -474,7 +519,6 @@ uint16_t _allocate_cores(struct job_record *job_ptr, bitstr_t *core_map,
goto fini;
}
-
/* Step 3: Compute task-related data: use ntasks_per_core,
* ntasks_per_node and cpus_per_task to determine
* the number of tasks to run on this node
@@ -482,7 +526,10 @@ uint16_t _allocate_cores(struct job_record *job_ptr, bitstr_t *core_map,
* Note: cpus_per_task and ntasks_per_core need to play nice
* 2 tasks_per_core vs. 2 cpus_per_task
*/
- threads_per_core = MIN(threads_per_core, ntasks_per_core);
+ if (cpu_type)
+ max_threads = threads_per_core;
+ else
+ threads_per_core = MIN(threads_per_core, max_threads);
num_tasks = avail_cpus = threads_per_core;
i = job_ptr->details->mc_ptr->ntasks_per_core;
if (!cpu_type && i > 0)
@@ -490,9 +537,12 @@ uint16_t _allocate_cores(struct job_record *job_ptr, bitstr_t *core_map,
/* convert from PER_CORE to TOTAL_FOR_NODE */
avail_cpus *= free_core_count;
- num_tasks *= free_core_count;
+ num_tasks *= free_core_count;
- if (job_ptr->details->ntasks_per_node)
+ /* If job requested exclusive rights to the node don't do the
+ min here since it will make it so we don't allocate the
+ entire node */
+ if (job_ptr->details->ntasks_per_node && job_ptr->details->shared)
num_tasks = MIN(num_tasks, job_ptr->details->ntasks_per_node);
if (cpus_per_task < 2) {
@@ -525,6 +575,7 @@ uint16_t _allocate_cores(struct job_record *job_ptr, bitstr_t *core_map,
avail_cpus = 0;
}
}
+
/* clear leftovers */
if (c < core_end)
bit_nclear(core_map, c, core_end-1);
@@ -1690,9 +1741,12 @@ extern int cr_job_test(struct job_record *job_ptr, bitstr_t *bitmap,
/* This is the case if -O/--overcommit is true */
if (job_ptr->num_procs == job_ptr->details->min_nodes) {
struct multi_core_data *mc_ptr = job_ptr->details->mc_ptr;
- job_ptr->num_procs *= MAX(1, mc_ptr->min_threads);
- job_ptr->num_procs *= MAX(1, mc_ptr->min_cores);
- job_ptr->num_procs *= MAX(1, mc_ptr->min_sockets);
+ if (mc_ptr->min_threads != (uint16_t) NO_VAL)
+ job_ptr->num_procs *= MAX(1, mc_ptr->min_threads);
+ if (mc_ptr->min_cores != (uint16_t) NO_VAL)
+ job_ptr->num_procs *= MAX(1, mc_ptr->min_cores);
+ if (mc_ptr->min_sockets != (uint16_t) NO_VAL)
+ job_ptr->num_procs *= MAX(1, mc_ptr->min_sockets);
}
debug3("cons_res: cr_job_test: evaluating job %u on %u nodes",
diff --git a/src/plugins/select/cons_res/select_cons_res.c b/src/plugins/select/cons_res/select_cons_res.c
index 62500a7b1..7278eea88 100644
--- a/src/plugins/select/cons_res/select_cons_res.c
+++ b/src/plugins/select/cons_res/select_cons_res.c
@@ -1065,9 +1065,9 @@ static struct multi_core_data * _create_default_mc(void)
{
struct multi_core_data *mc_ptr;
mc_ptr = xmalloc(sizeof(struct multi_core_data));
- mc_ptr->min_sockets = 1;
- mc_ptr->min_cores = 1;
- mc_ptr->min_threads = 1;
+ mc_ptr->min_sockets = (uint16_t) NO_VAL;
+ mc_ptr->min_cores = (uint16_t) NO_VAL;
+ mc_ptr->min_threads = (uint16_t) NO_VAL;
/* mc_ptr is initialized to zero by xmalloc*/
/* mc_ptr->ntasks_per_socket = 0; */
/* mc_ptr->ntasks_per_core = 0; */
diff --git a/src/plugins/select/linear/select_linear.c b/src/plugins/select/linear/select_linear.c
index 59a6c700c..69e76d730 100644
--- a/src/plugins/select/linear/select_linear.c
+++ b/src/plugins/select/linear/select_linear.c
@@ -392,10 +392,9 @@ static uint16_t _get_avail_cpus(struct job_record *job_ptr, int index)
uint16_t avail_cpus;
uint16_t cpus, sockets, cores, threads;
uint16_t cpus_per_task = 1;
- uint16_t ntasks_per_node = 0, ntasks_per_socket = 0;
- uint16_t ntasks_per_core = 0;
+ uint16_t ntasks_per_node = 0, ntasks_per_socket, ntasks_per_core;
+ uint16_t min_sockets, min_cores, min_threads;
multi_core_data_t *mc_ptr = NULL;
- int min_sockets = 0, min_cores = 0;
if (job_ptr->details == NULL)
return (uint16_t) 0;
@@ -407,6 +406,15 @@ static uint16_t _get_avail_cpus(struct job_record *job_ptr, int index)
if ((mc_ptr = job_ptr->details->mc_ptr)) {
ntasks_per_socket = mc_ptr->ntasks_per_socket;
ntasks_per_core = mc_ptr->ntasks_per_core;
+ min_sockets = mc_ptr->min_sockets;
+ min_cores = mc_ptr->min_cores;
+ min_threads = mc_ptr->min_threads;
+ } else {
+ ntasks_per_socket = 0;
+ ntasks_per_core = 0;
+ min_sockets = (uint16_t) NO_VAL;
+ min_cores = (uint16_t) NO_VAL;
+ min_threads = (uint16_t) NO_VAL;
}
node_ptr = &(select_node_ptr[index]);
@@ -428,7 +436,7 @@ static uint16_t _get_avail_cpus(struct job_record *job_ptr, int index)
#endif
avail_cpus = slurm_get_avail_procs(
- min_sockets, min_cores, cpus_per_task,
+ min_sockets, min_cores, min_threads, cpus_per_task,
ntasks_per_node, ntasks_per_socket, ntasks_per_core,
&cpus, &sockets, &cores, &threads, NULL,
SELECT_TYPE_INFO_NONE,
diff --git a/src/sacct/options.c b/src/sacct/options.c
index c5959b729..77d4d41be 100644
--- a/src/sacct/options.c
+++ b/src/sacct/options.c
@@ -484,6 +484,8 @@ sacct [<OPTION>] \n\
-V, --version: Print version. \n\
-W, --wckeys: \n\
Only send data about these wckeys. Default is all. \n\
+ -x, --associations: \n\
+ Only send data about these association id. Default is all.\n\
-X, --allocations: \n\
Only show cumulative statistics for each job, not the \n\
intermediate steps. \n\
@@ -650,6 +652,7 @@ void parse_command_line(int argc, char **argv)
{"verbose", 0, 0, 'v'},
{"version", 0, 0, 'V'},
{"wckeys", 1, 0, 'W'},
+ {"associations", 1, 0, 'x'},
{0, 0, 0, 0}};
params.opt_uid = getuid();
@@ -661,7 +664,7 @@ void parse_command_line(int argc, char **argv)
while (1) { /* now cycle through the command line */
c = getopt_long(argc, argv,
- "aA:bcC:dDeE:f:g:hi:I:j:lLnN:o:OpPr:s:S:Ttu:vVW:X",
+ "aA:bcC:dDeE:f:g:hi:I:j:lLnN:o:OpPr:s:S:Ttu:vVW:x:X",
long_options, &optionIndex);
if (c == -1)
break;
@@ -839,6 +842,13 @@ void parse_command_line(int argc, char **argv)
case 'V':
print_slurm_version();
exit(0);
+ case 'x':
+ if(!job_cond->associd_list)
+ job_cond->associd_list =
+ list_create(slurm_destroy_char);
+ slurm_addto_char_list(job_cond->associd_list, optarg);
+ break;
+
case 't':
case 'X':
params.opt_allocs = 1;
diff --git a/src/sacctmgr/account_functions.c b/src/sacctmgr/account_functions.c
index cb415032e..133b00e57 100644
--- a/src/sacctmgr/account_functions.c
+++ b/src/sacctmgr/account_functions.c
@@ -878,7 +878,8 @@ extern int sacctmgr_add_account(int argc, char *argv[])
assoc_list);
} else {
exit_code=1;
- fprintf(stderr, " Problem adding accounts\n");
+ fprintf(stderr, " Problem adding accounts: %s\n",
+ slurm_strerror(rc));
rc = SLURM_ERROR;
notice_thread_fini();
goto end_it;
@@ -895,7 +896,8 @@ extern int sacctmgr_add_account(int argc, char *argv[])
} else {
exit_code=1;
fprintf(stderr,
- " error: Problem adding account associations\n");
+ " error: Problem adding account associations: %s\n",
+ slurm_strerror(rc));
rc = SLURM_ERROR;
}
@@ -907,7 +909,6 @@ end_it:
destroy_acct_association_rec(start_assoc);
destroy_acct_account_rec(start_acct);
-
return rc;
}
diff --git a/src/sacctmgr/archive_functions.c b/src/sacctmgr/archive_functions.c
index ba57aaddd..bf7e0be1a 100644
--- a/src/sacctmgr/archive_functions.c
+++ b/src/sacctmgr/archive_functions.c
@@ -425,7 +425,8 @@ extern int sacctmgr_archive_dump(int argc, char *argv[])
}
} else {
exit_code=1;
- fprintf(stderr, " Problem dumping archive\n");
+ fprintf(stderr, " Problem dumping archive: %s\n",
+ slurm_strerror(rc));
rc = SLURM_ERROR;
}
destroy_acct_archive_cond(arch_cond);
@@ -504,7 +505,8 @@ extern int sacctmgr_archive_load(int argc, char *argv[])
}
} else {
exit_code=1;
- fprintf(stderr, " Problem loading archive file\n");
+ fprintf(stderr, " Problem loading archive file: %s\n",
+ slurm_strerror(rc));
rc = SLURM_ERROR;
}
diff --git a/src/sacctmgr/cluster_functions.c b/src/sacctmgr/cluster_functions.c
index a7b25e8e6..60bf6252f 100644
--- a/src/sacctmgr/cluster_functions.c
+++ b/src/sacctmgr/cluster_functions.c
@@ -404,7 +404,8 @@ extern int sacctmgr_add_cluster(int argc, char *argv[])
}
} else {
exit_code=1;
- fprintf(stderr, " Problem adding clusters\n");
+ fprintf(stderr, " Problem adding clusters: %s\n",
+ slurm_strerror(rc));
}
end_it:
list_destroy(cluster_list);
diff --git a/src/sacctmgr/file_functions.c b/src/sacctmgr/file_functions.c
index ff461f86f..f9a49389d 100644
--- a/src/sacctmgr/file_functions.c
+++ b/src/sacctmgr/file_functions.c
@@ -2278,7 +2278,8 @@ extern void load_sacctmgr_cfg_file (int argc, char *argv[])
if(rc != SLURM_SUCCESS) {
exit_code=1;
fprintf(stderr,
- " Problem adding cluster\n");
+ " Problem adding cluster: %s\n",
+ slurm_strerror(rc));
rc = SLURM_ERROR;
_destroy_sacctmgr_file_opts(file_opts);
break;
@@ -2698,7 +2699,8 @@ extern void load_sacctmgr_cfg_file (int argc, char *argv[])
}
} else {
exit_code=1;
- fprintf(stderr, " Problem with requests.\n");
+ fprintf(stderr, " Problem with requests: %s\n",
+ slurm_strerror(rc));
}
list_destroy(format_list);
diff --git a/src/sacctmgr/qos_functions.c b/src/sacctmgr/qos_functions.c
index 696e211ac..faa0769f6 100644
--- a/src/sacctmgr/qos_functions.c
+++ b/src/sacctmgr/qos_functions.c
@@ -448,7 +448,8 @@ extern int sacctmgr_add_qos(int argc, char *argv[])
}
} else {
exit_code=1;
- fprintf(stderr, " Problem adding QOS.\n");
+ fprintf(stderr, " Problem adding QOS: %s\n",
+ slurm_strerror(rc));
rc = SLURM_ERROR;
}
diff --git a/src/sacctmgr/user_functions.c b/src/sacctmgr/user_functions.c
index e686b97a7..6f47691d1 100644
--- a/src/sacctmgr/user_functions.c
+++ b/src/sacctmgr/user_functions.c
@@ -1045,8 +1045,11 @@ extern int sacctmgr_add_user(int argc, char *argv[])
xstrdup(list_peek(wckey_cond->name_list));
wckey_cond->cluster_list = assoc_cond->cluster_list;
wckey_cond->user_list = assoc_cond->user_list;
- local_wckey_list = acct_storage_g_get_wckeys(
- db_conn, my_uid, wckey_cond);
+ if(!(local_wckey_list = acct_storage_g_get_wckeys(
+ db_conn, my_uid, wckey_cond)))
+ info("If you are a coordinator ignore "
+ "the previous error");
+
wckey_cond->cluster_list = NULL;
wckey_cond->user_list = NULL;
@@ -1371,7 +1374,8 @@ no_default:
wckey_list);
} else {
exit_code=1;
- fprintf(stderr, " Problem adding users\n");
+ fprintf(stderr, " Problem adding users: %s\n",
+ slurm_strerror(rc));
rc = SLURM_ERROR;
notice_thread_fini();
goto end_it;
@@ -1388,7 +1392,8 @@ no_default:
}
} else {
exit_code=1;
- fprintf(stderr, " Problem adding user associations\n");
+ fprintf(stderr, " Problem adding user associations: %s\n",
+ slurm_strerror(rc));
rc = SLURM_ERROR;
}
@@ -1470,7 +1475,8 @@ extern int sacctmgr_add_coord(int argc, char *argv[])
}
} else {
exit_code=1;
- fprintf(stderr, " Problem adding coordinator\n");
+ fprintf(stderr, " Problem adding coordinator: %s\n",
+ slurm_strerror(rc));
rc = SLURM_ERROR;
}
diff --git a/src/salloc/opt.c b/src/salloc/opt.c
index aab3a2c75..3253ec573 100644
--- a/src/salloc/opt.c
+++ b/src/salloc/opt.c
@@ -562,7 +562,7 @@ _get_int(const char *arg, const char *what)
void set_options(const int argc, char **argv)
{
- int opt_char, option_index = 0;
+ int opt_char, option_index = 0, max_val = 0;
char *tmp;
static struct option long_options[] = {
{"account", required_argument, 0, 'A'},
@@ -982,17 +982,26 @@ void set_options(const int argc, char **argv)
case LONG_OPT_SOCKETSPERNODE:
get_resource_arg_range( optarg, "sockets-per-node",
&opt.min_sockets_per_node,
- NULL, true );
+ &max_val, true );
+ if ((opt.min_sockets_per_node == 1) &&
+ (max_val == INT_MAX))
+ opt.min_sockets_per_node = NO_VAL;
break;
case LONG_OPT_CORESPERSOCKET:
get_resource_arg_range( optarg, "cores-per-socket",
&opt.min_cores_per_socket,
- NULL, true );
+ &max_val, true );
+ if ((opt.min_cores_per_socket == 1) &&
+ (max_val == INT_MAX))
+ opt.min_cores_per_socket = NO_VAL;
break;
case LONG_OPT_THREADSPERCORE:
get_resource_arg_range( optarg, "threads-per-core",
&opt.min_threads_per_core,
- NULL, true );
+ &max_val, true );
+ if ((opt.min_threads_per_core == 1) &&
+ (max_val == INT_MAX))
+ opt.min_threads_per_core = NO_VAL;
break;
case LONG_OPT_NTASKSPERNODE:
opt.ntasks_per_node = _get_int(optarg,
diff --git a/src/salloc/salloc.c b/src/salloc/salloc.c
index 567f61893..09d639839 100644
--- a/src/salloc/salloc.c
+++ b/src/salloc/salloc.c
@@ -102,6 +102,7 @@ static void _signal_while_allocating(int signo);
static void _job_complete_handler(srun_job_complete_msg_t *msg);
static void _set_exit_code(void);
static void _set_rlimits(char **env);
+static void _set_spank_env(void);
static void _set_submit_dir_env(void);
static void _timeout_handler(srun_timeout_msg_t *msg);
static void _user_msg_handler(srun_user_msg_t *msg);
@@ -167,6 +168,7 @@ int main(int argc, char *argv[])
exit(error_exit);
}
+ _set_spank_env();
_set_submit_dir_env();
if (opt.cwd && chdir(opt.cwd)) {
error("chdir(%s): %m", opt.cwd);
@@ -428,6 +430,19 @@ static void _set_exit_code(void)
}
}
+/* Propagate SPANK environment via SLURM_SPANK_ environment variables */
+static void _set_spank_env(void)
+{
+ int i;
+
+ for (i=0; i<opt.spank_job_env_size; i++) {
+ if (setenvfs("SLURM_SPANK_%s", opt.spank_job_env[i]) < 0) {
+ error("unable to set %s in environment",
+ opt.spank_job_env[i]);
+ }
+ }
+}
+
/* Set SLURM_SUBMIT_DIR environment variable with current state */
static void _set_submit_dir_env(void)
{
@@ -553,11 +568,11 @@ static int _fill_job_desc_from_opts(job_desc_msg_t *desc)
desc->ntasks_per_core = opt.ntasks_per_core;
/* node constraints */
- if (opt.min_sockets_per_node > -1)
+ if (opt.min_sockets_per_node != NO_VAL)
desc->min_sockets = opt.min_sockets_per_node;
- if (opt.min_cores_per_socket > -1)
+ if (opt.min_cores_per_socket != NO_VAL)
desc->min_cores = opt.min_cores_per_socket;
- if (opt.min_threads_per_core > -1)
+ if (opt.min_threads_per_core != NO_VAL)
desc->min_threads = opt.min_threads_per_core;
if (opt.no_kill)
diff --git a/src/sbatch/opt.c b/src/sbatch/opt.c
index e6cd19540..8c963d051 100644
--- a/src/sbatch/opt.c
+++ b/src/sbatch/opt.c
@@ -1068,7 +1068,7 @@ static void _opt_pbs_batch_script(const char *file, const void *body, int size)
static void _set_options(int argc, char **argv)
{
- int opt_char, option_index = 0;
+ int opt_char, option_index = 0, max_val = 0;
char *tmp;
struct option *optz = spank_option_table_create(long_options);
@@ -1423,17 +1423,26 @@ static void _set_options(int argc, char **argv)
case LONG_OPT_SOCKETSPERNODE:
get_resource_arg_range( optarg, "sockets-per-node",
&opt.min_sockets_per_node,
- NULL, true );
+ &max_val, true );
+ if ((opt.min_sockets_per_node == 1) &&
+ (max_val == INT_MAX))
+ opt.min_sockets_per_node = NO_VAL;
break;
case LONG_OPT_CORESPERSOCKET:
get_resource_arg_range( optarg, "cores-per-socket",
&opt.min_cores_per_socket,
- NULL, true );
+ &max_val, true );
+ if ((opt.min_cores_per_socket == 1) &&
+ (max_val == INT_MAX))
+ opt.min_cores_per_socket = NO_VAL;
break;
case LONG_OPT_THREADSPERCORE:
get_resource_arg_range( optarg, "threads-per-core",
&opt.min_threads_per_core,
- NULL, true );
+ &max_val, true );
+ if ((opt.min_threads_per_core == 1) &&
+ (max_val == INT_MAX))
+ opt.min_threads_per_core = NO_VAL;
break;
case LONG_OPT_NTASKSPERNODE:
opt.ntasks_per_node = _get_int(optarg,
diff --git a/src/sbatch/sbatch.c b/src/sbatch/sbatch.c
index 656041509..9c60d4ec8 100644
--- a/src/sbatch/sbatch.c
+++ b/src/sbatch/sbatch.c
@@ -69,6 +69,7 @@ static char *script_wrap(char *command_string);
static void _set_exit_code(void);
static void _set_prio_process_env(void);
static int _set_rlimit_env(void);
+static void _set_spank_env(void);
static void _set_submit_dir_env(void);
static int _set_umask_env(void);
@@ -134,6 +135,7 @@ int main(int argc, char *argv[])
}
_set_prio_process_env();
+ _set_spank_env();
_set_submit_dir_env();
_set_umask_env();
slurm_init_job_desc_msg(&desc);
@@ -283,11 +285,11 @@ static int fill_job_desc_from_opts(job_desc_msg_t *desc)
desc->ntasks_per_core = opt.ntasks_per_core;
/* node constraints */
- if (opt.min_sockets_per_node > -1)
+ if (opt.min_sockets_per_node != NO_VAL)
desc->min_sockets = opt.min_sockets_per_node;
- if (opt.min_cores_per_socket > -1)
+ if (opt.min_cores_per_socket != NO_VAL)
desc->min_cores = opt.min_cores_per_socket;
- if (opt.min_threads_per_core > -1)
+ if (opt.min_threads_per_core != NO_VAL)
desc->min_threads = opt.min_threads_per_core;
if (opt.no_kill)
@@ -353,6 +355,19 @@ static void _set_exit_code(void)
}
}
+/* Propagate SPANK environment via SLURM_SPANK_ environment variables */
+static void _set_spank_env(void)
+{
+ int i;
+
+ for (i=0; i<opt.spank_job_env_size; i++) {
+ if (setenvfs("SLURM_SPANK_%s", opt.spank_job_env[i]) < 0) {
+ error("unable to set %s in environment",
+ opt.spank_job_env[i]);
+ }
+ }
+}
+
/* Set SLURM_SUBMIT_DIR environment variable with current state */
static void _set_submit_dir_env(void)
{
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index e3b15059c..f7c373206 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -2928,11 +2928,6 @@ static int _validate_job_create_req(job_desc_msg_t * job_desc)
strlen(job_desc->work_dir));
return ESLURM_PATHNAME_TOO_LONG;
}
- if (!valid_spank_job_env(job_desc->spank_job_env,
- job_desc->spank_job_env_size,
- job_desc->user_id)) {
- return EINVAL;
- }
return SLURM_SUCCESS;
}
@@ -3283,30 +3278,18 @@ _set_multi_core_data(job_desc_msg_t * job_desc)
{
multi_core_data_t * mc_ptr;
- if (((job_desc->min_sockets == (uint16_t) NO_VAL) ||
- (job_desc->min_sockets == (uint16_t) 1)) &&
- ((job_desc->min_cores == (uint16_t) NO_VAL) ||
- (job_desc->min_cores == (uint16_t) 1)) &&
- ((job_desc->min_threads == (uint16_t) NO_VAL) ||
- (job_desc->min_threads == (uint16_t) 1)) &&
+ if ((job_desc->min_sockets == (uint16_t) NO_VAL) &&
+ (job_desc->min_cores == (uint16_t) NO_VAL) &&
+ (job_desc->min_threads == (uint16_t) NO_VAL) &&
(job_desc->ntasks_per_socket == (uint16_t) NO_VAL) &&
(job_desc->ntasks_per_core == (uint16_t) NO_VAL) &&
(job_desc->plane_size == (uint16_t) NO_VAL))
return NULL;
mc_ptr = xmalloc(sizeof(multi_core_data_t));
- if (job_desc->min_sockets != (uint16_t) NO_VAL)
- mc_ptr->min_sockets = job_desc->min_sockets;
- else
- mc_ptr->min_sockets = 1;
- if (job_desc->min_cores != (uint16_t) NO_VAL)
- mc_ptr->min_cores = job_desc->min_cores;
- else
- mc_ptr->min_cores = 1;
- if (job_desc->min_threads != (uint16_t) NO_VAL)
- mc_ptr->min_threads = job_desc->min_threads;
- else
- mc_ptr->min_threads = 1;
+ mc_ptr->min_sockets = job_desc->min_sockets;
+ mc_ptr->min_cores = job_desc->min_cores;
+ mc_ptr->min_threads = job_desc->min_threads;
if (job_desc->ntasks_per_socket != (uint16_t) NO_VAL)
mc_ptr->ntasks_per_socket = job_desc->ntasks_per_socket;
else
@@ -3983,22 +3966,10 @@ static int _validate_job_desc(job_desc_msg_t * job_desc_msg, int allocate,
} else if (!_valid_job_min_mem(job_desc_msg))
return ESLURM_INVALID_TASK_MEMORY;
- if (job_desc_msg->min_sockets == (uint16_t) NO_VAL)
- job_desc_msg->min_sockets = 1; /* default socket count of 1 */
- if (job_desc_msg->min_cores == (uint16_t) NO_VAL)
- job_desc_msg->min_cores = 1; /* default core count of 1 */
- if (job_desc_msg->min_threads == (uint16_t) NO_VAL)
- job_desc_msg->min_threads = 1; /* default thread count of 1 */
if (job_desc_msg->min_nodes == NO_VAL)
job_desc_msg->min_nodes = 1; /* default node count of 1 */
if (job_desc_msg->num_procs == NO_VAL)
job_desc_msg->num_procs = job_desc_msg->min_nodes;
- if (job_desc_msg->min_sockets == (uint16_t) NO_VAL)
- job_desc_msg->min_sockets = 1; /* default socket count of 1 */
- if (job_desc_msg->min_cores == (uint16_t) NO_VAL)
- job_desc_msg->min_cores = 1; /* default core count of 1 */
- if (job_desc_msg->min_threads == (uint16_t) NO_VAL)
- job_desc_msg->min_threads = 1; /* default thread count of 1 */
if (job_desc_msg->job_min_cpus == (uint16_t) NO_VAL)
job_desc_msg->job_min_cpus = 1; /* default 1 cpu per node */
@@ -6434,7 +6405,27 @@ extern bool job_independent(struct job_record *job_ptr, int will_run)
{
struct job_details *detail_ptr = job_ptr->details;
time_t now = time(NULL);
- int rc;
+ int depend_rc;
+ bool send_acct_rec = false;
+
+ /* Test dependencies first so we can cancel jobs before dependent
+ * jobs records get purged (e.g. afterok, afternotok) */
+ depend_rc = test_job_dependency(job_ptr);
+ if (depend_rc == 1) {
+ job_ptr->state_reason = WAIT_DEPENDENCY;
+ xfree(job_ptr->state_desc);
+ return false;
+ } else if (depend_rc == 2) {
+ time_t now = time(NULL);
+ info("Job dependency can't be satisfied, cancelling job %u",
+ job_ptr->job_id);
+ job_ptr->job_state = JOB_CANCELLED;
+ xfree(job_ptr->state_desc);
+ job_ptr->start_time = now;
+ job_ptr->end_time = now;
+ job_completion_logger(job_ptr);
+ return false;
+ }
if (detail_ptr && (detail_ptr->begin_time > now)) {
job_ptr->state_reason = WAIT_TIME;
@@ -6448,45 +6439,27 @@ extern bool job_independent(struct job_record *job_ptr, int will_run)
return false; /* not yet time */
}
- rc = test_job_dependency(job_ptr);
- if (rc == 0) {
- bool send_acct_rec = false;
- if (job_ptr->state_reason == WAIT_DEPENDENCY) {
- job_ptr->state_reason = WAIT_NO_REASON;
- xfree(job_ptr->state_desc);
- }
- if (detail_ptr && (detail_ptr->begin_time == 0)) {
- detail_ptr->begin_time = now;
- send_acct_rec = true;
- } else if (job_ptr->state_reason == WAIT_TIME) {
- job_ptr->state_reason = WAIT_NO_REASON;
- xfree(job_ptr->state_desc);
- send_acct_rec = true;
- }
- if (send_acct_rec && !will_run) {
- /* We want to record when a job becomes eligible in
- * order to calculate reserved time (a measure of
- * system over-subscription), job really is not
- * starting now */
- jobacct_storage_g_job_start(
- acct_db_conn, slurmctld_cluster_name, job_ptr);
- }
- return true;
- } else if (rc == 1) {
- job_ptr->state_reason = WAIT_DEPENDENCY;
+ /* Job is eligible to start now */
+ if (job_ptr->state_reason == WAIT_DEPENDENCY) {
+ job_ptr->state_reason = WAIT_NO_REASON;
xfree(job_ptr->state_desc);
- return false;
- } else { /* rc == 2 */
- time_t now = time(NULL);
- info("Job dependency can't be satisfied, cancelling job %u",
- job_ptr->job_id);
- job_ptr->job_state = JOB_CANCELLED;
+ }
+ if (detail_ptr && (detail_ptr->begin_time == 0)) {
+ detail_ptr->begin_time = now;
+ send_acct_rec = true;
+ } else if (job_ptr->state_reason == WAIT_TIME) {
+ job_ptr->state_reason = WAIT_NO_REASON;
xfree(job_ptr->state_desc);
- job_ptr->start_time = now;
- job_ptr->end_time = now;
- job_completion_logger(job_ptr);
- return false;
+ send_acct_rec = true;
+ }
+ if (send_acct_rec && !will_run) {
+ /* We want to record when a job becomes eligible in order to
+ * calculate reserved time (a measure ofs ystem over-
+ * subscription), job really is not starting now */
+ jobacct_storage_g_job_start(acct_db_conn,
+ slurmctld_cluster_name, job_ptr);
}
+ return true;
}
/*
@@ -7879,7 +7852,7 @@ _copy_job_record_to_job_desc(struct job_record *job_ptr)
job_desc->max_nodes = details->max_nodes;
job_desc->min_sockets = mc_ptr->min_sockets;
job_desc->min_cores = mc_ptr->min_cores;
- job_desc->min_threads = mc_ptr->min_threads;;
+ job_desc->min_threads = mc_ptr->min_threads;
job_desc->cpus_per_task = details->cpus_per_task;
job_desc->ntasks_per_node = details->ntasks_per_node;
job_desc->ntasks_per_socket = mc_ptr->ntasks_per_socket;
diff --git a/src/slurmctld/job_scheduler.c b/src/slurmctld/job_scheduler.c
index bf70191f6..a2ffe2c0f 100644
--- a/src/slurmctld/job_scheduler.c
+++ b/src/slurmctld/job_scheduler.c
@@ -704,7 +704,7 @@ extern int test_job_dependency(struct job_record *job_ptr)
{
ListIterator depend_iter;
struct depend_spec *dep_ptr;
- bool failure = false;
+ bool failure = false, depends = false;
struct job_queue *job_queue = NULL;
int i, now, job_queue_size = 0;
struct job_record *qjob_ptr;
@@ -742,7 +742,7 @@ extern int test_job_dependency(struct job_record *job_ptr)
if (now)
list_delete_item(depend_iter);
else
- break;
+ depends = true;
} else if (dep_ptr->job_ptr->job_id != dep_ptr->job_id) {
/* job is gone, dependency lifted */
list_delete_item(depend_iter);
@@ -750,16 +750,16 @@ extern int test_job_dependency(struct job_record *job_ptr)
if (!IS_JOB_PENDING(dep_ptr->job_ptr))
list_delete_item(depend_iter);
else
- break;
+ depends = true;
} else if (dep_ptr->depend_type == SLURM_DEPEND_AFTER_ANY) {
if (IS_JOB_FINISHED(dep_ptr->job_ptr))
list_delete_item(depend_iter);
else
- break;
+ depends = true;
} else if (dep_ptr->depend_type == SLURM_DEPEND_AFTER_NOT_OK) {
if (!IS_JOB_FINISHED(dep_ptr->job_ptr))
- break;
- if (!IS_JOB_COMPLETE(dep_ptr->job_ptr))
+ depends = true;
+ else if (!IS_JOB_COMPLETE(dep_ptr->job_ptr))
list_delete_item(depend_iter);
else {
failure = true;
@@ -767,8 +767,8 @@ extern int test_job_dependency(struct job_record *job_ptr)
}
} else if (dep_ptr->depend_type == SLURM_DEPEND_AFTER_OK) {
if (!IS_JOB_FINISHED(dep_ptr->job_ptr))
- break;
- if (IS_JOB_COMPLETE(dep_ptr->job_ptr))
+ depends = true;
+ else if (IS_JOB_COMPLETE(dep_ptr->job_ptr))
list_delete_item(depend_iter);
else {
failure = true;
@@ -781,7 +781,7 @@ extern int test_job_dependency(struct job_record *job_ptr)
if (failure)
return 2;
- if (dep_ptr)
+ if (depends)
return 1;
return 0;
}
@@ -1148,6 +1148,7 @@ static char **_build_env(struct job_record *job_ptr)
name = gid_to_string((uid_t) job_ptr->group_id);
setenvf(&my_env, "SLURM_JOB_GROUP", "%s", name);
xfree(name);
+ setenvf(&my_env, "SLURM_JOBID", "%u", job_ptr->job_id);
setenvf(&my_env, "SLURM_JOB_ID", "%u", job_ptr->job_id);
setenvf(&my_env, "SLURM_JOB_NAME", "%s", job_ptr->name);
setenvf(&my_env, "SLURM_JOB_NODELIST", "%s", job_ptr->nodes);
diff --git a/src/slurmctld/node_mgr.c b/src/slurmctld/node_mgr.c
index 0cb2f6886..34130d412 100644
--- a/src/slurmctld/node_mgr.c
+++ b/src/slurmctld/node_mgr.c
@@ -2217,6 +2217,7 @@ extern int send_nodes_to_accounting(time_t event_time)
int rc = SLURM_SUCCESS, i = 0;
struct node_record *node_ptr = NULL;
uint32_t node_scaling = 0;
+ char *reason = NULL;
slurmctld_lock_t node_read_lock = {
READ_LOCK, NO_LOCK, READ_LOCK, WRITE_LOCK };
@@ -2226,10 +2227,15 @@ extern int send_nodes_to_accounting(time_t event_time)
/* send nodes not in not 'up' state */
node_ptr = node_record_table_ptr;
for (i = 0; i < node_record_count; i++, node_ptr++) {
+ if(node_ptr->reason)
+ reason = node_ptr->reason;
+ else
+ reason = "First Registration";
if (node_ptr->name == '\0' ||
(!IS_NODE_DRAIN(node_ptr) && !IS_NODE_FAIL(node_ptr) &&
!IS_NODE_DOWN(node_ptr))) {
- /* on some systems we need to make sure there
+ /* At this point, the node appears to be up,
+ but on some systems we need to make sure there
aren't some part of a node in an error state. */
if(node_ptr->select_nodeinfo) {
uint16_t err_cpus = 0;
@@ -2266,17 +2272,15 @@ extern int send_nodes_to_accounting(time_t event_time)
acct_db_conn,
slurmctld_cluster_name,
&send_node, event_time,
- NULL);
-
- continue;
+ reason);
}
- } else
- continue;
- }
- rc = clusteracct_storage_g_node_down(acct_db_conn,
- slurmctld_cluster_name,
- node_ptr, event_time,
- NULL);
+ }
+ } else
+ rc = clusteracct_storage_g_node_down(
+ acct_db_conn,
+ slurmctld_cluster_name,
+ node_ptr, event_time,
+ reason);
if (rc == SLURM_ERROR)
break;
}
diff --git a/src/slurmctld/node_scheduler.c b/src/slurmctld/node_scheduler.c
index cfa3db85c..77052471f 100644
--- a/src/slurmctld/node_scheduler.c
+++ b/src/slurmctld/node_scheduler.c
@@ -1414,25 +1414,31 @@ extern int job_req_node_filter(struct job_record *job_ptr,
continue;
}
if (mc_ptr &&
- ((mc_ptr->min_sockets > config_ptr->sockets) ||
- (mc_ptr->min_cores > config_ptr->cores) ||
- (mc_ptr->min_threads > config_ptr->threads))){
+ (((mc_ptr->min_sockets > config_ptr->sockets) &&
+ (mc_ptr->min_sockets != (uint16_t) NO_VAL)) ||
+ ((mc_ptr->min_cores > config_ptr->cores) &&
+ (mc_ptr->min_cores != (uint16_t) NO_VAL)) ||
+ ((mc_ptr->min_threads > config_ptr->threads) &&
+ (mc_ptr->min_threads != (uint16_t) NO_VAL)))) {
bit_clear(avail_bitmap, i);
continue;
}
} else {
if ((detail_ptr->job_min_cpus > node_ptr->cpus) ||
((detail_ptr->job_min_memory & (~MEM_PER_CPU)) >
- node_ptr->real_memory) ||
+ node_ptr->real_memory) ||
(detail_ptr->job_min_tmp_disk >
node_ptr->tmp_disk)) {
bit_clear(avail_bitmap, i);
continue;
}
if (mc_ptr &&
- ((mc_ptr->min_sockets > node_ptr->sockets) ||
- (mc_ptr->min_cores > node_ptr->cores) ||
- (mc_ptr->min_threads > node_ptr->threads))) {
+ (((mc_ptr->min_sockets > node_ptr->sockets) &&
+ (mc_ptr->min_sockets != (uint16_t) NO_VAL)) ||
+ ((mc_ptr->min_cores > node_ptr->cores) &&
+ (mc_ptr->min_cores != (uint16_t) NO_VAL)) ||
+ ((mc_ptr->min_threads > node_ptr->threads) &&
+ (mc_ptr->min_threads != (uint16_t) NO_VAL)))) {
bit_clear(avail_bitmap, i);
continue;
}
@@ -1539,10 +1545,13 @@ static int _build_node_list(struct job_record *job_ptr,
config_ptr->real_memory) ||
(detail_ptr->job_min_tmp_disk > config_ptr->tmp_disk))
config_filter = 1;
- if (mc_ptr &&
- ((mc_ptr->min_sockets > config_ptr->sockets ) ||
- (mc_ptr->min_cores > config_ptr->cores ) ||
- (mc_ptr->min_threads > config_ptr->threads )))
+ if (mc_ptr &&
+ (((mc_ptr->min_sockets > config_ptr->sockets) &&
+ (mc_ptr->min_sockets != (uint16_t) NO_VAL)) ||
+ ((mc_ptr->min_cores > config_ptr->cores) &&
+ (mc_ptr->min_cores != (uint16_t) NO_VAL)) ||
+ ((mc_ptr->min_threads > config_ptr->threads) &&
+ (mc_ptr->min_threads != (uint16_t) NO_VAL))))
config_filter = 1;
/* since nodes can register with more resources than defined */
@@ -1695,10 +1704,13 @@ static void _filter_nodes_in_set(struct node_set *node_set_ptr,
node_con->real_memory) &&
(job_con->job_min_tmp_disk <= node_con->tmp_disk))
job_ok = 1;
- if (mc_ptr &&
- ((mc_ptr->min_sockets <= node_con->sockets) &&
- (mc_ptr->min_cores <= node_con->cores ) &&
- (mc_ptr->min_threads <= node_con->threads)))
+ if (mc_ptr &&
+ (((mc_ptr->min_sockets <= node_con->sockets) ||
+ (mc_ptr->min_sockets == (uint16_t) NO_VAL)) &&
+ ((mc_ptr->min_cores <= node_con->cores) ||
+ (mc_ptr->min_cores == (uint16_t) NO_VAL)) &&
+ ((mc_ptr->min_threads <= node_con->threads) ||
+ (mc_ptr->min_threads == (uint16_t) NO_VAL))))
job_mc_ptr_ok = 1;
if (job_ok && (!mc_ptr || job_mc_ptr_ok))
continue;
@@ -1721,10 +1733,13 @@ static void _filter_nodes_in_set(struct node_set *node_set_ptr,
node_ptr->real_memory) &&
(job_con->job_min_tmp_disk <= node_ptr->tmp_disk))
job_ok = 1;
- if (mc_ptr &&
- ((mc_ptr->min_sockets <= node_ptr->sockets) &&
- (mc_ptr->min_cores <= node_ptr->cores ) &&
- (mc_ptr->min_threads <= node_ptr->threads)))
+ if (mc_ptr &&
+ (((mc_ptr->min_sockets <= node_ptr->sockets) ||
+ (mc_ptr->min_sockets == (uint16_t) NO_VAL)) &&
+ ((mc_ptr->min_cores <= node_ptr->cores) ||
+ (mc_ptr->min_cores == (uint16_t) NO_VAL)) &&
+ ((mc_ptr->min_threads <= node_ptr->threads) ||
+ (mc_ptr->min_threads == (uint16_t) NO_VAL))))
job_mc_ptr_ok = 1;
if (job_ok && (!mc_ptr || job_mc_ptr_ok))
continue;
diff --git a/src/squeue/sort.c b/src/squeue/sort.c
index 1c7e0c0eb..d05390cd7 100644
--- a/src/squeue/sort.c
+++ b/src/squeue/sort.c
@@ -165,6 +165,9 @@ void sort_job_list(List job_list)
list_sort(job_list, _sort_job_by_reservation);
else if (params.sort[i] == 'z')
list_sort(job_list, _sort_job_by_num_sct);
+ else
+ error("Invalid sort specification: %c",
+ params.sort[i]);
}
}
diff --git a/src/srun/opt.c b/src/srun/opt.c
index f730ac6fd..430588a1f 100644
--- a/src/srun/opt.c
+++ b/src/srun/opt.c
@@ -182,6 +182,8 @@
#define LONG_OPT_SIGNAL 0x14e
#define LONG_OPT_DEBUG_SLURMD 0x14f
+extern char **environ;
+
/*---- global variables, defined in opt.h ----*/
int _verbose;
opt_t opt;
@@ -708,7 +710,7 @@ _get_int(const char *arg, const char *what, bool positive)
static void set_options(const int argc, char **argv)
{
- int opt_char, option_index = 0;
+ int opt_char, option_index = 0, max_val = 0;
struct utsname name;
static struct option long_options[] = {
{"attach", no_argument, 0, 'a'},
@@ -1302,17 +1304,26 @@ static void set_options(const int argc, char **argv)
case LONG_OPT_SOCKETSPERNODE:
get_resource_arg_range( optarg, "sockets-per-node",
&opt.min_sockets_per_node,
- NULL, true );
+ &max_val, true );
+ if ((opt.min_sockets_per_node == 1) &&
+ (max_val == INT_MAX))
+ opt.min_sockets_per_node = NO_VAL;
break;
case LONG_OPT_CORESPERSOCKET:
get_resource_arg_range( optarg, "cores-per-socket",
&opt.min_cores_per_socket,
- NULL, true);
+ &max_val, true );
+ if ((opt.min_cores_per_socket == 1) &&
+ (max_val == INT_MAX))
+ opt.min_cores_per_socket = NO_VAL;
break;
case LONG_OPT_THREADSPERCORE:
get_resource_arg_range( optarg, "threads-per-core",
&opt.min_threads_per_core,
- NULL, true );
+ &max_val, true );
+ if ((opt.min_threads_per_core == 1) &&
+ (max_val == INT_MAX))
+ opt.min_threads_per_core = NO_VAL;
break;
case LONG_OPT_NTASKSPERNODE:
opt.ntasks_per_node = _get_int(optarg, "ntasks-per-node",
@@ -1936,6 +1947,33 @@ static bool _opt_verify(void)
return verified;
}
+/* Initialize the the spank_job_env based upon environment variables set
+ * via salloc or sbatch commands */
+extern void init_spank_env(void)
+{
+ int i;
+ char *name, *eq, *value;
+
+ if (environ == NULL)
+ return;
+
+ for (i=0; environ[i]; i++) {
+ if (strncmp(environ[i], "SLURM_SPANK_", 12))
+ continue;
+ name = xstrdup(environ[i] + 12);
+ eq = strchr(name, (int)'=');
+ if (eq == NULL) {
+ xfree(name);
+ break;
+ }
+ eq[0] = '\0';
+ value = eq + 1;
+ spank_set_job_env(name, value, 1);
+ xfree(name);
+ }
+
+}
+
/* Functions used by SPANK plugins to read and write job environment
* variables for use within job's Prolog and/or Epilog */
extern char *spank_get_job_env(const char *name)
diff --git a/src/srun/opt.h b/src/srun/opt.h
index f0c1213ce..8ad26cdc7 100644
--- a/src/srun/opt.h
+++ b/src/srun/opt.h
@@ -1,6 +1,6 @@
/*****************************************************************************\
* opt.h - definitions for srun option processing
- * $Id: opt.h 19316 2010-01-26 00:12:13Z jette $
+ * $Id: opt.h 20173 2010-04-29 16:14:33Z jette $
*****************************************************************************
* Copyright (C) 2002-2007 The Regents of the University of California.
* Copyright (C) 2008-2009 Lawrence Livermore National Security.
@@ -257,4 +257,8 @@ extern int spank_set_job_env(const char *name, const char *value,
int overwrite);
extern int spank_unset_job_env(const char *name);
+/* Initialize the the spank_job_env based upon environment variables set
+ * via salloc or sbatch commands */
+extern void init_spank_env(void);
+
#endif /* _HAVE_OPT_H */
diff --git a/src/srun/srun.c b/src/srun/srun.c
index abfc55af8..8182ba607 100644
--- a/src/srun/srun.c
+++ b/src/srun/srun.c
@@ -209,6 +209,7 @@ int srun(int ac, char **av)
/* Initialize plugin stack, read options from plugins, etc.
*/
+ init_spank_env();
if (spank_init(NULL) < 0) {
error("Plug-in initialization failed");
exit(error_exit);
@@ -754,11 +755,11 @@ static int _validate_relative(resource_allocation_response_msg_t *resp)
if (opt.relative_set &&
((opt.relative + opt.min_nodes) > resp->node_cnt)) {
if (opt.nodes_set_opt) { /* -N command line option used */
- error("--relative and --nodes option incompatable "
+ error("--relative and --nodes option incompatible "
"with count of allocated nodes (%d+%d>%d)",
opt.relative, opt.min_nodes, resp->node_cnt);
} else { /* SLURM_NNODES option used */
- error("--relative and SLURM_NNODES option incompatable "
+ error("--relative and SLURM_NNODES option incompatible "
"with count of allocated nodes (%d+%d>%d)",
opt.relative, opt.min_nodes, resp->node_cnt);
}
diff --git a/src/sview/block_info.c b/src/sview/block_info.c
index f45e2d0be..0c681b018 100644
--- a/src/sview/block_info.c
+++ b/src/sview/block_info.c
@@ -521,6 +521,13 @@ static List _create_block_list(partition_info_msg_t *part_info_ptr,
block_ptr->color_inx =
atoi(block_ptr->bg_block_name+3);
#endif
+ /* on some systems they make there own blocks named
+ whatever they want, so doing this fixes what could
+ be a negative number.
+ */
+ if(block_ptr->color_inx < 0)
+ block_ptr->color_inx = i;
+
block_ptr->color_inx %= sview_colors_cnt;
block_ptr->nodes
diff --git a/testsuite/expect/test7.7.prog.c b/testsuite/expect/test7.7.prog.c
index 1a0bed977..2c29c5b8b 100644
--- a/testsuite/expect/test7.7.prog.c
+++ b/testsuite/expect/test7.7.prog.c
@@ -370,7 +370,7 @@ static void _modify_job(long my_job_id)
/* "NODES=2 " */
/* "DEPEND=afterany:3 " */
/* "INVALID=123 " */
- /* "VARIABLELIST=TEST_ENV1=test_val1 " */
+ "COMMENT=test123 "
"VARIABLELIST=TEST_ENV1=test_val1,TEST_ENV2=test_val2 "
"TIMELIMIT=10 BANK=test_bank",
(uint32_t) now, my_job_id);
--
GitLab