HTCondor Class Reference

Inheritance diagram for HTCondor:

Classes
class	HTCondorResult

Public Member Functions
	get_batch_submit_script_path (self, job)
	can_submit (self, njobs=1)
	condor_q (cls, class_ads=None, job_id="", username="")
	condor_history (cls, class_ads=None, job_id="", username="")
	submit (self, job, check_can_submit=True, jobs_per_check=100)
	get_submit_script_path (self, job)

Public Attributes
int	global_job_limit = self.default_global_job_limit
	The active job limit.
int	sleep_between_submission_checks = self.default_sleep_between_submission_checks
	Seconds we wait before checking if we can submit a list of jobs.
dict	backend_args = {**self.default_backend_args, **backend_args}
	The backend args that will be applied to jobs unless the job specifies them itself.

Static Public Attributes
str	batch_submit_script = "submit.sub"
	HTCondor batch script (different to the wrapper script of Backend.submit_script)
list	default_class_ads = ["GlobalJobId", "JobStatus", "Owner"]
	Default ClassAd attributes to return from commands like condor_q.
list	submission_cmds = []
	Shell command to submit a script, should be implemented in the derived class.
int	default_global_job_limit = 1000
	Default global limit on the total number of submitted/running jobs that the user can have.
int	default_sleep_between_submission_checks = 30
	Default time betweeon re-checking if the active jobs is below the global job limit.
str	submit_script = "submit.sh"
	Default submission script name.
str	exit_code_file = "__BACKEND_CMD_EXIT_STATUS__"
	Default exit code file name.
dict	default_backend_args = {}
	Default backend_args.

Protected Member Functions
	_make_submit_file (self, job, submit_file_path)
	_add_batch_directives (self, job, batch_file)
	_create_cmd (self, script_path)
	_submit_to_batch (cls, cmd)
	_create_job_result (cls, job, job_id)
	_create_parent_job_result (cls, parent)
	_add_wrapper_script_setup (self, job, batch_file)
	_add_wrapper_script_teardown (self, job, batch_file)

Static Protected Member Functions
	_add_setup (job, batch_file)

Detailed Description

Backend for submitting calibration processes to a HTCondor batch system.

Definition at line 1929 of file backends.py.

Member Function Documentation

_add_batch_directives()

_add_batch_directives ( self, job, batch_file )

protected

For HTCondor leave empty as the directives are already included in the submit file.

Reimplemented from Batch.

Definition at line 1976 of file backends.py.

    def _add_batch_directives(self, job, batch_file):
        """
        For HTCondor leave empty as the directives are already included in the submit file.
        """
        print('#!/bin/bash', file=batch_file)
 

_add_setup()

_add_setup ( job, batch_file )

staticprotectedinherited

Adds setup lines to the shell script file.

Definition at line 807 of file backends.py.

    def _add_setup(job, batch_file):
        """
        Adds setup lines to the shell script file.
        """
        for line in job.setup_cmds:
            print(line, file=batch_file)
 

_add_wrapper_script_setup()

_add_wrapper_script_setup ( self, job, batch_file )

protectedinherited

Adds lines to the submitted script that help with job monitoring/setup. Mostly here so that we can insert
`trap` statements for Ctrl-C situations.

Definition at line 814 of file backends.py.

    def _add_wrapper_script_setup(self, job, batch_file):
        """
        Adds lines to the submitted script that help with job monitoring/setup. Mostly here so that we can insert
        `trap` statements for Ctrl-C situations.
        """
        start_wrapper = f"""# ---
# trap ctrl-c and call ctrl_c()
trap '(ctrl_c 130)' SIGINT
trap '(ctrl_c 143)' SIGTERM
 
function write_exit_code() {{
  echo "Writing $1 to exit status file"
  echo "$1" > {self.exit_code_file}
  exit $1
}}
 
function ctrl_c() {{
  trap '' SIGINT SIGTERM
  echo "** Trapped Ctrl-C **"
  echo "$1" > {self.exit_code_file}
  exit $1
}}
# ---"""
        print(start_wrapper, file=batch_file)
 

_add_wrapper_script_teardown()

_add_wrapper_script_teardown ( self, job, batch_file )

protectedinherited

Adds lines to the submitted script that help with job monitoring/teardown. Mostly here so that we can insert
an exit code of the job cmd being written out to a file. Which means that we can know if the command was
successful or not even if the backend server/monitoring database purges the data about our job i.e. If PBS
removes job information too quickly we may never know if a job succeeded or failed without some kind of exit
file.

Definition at line 839 of file backends.py.

    def _add_wrapper_script_teardown(self, job, batch_file):
        """
        Adds lines to the submitted script that help with job monitoring/teardown. Mostly here so that we can insert
        an exit code of the job cmd being written out to a file. Which means that we can know if the command was
        successful or not even if the backend server/monitoring database purges the data about our job i.e. If PBS
        removes job information too quickly we may never know if a job succeeded or failed without some kind of exit
        file.
        """
        end_wrapper = """# ---
write_exit_code $?"""
        print(end_wrapper, file=batch_file)
 

_create_cmd()

_create_cmd ( self, script_path )

protected

Reimplemented from Batch.

Definition at line 1982 of file backends.py.

    def _create_cmd(self, script_path):
        """
        """
        submission_cmd = self.submission_cmds[:]
        submission_cmd.append(script_path.as_posix())
        return submission_cmd
 

_create_job_result()

_create_job_result ( cls, job, job_id )

protected

Reimplemented from Batch.

Definition at line 2126 of file backends.py.

    def _create_job_result(cls, job, job_id):
        """
        """
        B2INFO(f"Job ID of {job} recorded as: {job_id}")
        job.result = cls.HTCondorResult(job, job_id)
 

_create_parent_job_result()

_create_parent_job_result ( cls, parent )

protected

We want to be able to call `ready()` on the top level `Job.result`. So this method needs to exist
so that a Job.result object actually exists. It will be mostly empty and simply updates subjob
statuses and allows the use of ready().

Reimplemented from Backend.

Definition at line 2133 of file backends.py.

    def _create_parent_job_result(cls, parent):
        parent.result = cls.HTCondorResult(parent, None)
 

_make_submit_file()

_make_submit_file ( self, job, submit_file_path )

protected

Fill HTCondor submission file.

Reimplemented from Batch.

Definition at line 1950 of file backends.py.

    def _make_submit_file(self, job, submit_file_path):
        """
        Fill HTCondor submission file.
        """
        # Find all files/directories in the working directory to copy on the worker node
 
        files_to_transfer = [i.as_posix() for i in job.working_dir.iterdir()]
 
        job_backend_args = {**self.backend_args, **job.backend_args}  # Merge the two dictionaries, with the job having priority
 
        with open(submit_file_path, "w") as submit_file:
            print(f'executable = {self.get_submit_script_path(job)}', file=submit_file)
            print(f'log = {Path(job.output_dir, "htcondor.log").as_posix()}', file=submit_file)
            print(f'output = {Path(job.working_dir, _STDOUT_FILE).as_posix()}', file=submit_file)
            print(f'error = {Path(job.working_dir, _STDERR_FILE).as_posix()}', file=submit_file)
            print('transfer_input_files = ', ','.join(files_to_transfer), file=submit_file)
            print(f'universe = {job_backend_args["universe"]}', file=submit_file)
            print(f'getenv = {job_backend_args["getenv"]}', file=submit_file)
            print(f'request_memory = {job_backend_args["request_memory"]}', file=submit_file)
            print('should_transfer_files = Yes', file=submit_file)
            print('when_to_transfer_output = ON_EXIT', file=submit_file)
            # Any other lines in the backend args that we don't deal with explicitly but maybe someone wants to insert something
            for line in job_backend_args["extra_lines"]:
                print(line, file=submit_file)
            print('queue', file=submit_file)
 

_submit_to_batch()

_submit_to_batch ( cls, cmd )

protected

Do the actual batch submission command and collect the output to find out the job id for later monitoring.

Reimplemented from Batch.

Definition at line 1996 of file backends.py.

    def _submit_to_batch(cls, cmd):
        """
        Do the actual batch submission command and collect the output to find out the job id for later monitoring.
        """
        job_dir = Path(cmd[-1]).parent.as_posix()
        sub_out = ""
        attempt = 0
        sleep_time = 30
 
        while attempt < 3:
            try:
                sub_out = subprocess.check_output(cmd, stderr=subprocess.STDOUT, universal_newlines=True, cwd=job_dir)
                break
            except subprocess.CalledProcessError as e:
                attempt += 1
                if attempt == 3:
                    B2ERROR(f"Error during condor_submit: {str(e)} occurred more than 3 times.")
                    raise e
                else:
                    B2ERROR(f"Error during condor_submit: {str(e)}, sleeping for {sleep_time} seconds.")
                    time.sleep(30)
        return re.search(r"(\d+\.\d+) - \d+\.\d+", sub_out).groups()[0]
 

can_submit()

can_submit	(		self,
			njobs = 1 )

Checks the global number of jobs in HTCondor right now (submitted or running) for this user.
Returns True if the number is lower that the limit, False if it is higher.

Parameters:
    njobs (int): The number of jobs that we want to submit before checking again. Lets us check if we
        are sufficiently below the limit in order to (somewhat) safely submit. It is slightly dangerous to
        assume that it is safe to submit too many jobs since there might be other processes also submitting jobs.
        So njobs really shouldn't be abused when you might be getting close to the limit i.e. keep it <=250
        and check again before submitting more.

Reimplemented from Batch.

Definition at line 2136 of file backends.py.

    def can_submit(self, njobs=1):
        """
        Checks the global number of jobs in HTCondor right now (submitted or running) for this user.
        Returns True if the number is lower that the limit, False if it is higher.
 
        Parameters:
            njobs (int): The number of jobs that we want to submit before checking again. Lets us check if we
                are sufficiently below the limit in order to (somewhat) safely submit. It is slightly dangerous to
                assume that it is safe to submit too many jobs since there might be other processes also submitting jobs.
                So njobs really shouldn't be abused when you might be getting close to the limit i.e. keep it <=250
                and check again before submitting more.
        """
        B2DEBUG(29, "Calling HTCondor().can_submit()")
        jobs_info = self.condor_q()
        total_jobs = jobs_info["NJOBS"]
        B2INFO(f"Total jobs active in the HTCondor system is currently {total_jobs}")
        if (total_jobs + njobs) > self.global_job_limit:
            B2INFO(f"Since the global limit is {self.global_job_limit} we cannot submit {njobs} jobs until some complete.")
            return False
        else:
            B2INFO("There is enough space to submit more jobs.")
            return True
 

condor_history()

condor_history	(		cls,
			class_ads = None,
			job_id = "",
			username = "" )

Simplistic interface to the ``condor_history`` command. lets you request information about all jobs matching the filters
``job_id`` and ``username``. Note that setting job_id negates username so it is ignored.
The result is a JSON dictionary filled by output of the ``-json`` ``condor_history`` option.

Parameters:
    class_ads (list[str]): A list of condor_history ClassAds that you would like information about.
        By default we give {cls.default_class_ads}, increasing the amount of class_ads increase the time taken
        by the condor_q call.
    job_id (str): String representation of the Job ID given by condor_submit during submission.
        If this argument is given then the output of this function will be only information about this job.
        If this argument is not given, then all jobs matching the other filters will be returned.
    username (str): By default we return information about only the current user's jobs. By giving
        a username you can access the job information of a specific user's jobs. By giving ``username='all'`` you will
        receive job information from all known user jobs matching the other filters. This is limited to 10000 records
        and isn't recommended.

Returns:
    dict: JSON dictionary of the form:

    .. code-block:: python

      {
        "NJOBS":<number of records returned by command>,
        "JOBS":[
                {
                 <ClassAd: value>, ...
                }, ...
               ]
      }

Definition at line 2227 of file backends.py.

    def condor_history(cls, class_ads=None, job_id="", username=""):
        """
        Simplistic interface to the ``condor_history`` command. lets you request information about all jobs matching the filters
        ``job_id`` and ``username``. Note that setting job_id negates username so it is ignored.
        The result is a JSON dictionary filled by output of the ``-json`` ``condor_history`` option.
 
        Parameters:
            class_ads (list[str]): A list of condor_history ClassAds that you would like information about.
                By default we give {cls.default_class_ads}, increasing the amount of class_ads increase the time taken
                by the condor_q call.
            job_id (str): String representation of the Job ID given by condor_submit during submission.
                If this argument is given then the output of this function will be only information about this job.
                If this argument is not given, then all jobs matching the other filters will be returned.
            username (str): By default we return information about only the current user's jobs. By giving
                a username you can access the job information of a specific user's jobs. By giving ``username='all'`` you will
                receive job information from all known user jobs matching the other filters. This is limited to 10000 records
                and isn't recommended.
 
        Returns:
            dict: JSON dictionary of the form:
 
            .. code-block:: python
 
              {
                "NJOBS":<number of records returned by command>,
                "JOBS":[
                        {
                         <ClassAd: value>, ...
                        }, ...
                       ]
              }
        """
        B2DEBUG(29, f"Calling HTCondor.condor_history(class_ads={class_ads}, job_id={job_id}, username={username})")
        if not class_ads:
            class_ads = cls.default_class_ads
        # Output fields should be comma separated.
        field_list_cmd = ",".join(class_ads)
        cmd_list = ["condor_history", "-json", "-attributes", field_list_cmd]
        # If job_id is set then we ignore all other filters
        if job_id:
            cmd_list.append(job_id)
        else:
            if not username:
                username = os.environ["USER"]
            # If the username is set to all it is a special case
            if username != "all":
                cmd_list.append(username)
        # We get a JSON serialisable summary from condor_q. But we will alter it slightly to be more similar to other backends
        cmd = " ".join(cmd_list)
        B2DEBUG(29, f"Calling subprocess with command = '{cmd}'")
        try:
            records = subprocess.check_output(cmd, stderr=subprocess.STDOUT, universal_newlines=True, shell=True)
        except BaseException:
            records = None
 
        if records:
            records = decode_json_string(records)
        else:
            records = []
 
        jobs_info = {"JOBS": records}
        jobs_info["NJOBS"] = len(jobs_info["JOBS"])  # Just to avoid having to len() it in the future
        return jobs_info
 
 

condor_q()

condor_q	(		cls,
			class_ads = None,
			job_id = "",
			username = "" )

Simplistic interface to the `condor_q` command. lets you request information about all jobs matching the filters
'job_id' and 'username'. Note that setting job_id negates username so it is ignored.
The result is the JSON dictionary returned by output of the ``-json`` condor_q option.

Parameters:
    class_ads (list[str]): A list of condor_q ClassAds that you would like information about.
        By default we give {cls.default_class_ads}, increasing the amount of class_ads increase the time taken
        by the condor_q call.
    job_id (str): String representation of the Job ID given by condor_submit during submission.
        If this argument is given then the output of this function will be only information about this job.
        If this argument is not given, then all jobs matching the other filters will be returned.
    username (str): By default we return information about only the current user's jobs. By giving
        a username you can access the job information of a specific user's jobs. By giving ``username='all'`` you will
        receive job information from all known user jobs matching the other filters. This may be a LOT of jobs
        so it isn't recommended.

Returns:
    dict: JSON dictionary of the form:

    .. code-block:: python

      {
        "NJOBS":<number of records returned by command>,
        "JOBS":[
                {
                 <ClassAd: value>, ...
                }, ...
               ]
      }

Definition at line 2160 of file backends.py.

    def condor_q(cls, class_ads=None, job_id="", username=""):
        """
        Simplistic interface to the `condor_q` command. lets you request information about all jobs matching the filters
        'job_id' and 'username'. Note that setting job_id negates username so it is ignored.
        The result is the JSON dictionary returned by output of the ``-json`` condor_q option.
 
        Parameters:
            class_ads (list[str]): A list of condor_q ClassAds that you would like information about.
                By default we give {cls.default_class_ads}, increasing the amount of class_ads increase the time taken
                by the condor_q call.
            job_id (str): String representation of the Job ID given by condor_submit during submission.
                If this argument is given then the output of this function will be only information about this job.
                If this argument is not given, then all jobs matching the other filters will be returned.
            username (str): By default we return information about only the current user's jobs. By giving
                a username you can access the job information of a specific user's jobs. By giving ``username='all'`` you will
                receive job information from all known user jobs matching the other filters. This may be a LOT of jobs
                so it isn't recommended.
 
        Returns:
            dict: JSON dictionary of the form:
 
            .. code-block:: python
 
              {
                "NJOBS":<number of records returned by command>,
                "JOBS":[
                        {
                         <ClassAd: value>, ...
                        }, ...
                       ]
              }
        """
        B2DEBUG(29, f"Calling HTCondor.condor_q(class_ads={class_ads}, job_id={job_id}, username={username})")
        if not class_ads:
            class_ads = cls.default_class_ads
        # Output fields should be comma separated.
        field_list_cmd = ",".join(class_ads)
        cmd_list = ["condor_q", "-json", "-attributes", field_list_cmd]
        # If job_id is set then we ignore all other filters
        if job_id:
            cmd_list.append(job_id)
        else:
            if not username:
                username = os.environ["USER"]
            # If the username is set to all it is a special case
            if username == "all":
                cmd_list.append("-allusers")
            else:
                cmd_list.append(username)
        # We get a JSON serialisable summary from condor_q. But we will alter it slightly to be more similar to other backends
        cmd = " ".join(cmd_list)
        B2DEBUG(29, f"Calling subprocess with command = '{cmd}'")
        # condor_q occasionally fails
        try:
            records = subprocess.check_output(cmd, stderr=subprocess.STDOUT, universal_newlines=True, shell=True)
        except BaseException:
            records = None
 
        if records:
            records = decode_json_string(records)
        else:
            records = []
        jobs_info = {"JOBS": records}
        jobs_info["NJOBS"] = len(jobs_info["JOBS"])  # Just to avoid having to len() it in the future
        return jobs_info
 

get_batch_submit_script_path()

get_batch_submit_script_path	(		self,
			job )

Construct the Path object of the .sub file that we will use to describe the job.

Reimplemented from Batch.

Definition at line 1989 of file backends.py.

    def get_batch_submit_script_path(self, job):
        """
        Construct the Path object of the .sub file that we will use to describe the job.
        """
        return Path(job.working_dir, self.batch_submit_script)
 

get_submit_script_path()

get_submit_script_path ( self, job )

inherited

Construct the Path object of the bash script file that we will submit. It will contain
the actual job command, wrapper commands, setup commands, and any batch directives

Definition at line 860 of file backends.py.

    def get_submit_script_path(self, job):
        """
        Construct the Path object of the bash script file that we will submit. It will contain
        the actual job command, wrapper commands, setup commands, and any batch directives
        """
        return Path(job.working_dir, self.submit_script)
 
 

submit()

submit ( self, job, check_can_submit = True, jobs_per_check = 100 )

inherited

Reimplemented from Backend.

Definition at line 1205 of file backends.py.

    def submit(self, job, check_can_submit=True, jobs_per_check=100):
        """
        """
        raise NotImplementedError("This is an abstract submit(job) method that shouldn't have been called. "
                                  "Did you submit a (Sub)Job?")
 

Member Data Documentation

backend_args

dict backend_args = {**self.default_backend_args, **backend_args}

inherited

The backend args that will be applied to jobs unless the job specifies them itself.

Definition at line 797 of file backends.py.

batch_submit_script

batch_submit_script = "submit.sub"

static

HTCondor batch script (different to the wrapper script of Backend.submit_script)

Definition at line 1934 of file backends.py.

default_backend_args

dict default_backend_args = {}

staticinherited

Default backend_args.

Definition at line 789 of file backends.py.

default_class_ads

list default_class_ads = ["GlobalJobId", "JobStatus", "Owner"]

static

Default ClassAd attributes to return from commands like condor_q.

Definition at line 1948 of file backends.py.

default_global_job_limit

int default_global_job_limit = 1000

staticinherited

Default global limit on the total number of submitted/running jobs that the user can have.

This limit will not affect the total number of jobs that are eventually submitted. But the jobs won't actually be submitted until this limit can be respected i.e. until the number of total jobs in the Batch system goes down. Since we actually submit in chunks of N jobs, before checking this limit value again, this value needs to be a little lower than the real batch system limit. Otherwise you could accidentally go over during the N job submission if other processes are checking and submitting concurrently. This is quite common for the first submission of jobs from parallel calibrations.

Note that if there are other jobs already submitted for your account, then these will count towards this limit.

Definition at line 1156 of file backends.py.

default_sleep_between_submission_checks

int default_sleep_between_submission_checks = 30

staticinherited

Default time betweeon re-checking if the active jobs is below the global job limit.

Definition at line 1158 of file backends.py.

exit_code_file

str exit_code_file = "__BACKEND_CMD_EXIT_STATUS__"

staticinherited

Default exit code file name.

Definition at line 787 of file backends.py.

global_job_limit

int global_job_limit = self.default_global_job_limit

inherited

The active job limit.

This is 'global' because we want to prevent us accidentally submitting too many jobs from all current and previous submission scripts.

Definition at line 1167 of file backends.py.

sleep_between_submission_checks

int sleep_between_submission_checks = self.default_sleep_between_submission_checks

inherited

Seconds we wait before checking if we can submit a list of jobs.

Only relevant once we hit the global limit of active jobs, which is a lot usually.

Definition at line 1170 of file backends.py.

submission_cmds

list submission_cmds = []

staticinherited

Shell command to submit a script, should be implemented in the derived class.

Definition at line 1143 of file backends.py.

submit_script

submit_script = "submit.sh"

staticinherited

Default submission script name.

Definition at line 785 of file backends.py.

---

The documentation for this class was generated from the following file:

• calibration/scripts/caf/backends.py