Coverage for python/lsst/ctrl/bps/wms/htcondor/htcondor_service.py : 1%

Hot-keys on this page
r m x p toggle line displays
j k next/prev highlighted chunk
0 (zero) top of page
1 (one) first highlighted chunk
1# This file is part of ctrl_bps.
2#
3# Developed for the LSST Data Management System.
4# This product includes software developed by the LSST Project
5# (https://www.lsst.org).
6# See the COPYRIGHT file at the top-level directory of this distribution
7# for details of code ownership.
8#
9# This program is free software: you can redistribute it and/or modify
10# it under the terms of the GNU General Public License as published by
11# the Free Software Foundation, either version 3 of the License, or
12# (at your option) any later version.
13#
14# This program is distributed in the hope that it will be useful,
15# but WITHOUT ANY WARRANTY; without even the implied warranty of
16# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17# GNU General Public License for more details.
18#
19# You should have received a copy of the GNU General Public License
20# along with this program. If not, see <https://www.gnu.org/licenses/>.
22"""Interface between generic workflow to HTCondor workflow system.
23"""
25__all__ = ["HTCondorService", "HTCondorWorkflow"]
28import dataclasses
29import os
30import re
31import logging
32from datetime import datetime, timedelta
33from pathlib import Path
35import htcondor
37from ... import (
38 BaseWmsWorkflow,
39 BaseWmsService,
40 GenericWorkflow,
41 GenericWorkflowJob,
42 WmsRunReport,
43 WmsJobReport,
44 WmsStates
45)
46from ...bps_utils import chdir
47from .lssthtc import (
48 HTCDag,
49 HTCJob,
50 MISSING_ID,
51 JobStatus,
52 NodeStatus,
53 htc_check_dagman_output,
54 htc_escape,
55 htc_submit_dag,
56 read_dag_log,
57 read_dag_status,
58 read_node_status,
59 condor_history,
60 condor_q,
61 condor_status,
62 pegasus_name_to_label,
63 summary_from_dag,
64)
67DEFAULT_HTC_EXEC_PATT = ".*worker.*"
68"""Default pattern for searching execute machines in an HTCondor pool.
69"""
71_LOG = logging.getLogger(__name__)
74class HTCondorService(BaseWmsService):
75 """HTCondor version of WMS service.
76 """
77 def prepare(self, config, generic_workflow, out_prefix=None):
78 """Convert generic workflow to an HTCondor DAG ready for submission.
80 Parameters
81 ----------
82 config : `lsst.ctrl.bps.BpsConfig`
83 BPS configuration that includes necessary submit/runtime
84 information.
85 generic_workflow : `lsst.ctrl.bps.GenericWorkflow`
86 The generic workflow (e.g., has executable name and arguments).
87 out_prefix : `str`
88 The root directory into which all WMS-specific files are written.
90 Returns
91 ----------
92 workflow : `lsst.ctrl.bps.wms.htcondor.HTCondorWorkflow`
93 HTCondor workflow ready to be run.
94 """
95 _LOG.debug("out_prefix = '%s'", out_prefix)
96 workflow = HTCondorWorkflow.from_generic_workflow(config, generic_workflow, out_prefix,
97 f"{self.__class__.__module__}."
98 f"{self.__class__.__name__}")
99 workflow.write(out_prefix)
100 return workflow
102 def submit(self, workflow):
103 """Submit a single HTCondor workflow.
105 Parameters
106 ----------
107 workflow : `lsst.ctrl.bps.BaseWorkflow`
108 A single HTCondor workflow to submit. run_id is updated after
109 successful submission to WMS.
110 """
111 # For workflow portability, internal paths are all relative. Hence
112 # the DAG needs to be submitted to HTCondor from inside the submit
113 # directory.
114 with chdir(workflow.submit_path):
115 _LOG.info("Submitting from directory: %s", os.getcwd())
116 htc_submit_dag(workflow.dag, dict())
117 workflow.run_id = workflow.dag.run_id
119 def list_submitted_jobs(self, wms_id=None, user=None, require_bps=True, pass_thru=None):
120 """Query WMS for list of submitted WMS workflows/jobs.
122 This should be a quick lookup function to create list of jobs for
123 other functions.
125 Parameters
126 ----------
127 wms_id : `int` or `str`, optional
128 Id or path that can be used by WMS service to look up job.
129 user : `str`, optional
130 User whose submitted jobs should be listed.
131 require_bps : `bool`, optional
132 Whether to require jobs returned in list to be bps-submitted jobs.
133 pass_thru : `str`, optional
134 Information to pass through to WMS.
136 Returns
137 -------
138 job_ids : `list` [`Any`]
139 Only job ids to be used by cancel and other functions. Typically
140 this means top-level jobs (i.e., not children jobs).
141 """
142 _LOG.debug("list_submitted_jobs params: wms_id=%s, user=%s, require_bps=%s, pass_thru=%s",
143 wms_id, user, require_bps, pass_thru)
144 constraint = ""
146 if wms_id is None:
147 if user is not None:
148 constraint = f'(Owner == "{user}")'
149 else:
150 cluster_id = _wms_id_to_cluster(wms_id)
151 if cluster_id != 0:
152 constraint = f"(DAGManJobId == {cluster_id} || ClusterId == {cluster_id})"
154 if require_bps:
155 constraint += ' && (bps_isjob == "True")'
157 if pass_thru:
158 if "-forcex" in pass_thru:
159 pass_thru_2 = pass_thru.replace("-forcex", "")
160 if pass_thru_2 and not pass_thru_2.isspace():
161 constraint += f"&& ({pass_thru_2})"
162 else:
163 constraint += f" && ({pass_thru})"
165 _LOG.debug("constraint = %s", constraint)
166 jobs = condor_q(constraint)
168 # Prune child jobs where DAG job is in queue (i.e., aren't orphans).
169 job_ids = []
170 for job_id, job_info in jobs.items():
171 _LOG.debug("job_id=%s DAGManJobId=%s", job_id, job_info.get("DAGManJobId", "None"))
172 if "DAGManJobId" not in job_info: # orphaned job
173 job_ids.append(job_id)
174 else:
175 _LOG.debug("Looking for %s", f"{job_info['DAGManJobId']}.0")
176 _LOG.debug("\tin jobs.keys() = %s", jobs.keys())
177 if f"{job_info['DAGManJobId']}.0" not in jobs:
178 job_ids.append(job_id)
180 _LOG.debug("job_ids = %s", job_ids)
181 return job_ids
183 def report(self, wms_workflow_id=None, user=None, hist=0, pass_thru=None):
184 """Return run information based upon given constraints.
186 Parameters
187 ----------
188 wms_workflow_id : `str`
189 Limit to specific run based on id.
190 user : `str`
191 Limit results to runs for this user.
192 hist : `float`
193 Limit history search to this many days.
194 pass_thru : `str`
195 Constraints to pass through to HTCondor.
197 Returns
198 -------
199 runs : `list` [`lsst.ctrl.bps.WmsRunReport`]
200 Information about runs from given job information.
201 message : `str`
202 Extra message for report command to print. This could be pointers
203 to documentation or to WMS specific commands.
204 """
205 message = ""
207 if wms_workflow_id:
208 # Explicitly checking if wms_workflow_id can be converted to a
209 # float instead of using try/except to avoid catching a different
210 # ValueError from _report_from_id
211 try:
212 float(wms_workflow_id)
213 is_float = True
214 except ValueError: # Don't need TypeError here as None goes to else branch.
215 is_float = False
217 if is_float:
218 run_reports, message = _report_from_id(float(wms_workflow_id), hist)
219 else:
220 run_reports, message = _report_from_path(wms_workflow_id)
221 else:
222 run_reports, message = _summary_report(user, hist, pass_thru)
223 _LOG.debug("report: %s, %s", run_reports, message)
225 return list(run_reports.values()), message
227 def cancel(self, wms_id, pass_thru=None):
228 """Cancel submitted workflows/jobs.
230 Parameters
231 ----------
232 wms_id : `str`
233 ID or path of job that should be canceled.
234 pass_thru : `str`, optional
235 Information to pass through to WMS.
237 Returns
238 --------
239 deleted : `bool`
240 Whether successful deletion or not. Currently, if any doubt or any
241 individual jobs not deleted, return False.
242 message : `str`
243 Any message from WMS (e.g., error details).
244 """
245 _LOG.debug("Canceling wms_id = %s", wms_id)
247 cluster_id = _wms_id_to_cluster(wms_id)
248 if cluster_id == 0:
249 deleted = False
250 message = "Invalid id"
251 else:
252 _LOG.debug("Canceling cluster_id = %s", cluster_id)
253 schedd = htcondor.Schedd()
254 constraint = f"ClusterId == {cluster_id}"
255 if pass_thru is not None and "-forcex" in pass_thru:
256 pass_thru_2 = pass_thru.replace("-forcex", "")
257 if pass_thru_2 and not pass_thru_2.isspace():
258 constraint += f"&& ({pass_thru_2})"
259 _LOG.debug("JobAction.RemoveX constraint = %s", constraint)
260 results = schedd.act(htcondor.JobAction.RemoveX, constraint)
261 else:
262 if pass_thru:
263 constraint += f"&& ({pass_thru})"
264 _LOG.debug("JobAction.Remove constraint = %s", constraint)
265 results = schedd.act(htcondor.JobAction.Remove, constraint)
266 _LOG.debug("Remove results: %s", results)
268 if results["TotalSuccess"] > 0 and results["TotalError"] == 0:
269 deleted = True
270 message = ""
271 else:
272 deleted = False
273 if results["TotalSuccess"] == 0 and results["TotalError"] == 0:
274 message = "no such bps job in batch queue"
275 else:
276 message = f"unknown problems deleting: {results}"
278 _LOG.debug("deleted: %s; message = %s", deleted, message)
279 return deleted, message
282class HTCondorWorkflow(BaseWmsWorkflow):
283 """Single HTCondor workflow.
285 Parameters
286 ----------
287 name : `str`
288 Unique name for Workflow used when naming files.
289 config : `lsst.ctrl.bps.BpsConfig`
290 BPS configuration that includes necessary submit/runtime information.
291 """
292 def __init__(self, name, config=None):
293 super().__init__(name, config)
294 self.dag = None
296 @classmethod
297 def from_generic_workflow(cls, config, generic_workflow, out_prefix, service_class):
298 # Docstring inherited
299 htc_workflow = cls(generic_workflow.name, config)
300 htc_workflow.dag = HTCDag(name=generic_workflow.name)
302 _LOG.debug("htcondor dag attribs %s", generic_workflow.run_attrs)
303 htc_workflow.dag.add_attribs(generic_workflow.run_attrs)
304 htc_workflow.dag.add_attribs({"bps_wms_service": service_class,
305 "bps_wms_workflow": f"{cls.__module__}.{cls.__name__}"})
307 # Determine pool specific settings for future reference.
308 search_opts = {"default": DEFAULT_HTC_EXEC_PATT}
309 _, site = config.search("computeSite")
310 if site:
311 search_opts["curvals"] = {"curr_site": site}
312 _, patt = config.search("executeMachinesPattern", opt=search_opts)
314 # Determine the hard limit for the memory requirement.
315 #
316 # Note:
317 # To reduce the number of data that need to be dealt with we are
318 # ignoring dynamic slots (if any) as, by definition, they cannot have
319 # more memory than the partitionable slot they are the part of.
320 constraint = f'SlotType != "Dynamic" && regexp("{patt}", Machine)'
321 pool_info = condor_status(constraint=constraint)
322 if not pool_info:
323 raise RuntimeError(f"No nodes in the HTCondor pool matches pattern '{patt}'")
324 config["bps_mem_limit"] = max(int(info["TotalSlotMemory"]) for info in pool_info.values())
326 # Create all DAG jobs
327 for job_name in generic_workflow:
328 gwjob = generic_workflow.get_job(job_name)
329 htc_job = HTCondorWorkflow._create_job(config, generic_workflow, gwjob, out_prefix)
330 htc_workflow.dag.add_job(htc_job)
332 # Add job dependencies to the DAG
333 for job_name in generic_workflow:
334 htc_workflow.dag.add_job_relationships([job_name], generic_workflow.successors(job_name))
336 # If final job exists in generic workflow, create DAG final job
337 final = generic_workflow.get_final()
338 if final and isinstance(final, GenericWorkflowJob):
339 final_htjob = HTCondorWorkflow._create_job(config, generic_workflow, final, out_prefix)
340 if "post" not in final_htjob.dagcmds:
341 final_htjob.dagcmds["post"] = f"{os.path.dirname(__file__)}/final_post.sh" \
342 f" {final.name} $DAG_STATUS $RETURN"
343 htc_workflow.dag.add_final_job(final_htjob)
344 elif final and isinstance(final, GenericWorkflow):
345 raise NotImplementedError("HTCondor plugin does not support a workflow as the final job")
346 elif final:
347 return TypeError(f"Invalid type for GenericWorkflow.get_final() results ({type(final)})")
349 return htc_workflow
351 @staticmethod
352 def _create_job(config, generic_workflow, gwjob, out_prefix):
353 """Convert GenericWorkflow job nodes to DAG jobs.
355 Parameters
356 ----------
357 config : `lsst.ctrl.bps.BpsConfig`
358 BPS configuration that includes necessary submit/runtime
359 information.
360 generic_workflow : `lsst.ctrl.bps.GenericWorkflow`
361 Generic workflow that is being converted.
362 gwjob : `lsst.ctrl.bps.GenericWorkflowJob`
363 The generic job to convert to a HTCondor job.
364 out_prefix : `str`
365 Directory prefix for HTCondor files.
367 Returns
368 -------
369 htc_job : `lsst.ctrl.bps.wms.htcondor.HTCJob`
370 The HTCondor job equivalent to the given generic job.
371 """
372 htc_job = HTCJob(gwjob.name, label=gwjob.label)
374 curvals = dataclasses.asdict(gwjob)
375 if gwjob.tags:
376 curvals.update(gwjob.tags)
377 found, subdir = config.search("subDirTemplate", opt={'curvals': curvals})
378 if not found:
379 subdir = "jobs"
380 htc_job.subfile = Path("jobs") / subdir / f"{gwjob.name}.sub"
382 htc_job_cmds = {
383 "universe": "vanilla",
384 "should_transfer_files": "YES",
385 "when_to_transfer_output": "ON_EXIT_OR_EVICT",
386 "transfer_executable": "False",
387 "getenv": "True",
389 # Exceeding memory sometimes triggering SIGBUS error. Tell htcondor
390 # to put SIGBUS jobs on hold.
391 "on_exit_hold": "(ExitBySignal == true) && (ExitSignal == 7)",
392 "on_exit_hold_reason": '"Job raised a signal 7. Usually means job has gone over memory limit."',
393 "on_exit_hold_subcode": "34"
394 }
396 htc_job_cmds.update(_translate_job_cmds(config, generic_workflow, gwjob))
398 # job stdout, stderr, htcondor user log.
399 for key in ("output", "error", "log"):
400 htc_job_cmds[key] = htc_job.subfile.with_suffix(f".$(Cluster).{key[:3]}")
401 _LOG.debug("HTCondor %s = %s", key, htc_job_cmds[key])
403 _, use_shared = config.search("bpsUseShared", opt={"default": False})
404 htc_job_cmds.update(_handle_job_inputs(generic_workflow, gwjob.name, use_shared, out_prefix))
406 # Add the job cmds dict to the job object.
407 htc_job.add_job_cmds(htc_job_cmds)
409 htc_job.add_dag_cmds(_translate_dag_cmds(gwjob))
411 # Add run level attributes to job.
412 htc_job.add_job_attrs(generic_workflow.run_attrs)
414 # Add job attributes to job.
415 _LOG.debug("gwjob.attrs = %s", gwjob.attrs)
416 htc_job.add_job_attrs(gwjob.attrs)
417 if gwjob.tags:
418 htc_job.add_job_attrs({"bps_job_quanta": gwjob.tags.get("quanta_summary", "")})
419 htc_job.add_job_attrs({"bps_job_name": gwjob.name,
420 "bps_job_label": gwjob.label})
422 return htc_job
424 def write(self, out_prefix):
425 """Output HTCondor DAGMan files needed for workflow submission.
427 Parameters
428 ----------
429 out_prefix : `str`
430 Directory prefix for HTCondor files.
431 """
432 self.submit_path = out_prefix
433 os.makedirs(out_prefix, exist_ok=True)
435 # Write down the workflow in HTCondor format.
436 self.dag.write(out_prefix, "jobs/{self.label}")
439def _translate_job_cmds(config, generic_workflow, gwjob):
440 """Translate the job data that are one to one mapping
442 Parameters
443 ----------
444 config : `lsst.ctrl.bps.BpsConfig`
445 BPS configuration that includes necessary submit/runtime
446 information.
447 generic_workflow : `lsst.ctrl.bps.GenericWorkflow`
448 Generic workflow that contains job to being converted.
449 gwjob : `lsst.ctrl.bps.GenericWorkflowJob`
450 Generic workflow job to be converted.
452 Returns
453 -------
454 htc_job_commands : `dict` [`str`, `Any`]
455 Contains commands which can appear in the HTCondor submit description
456 file.
457 """
458 # Values in the job script that just are name mappings.
459 job_translation = {"mail_to": "notify_user",
460 "when_to_mail": "notification",
461 "request_cpus": "request_cpus",
462 "priority": "priority",
463 "category": "category"}
465 jobcmds = {}
466 for gwkey, htckey in job_translation.items():
467 jobcmds[htckey] = getattr(gwjob, gwkey, None)
469 # job commands that need modification
470 if gwjob.number_of_retries:
471 jobcmds["max_retries"] = f"{gwjob.number_of_retries}"
473 if gwjob.retry_unless_exit:
474 jobcmds["retry_until"] = f"{gwjob.retry_unless_exit}"
476 if gwjob.request_disk:
477 jobcmds["request_disk"] = f"{gwjob.request_disk}MB"
479 if gwjob.request_memory:
480 jobcmds["request_memory"] = f"{gwjob.request_memory}"
482 if gwjob.memory_multiplier:
483 _, memory_limit = config.search("bps_mem_limit")
484 jobcmds["request_memory"] = _create_request_memory_expr(gwjob.request_memory, gwjob.memory_multiplier)
486 # Periodically release jobs which are being held due to exceeding
487 # memory. Stop doing that (by removing the job from the HTCondor queue)
488 # after the maximal number of retries has been reached or the memory
489 # requirements cannot be satisfied.
490 jobcmds["periodic_release"] = \
491 "NumJobStarts <= JobMaxRetries && (HoldReasonCode == 34 || HoldReasonSubCode == 34)"
492 jobcmds["periodic_remove"] = \
493 f"JobStatus == 1 && RequestMemory > {memory_limit} || " \
494 f"JobStatus == 5 && NumJobStarts > JobMaxRetries"
496 # Assume concurrency_limit implemented using HTCondor concurrency limits.
497 # May need to move to special site-specific implementation if sites use
498 # other mechanisms.
499 if gwjob.concurrency_limit:
500 jobcmds["concurrency_limit"] = ",".join(gwjob.concurrency_limit)
502 # Handle command line
503 if gwjob.executable.transfer_executable:
504 jobcmds["transfer_executable"] = "True"
505 jobcmds["executable"] = os.path.basename(gwjob.executable.src_uri)
506 else:
507 jobcmds["executable"] = _fix_env_var_syntax(gwjob.executable.src_uri)
509 if gwjob.arguments:
510 arguments = gwjob.arguments
511 arguments = _replace_cmd_vars(arguments, gwjob)
512 arguments = _replace_file_vars(config, arguments, generic_workflow, gwjob)
513 arguments = _fix_env_var_syntax(arguments)
514 jobcmds["arguments"] = arguments
516 # Add extra "pass-thru" job commands
517 if gwjob.profile:
518 for key, val in gwjob.profile.items():
519 jobcmds[key] = htc_escape(val)
521 return jobcmds
524def _translate_dag_cmds(gwjob):
525 """Translate job values into DAGMan commands.
527 Parameters
528 ----------
529 gwjob : `lsst.ctrl.bps.GenericWorkflowJob`
530 Job containing values to be translated.
532 Returns
533 -------
534 dagcmds : `dict` [`str`, `Any`]
535 DAGMan commands for the job.
536 """
537 # Values in the dag script that just are name mappings.
538 dag_translation = {"abort_on_value": "abort_dag_on",
539 "abort_return_value": "abort_exit"}
541 dagcmds = {}
542 for gwkey, htckey in dag_translation.items():
543 dagcmds[htckey] = getattr(gwjob, gwkey, None)
545 # Still to be coded: vars "pre_cmdline", "post_cmdline"
546 return dagcmds
549def _fix_env_var_syntax(oldstr):
550 """Change ENV place holders to HTCondor Env var syntax.
552 Parameters
553 ----------
554 oldstr : `str`
555 String in which environment variable syntax is to be fixed.
557 Returns
558 -------
559 newstr : `str`
560 Given string with environment variable syntax fixed.
561 """
562 newstr = oldstr
563 for key in re.findall(r"<ENV:([^>]+)>", oldstr):
564 newstr = newstr.replace(rf"<ENV:{key}>", f"$ENV({key})")
565 return newstr
568def _replace_file_vars(config, arguments, workflow, gwjob):
569 """Replace file placeholders in command line arguments with correct
570 physical file names.
572 Parameters
573 ----------
574 config : `lsst.ctrl.bps.BpsConfig`
575 BPS configuration that includes necessary submit/runtime
576 information.
577 arguments : `str`
578 Arguments string in which to replace file placeholders.
579 workflow : `lsst.ctrl.bps.GenericWorkflow`
580 Generic workflow that contains file information.
581 gwjob : `lsst.ctrl.bps.GenericWorkflowJob`
582 The job corresponding to the arguments.
584 Returns
585 -------
586 arguments : `str`
587 Given arguments string with file placeholders replaced.
588 """
589 _, use_shared = config.search("bpsUseShared", opt={"default": False})
591 # Replace input file placeholders with paths.
592 for gwfile in workflow.get_job_inputs(gwjob.name, data=True, transfer_only=False):
593 if gwfile.wms_transfer and not use_shared or not gwfile.job_shared:
594 uri = os.path.basename(gwfile.src_uri)
595 else:
596 uri = gwfile.src_uri
597 arguments = arguments.replace(f"<FILE:{gwfile.name}>", uri)
599 # Replace output file placeholders with paths.
600 for gwfile in workflow.get_job_outputs(gwjob.name, data=True, transfer_only=False):
601 if gwfile.wms_transfer and not use_shared or not gwfile.job_shared:
602 uri = os.path.basename(gwfile.src_uri)
603 else:
604 uri = gwfile.src_uri
605 arguments = arguments.replace(f"<FILE:{gwfile.name}>", uri)
606 return arguments
609def _replace_cmd_vars(arguments, gwjob):
610 """Replace format-style placeholders in arguments.
612 Parameters
613 ----------
614 arguments : `str`
615 Arguments string in which to replace placeholders.
616 gwjob : `lsst.ctrl.bps.GenericWorkflowJob`
617 Job containing values to be used to replace placeholders
618 (in particular gwjob.cmdvals).
620 Returns
621 -------
622 arguments : `str`
623 Given arguments string with placeholders replaced.
624 """
625 try:
626 arguments = arguments.format(**gwjob.cmdvals)
627 except (KeyError, TypeError): # TypeError in case None instead of {}
628 _LOG.error("Could not replace command variables:\n"
629 "arguments: %s\n"
630 "cmdvals: %s", arguments, gwjob.cmdvals)
631 raise
632 return arguments
635def _handle_job_inputs(generic_workflow: GenericWorkflow, job_name: str, use_shared: bool, out_prefix: str):
636 """Add job input files from generic workflow to job.
638 Parameters
639 ----------
640 generic_workflow : `lsst.ctrl.bps.GenericWorkflow`
641 The generic workflow (e.g., has executable name and arguments).
642 job_name : `str`
643 Unique name for the job.
644 use_shared : `bool`
645 Whether job has access to files via shared filesystem.
646 out_prefix : `str`
647 The root directory into which all WMS-specific files are written.
649 Returns
650 -------
651 htc_commands : `dict` [`str`, `str`]
652 HTCondor commands for the job submission script.
653 """
654 htc_commands = {}
655 inputs = []
656 for gwf_file in generic_workflow.get_job_inputs(job_name, data=True, transfer_only=True):
657 _LOG.debug("src_uri=%s", gwf_file.src_uri)
658 if not use_shared or not gwf_file.job_shared:
659 inputs.append(os.path.relpath(gwf_file.src_uri, out_prefix))
661 if inputs:
662 htc_commands["transfer_input_files"] = ",".join(inputs)
663 _LOG.debug("transfer_input_files=%s", htc_commands["transfer_input_files"])
664 return htc_commands
667def _report_from_path(wms_path):
668 """Gather run information from a given run directory.
670 Parameters
671 ----------
672 wms_path : `str`
673 The directory containing the submit side files (e.g., HTCondor files).
675 Returns
676 -------
677 run_reports : `dict` [`str`, `lsst.ctrl.bps.WmsRunReport`]
678 Run information for the detailed report. The key is the HTCondor id
679 and the value is a collection of report information for that run.
680 message : `str`
681 Message to be printed with the summary report.
682 """
683 wms_workflow_id, jobs, message = _get_info_from_path(wms_path)
684 if wms_workflow_id == MISSING_ID:
685 run_reports = {}
686 else:
687 run_reports = _create_detailed_report_from_jobs(wms_workflow_id, jobs)
688 return run_reports, message
691def _report_from_id(wms_workflow_id, hist):
692 """Gather run information from a given run directory.
694 Parameters
695 ----------
696 wms_workflow_id : `int` or `str`
697 Limit to specific run based on id.
698 hist : `float`
699 Limit history search to this many days.
701 Returns
702 -------
703 run_reports : `dict` [`str`, `lsst.ctrl.bps.WmsRunReport`]
704 Run information for the detailed report. The key is the HTCondor id
705 and the value is a collection of report information for that run.
706 message : `str`
707 Message to be printed with the summary report.
708 """
709 constraint = f"(DAGManJobId == {int(float(wms_workflow_id))} || ClusterId == " \
710 f"{int(float(wms_workflow_id))})"
711 jobs = condor_q(constraint)
712 if hist:
713 epoch = (datetime.now() - timedelta(days=hist)).timestamp()
714 constraint += f" && (CompletionDate >= {epoch} || JobFinishedHookDone >= {epoch})"
715 hist_jobs = condor_history(constraint)
716 _update_jobs(jobs, hist_jobs)
718 # keys in dictionary will be strings of format "ClusterId.ProcId"
719 wms_workflow_id = str(wms_workflow_id)
720 if not wms_workflow_id.endswith(".0"):
721 wms_workflow_id += ".0"
723 if wms_workflow_id in jobs:
724 _, path_jobs, message = _get_info_from_path(jobs[wms_workflow_id]["Iwd"])
725 _update_jobs(jobs, path_jobs)
726 run_reports = _create_detailed_report_from_jobs(wms_workflow_id, jobs)
727 else:
728 run_reports = {}
729 message = f"Found 0 records for run id {wms_workflow_id}"
730 return run_reports, message
733def _get_info_from_path(wms_path):
734 """Gather run information from a given run directory.
736 Parameters
737 ----------
738 wms_path : `str`
739 Directory containing HTCondor files.
741 Returns
742 -------
743 wms_workflow_id : `str`
744 The run id which is a DAGman job id.
745 jobs : `dict` [`str`, `dict` [`str`, `Any`]]
746 Information about jobs read from files in the given directory.
747 The key is the HTCondor id and the value is a dictionary of HTCondor
748 keys and values.
749 message : `str`
750 Message to be printed with the summary report.
751 """
752 try:
753 wms_workflow_id, jobs = read_dag_log(wms_path)
754 _LOG.debug("_get_info_from_path: from dag log %s = %s", wms_workflow_id, jobs)
755 _update_jobs(jobs, read_node_status(wms_path))
756 _LOG.debug("_get_info_from_path: after node status %s = %s", wms_workflow_id, jobs)
758 # Add more info for DAGman job
759 job = jobs[wms_workflow_id]
760 job.update(read_dag_status(wms_path))
761 job["total_jobs"], job["state_counts"] = _get_state_counts_from_jobs(wms_workflow_id, jobs)
762 if "bps_run" not in job:
763 _add_run_info(wms_path, job)
765 message = htc_check_dagman_output(wms_path)
766 _LOG.debug("_get_info: id = %s, total_jobs = %s", wms_workflow_id,
767 jobs[wms_workflow_id]["total_jobs"])
768 except StopIteration:
769 message = f"Could not find HTCondor files in {wms_path}"
770 _LOG.warning(message)
771 wms_workflow_id = MISSING_ID
772 jobs = {}
774 return wms_workflow_id, jobs, message
777def _create_detailed_report_from_jobs(wms_workflow_id, jobs):
778 """Gather run information to be used in generating summary reports.
780 Parameters
781 ----------
782 wms_workflow_id : `str`
783 Run lookup restricted to given user.
784 jobs : `float`
785 How many previous days to search for run information.
787 Returns
788 -------
789 run_reports : `dict` [`str`, `lsst.ctrl.bps.WmsRunReport`]
790 Run information for the detailed report. The key is the given HTCondor
791 id and the value is a collection of report information for that run.
792 """
793 _LOG.debug("_create_detailed_report: id = %s, job = %s", wms_workflow_id, jobs[wms_workflow_id])
794 dag_job = jobs[wms_workflow_id]
795 if "total_jobs" not in dag_job or "DAGNodeName" in dag_job:
796 _LOG.error("Job ID %s is not a DAG job.", wms_workflow_id)
797 return {}
798 report = WmsRunReport(wms_id=wms_workflow_id,
799 path=dag_job["Iwd"],
800 label=dag_job.get("bps_job_label", "MISS"),
801 run=dag_job.get("bps_run", "MISS"),
802 project=dag_job.get("bps_project", "MISS"),
803 campaign=dag_job.get("bps_campaign", "MISS"),
804 payload=dag_job.get("bps_payload", "MISS"),
805 operator=_get_owner(dag_job),
806 run_summary=_get_run_summary(dag_job),
807 state=_htc_status_to_wms_state(dag_job),
808 jobs=[],
809 total_number_jobs=dag_job["total_jobs"],
810 job_state_counts=dag_job["state_counts"])
812 try:
813 for job in jobs.values():
814 if job["ClusterId"] != int(float(wms_workflow_id)):
815 job_report = WmsJobReport(wms_id=job["ClusterId"],
816 name=job.get("DAGNodeName", str(job["ClusterId"])),
817 label=job.get("bps_job_label",
818 pegasus_name_to_label(job["DAGNodeName"])),
819 state=_htc_status_to_wms_state(job))
820 if job_report.label == "init":
821 job_report.label = "pipetaskInit"
822 report.jobs.append(job_report)
823 except KeyError as ex:
824 _LOG.error("Job missing key '%s': %s", str(ex), job)
825 raise
827 run_reports = {report.wms_id: report}
828 _LOG.debug("_create_detailed_report: run_reports = %s", run_reports)
829 return run_reports
832def _summary_report(user, hist, pass_thru):
833 """Gather run information to be used in generating summary reports.
835 Parameters
836 ----------
837 user : `str`
838 Run lookup restricted to given user.
839 hist : `float`
840 How many previous days to search for run information.
841 pass_thru : `str`
842 Advanced users can define the HTCondor constraint to be used
843 when searching queue and history.
845 Returns
846 -------
847 run_reports : `dict` [`str`, `lsst.ctrl.bps.WmsRunReport`]
848 Run information for the summary report. The keys are HTCondor ids and
849 the values are collections of report information for each run.
850 message : `str`
851 Message to be printed with the summary report.
852 """
853 # only doing summary report so only look for dagman jobs
854 if pass_thru:
855 constraint = pass_thru
856 else:
857 # Notes:
858 # * bps_isjob == 'True' isn't getting set for DAG jobs that are
859 # manually restarted.
860 # * Any job with DAGManJobID isn't a DAG job
861 constraint = 'bps_isjob == "True" && JobUniverse == 7'
862 if user:
863 constraint += f' && (Owner == "{user}" || bps_operator == "{user}")'
865 # Check runs in queue.
866 jobs = condor_q(constraint)
868 if hist:
869 epoch = (datetime.now() - timedelta(days=hist)).timestamp()
870 constraint += f" && (CompletionDate >= {epoch} || JobFinishedHookDone >= {epoch})"
871 hist_jobs = condor_history(constraint)
872 _update_jobs(jobs, hist_jobs)
874 _LOG.debug("Job ids from queue and history %s", jobs.keys())
876 # Have list of DAGMan jobs, need to get run_report info.
877 run_reports = {}
878 for job in jobs.values():
879 total_jobs, state_counts = _get_state_counts_from_dag_job(job)
880 # If didn't get from queue information (e.g., Kerberos bug),
881 # try reading from file.
882 if total_jobs == 0:
883 try:
884 job.update(read_dag_status(job["Iwd"]))
885 total_jobs, state_counts = _get_state_counts_from_dag_job(job)
886 except StopIteration:
887 pass # don't kill report can't find htcondor files
889 if "bps_run" not in job:
890 _add_run_info(job["Iwd"], job)
891 report = WmsRunReport(wms_id=str(job.get("ClusterId", MISSING_ID)),
892 path=job["Iwd"],
893 label=job.get("bps_job_label", "MISS"),
894 run=job.get("bps_run", "MISS"),
895 project=job.get("bps_project", "MISS"),
896 campaign=job.get("bps_campaign", "MISS"),
897 payload=job.get("bps_payload", "MISS"),
898 operator=_get_owner(job),
899 run_summary=_get_run_summary(job),
900 state=_htc_status_to_wms_state(job),
901 jobs=[],
902 total_number_jobs=total_jobs,
903 job_state_counts=state_counts)
905 run_reports[report.wms_id] = report
907 return run_reports, ""
910def _add_run_info(wms_path, job):
911 """Find BPS run information elsewhere for runs without bps attributes.
913 Parameters
914 ----------
915 wms_path : `str`
916 Path to submit files for the run.
917 job : `dict` [`str`, `Any`]
918 HTCondor dag job information.
920 Raises
921 ------
922 StopIteration
923 If cannot find file it is looking for. Permission errors are
924 caught and job's run is marked with error.
925 """
926 path = Path(wms_path) / "jobs"
927 try:
928 jobdir = next(path.glob("*"), Path(wms_path))
929 try:
930 subfile = next(jobdir.glob("*.sub"))
931 _LOG.debug("_add_run_info: subfile = %s", subfile)
932 with open(subfile, "r") as fh:
933 for line in fh:
934 if line.startswith("+bps_"):
935 m = re.match(r"\+(bps_[^\s]+)\s*=\s*(.+)$", line)
936 if m:
937 _LOG.debug("Matching line: %s", line)
938 job[m.group(1)] = m.group(2).replace('"', "")
939 else:
940 _LOG.debug("Could not parse attribute: %s", line)
941 except StopIteration:
942 job["bps_run"] = "Missing"
944 except PermissionError:
945 job["bps_run"] = "PermissionError"
946 _LOG.debug("After adding job = %s", job)
949def _get_owner(job):
950 """Get the owner of a dag job.
952 Parameters
953 ----------
954 job : `dict` [`str`, `Any`]
955 HTCondor dag job information.
957 Returns
958 -------
959 owner : `str`
960 Owner of the dag job.
961 """
962 owner = job.get("bps_operator", None)
963 if not owner:
964 owner = job.get("Owner", None)
965 if not owner:
966 _LOG.warning("Could not get Owner from htcondor job: %s", job)
967 owner = "MISS"
968 return owner
971def _get_run_summary(job):
972 """Get the run summary for a job.
974 Parameters
975 ----------
976 job : `dict` [`str`, `Any`]
977 HTCondor dag job information.
979 Returns
980 -------
981 summary : `str`
982 Number of jobs per PipelineTask label in approximate pipeline order.
983 Format: <label>:<count>[;<label>:<count>]+
984 """
985 summary = job.get("bps_run_summary", None)
986 if not summary:
987 summary, _ = summary_from_dag(job["Iwd"])
988 if not summary:
989 _LOG.warning("Could not get run summary for htcondor job: %s", job)
990 _LOG.debug("_get_run_summary: summary=%s", summary)
992 # Workaround sometimes using init vs pipetaskInit
993 summary = summary.replace("init:", "pipetaskInit:")
995 if "pegasus_version" in job and "pegasus" not in summary:
996 summary += ";pegasus:0"
998 return summary
1001def _get_state_counts_from_jobs(wms_workflow_id, jobs):
1002 """Count number of jobs per WMS state.
1004 Parameters
1005 ----------
1006 wms_workflow_id : `str`
1007 HTCondor job id.
1008 jobs : `dict` [`str`, `Any`]
1009 HTCondor dag job information.
1011 Returns
1012 -------
1013 total_count : `int`
1014 Total number of dag nodes.
1015 state_counts : `dict` [`lsst.ctrl.bps.WmsStates`, `int`]
1016 Keys are the different WMS states and values are counts of jobs
1017 that are in that WMS state.
1018 """
1019 state_counts = dict.fromkeys(WmsStates, 0)
1021 for jid, jinfo in jobs.items():
1022 if jid != wms_workflow_id:
1023 state_counts[_htc_status_to_wms_state(jinfo)] += 1
1025 total_counted = sum(state_counts.values())
1026 if "NodesTotal" in jobs[wms_workflow_id]:
1027 total_count = jobs[wms_workflow_id]["NodesTotal"]
1028 else:
1029 total_count = total_counted
1031 state_counts[WmsStates.UNREADY] += total_count - total_counted
1033 return total_count, state_counts
1036def _get_state_counts_from_dag_job(job):
1037 """Count number of jobs per WMS state.
1039 Parameters
1040 ----------
1041 job : `dict` [`str`, `Any`]
1042 HTCondor dag job information.
1044 Returns
1045 -------
1046 total_count : `int`
1047 Total number of dag nodes.
1048 state_counts : `dict` [`lsst.ctrl.bps.WmsStates`, `int`]
1049 Keys are the different WMS states and values are counts of jobs
1050 that are in that WMS state.
1051 """
1052 _LOG.debug("_get_state_counts_from_dag_job: job = %s %s", type(job), len(job))
1053 state_counts = dict.fromkeys(WmsStates, 0)
1054 if "DAG_NodesReady" in job:
1055 state_counts = {
1056 WmsStates.UNREADY: job.get("DAG_NodesUnready", 0),
1057 WmsStates.READY: job.get("DAG_NodesReady", 0),
1058 WmsStates.HELD: job.get("JobProcsHeld", 0),
1059 WmsStates.SUCCEEDED: job.get("DAG_NodesDone", 0),
1060 WmsStates.FAILED: job.get("DAG_NodesFailed", 0),
1061 WmsStates.MISFIT: job.get("DAG_NodesPre", 0) + job.get("DAG_NodesPost", 0)}
1062 total_jobs = job.get("DAG_NodesTotal")
1063 _LOG.debug("_get_state_counts_from_dag_job: from DAG_* keys, total_jobs = %s", total_jobs)
1064 elif "NodesFailed" in job:
1065 state_counts = {
1066 WmsStates.UNREADY: job.get("NodesUnready", 0),
1067 WmsStates.READY: job.get("NodesReady", 0),
1068 WmsStates.HELD: job.get("JobProcsHeld", 0),
1069 WmsStates.SUCCEEDED: job.get("NodesDone", 0),
1070 WmsStates.FAILED: job.get("NodesFailed", 0),
1071 WmsStates.MISFIT: job.get("NodesPre", 0) + job.get("NodesPost", 0)}
1072 try:
1073 total_jobs = job.get("NodesTotal")
1074 except KeyError as ex:
1075 _LOG.error("Job missing %s. job = %s", str(ex), job)
1076 raise
1077 _LOG.debug("_get_state_counts_from_dag_job: from NODES* keys, total_jobs = %s", total_jobs)
1078 else:
1079 # With Kerberos job auth and Kerberos bug, if warning would be printed
1080 # for every DAG.
1081 _LOG.debug("Can't get job state counts %s", job["Iwd"])
1082 total_jobs = 0
1084 _LOG.debug("total_jobs = %s, state_counts: %s", total_jobs, state_counts)
1085 return total_jobs, state_counts
1088def _htc_status_to_wms_state(job):
1089 """Convert HTCondor job status to generic wms state.
1091 Parameters
1092 ----------
1093 job : `dict` [`str`, `Any`]
1094 HTCondor job information.
1096 Returns
1097 -------
1098 wms_state : `WmsStates`
1099 The equivalent WmsState to given job's status.
1100 """
1101 wms_state = WmsStates.MISFIT
1102 if "JobStatus" in job:
1103 wms_state = _htc_job_status_to_wms_state(job)
1104 elif "NodeStatus" in job:
1105 wms_state = _htc_node_status_to_wms_state(job)
1106 return wms_state
1109def _htc_job_status_to_wms_state(job):
1110 """Convert HTCondor job status to generic wms state.
1112 Parameters
1113 ----------
1114 job : `dict` [`str`, `Any`]
1115 HTCondor job information.
1117 Returns
1118 -------
1119 wms_state : `lsst.ctrl.bps.WmsStates`
1120 The equivalent WmsState to given job's status.
1121 """
1122 _LOG.debug("htc_job_status_to_wms_state: %s=%s, %s", job["ClusterId"], job["JobStatus"],
1123 type(job["JobStatus"]))
1124 job_status = int(job["JobStatus"])
1125 wms_state = WmsStates.MISFIT
1127 _LOG.debug("htc_job_status_to_wms_state: job_status = %s", job_status)
1128 if job_status == JobStatus.IDLE:
1129 wms_state = WmsStates.PENDING
1130 elif job_status == JobStatus.RUNNING:
1131 wms_state = WmsStates.RUNNING
1132 elif job_status == JobStatus.REMOVED:
1133 wms_state = WmsStates.DELETED
1134 elif job_status == JobStatus.COMPLETED:
1135 if job.get("ExitBySignal", False) or job.get("ExitCode", 0) or \
1136 job.get("ExitSignal", 0) or job.get("DAG_Status", 0) or \
1137 job.get("ReturnValue", 0):
1138 wms_state = WmsStates.FAILED
1139 else:
1140 wms_state = WmsStates.SUCCEEDED
1141 elif job_status == JobStatus.HELD:
1142 wms_state = WmsStates.HELD
1144 return wms_state
1147def _htc_node_status_to_wms_state(job):
1148 """Convert HTCondor status to generic wms state.
1150 Parameters
1151 ----------
1152 job : `dict` [`str`, `Any`]
1153 HTCondor job information.
1155 Returns
1156 -------
1157 wms_state : `lsst.ctrl.bps.WmsStates`
1158 The equivalent WmsState to given node's status.
1159 """
1160 wms_state = WmsStates.MISFIT
1162 status = job["NodeStatus"]
1163 if status == NodeStatus.NOT_READY:
1164 wms_state = WmsStates.UNREADY
1165 elif status == NodeStatus.READY:
1166 wms_state = WmsStates.READY
1167 elif status == NodeStatus.PRERUN:
1168 wms_state = WmsStates.MISFIT
1169 elif status == NodeStatus.SUBMITTED:
1170 if job["JobProcsHeld"]:
1171 wms_state = WmsStates.HELD
1172 elif job["StatusDetails"] == "not_idle":
1173 wms_state = WmsStates.RUNNING
1174 elif job["JobProcsQueued"]:
1175 wms_state = WmsStates.PENDING
1176 elif status == NodeStatus.POSTRUN:
1177 wms_state = WmsStates.MISFIT
1178 elif status == NodeStatus.DONE:
1179 wms_state = WmsStates.SUCCEEDED
1180 elif status == NodeStatus.ERROR:
1181 wms_state = WmsStates.FAILED
1183 return wms_state
1186def _update_jobs(jobs1, jobs2):
1187 """Update jobs1 with info in jobs2.
1189 (Basically an update for nested dictionaries.)
1191 Parameters
1192 ----------
1193 jobs1 : `dict` [`str`, `dict` [`str`, `Any`]]
1194 HTCondor job information to be updated.
1195 jobs2 : `dict` [`str`, `dict` [`str`, `Any`]]
1196 Additional HTCondor job information.
1197 """
1198 for jid, jinfo in jobs2.items():
1199 if jid in jobs1:
1200 jobs1[jid].update(jinfo)
1201 else:
1202 jobs1[jid] = jinfo
1205def _wms_id_to_cluster(wms_id):
1206 """Convert WMS ID to cluster ID.
1208 Parameters
1209 ----------
1210 wms_id : `int` or `float` or `str`
1211 HTCondor job id or path.
1213 Returns
1214 -------
1215 cluster_id : `int`
1216 HTCondor cluster id.
1217 """
1218 # If wms_id represents path, get numeric id.
1219 try:
1220 cluster_id = int(float(wms_id))
1221 except ValueError:
1222 wms_path = Path(wms_id)
1223 if wms_path.exists():
1224 try:
1225 cluster_id, _ = read_dag_log(wms_id)
1226 cluster_id = int(float(cluster_id))
1227 except StopIteration:
1228 cluster_id = 0
1229 else:
1230 cluster_id = 0
1231 return cluster_id
1234def _create_request_memory_expr(memory, multiplier):
1235 """Construct an HTCondor ClassAd expression for safe memory scaling.
1237 Parameters
1238 ----------
1239 memory : `int`
1240 Requested memory in MB.
1241 multiplier : `float`
1242 Memory growth rate between retires.
1244 Returns
1245 -------
1246 ad : `str`
1247 A string representing an HTCondor ClassAd expression enabling safe
1248 memory scaling between job retries.
1249 """
1250 was_mem_exceeded = "LastJobStatus =?= 5 " \
1251 "&& (LastHoldReasonCode =?= 34 && LastHoldReasonSubCode =?= 0 " \
1252 "|| LastHoldReasonCode =?= 3 && LastHoldReasonSubCode =?= 34)"
1254 # If job runs the first time ('MemoryUsage' is not defined), set the
1255 # required memory to a given value.
1256 ad = f"ifThenElse({was_mem_exceeded}, " \
1257 f"ifThenElse(isUndefined(MemoryUsage), {memory}, int({multiplier} * MemoryUsage)), " \
1258 f"ifThenElse(isUndefined(MemoryUsage), {memory}, max({memory}, MemoryUsage)))"
1259 return ad