pool/slurm
SHA256
2
0
Fork 1
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Accepting request 1238577 from network:cluster

Browse Source 
* `slurmrestd` - Remove deprecated fields from the following
     `.result` from `POST /slurm/v0.0.42/job/submit`.  
     `.job_id`, `.step_id`, `.job_submit_user_msg` from `POST /slurm/v0.0.42/job/{job_id}`.  
     `.job.exclusive`, `.jobs[].exclusive` to `POST /slurm/v0.0.42/job/submit`.  
     `.jobs[].exclusive` from `GET /slurm/v0.0.42/job/{job_id}`.  
     `.jobs[].exclusive` from `GET /slurm/v0.0.42/jobs`.  
     `.job.oversubscribe`, `.jobs[].oversubscribe` to `POST /slurm/v0.0.42/job/submit`.  
     `.jobs[].oversubscribe` from `GET /slurm/v0.0.42/job/{job_id}`.  
     `.jobs[].oversubscribe` from `GET /slurm/v0.0.42/jobs`.  
     `DELETE /slurm/v0.0.40/jobs`  
     `DELETE /slurm/v0.0.41/jobs`  
     `DELETE /slurm/v0.0.42/jobs`  
    allocation is granted.
    `job|socket|task` or `cpus|mem` per GRES.
    node update whereas previously only single nodes could be
    updated through `/node/<nodename>` endpoint:
    `POST /slurm/v0.0.42/nodes`
    partition as this is a cluster-wide option.
    `REQUEST_NODE_INFO RPC`.
    the db server is not reachable.
    (`.jobs[].priority_by_partition`) to JSON and YAML output.
    connection` error if the error was the result of an
    authentication failure.
    errors with the `SLURM_PROTOCOL_AUTHENTICATION_ERROR` error
    code.
    of `Unspecified error` if querying the following endpoints
    fails:  
    `GET /slurm/v0.0.40/diag/`  
    `GET /slurm/v0.0.41/diag/`  
    `GET /slurm/v0.0.42/diag/` (forwarded request 1238576 from eeich)

OBS-URL: https://build.opensuse.org/request/show/1238577
OBS-URL: https://build.opensuse.org/package/show/openSUSE:Factory/slurm?expand=0&rev=111
This commit is contained in:
Dominique Leuenberger 2025-01-18 12:18:25 +00:00 committed by Git OBS Bridge
commit 8a2be70840
4 changed files with 575 additions and 187 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

369
regression.py.sle12 Normal file
View File

@ -0,0 +1,369 @@
#!/usr/bin/env python3
############################################################################
# Copyright (C) 2006 The Regents of the University of California.
# Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
# Written by Christopher J. Morrone <morrone2@llnl.gov>
# CODE-OCEC-09-009. All rights reserved.
#
# This file is part of Slurm, a resource management program.
# For details, see <https://slurm.schedmd.com/>.
# Please also read the supplied file: DISCLAIMER.
#
# Slurm is free software; you can redistribute it and/or modify it under
# the terms of the GNU General Public License as published by the Free
# Software Foundation; either version 2 of the License, or (at your option)
# any later version.
#
# Slurm is distributed in the hope that it will be useful, but WITHOUT ANY
# WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
# FOR A PARTICULAR PURPOSE. See the GNU General Public License for more
# details.
#
# You should have received a copy of the GNU General Public License along
# with Slurm; if not, write to the Free Software Foundation, Inc.,
# 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
############################################################################
"""This script makes it easier to run the Slurm expect test scripts."""
from __future__ import print_function
import json
import os
import re
import sys
import time
import signal
from optparse import OptionParser
from optparse import OptionValueError
from subprocess import Popen
def main(argv=None):
# "tests" is a list containing tuples of length 3 of the form
# (test major number, test minor number, test filename)
tests = []
failed_tests = []
passed_tests = []
skipped_tests = []
begin = (1, 1)
abort = False
# Handle command line parameters
if argv is None:
argv = sys.argv
parser = OptionParser()
parser.add_option(
"-t",
"--time-individual",
action="store_true",
dest="time_individual",
default=False,
)
parser.add_option(
"-e",
"--exclude",
type="string",
dest="exclude_tests",
action="callback",
callback=test_parser,
help="comma or space separated string of tests to skip",
)
parser.add_option(
"-i",
"--include",
type="string",
dest="include_tests",
action="callback",
callback=test_parser,
help="comma or space separated string of tests to include",
)
parser.add_option("-k", "--keep-logs", action="store_true", default=False)
parser.add_option("-s", "--stop-on-first-fail", action="store_true", default=False)
parser.add_option(
"-b",
"--begin-from-test",
type="string",
dest="begin_from_test",
action="callback",
callback=test_parser,
)
parser.add_option(
"-f",
"--results-file",
type="string",
help="write json result to specified file name",
)
(options, args) = parser.parse_args(args=argv)
# Sanity check
if not os.path.isfile("globals"):
print('ERROR: "globals" not here as needed', file=sys.stderr)
return -1
# Clear any environment variables that could break the tests.
# Cray sets some squeue format options that break tests
del os.environ["SQUEUE_ALL"]
del os.environ["SQUEUE_SORT"]
del os.environ["SQUEUE_FORMAT"]
del os.environ["SQUEUE_FORMAT2"]
# Read the current working directory and build a sorted list
# of the available tests.
test_re = re.compile(r"test(\d+)\.(\d+)$")
for filename in os.listdir("."):
match = test_re.match(filename)
if match:
major = int(match.group(1))
minor = int(match.group(2))
if not test_in_list(major, minor, options.exclude_tests) and (
not options.include_tests
or test_in_list(major, minor, options.include_tests)
):
tests.append((major, minor, filename))
if not tests:
print(
"ERROR: no test files found in current working directory", file=sys.stderr
)
return -1
# sory by major, minor
tests.sort(key=lambda t: (t[0], t[1]))
# Set begin value
if options.begin_from_test is not None:
begin = options.begin_from_test[0]
# Now run the tests
start_time = time.time()
test_env = os.environ.copy()
if options.stop_on_first_fail:
test_env["SLURM_TESTSUITE_CLEANUP_ON_FAILURE"] = "false"
else:
test_env["SLURM_TESTSUITE_CLEANUP_ON_FAILURE"] = "true"
print("Started:", time.asctime(time.localtime(start_time)), file=sys.stdout)
sys.stdout.flush()
results_list = []
for test in tests:
if begin[0] > test[0] or (begin[0] == test[0] and begin[1] > test[1]):
continue
test_id = "{0}.{1}".format(test[0],test[1])
sys.stdout.write("Running test %s " % test_id)
sys.stdout.flush()
test_dict = {}
test_dict["id"] = test_id
testlog_name = "test{test_id}.log"
try:
os.remove(testlog_name + ".failed")
except:
pass
testlog = open(testlog_name, "w+")
if options.time_individual:
t1 = time.time()
test_dict["start_time"] = float("%.03f" % t1)
try:
child = Popen(
("expect", test[2]),
shell=False,
env=test_env,
stdout=testlog,
stderr=testlog,
)
retcode = child.wait()
except KeyboardInterrupt:
child.send_signal(signal.SIGINT)
retcode = child.wait()
abort = True
if options.time_individual:
t2 = time.time()
minutes = int(int(t2 - t1) / 60)
seconds = (int(t2 - t1)) % 60
if minutes > 0:
sys.stdout.write("%d min " % (minutes))
sys.stdout.write("%.2f sec " % (seconds))
test_dict["duration"] = float("%.03f" % (t2 - t1))
if retcode == 0:
status = "pass"
elif retcode > 127:
status = "skip"
else:
status = "fail"
test_dict["status"] = status
# Determine the reason if requesting a json results file
if status != "pass" and options.results_file:
testlog.flush()
testlog.seek(0)
test_output = testlog.read()
sections = [s for s in test_output.split("=" * 78 + "\n")]
header = sections[1]
body = sections[2]
footer = "".join(sections[3:])
fatals = re.findall(
r"(?ms)\[[^\]]+\][ \[]+Fatal[ \]:]+(.*?) \(fail[^\)]+\)$", body
)
errors = re.findall(
r"(?ms)\[[^\]]+\][ \[]+Error[ \]:]+(.*?) \(subfail[^\)]+\)$", body
)
warnings = re.findall(
r"(?ms)\[[^\]]+\][ \[]+Warning[ \]:]+((?:(?!Warning).)*) \((?:sub)?skip[^\)]+\)$",
body,
)
if fatals:
test_dict["reason"] = fatals[0]
elif errors:
test_dict["reason"] = errors[0]
elif warnings:
test_dict["reason"] = warnings[0]
results_list.append(test_dict)
testlog.close()
if status == "pass":
passed_tests.append(test)
sys.stdout.write("\n")
if not options.keep_logs:
try:
os.remove(testlog_name)
except IOError as e:
print(
"ERROR failed to close %s %s" % (testlog_name, e),
file=sys.stederr,
)
elif status == "skip":
skipped_tests.append(test)
sys.stdout.write("SKIPPED\n")
if not options.keep_logs:
try:
os.remove(testlog_name)
except IOError as e:
print(
"ERROR failed to close %s %s" % (testlog_name, e),
file=sys.stederr,
)
else:
failed_tests.append(test)
os.rename(testlog_name, testlog_name + ".failed")
sys.stdout.write("FAILED!\n")
if options.stop_on_first_fail:
break
sys.stdout.flush()
if abort:
sys.stdout.write("\nRegression interrupted!\n")
break
end_time = time.time()
print("Ended:", time.asctime(time.localtime(end_time)), file=sys.stdout)
print(
"\nTestsuite ran for %d minutes %d seconds"
% ((end_time - start_time) / 60, (end_time - start_time) % 60),
file=sys.stdout,
)
if options.results_file:
with open(options.results_file, "w") as results_file:
json.dump(results_list, results_file)
print("Completions :", len(passed_tests), file=sys.stdout)
print("Failures :", len(failed_tests), file=sys.stdout)
print("Skipped :", len(skipped_tests), file=sys.stdout)
if len(failed_tests) > 0:
print("Failed tests : ", file=sys.stdout)
first = True
for test in failed_tests:
if first:
first = False
else:
sys.stdout.write(",")
sys.stdout.write("%d.%d" % (test[0], test[1]))
sys.stdout.write("\n")
sys.stdout.flush()
if abort:
print("INCOMPLETE", file=sys.stdout)
if len(failed_tests) > 0:
return 1
def test_in_list(major, minor, test_list):
"""Test for whether a test numbered major.minor is in test_list.
"major" and "minor" must be integers. "test_list" is a list of
tuples, each tuple representing one test. The tuples are of the
form:
(major, minor, filename)
Returns True if the test is in the list, and False otherwise.
"""
if not test_list:
return False
for test in test_list:
if (test[0] == "*" or test[0] == major) and (
test[1] == "*" or test[1] == minor
):
return True
return False
def test_parser(option, opt_str, value, parser):
"""Option callback function for the optparse.OptionParser class.
Will take a string representing one or more test names and append
a tuple representing the test into a list in the options's destination
variable.
A string representing test names must patch the regular expression
named "test_re" below. Some examples of exceptable options are:
'1.5'
'test9.8'
'2.6 test3.1 14.2'
'3.4,6.7,8.3'
'1.*'
'*.2'
'1.*,3.8,9.2'
Raises OptionValueError on error.
"""
# Initialize the option's destination array, if is does not already exist.
if not hasattr(parser.values, option.dest):
setattr(parser.values, option.dest, [])
if getattr(parser.values, option.dest) is None:
setattr(parser.values, option.dest, [])
# Get a pointer to the option's destination array.
l = getattr(parser.values, option.dest)
# Split the user's option string into a series of tuples that represent
# each test, and add each tuple to the destination array.
splitter = re.compile(r"[,\s]+")
val = splitter.split(value)
test_re = re.compile(r"(test)?((\d+)|\*)\.((\d+)|\*)$")
for v in val:
m = test_re.match(v)
if not m:
raise OptionValueError
major = m.group(2)
if major != "*":
major = int(major)
minor = m.group(4)
if minor != "*":
minor = int(minor)
l.append((major, minor))
if __name__ == "__main__":
sys.exit(main())

375
slurm.changes
View File

@ -1,3 +1,8 @@
-------------------------------------------------------------------
Fri Jan 17 14:19:10 UTC 2025 - Egbert Eich <eich@suse.com>
- Make test suite package work on SLE-12.
-------------------------------------------------------------------
Thu Jan 9 08:35:38 UTC 2025 - Egbert Eich <eich@suse.com>
@ -75,16 +80,16 @@ Mon Jan 6 12:40:31 UTC 2025 - Egbert Eich <eich@suse.com>
* Increase efficency of sending logs to syslog.
* Switch to new official YAML mime type `application/yaml` in
compliance with RFC9512 as primary mime type for YAML formatting.
* `slurmrestd` - Removed deprecated fields from the following
* `slurmrestd` - Remove deprecated fields from the following
endpoints:
`.result' from `POST /slurm/v0.0.42/job/submit`.
`.job_id`, `.step_id`, `.job_submit_user_msg` from `POST /slurm/v0.0.42/job/{job_id}`.
`.job.exclusive`, `.jobs[].exclusive` to `POST /slurm/v0.0.42/job/submit`.
`.jobs[].exclusive` from `GET /slurm/v0.0.42/job/{job_id}`.
`.jobs[].exclusive` from `GET /slurm/v0.0.42/jobs`.
`.job.oversubscribe`, `.jobs[].oversubscribe` to `POST /slurm/v0.0.42/job/submit`.
`.jobs[].oversubscribe` from `GET /slurm/v0.0.42/job/{job_id}`.
`.jobs[].oversubscribe` from `GET /slurm/v0.0.42/jobs`.
`.result` from `POST /slurm/v0.0.42/job/submit`.
`.job_id`, `.step_id`, `.job_submit_user_msg` from `POST /slurm/v0.0.42/job/{job_id}`.
`.job.exclusive`, `.jobs[].exclusive` to `POST /slurm/v0.0.42/job/submit`.
`.jobs[].exclusive` from `GET /slurm/v0.0.42/job/{job_id}`.
`.jobs[].exclusive` from `GET /slurm/v0.0.42/jobs`.
`.job.oversubscribe`, `.jobs[].oversubscribe` to `POST /slurm/v0.0.42/job/submit`.
`.jobs[].oversubscribe` from `GET /slurm/v0.0.42/job/{job_id}`.
`.jobs[].oversubscribe` from `GET /slurm/v0.0.42/jobs`.
* `scontrol` - Removed deprecated fields `.jobs[].exclusive` and
`.jobs[].oversubscribe` from `scontrol show jobs --{json|yaml}`.
* `squeue` - Removed deprecated fields `.jobs[].exclusive` and
@ -100,297 +105,297 @@ Mon Jan 6 12:40:31 UTC 2025 - Egbert Eich <eich@suse.com>
to the drivers.
* Limit `SwitchName` to `HOST_NAME_MAX` chars length.
* For `scancel --ctld` and the following rest api endpoints:
`DELETE /slurm/v0.0.40/jobs`
`DELETE /slurm/v0.0.41/jobs`
`DELETE /slurm/v0.0.42/jobs`
`DELETE /slurm/v0.0.40/jobs`
`DELETE /slurm/v0.0.41/jobs`
`DELETE /slurm/v0.0.42/jobs`
Support array expressions in the responses to the client.
* `salloc` - Always output node names to the user when an
allocation is granted.
allocation is granted.
* `slurmrestd` - Removed all v0.0.39 endpoints.
* `select/linear` - Reject jobs asking for GRES per
`job|socket|task` or `cpus|mem` per GRES.
`job|socket|task` or `cpus|mem` per GRES.
* Add `/nodes` POST endpoint to REST API, supports multiple
node update whereas previously only single nodes could be
updated through `/node/<nodename>` endpoint:
`POST /slurm/v0.0.42/nodes`
node update whereas previously only single nodes could be
updated through `/node/<nodename>` endpoint:
`POST /slurm/v0.0.42/nodes`
* Do not allow changing or setting `PreemptMode=GANG` to a
partition as this is a cluster-wide option.
partition as this is a cluster-wide option.
* Add `%b` as a file name pattern for the array task id modulo 10.
* Skip packing empty nodes when they are hidden during
`REQUEST_NODE_INFO RPC`.
`REQUEST_NODE_INFO RPC`.
* `accounting_storage/mysql` - Avoid a fatal condition when
the db server is not reachable.
the db server is not reachable.
* Always lay out steps cyclically on nodes in an allocation.
* `squeue` - add priority by partition
(`.jobs[].priority_by_partition`) to JSON and YAML output.
(`.jobs[].priority_by_partition`) to JSON and YAML output.
* `slurmrestd` - Add clarification to `failed to open slurmdbd
connection` error if the error was the result of an
authentication failure.
connection` error if the error was the result of an
authentication failure.
* Make it so `slurmctld` responds to RPCs that have authentication
errors with the `SLURM_PROTOCOL_AUTHENTICATION_ERROR` error
code.
errors with the `SLURM_PROTOCOL_AUTHENTICATION_ERROR` error
code.
* `openapi/slurmctld` - Display the correct error code instead
of `Unspecified error` if querying the following endpoints
fails:
`GET /slurm/v0.0.40/diag/`
`GET /slurm/v0.0.41/diag/`
`GET /slurm/v0.0.42/diag/`
`GET /slurm/v0.0.40/licenses/`
`GET /slurm/v0.0.41/licenses/`
`GET /slurm/v0.0.42/licenses/`
`GET /slurm/v0.0.40/reconfigure`
`GET /slurm/v0.0.41/reconfigure`
`GET /slurm/v0.0.42/reconfigure`
of `Unspecified error` if querying the following endpoints
fails:
`GET /slurm/v0.0.40/diag/`
`GET /slurm/v0.0.41/diag/`
`GET /slurm/v0.0.42/diag/`
`GET /slurm/v0.0.40/licenses/`
`GET /slurm/v0.0.41/licenses/`
`GET /slurm/v0.0.42/licenses/`
`GET /slurm/v0.0.40/reconfigure`
`GET /slurm/v0.0.41/reconfigure`
`GET /slurm/v0.0.42/reconfigure`
* Fix how used CPUs are tracked in a job allocation to allow the
max number of concurrent steps to run at a time if threads per
core is greater than 1.
max number of concurrent steps to run at a time if threads per
core is greater than 1.
* In existing allocations SLURM_GPUS_PER_NODE environment
variable will be ignored by srun if `--gpus` is specified.
variable will be ignored by srun if `--gpus` is specified.
* When using `--get-user-env` explicitly or implicitly, check
if PID or mnt namespaces are disabled and fall back to old
logic that does not rely on them when they are not available.
if PID or mnt namespaces are disabled and fall back to old
logic that does not rely on them when they are not available.
* Removed non-functional option `SLURM_PROLOG_CPU_MASK` from
`TaskProlog` which was used to reset the affinity of a task
based on the mask given.
`TaskProlog` which was used to reset the affinity of a task
based on the mask given.
* `slurmrestd` - Support passing of `-d latest` to load latest
version of `data_parser` plugin.
* `sacct`,`sacctmgr`,`scontrol`,`sdiag`,`sinfo`,`squeue`,`sshare`
- Change response to `--json=list` or `--yaml=list` to send
list of plugins to stdout and descriptive header to stderr to
allow for easier parsing.
- Change response to `--json=list` or `--yaml=list` to send
list of plugins to stdout and descriptive header to stderr to
allow for easier parsing.
* `slurmrestd` - Change response to `-d list`, `-a list` or
`-s list` to send list of plugins to stdout and descriptive
header to stderr to allow for easier parsing.
`-s list` to send list of plugins to stdout and descriptive
header to stderr to allow for easier parsing.
* `sacct`,`sacctmgr`,`scontrol`,`sdiag`,`sinfo`,`squeue`,
`sshare`,`slurmrestd` - Avoid crash when loading `data_parser`
plugins fail due to NULL dereference.
plugins fail due to NULL dereference.
* Add autodetected GPUs to the output of `slurmd -C`
* Remove `burst_buffer/lua` call `slurm.job_info_to_string()`.
* Add `SchedulerParameters=bf_allow_magnetic_slot` option. It
allows jobs in magnetic reservations to be planned by backfill
scheduler.
allows jobs in magnetic reservations to be planned by backfill
scheduler.
* `slurmrestd` - Refuse to run as root, `SlurmUser`, and
`nobody(99)`.
`nobody(99)`.
* `openapi/slurmctld` - Revert regression that caused signaling
jobs to cancel entire job arrays instead of job array tasks:
`DELETE /slurm/v0.0.40/{job_id}`
`DELETE /slurm/v0.0.41/{job_id}`
`DELETE /slurm/v0.0.42/{job_id}`
jobs to cancel entire job arrays instead of job array tasks:
`DELETE /slurm/v0.0.40/{job_id}`
`DELETE /slurm/v0.0.41/{job_id}`
`DELETE /slurm/v0.0.42/{job_id}`
* `openapi/slurmctld` - Support more formats for `{job_id}`
including job steps:
`DELETE /slurm/v0.0.40/{job_id}`
`DELETE /slurm/v0.0.41/{job_id}`
`DELETE /slurm/v0.0.42/{job_id}`
including job steps:
`DELETE /slurm/v0.0.40/{job_id}`
`DELETE /slurm/v0.0.41/{job_id}`
`DELETE /slurm/v0.0.42/{job_id}`
* Alter scheduling of jobs at submission time to consider job
submission time and job id. This makes it so that that
interactive jobs aren't allocated resources before batch jobs
when they have the same priority at submit time.
submission time and job id. This makes it so that that
interactive jobs aren't allocated resources before batch jobs
when they have the same priority at submit time.
* Fix multi-cluster submissions with differing Switch plugins.
* `slurmrestd` - Change `+prefer_refs` flag to default in
`data_parser/v0.0.42` plugin. Add `+minimize_refs` flag to
inline single referenced schemas in the OpenAPI schema. This
sets the default OpenAPI schema generation behavior of
`data_parser/v0.0.42` plugin. Add `+minimize_refs` flag to
inline single referenced schemas in the OpenAPI schema. This
sets the default OpenAPI schema generation behavior of
`data_parser/v0.0.42` to match v0.0.41 `+prefer_refs` and
v0.0.40 (without flags).
v0.0.40 (without flags).
* Fix `LaunchParameters=batch_step_set_cpu_freq`.
* Clearer `seff` warning message for running jobs.
* `data_parser/v0.0.42` - Rename `JOB_INFO` field
`minimum_switches` to `required_switches` to reflect the
actual behavior.
`minimum_switches` to `required_switches` to reflect the
actual behavior.
* `data_parser/v0.0.42` - Rename `ACCOUNT_CONDITION` field
`assocation` to `association` to fix typo.
`assocation` to `association` to fix typo.
* `cgroup/v2` - fix cgroup cleanup when running inside a
container without write permissions to `/sys/fs/cgroup`.
container without write permissions to `/sys/fs/cgroup`.
* `cgroup/v2` - fix accounting of swap events detection.
* Fix gathering MaxRSS for jobs that run shorter than two
`jobacctgather` intervals. Get the metrics from cgroups
`memory.peak` or `memory.max_usage_in_bytes` where available.
`jobacctgather` intervals. Get the metrics from cgroups
`memory.peak` or `memory.max_usage_in_bytes` where available.
* `openapi/slurmctld` - Set complex number support for the
following fields:
`.shares[][].fairshare.factor`
`.shares[][].fairshare.level`
for endpoints:
`GET /slurm/v0.0.42/shares`
and for commands:
`sshare --json`
`sshare --yaml`
following fields:
`.shares[][].fairshare.factor`
`.shares[][].fairshare.level`
for endpoints:
`GET /slurm/v0.0.42/shares`
and for commands:
`sshare --json`
`sshare --yaml`
* `data_parser/v0.0.42` - Avoid dumping `Infinity` for `NO_VAL`
tagged `number` fields.
tagged `number` fields.
* Add `TopologyParam=TopoMaxSizeUnroll=#` to allow
`--nodes=<min>-<max>` for `topology/block`.
`--nodes=<min>-<max>` for `topology/block`.
* `sacct` - Respect `--noheader` for `--batch-script` and
`--env-vars`.
`--env-vars`.
* `sacct` - Remove extra newline in output from `--batch-script`
and --env-vars.
and --env-vars.
* Add `sacctmgr ping` command to query status of `slurmdbd`.
* Generate an error message when a `NodeSet` name conflicts with
a `NodeName`, and prevent the controller from starting if such
a conflict exists.
a `NodeName`, and prevent the controller from starting if such
a conflict exists.
* `slurmd` - properly detect slurmd restarts in the energy
gathering logic which caused bad numbers in accounting.
gathering logic which caused bad numbers in accounting.
* `sackd` - retry fetching slurm configs indefinately in
configless mode.
configless mode.
* `job_submit/lua` - Add `assoc_qos` attribute to `job_desc`
to display all potential QOS's for a job's association.
to display all potential QOS's for a job's association.
* `job_submit/lua` - Add `slurm.get_qos_priority()` function
to retrieve the given QOS's priority.
to retrieve the given QOS's priority.
* `sbcast` - Add `--nodelist` option to specify where files are
transmitted to.
transmitted to.
* `sbcast` - Add `--no-allocation` option to transmit files to
nodes outside of a job allocation
nodes outside of a job allocation
* Add `DataParserParameters` `slurm.conf` parameter to allow
setting default value for CLI `--json` and `--yaml` arguments.
setting default value for CLI `--json` and `--yaml` arguments.
* `seff` - improve step's max memory consumption report by using
`TresUsageInTot` and `TresUsageInAve` instead of overestimating
the values.
`TresUsageInTot` and `TresUsageInAve` instead of overestimating
the values.
* Enable RPC queueing for `REQUEST_KILL_JOBS`, which is used when
`scancel` is executed with `--ctld` flag.
`scancel` is executed with `--ctld` flag.
* `slurmdbd` - Add `-u` option. This is used to determine if
restarting the DBD will result in database conversion.
restarting the DBD will result in database conversion.
* Fix `srun` inside an `salloc` in a federated cluster when using
IPv6.
IPv6.
* Calculate the forwarding timeouts according to tree depth
rather than node count / tree width for each level. Fixes race
conditions with same timeouts between two consecutive node
levels.
rather than node count / tree width for each level. Fixes race
conditions with same timeouts between two consecutive node
levels.
* Add ability to submit jobs with multiple QOS.
* Fix difference in behavior when swapping partition order in job
submission.
submission.
* Improve `PLANNED` state detection for mixed nodes and updating
state before yielding backfill locks.
state before yielding backfill locks.
* Always consider partition priority tiers when deciding to try
scheduling jobs on submit.
scheduling jobs on submit.
* Prevent starting jobs without reservations on submit when there
are pending jobs with reservations that have flags `FLEX` or
`ANY_NODES` that can be scheduled on overlapping nodes.
are pending jobs with reservations that have flags `FLEX` or
`ANY_NODES` that can be scheduled on overlapping nodes.
* Prevent jobs that request both high and low priority tier
partitions from starting on submit in lower priority tier
partitions if it could delay pending jobs in higher priority
tier partitions.
partitions from starting on submit in lower priority tier
partitions if it could delay pending jobs in higher priority
tier partitions.
* `scontrol` - Wait for `slurmctld` to start reconfigure in
foreground mode before returning.
foreground mode before returning.
* Improve reconfigure handling on Linux to only close open file
descriptors to avoid long delays on systems with large
`RLIMIT_NOFILE` settings.
descriptors to avoid long delays on systems with large
`RLIMIT_NOFILE` settings.
* `salloc` - Removed `--get-user-env` option.
* Removed the instant on feature from `switch/hpe_slingshot`.
* Hardware collectives in `switch/hpe_slingshot` now requires
`enable_stepmgr`.
`enable_stepmgr`.
* Allow backfill to plan jobs on nodes currently being used by
exclusive user or mcs jobs.
exclusive user or mcs jobs.
* Avoid miscaching IPv6 address to hostname lookups that could
have caused logs to have the incorrect hostname.
have caused logs to have the incorrect hostname.
* `scontrol` - Add `--json`/`--yaml` support to `listpids`
* `scontrol` - Add `liststeps`
* `scontrol` - Add `listjobs`
* `slurmrestd` - Avoid connection to slurmdbd for the following
endpoints:
`GET /slurm/v0.0.42/jobs`
`GET /slurm/v0.0.42/job/{job_id}`
endpoints:
`GET /slurm/v0.0.42/jobs`
`GET /slurm/v0.0.42/job/{job_id}`
* `slurmctld` - Changed incoming RPC handling to dedicated thread
pool.
pool.
* `job_container/tmpfs` - Add `EntireStepInNS` option that will
place the `slurmstepd` process within the constructed namespace
directly.
place the `slurmstepd` process within the constructed namespace
directly.
* `scontrol show topo` - Show aggregated block sizes when using
`topology/block`.
`topology/block`.
* `slurmrestd` - Add more descriptive HTTP status for
authentication failure and connectivity errors with controller.
authentication failure and connectivity errors with controller.
* `slurmrestd` - Improve reporting errors from `slurmctld` for
job queries:
`GET /slurm/v0.0.41/{job_id}`
`GET /slurm/v0.0.41/jobs/`
job queries:
`GET /slurm/v0.0.41/{job_id}`
`GET /slurm/v0.0.41/jobs/`
* Avoid rejecting a step request that needs fewer GRES than nodes
in the job allocation.
in the job allocation.
* `slurmrestd` - Tag the never populated `.jobs[].pid` field as
deprecated for the following endpoints:
`GET /slurm/v0.0.42/{job_id}`
`GET /slurm/v0.0.42/jobs/`
deprecated for the following endpoints:
`GET /slurm/v0.0.42/{job_id}`
`GET /slurm/v0.0.42/jobs/`
* `scontrol`,`squeue` - Tag the never populated `.jobs[].pid` field
as deprecated for the following:
`scontrol show jobs --json`
`scontrol show jobs --yaml`
`scontrol show job ${JOB_ID} --json`
`scontrol show job ${JOB_ID} --yaml`
`squeue --json`
`squeue --yaml`
as deprecated for the following:
`scontrol show jobs --json`
`scontrol show jobs --yaml`
`scontrol show job ${JOB_ID} --json`
`scontrol show job ${JOB_ID} --yaml`
`squeue --json`
`squeue --yaml`
* `data_parser` v0.0.42 - fix timestamp parsing regression
introduced in in v0.0.40 (eaf3b6631f), parsing of non iso 8601
style timestamps
introduced in in v0.0.40 (eaf3b6631f), parsing of non iso 8601
style timestamps
* `cgroup/v2` will detect some special container and namespaced
setups and will work with it.
setups and will work with it.
* Support IPv6 in configless mode.
* Add `SlurmctldParamters=ignore_constraint_validation` to ignore
`constraint/feature` validation at submission.
* `slurmrestd` - Set `.pings[].mode` field as deprecated in the
following endpoints:
`GET /slurm/v0.0.42/ping`
following endpoints:
`GET /slurm/v0.0.42/ping`
* `scontrol` - Set `.pings[].mode` field as deprecated in the
following commands:
`scontrol ping --json`
`scontrol ping --yaml`
following commands:
`scontrol ping --json`
`scontrol ping --yaml`
* `slurmrestd` - Set `.pings[].pinged` field as deprecated in
the following endpoints:
`GET /slurm/v0.0.42/ping`
the following endpoints:
`GET /slurm/v0.0.42/ping`
* `scontrol` - Set `.pings[].pinged` field as deprecated in the
following commands:
`scontrol ping --json`
`scontrol ping --yaml`
following commands:
`scontrol ping --json`
`scontrol ping --yaml`
* `slurmrestd` - Add `.pings[].primary` field to the following
endpoints:
`GET /slurm/v0.0.42/ping`
endpoints:
`GET /slurm/v0.0.42/ping`
* `scontrol` - Add `.pings[].primary` field to the following
commands:
`scontrol ping --json`
`scontrol ping --yaml`
commands:
`scontrol ping --json`
`scontrol ping --yaml`
* `slurmrestd` - Add `.pings[].responding` field to the following
endpoints:
`GET /slurm/v0.0.42/ping`
endpoints:
`GET /slurm/v0.0.42/ping`
* `scontrol` - Add `.pings[].responding` field to the following
commands:
`scontrol ping --json`
`scontrol ping --yaml`
commands:
`scontrol ping --json`
`scontrol ping --yaml`
* Prevent jobs without reservations from delaying jobs in
reservations with flags `FLEX` or `ANY_NODES` in the main
scheduler.
reservations with flags `FLEX` or `ANY_NODES` in the main
scheduler.
* Fix allowing to ask for multiple different types of TRES
when one of them has a value of 0.
when one of them has a value of 0.
* `slurmctld` - Add a grace period to ensure the agent retry
queue is properly flushed during shutdown.
queue is properly flushed during shutdown.
* Don't ship `src/slurmrestd/plugins/openapi/slurmdbd/openapi.json`
`slurmrest` should always be used to enerate a new OpenAPI
schema (aka openapi.json or openapi.yaml).
`slurmrest` should always be used to enerate a new OpenAPI
schema (aka openapi.json or openapi.yaml).
* `mpi/pmix` - Fix potential deadlock and races with het jobs,
and fix potential memory and FDs leaks.
and fix potential memory and FDs leaks.
* Fix jobs with `--gpus` being rejected in some edge cases for
partitions where not all nodes have the same amount of GPUs
and CPUs configured.
partitions where not all nodes have the same amount of GPUs
and CPUs configured.
* In an extra constraints expression in a job request, do not
allow an empty string for a key or value.
allow an empty string for a key or value.
* In an extra constraints expression in a job request, fix
validation that requests are separated by boolean operators.
validation that requests are separated by boolean operators.
* Add `TaskPluginParam=OOMKillStep` to kill the step as a whole
when one task OOMs.
when one task OOMs.
* Fix `scontrol` show conf not showing all `TaskPluginParam`
elements.
elements.
* `slurmrestd` - Add fields `.job.oom_kill_step`
`.jobs[].oom_kill_step` to `POST /slurm/v0.0.42/job/submit`
and `POST /slurm/v0.0.42/job/allocate`.
`.jobs[].oom_kill_step` to `POST /slurm/v0.0.42/job/submit`
and `POST /slurm/v0.0.42/job/allocate`.
* Improve performance for `_will_run_test()`.
* Add `SchedulerParameters=bf_topopt_enable` option to enable
experimental hook to control backfill.
experimental hook to control backfill.
* If a step fails to launch under certain conditions, set the
step's state to `NODE_FAIL`.
step's state to `NODE_FAIL`.
* `sched/backfill` - Fix certain situations where a job would
not get a planned time, which could lead to it being delayed
by lower priority jobs.
not get a planned time, which could lead to it being delayed
by lower priority jobs.
* `slurmrestd` - Dump JSON `null` instead of `{}` (empty object)
for non-required fields in objects to avoid client
compatiblity issues for v0.0.42 version tagged endpoints.
for non-required fields in objects to avoid client
compatiblity issues for v0.0.42 version tagged endpoints.
* `sacct`,`sacctmgr`,`scontrol`,`sdiag`,`sinfo`,`squeue`,
`sshare` - Dump `null` instead `{}` (empty object) for
non-required fields in objects to avoid client compatiblity
issues when run with `--json` or `--yaml`.
`sshare` - Dump `null` instead `{}` (empty object) for
non-required fields in objects to avoid client compatiblity
issues when run with `--json` or `--yaml`.
-------------------------------------------------------------------
Fri Nov 1 12:50:27 UTC 2024 - Egbert Eich <eich@suse.com>

14
slurm.spec
View File

@ -174,6 +174,7 @@ Source12: slurmdbd.xml
# create: tar --owner=nobody --group=nogroup --exclude=*~ -cvzf test_setup.tar.gz test_setup
Source20: test_setup.tar.gz
Source21: README_Testsuite.md
Source22: regression.py.sle12
Patch0: Remove-rpath-from-build.patch
Patch2: pam_slurm-Initialize-arrays-and-pass-sizes.patch
Patch15: Fix-test7.2-to-find-libpmix-under-lib64-as-well.patch
@ -581,7 +582,9 @@ Requires: %{name}-lua = %version
Requires: %{name}-munge = %version
Requires: %{name}-node = %version
Requires: %{name}-openlava = %version
%if 0%{?build_slurmrestd}
Requires: %{name}-rest = %version
%endif
Requires: %{name}-seff = %version
Requires: %{name}-sjstat = %version
Requires: %{name}-slurmdbd = %version
@ -598,6 +601,7 @@ Requires: libnuma-devel
Requires: pam
Requires: pdsh
Requires: perl-%{name} = %version
Requires: readline-devel
Requires: sudo
Requires: tar
BuildRequires: sudo
@ -890,6 +894,10 @@ find -type f -name "*.[ao]" -print | while read f; do
# drop non-deterministic lto bits from .o files
strip -p --discard-locals -R .gnu.lto_* -R .gnu.debuglto_* -N __gnu_lto_v1 $f
done
# on versions < SLE15 replace regression.py with one compatible with py 3.4
%if 0%{?sle_version:1} && 0%{?sle_version} < 150000
install -m 755 %{S:22} %{buildroot}/srv/slurm-testsuite/testsuite/expect/regression.py
%endif
%if 0%{?suse_version} >= 1500
%define tar_sort --sort=name
%endif
@ -922,6 +930,12 @@ fi
sed -i -e '/ExecStart/aExecStartPre=/bin/bash -c "for i in 0 1 2 3; do test -e /dev/nvidia$i || mknod /dev/nvidia$i c 10 $((i+2)); done"' $SLURMD_SERVICE
tar -xzf %{S:20}
# on versions < SLE15 turn off AcctGatherProfileType and pmix
%if 0%{?sle_version:1} && 0%{?sle_version} < 150000
sed -i -e "/AcctGatherProfileType/s@^@#@" \
-e "/MpiDefault/s@pmix_v3@pmi2@" test_setup/slurm.conf
sed -i -e "/ProfileHDF5Dir/s@^@#@" test_setup/acct_gather.conf
%endif
mkdir -p %{buildroot}%{_pam_secconfdir}/limits.d
mv test_setup/slurm.conf.limits %{buildroot}%_pam_secconfdir/limits.d/slurm.conf
%if 0%{?sle_version} < 150200

4
test_setup.tar.gz
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:7a45706911924b06a2ec7d436d4e991d84dc459a505cbdfca244ac5fad2b9b60
size 3165
oid sha256:3c2249601135c2d6c2e6a8d7aa7318d50d354015ecf8a56fc467b43aa0059288
size 3201