3 # Copyright (C) 2010-2013 CEA/DEN
5 # This library is free software; you can redistribute it and/or
6 # modify it under the terms of the GNU Lesser General Public
7 # License as published by the Free Software Foundation; either
8 # version 2.1 of the License.
10 # This library is distributed in the hope that it will be useful,
11 # but WITHOUT ANY WARRANTY; without even the implied warranty of
12 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 # Lesser General Public License for more details.
15 # You should have received a copy of the GNU Lesser General Public
16 # License along with this library; if not, write to the Free Software
17 # Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
29 STYLESHEET_GLOBAL = "jobs_global_report.xsl"
30 STYLESHEET_BOARD = "jobs_board_report.xsl"
35 parser = src.options.Options()
37 parser.add_option('n', 'name', 'string', 'jobs_cfg',
38 _('The name of the config file that contains'
39 ' the jobs configuration'))
40 parser.add_option('o', 'only_jobs', 'list2', 'only_jobs',
41 _('Optional: the list of jobs to launch, by their name. '))
42 parser.add_option('l', 'list', 'boolean', 'list',
43 _('Optional: list all available config files.'))
44 parser.add_option('t', 'test_connection', 'boolean', 'test_connection',
45 _("Optional: try to connect to the machines. "
46 "Not executing the jobs."),
48 parser.add_option('p', 'publish', 'boolean', 'publish',
49 _("Optional: generate an xml file that can be read in a "
50 "browser to display the jobs status."),
52 parser.add_option('i', 'input_boards', 'string', 'input_boards', _("Optional: "
53 "the path to csv file that contain "
54 "the expected boards."),"")
55 parser.add_option('n', 'completion', 'boolean', 'no_label',
56 _("Optional (internal use): do not print labels, Works only "
60 class Machine(object):
61 '''Class to manage a ssh connection on a machine
69 sat_path="salomeTools"):
73 self.distribution = None # Will be filled after copying SAT on the machine
75 self.password = passwd
76 self.sat_path = sat_path
77 self.ssh = paramiko.SSHClient()
78 self._connection_successful = None
80 def connect(self, logger):
81 '''Initiate the ssh connection to the remote machine
83 :param logger src.logger.Logger: The logger instance
88 self._connection_successful = False
89 self.ssh.load_system_host_keys()
90 self.ssh.set_missing_host_key_policy(paramiko.AutoAddPolicy())
92 self.ssh.connect(self.host,
95 password = self.password)
96 except paramiko.AuthenticationException:
97 message = src.KO_STATUS + _("Authentication failed")
98 except paramiko.BadHostKeyException:
99 message = (src.KO_STATUS +
100 _("The server's host key could not be verified"))
101 except paramiko.SSHException:
102 message = ( _("SSHException error connecting or "
103 "establishing an SSH session"))
105 message = ( _("Error connecting or establishing an SSH session"))
107 self._connection_successful = True
111 def successfully_connected(self, logger):
112 '''Verify if the connection to the remote machine has succeed
114 :param logger src.logger.Logger: The logger instance
115 :return: True if the connection has succeed, False if not
118 if self._connection_successful == None:
119 message = _("Warning : trying to ask if the connection to "
120 "(name: %s host: %s, port: %s, user: %s) is OK whereas there were"
121 " no connection request" %
122 (self.name, self.host, self.port, self.user))
123 logger.write( src.printcolors.printcWarning(message))
124 return self._connection_successful
126 def copy_sat(self, sat_local_path, job_file):
127 '''Copy salomeTools to the remote machine in self.sat_path
131 # open a sftp connection
132 self.sftp = self.ssh.open_sftp()
133 # Create the sat directory on remote machine if it is not existing
134 self.mkdir(self.sat_path, ignore_existing=True)
136 self.put_dir(sat_local_path, self.sat_path, filters = ['.git'])
137 # put the job configuration file in order to make it reachable
138 # on the remote machine
139 self.sftp.put(job_file, os.path.join(".salomeTools",
141 ".jobs_command_file.pyconf"))
142 except Exception as e:
144 self._connection_successful = False
148 def put_dir(self, source, target, filters = []):
149 ''' Uploads the contents of the source directory to the target path. The
150 target directory needs to exists. All sub-directories in source are
151 created under target.
153 for item in os.listdir(source):
156 source_path = os.path.join(source, item)
157 destination_path = os.path.join(target, item)
158 if os.path.islink(source_path):
159 linkto = os.readlink(source_path)
161 self.sftp.symlink(linkto, destination_path)
162 self.sftp.chmod(destination_path,
163 os.stat(source_path).st_mode)
167 if os.path.isfile(source_path):
168 self.sftp.put(source_path, destination_path)
169 self.sftp.chmod(destination_path,
170 os.stat(source_path).st_mode)
172 self.mkdir(destination_path, ignore_existing=True)
173 self.put_dir(source_path, destination_path)
175 def mkdir(self, path, mode=511, ignore_existing=False):
176 ''' Augments mkdir by adding an option to not fail
180 self.sftp.mkdir(path, mode)
187 def exec_command(self, command, logger):
188 '''Execute the command on the remote machine
190 :param command str: The command to be run
191 :param logger src.logger.Logger: The logger instance
192 :return: the stdin, stdout, and stderr of the executing command,
194 :rtype: (paramiko.channel.ChannelFile, paramiko.channel.ChannelFile,
195 paramiko.channel.ChannelFile)
198 # Does not wait the end of the command
199 (stdin, stdout, stderr) = self.ssh.exec_command(command)
200 except paramiko.SSHException:
201 message = src.KO_STATUS + _(
202 ": the server failed to execute the command\n")
203 logger.write( src.printcolors.printcError(message))
204 return (None, None, None)
206 logger.write( src.printcolors.printcError(src.KO_STATUS + '\n'))
207 return (None, None, None)
209 return (stdin, stdout, stderr)
212 '''Close the ssh connection
218 def write_info(self, logger):
219 '''Prints the informations relative to the machine in the logger
220 (terminal traces and log file)
222 :param logger src.logger.Logger: The logger instance
226 logger.write("host : " + self.host + "\n")
227 logger.write("port : " + str(self.port) + "\n")
228 logger.write("user : " + str(self.user) + "\n")
229 if self.successfully_connected(logger):
230 status = src.OK_STATUS
232 status = src.KO_STATUS
233 logger.write("Connection : " + status + "\n\n")
237 '''Class to manage one job
239 def __init__(self, name, machine, application, board,
240 commands, timeout, config, logger, after=None):
243 self.machine = machine
245 self.timeout = timeout
246 self.application = application
250 # The list of log files to download from the remote machine
251 self.remote_log_files = []
253 # The remote command status
254 # -1 means that it has not been launched,
255 # 0 means success and 1 means fail
257 self.cancelled = False
261 self._has_begun = False
262 self._has_finished = False
263 self._has_timouted = False
264 self._stdin = None # Store the command inputs field
265 self._stdout = None # Store the command outputs field
266 self._stderr = None # Store the command errors field
271 self.commands = commands
272 self.command = (os.path.join(self.machine.sat_path, "sat") +
274 os.path.join(self.machine.sat_path,
275 "list_log_files.txt") +
276 " job --jobs_config .jobs_command_file" +
281 """ Get the pid(s) corresponding to the command that have been launched
282 On the remote machine
284 :return: The list of integers corresponding to the found pids
288 cmd_pid = 'ps aux | grep "' + self.command + '" | awk \'{print $2}\''
289 (_, out_pid, _) = self.machine.exec_command(cmd_pid, self.logger)
290 pids_cmd = out_pid.readlines()
291 pids_cmd = [str(src.only_numbers(pid)) for pid in pids_cmd]
295 def kill_remote_process(self, wait=1):
296 '''Kills the process on the remote machine.
298 :return: (the output of the kill, the error of the kill)
302 pids = self.get_pids()
303 cmd_kill = " ; ".join([("kill -2 " + pid) for pid in pids])
304 (_, out_kill, err_kill) = self.machine.exec_command(cmd_kill,
307 return (out_kill, err_kill)
310 '''Returns True if the job has already begun
312 :return: True if the job has already begun
315 return self._has_begun
317 def has_finished(self):
318 '''Returns True if the job has already finished
319 (i.e. all the commands have been executed)
320 If it is finished, the outputs are stored in the fields out and err.
322 :return: True if the job has already finished
326 # If the method has already been called and returned True
327 if self._has_finished:
330 # If the job has not begun yet
331 if not self.has_begun():
334 if self._stdout.channel.closed:
335 self._has_finished = True
336 # Store the result outputs
337 self.out += self._stdout.read().decode()
338 self.err += self._stderr.read().decode()
340 self._Tf = time.time()
341 # And get the remote command status and log files
344 return self._has_finished
346 def get_log_files(self):
347 """Get the log files produced by the command launched
348 on the remote machine, and put it in the log directory of the user,
349 so they can be accessible from
351 # Do not get the files if the command is not finished
352 if not self.has_finished():
353 msg = _("Trying to get log files whereas the job is not finished.")
354 self.logger.write(src.printcolors.printcWarning(msg))
357 # First get the file that contains the list of log files to get
358 tmp_file_path = src.get_tmp_filename(self.config, "list_log_files.txt")
359 remote_path = os.path.join(self.machine.sat_path, "list_log_files.txt")
360 self.machine.sftp.get(
364 # Read the file and get the result of the command and all the log files
366 fstream_tmp = open(tmp_file_path, "r")
367 file_lines = fstream_tmp.readlines()
368 file_lines = [line.replace("\n", "") for line in file_lines]
370 os.remove(tmp_file_path)
373 # The first line is the result of the command (0 success or 1 fail)
374 self.res_job = file_lines[0]
375 except Exception as e:
376 self.err += _("Unable to get status from remote file %s: %s" %
377 (remote_path, str(e)))
379 for i, job_path_remote in enumerate(file_lines[1:]):
381 # For each command, there is two files to get :
382 # 1- The xml file describing the command and giving the
384 # 2- The txt file containing the system command traces (like
385 # traces produced by the "make" command)
386 # 3- In case of the test command, there is another file to get :
387 # the xml board that contain the test results
388 dirname = os.path.basename(os.path.dirname(job_path_remote))
389 if dirname != 'OUT' and dirname != 'TEST':
391 local_path = os.path.join(os.path.dirname(
392 self.logger.logFilePath),
393 os.path.basename(job_path_remote))
394 if i==0: # The first is the job command
395 self.logger.add_link(os.path.basename(job_path_remote),
399 elif dirname == 'OUT':
401 local_path = os.path.join(os.path.dirname(
402 self.logger.logFilePath),
404 os.path.basename(job_path_remote))
405 elif dirname == 'TEST':
407 local_path = os.path.join(os.path.dirname(
408 self.logger.logFilePath),
410 os.path.basename(job_path_remote))
413 if not os.path.exists(local_path):
414 self.machine.sftp.get(job_path_remote, local_path)
415 self.remote_log_files.append(local_path)
416 except Exception as e:
417 self.err += _("Unable to get %s log file from remote: %s" %
418 (str(job_path_remote),
421 def has_failed(self):
422 '''Returns True if the job has failed.
423 A job is considered as failed if the machine could not be reached,
424 if the remote command failed,
425 or if the job finished with a time out.
427 :return: True if the job has failed
430 if not self.has_finished():
432 if not self.machine.successfully_connected(self.logger):
434 if self.is_timeout():
436 if self.res_job == "1":
441 """In case of a failing job, one has to cancel every job that depend
442 on it. This method put the job as failed and will not be executed.
444 self._has_begun = True
445 self._has_finished = True
446 self.cancelled = True
447 self.out += _("This job was not launched because its father has failed.")
448 self.err += _("This job was not launched because its father has failed.")
450 def is_running(self):
451 '''Returns True if the job commands are running
453 :return: True if the job is running
456 return self.has_begun() and not self.has_finished()
458 def is_timeout(self):
459 '''Returns True if the job commands has finished with timeout
461 :return: True if the job has finished with timeout
464 return self._has_timouted
466 def time_elapsed(self):
467 """Get the time elapsed since the job launching
469 :return: The number of seconds
472 if not self.has_begun():
475 return T_now - self._T0
477 def check_time(self):
478 """Verify that the job has not exceeded its timeout.
479 If it has, kill the remote command and consider the job as finished.
481 if not self.has_begun():
483 if self.time_elapsed() > self.timeout:
484 self._has_finished = True
485 self._has_timouted = True
486 self._Tf = time.time()
488 (out_kill, _) = self.kill_remote_process()
489 self.out += "TIMEOUT \n" + out_kill.read().decode()
490 self.err += "TIMEOUT : %s seconds elapsed\n" % str(self.timeout)
493 except Exception as e:
494 self.err += _("Unable to get remote log files: %s" % e)
496 def total_duration(self):
497 """Give the total duration of the job
499 :return: the total duration of the job in seconds
502 return self._Tf - self._T0
505 """Launch the job by executing the remote command.
508 # Prevent multiple run
510 msg = _("Warning: A job can only be launched one time")
511 msg2 = _("Trying to launch the job \"%s\" whereas it has "
512 "already been launched." % self.name)
513 self.logger.write(src.printcolors.printcWarning("%s\n%s\n" % (msg,
517 # Do not execute the command if the machine could not be reached
518 if not self.machine.successfully_connected(self.logger):
519 self._has_finished = True
521 self.err += ("Connection to machine (name : %s, host: %s, port:"
522 " %s, user: %s) has failed\nUse the log command "
523 "to get more information."
524 % (self.machine.name,
529 # Usual case : Launch the command on remote machine
530 self._T0 = time.time()
531 self._stdin, self._stdout, self._stderr = self.machine.exec_command(
534 # If the results are not initialized, finish the job
535 if (self._stdin, self._stdout, self._stderr) == (None, None, None):
536 self._has_finished = True
537 self._Tf = time.time()
539 self.err += "The server failed to execute the command"
541 # Put the beginning flag to true.
542 self._has_begun = True
544 def write_results(self):
545 """Display on the terminal all the job's information
547 self.logger.write("name : " + self.name + "\n")
549 self.logger.write("after : %s\n" % self.after)
550 self.logger.write("Time elapsed : %4imin %2is \n" %
551 (self.total_duration()//60 , self.total_duration()%60))
553 self.logger.write("Begin time : %s\n" %
554 time.strftime('%Y-%m-%d %H:%M:%S',
555 time.localtime(self._T0)) )
557 self.logger.write("End time : %s\n\n" %
558 time.strftime('%Y-%m-%d %H:%M:%S',
559 time.localtime(self._Tf)) )
561 machine_head = "Informations about connection :\n"
562 underline = (len(machine_head) - 2) * "-"
563 self.logger.write(src.printcolors.printcInfo(
564 machine_head+underline+"\n"))
565 self.machine.write_info(self.logger)
567 self.logger.write(src.printcolors.printcInfo("out : \n"))
569 self.logger.write("Unable to get output\n")
571 self.logger.write(self.out + "\n")
572 self.logger.write(src.printcolors.printcInfo("err : \n"))
573 self.logger.write(self.err + "\n")
575 def get_status(self):
576 """Get the status of the job (used by the Gui for xml display)
578 :return: The current status of the job
581 if not self.machine.successfully_connected(self.logger):
582 return "SSH connection KO"
583 if not self.has_begun():
584 return "Not launched"
587 if self.is_running():
588 return "running since " + time.strftime('%Y-%m-%d %H:%M:%S',
589 time.localtime(self._T0))
590 if self.has_finished():
591 if self.is_timeout():
592 return "Timeout since " + time.strftime('%Y-%m-%d %H:%M:%S',
593 time.localtime(self._Tf))
594 return "Finished since " + time.strftime('%Y-%m-%d %H:%M:%S',
595 time.localtime(self._Tf))
598 '''Class to manage the jobs to be run
605 lenght_columns = 20):
606 # The jobs configuration
607 self.cfg_jobs = config_jobs
608 self.job_file_path = job_file_path
609 # The machine that will be used today
611 # The list of machine (hosts, port) that will be used today
612 # (a same host can have several machine instances since there
613 # can be several ssh parameters)
615 # The jobs to be launched today
617 # The jobs that will not be launched today
618 self.ljobs_not_today = []
621 self.len_columns = lenght_columns
623 # the list of jobs that have not been run yet
624 self._l_jobs_not_started = []
625 # the list of jobs that have already ran
626 self._l_jobs_finished = []
627 # the list of jobs that are running
628 self._l_jobs_running = []
630 self.determine_jobs_and_machines()
632 def define_job(self, job_def, machine):
633 '''Takes a pyconf job definition and a machine (from class machine)
634 and returns the job instance corresponding to the definition.
636 :param job_def src.config.Mapping: a job definition
637 :param machine machine: the machine on which the job will run
638 :return: The corresponding job in a job class instance
642 cmmnds = job_def.commands
643 if not "timeout" in job_def:
644 timeout = 4*60*60 # default timeout = 4h
646 timeout = job_def.timeout
648 if 'after' in job_def:
649 after = job_def.after
651 if 'application' in job_def:
652 application = job_def.application
654 if 'board' in job_def:
655 board = job_def.board
667 def determine_jobs_and_machines(self):
668 '''Function that reads the pyconf jobs definition and instantiates all
669 the machines and jobs to be done today.
674 today = datetime.date.weekday(datetime.date.today())
677 for job_def in self.cfg_jobs.jobs :
679 if not "machine" in job_def:
680 msg = _('WARNING: The job "%s" do not have the key '
681 '"machine", this job is ignored.\n\n' % job_def.name)
682 self.logger.write(src.printcolors.printcWarning(msg))
684 name_machine = job_def.machine
687 for mach in self.lmachines:
688 if mach.name == name_machine:
692 if a_machine == None:
693 for machine_def in self.cfg_jobs.machines:
694 if machine_def.name == name_machine:
695 if 'host' not in machine_def:
696 host = self.runner.cfg.VARS.hostname
698 host = machine_def.host
700 if 'user' not in machine_def:
701 user = self.runner.cfg.VARS.user
703 user = machine_def.user
705 if 'port' not in machine_def:
708 port = machine_def.port
710 if 'password' not in machine_def:
713 passwd = machine_def.password
715 if 'sat_path' not in machine_def:
716 sat_path = "salomeTools"
718 sat_path = machine_def.sat_path
729 self.lmachines.append(a_machine)
730 if (host, port) not in host_list:
731 host_list.append((host, port))
733 if a_machine == None:
734 msg = _("WARNING: The job \"%(job_name)s\" requires the "
735 "machine \"%(machine_name)s\" but this machine "
736 "is not defined in the configuration file.\n"
737 "The job will not be launched")
738 self.logger.write(src.printcolors.printcWarning(msg))
740 a_job = self.define_job(job_def, a_machine)
742 if today in job_def.when:
743 self.ljobs.append(a_job)
744 else: # today in job_def.when
745 self.ljobs_not_today.append(a_job)
747 self.lhosts = host_list
749 def ssh_connection_all_machines(self, pad=50):
750 '''Function that do the ssh connection to every machine
756 self.logger.write(src.printcolors.printcInfo((
757 "Establishing connection with all the machines :\n")))
758 for machine in self.lmachines:
759 # little algorithm in order to display traces
760 begin_line = (_("Connection to %s: " % machine.name))
761 if pad - len(begin_line) < 0:
764 endline = (pad - len(begin_line)) * "." + " "
766 step = "SSH connection"
767 self.logger.write( begin_line + endline + step)
769 # the call to the method that initiate the ssh connection
770 msg = machine.connect(self.logger)
772 # Copy salomeTools to the remote machine
773 if machine.successfully_connected(self.logger):
775 self.logger.write('\r%s%s%s' % (begin_line, endline, 20 * " "),3)
776 self.logger.write('\r%s%s%s' % (begin_line, endline, step), 3)
778 res_copy = machine.copy_sat(self.runner.cfg.VARS.salometoolsway,
780 # get the remote machine distribution using a sat command
781 (__, out_dist, __) = machine.exec_command(
782 os.path.join(machine.sat_path,
783 "sat config --value VARS.dist --no_label"),
785 machine.distribution = out_dist.read().decode().replace("\n",
787 # Print the status of the copy
789 self.logger.write('\r%s' %
790 ((len(begin_line)+len(endline)+20) * " "), 3)
791 self.logger.write('\r%s%s%s' %
794 src.printcolors.printc(src.OK_STATUS)), 3)
796 self.logger.write('\r%s' %
797 ((len(begin_line)+len(endline)+20) * " "), 3)
798 self.logger.write('\r%s%s%s %s' %
801 src.printcolors.printc(src.OK_STATUS),
802 _("Copy of SAT failed")), 3)
804 self.logger.write('\r%s' %
805 ((len(begin_line)+len(endline)+20) * " "), 3)
806 self.logger.write('\r%s%s%s %s' %
809 src.printcolors.printc(src.KO_STATUS),
811 self.logger.write("\n", 3)
813 self.logger.write("\n")
816 def is_occupied(self, hostname):
817 '''Function that returns True if a job is running on
818 the machine defined by its host and its port.
820 :param hostname (str, int): the pair (host, port)
821 :return: the job that is running on the host,
822 or false if there is no job running on the host.
827 for jb in self.ljobs:
828 if jb.machine.host == host and jb.machine.port == port:
833 def update_jobs_states_list(self):
834 '''Function that updates the lists that store the currently
835 running jobs and the jobs that have already finished.
840 jobs_finished_list = []
841 jobs_running_list = []
842 for jb in self.ljobs:
844 jobs_running_list.append(jb)
846 if jb.has_finished():
847 jobs_finished_list.append(jb)
849 nb_job_finished_before = len(self._l_jobs_finished)
850 self._l_jobs_finished = jobs_finished_list
851 self._l_jobs_running = jobs_running_list
853 nb_job_finished_now = len(self._l_jobs_finished)
855 return nb_job_finished_now > nb_job_finished_before
857 def cancel_dependencies_of_failing_jobs(self):
858 '''Function that cancels all the jobs that depend on a failing one.
864 for job in self.ljobs:
865 if job.after is None:
867 father_job = self.find_job_that_has_name(job.after)
868 if father_job is not None and father_job.has_failed():
871 def find_job_that_has_name(self, name):
872 '''Returns the job by its name.
874 :param name str: a job name
875 :return: the job that has the name.
878 for jb in self.ljobs:
881 # the following is executed only if the job was not found
884 def str_of_length(self, text, length):
885 '''Takes a string text of any length and returns
886 the most close string of length "length".
888 :param text str: any string
889 :param length int: a length for the returned string
890 :return: the most close string of length "length"
893 if len(text) > length:
894 text_out = text[:length-3] + '...'
896 diff = length - len(text)
897 before = " " * (diff//2)
898 after = " " * (diff//2 + diff%2)
899 text_out = before + text + after
903 def display_status(self, len_col):
904 '''Takes a lenght and construct the display of the current status
905 of the jobs in an array that has a column for each host.
906 It displays the job that is currently running on the host
909 :param len_col int: the size of the column
915 for host_port in self.lhosts:
916 jb = self.is_occupied(host_port)
917 if not jb: # nothing running on the host
918 empty = self.str_of_length("empty", len_col)
919 display_line += "|" + empty
921 display_line += "|" + src.printcolors.printcInfo(
922 self.str_of_length(jb.name, len_col))
924 self.logger.write("\r" + display_line + "|")
929 '''The main method. Runs all the jobs on every host.
930 For each host, at a given time, only one job can be running.
931 The jobs that have the field after (that contain the job that has
932 to be run before it) are run after the previous job.
933 This method stops when all the jobs are finished.
940 self.logger.write(src.printcolors.printcInfo(
941 _('Executing the jobs :\n')))
943 for host_port in self.lhosts:
946 if port == 22: # default value
947 text_line += "|" + self.str_of_length(host, self.len_columns)
949 text_line += "|" + self.str_of_length(
950 "("+host+", "+str(port)+")", self.len_columns)
952 tiret_line = " " + "-"*(len(text_line)-1) + "\n"
953 self.logger.write(tiret_line)
954 self.logger.write(text_line + "|\n")
955 self.logger.write(tiret_line)
958 # The infinite loop that runs the jobs
959 l_jobs_not_started = src.deepcopy_list(self.ljobs)
960 while len(self._l_jobs_finished) != len(self.ljobs):
961 new_job_start = False
962 for host_port in self.lhosts:
964 if self.is_occupied(host_port):
967 for jb in l_jobs_not_started:
968 if (jb.machine.host, jb.machine.port) != host_port:
972 l_jobs_not_started.remove(jb)
976 jb_before = self.find_job_that_has_name(jb.after)
977 if jb_before is None:
979 msg = _("This job was not launched because its "
980 "father is not in the jobs list.")
984 if jb_before.has_finished():
986 l_jobs_not_started.remove(jb)
989 self.cancel_dependencies_of_failing_jobs()
990 new_job_finished = self.update_jobs_states_list()
992 if new_job_start or new_job_finished:
994 self.gui.update_xml_files(self.ljobs)
995 # Display the current status
996 self.display_status(self.len_columns)
998 # Make sure that the proc is not entirely busy
1001 self.logger.write("\n")
1002 self.logger.write(tiret_line)
1003 self.logger.write("\n\n")
1006 self.gui.update_xml_files(self.ljobs)
1007 self.gui.last_update()
1009 def write_all_results(self):
1010 '''Display all the jobs outputs.
1016 for jb in self.ljobs:
1017 self.logger.write(src.printcolors.printcLabel(
1018 "#------- Results for job %s -------#\n" % jb.name))
1020 self.logger.write("\n\n")
1023 '''Class to manage the the xml data that can be displayed in a browser to
1027 def __init__(self, xml_dir_path, l_jobs, l_jobs_not_today, file_boards=""):
1030 :param xml_dir_path str: The path to the directory where to put
1031 the xml resulting files
1032 :param l_jobs List: the list of jobs that run today
1033 :param l_jobs_not_today List: the list of jobs that do not run today
1034 :param file_boards str: the file path from which to read the
1037 # The path of the csv files to read to fill the expected boards
1038 self.file_boards = file_boards
1040 today = datetime.date.weekday(datetime.date.today())
1041 self.parse_csv_boards(today)
1043 # The path of the global xml file
1044 self.xml_dir_path = xml_dir_path
1045 # Initialize the xml files
1046 xml_global_path = os.path.join(self.xml_dir_path, "global_report.xml")
1047 self.xml_global_file = src.xmlManager.XmlLogFile(xml_global_path,
1049 # The xml files that corresponds to the boards.
1050 # {name_board : xml_object}}
1051 self.d_xml_board_files = {}
1052 # Create the lines and columns
1053 self.initialize_boards(l_jobs, l_jobs_not_today)
1055 # Write the xml file
1056 self.update_xml_files(l_jobs)
1058 def add_xml_board(self, name):
1059 xml_board_path = os.path.join(self.xml_dir_path, name + ".xml")
1060 self.d_xml_board_files[name] = src.xmlManager.XmlLogFile(
1063 self.d_xml_board_files[name].add_simple_node("distributions")
1064 self.d_xml_board_files[name].add_simple_node("applications")
1065 self.d_xml_board_files[name].add_simple_node("board", text=name)
1067 def initialize_boards(self, l_jobs, l_jobs_not_today):
1068 '''Get all the first information needed for each file and write the
1069 first version of the files
1070 :param l_jobs List: the list of jobs that run today
1071 :param l_jobs_not_today List: the list of jobs that do not run today
1073 # Get the boards to fill and put it in a dictionary
1074 # {board_name : xml instance corresponding to the board}
1075 for job in l_jobs + l_jobs_not_today:
1077 if (board is not None and
1078 board not in self.d_xml_board_files.keys()):
1079 self.add_xml_board(board)
1081 # Verify that the boards given as input are done
1082 for board in list(self.d_input_boards.keys()):
1083 if board not in self.d_xml_board_files:
1084 self.add_xml_board(board)
1085 root_node = self.d_xml_board_files[board].xmlroot
1086 src.xmlManager.append_node_attrib(root_node,
1087 {"input_file" : self.file_boards})
1089 # Loop over all jobs in order to get the lines and columns for each
1093 for board in self.d_xml_board_files:
1095 d_application[board] = []
1099 for job in l_jobs + l_jobs_not_today:
1101 if (job.machine.host, job.machine.port) not in l_hosts_ports:
1102 l_hosts_ports.append((job.machine.host, job.machine.port))
1104 distrib = job.machine.distribution
1105 application = job.application
1107 board_job = job.board
1110 for board in self.d_xml_board_files:
1111 if board_job == board:
1112 if distrib is not None and distrib not in d_dist[board]:
1113 d_dist[board].append(distrib)
1114 src.xmlManager.add_simple_node(
1115 self.d_xml_board_files[board].xmlroot.find(
1118 attrib={"name" : distrib})
1120 if board_job == board:
1121 if (application is not None and
1122 application not in d_application[board]):
1123 d_application[board].append(application)
1124 src.xmlManager.add_simple_node(
1125 self.d_xml_board_files[board].xmlroot.find(
1129 "name" : application})
1131 # Verify that there are no missing application or distribution in the
1132 # xml board files (regarding the input boards)
1133 for board in self.d_xml_board_files:
1134 l_dist = d_dist[board]
1135 if board not in self.d_input_boards.keys():
1137 for dist in self.d_input_boards[board]["rows"]:
1138 if dist not in l_dist:
1139 src.xmlManager.add_simple_node(
1140 self.d_xml_board_files[board].xmlroot.find(
1143 attrib={"name" : dist})
1144 l_appli = d_application[board]
1145 for appli in self.d_input_boards[board]["columns"]:
1146 if appli not in l_appli:
1147 src.xmlManager.add_simple_node(
1148 self.d_xml_board_files[board].xmlroot.find(
1151 attrib={"name" : appli})
1153 # Initialize the hosts_ports node for the global file
1154 self.xmlhosts_ports = self.xml_global_file.add_simple_node(
1156 for host, port in l_hosts_ports:
1157 host_port = "%s:%i" % (host, port)
1158 src.xmlManager.add_simple_node(self.xmlhosts_ports,
1160 attrib={"name" : host_port})
1162 # Initialize the jobs node in all files
1163 for xml_file in [self.xml_global_file] + list(
1164 self.d_xml_board_files.values()):
1165 xml_jobs = xml_file.add_simple_node("jobs")
1166 # Get the jobs present in the config file but
1167 # that will not be launched today
1168 self.put_jobs_not_today(l_jobs_not_today, xml_jobs)
1170 xml_file.add_simple_node("infos",
1171 attrib={"name" : "last update",
1172 "JobsCommandStatus" : "running"})
1174 # Find in each board the squares that needs to be filled regarding the
1175 # input csv files but that are not covered by a today job
1176 for board in self.d_input_boards.keys():
1177 xml_root_board = self.d_xml_board_files[board].xmlroot
1178 xml_missing = src.xmlManager.add_simple_node(xml_root_board,
1180 for row, column in self.d_input_boards[board]["jobs"]:
1183 if (job.application == column and
1184 job.machine.distribution == row):
1188 src.xmlManager.add_simple_node(xml_missing,
1190 attrib={"distribution" : row,
1191 "application" : column })
1193 def put_jobs_not_today(self, l_jobs_not_today, xml_node_jobs):
1194 '''Get all the first information needed for each file and write the
1195 first version of the files
1197 :param xml_node_jobs etree.Element: the node corresponding to a job
1198 :param l_jobs_not_today List: the list of jobs that do not run today
1200 for job in l_jobs_not_today:
1201 xmlj = src.xmlManager.add_simple_node(xml_node_jobs,
1203 attrib={"name" : job.name})
1204 src.xmlManager.add_simple_node(xmlj, "application", job.application)
1205 src.xmlManager.add_simple_node(xmlj,
1207 job.machine.distribution)
1208 src.xmlManager.add_simple_node(xmlj, "board", job.board)
1209 src.xmlManager.add_simple_node(xmlj,
1210 "commands", " ; ".join(job.commands))
1211 src.xmlManager.add_simple_node(xmlj, "state", "Not today")
1212 src.xmlManager.add_simple_node(xmlj, "machine", job.machine.name)
1213 src.xmlManager.add_simple_node(xmlj, "host", job.machine.host)
1214 src.xmlManager.add_simple_node(xmlj, "port", str(job.machine.port))
1215 src.xmlManager.add_simple_node(xmlj, "user", job.machine.user)
1216 src.xmlManager.add_simple_node(xmlj, "sat_path",
1217 job.machine.sat_path)
1219 def parse_csv_boards(self, today):
1220 """ Parse the csv file that describes the boards to produce and fill
1221 the dict d_input_boards that contain the csv file contain
1223 :param today int: the current day of the week
1225 # open the csv file and read its content
1227 with open(self.file_boards, 'r') as f:
1228 reader = csv.reader(f,delimiter=CSV_DELIMITER)
1231 # get the delimiter for the boards (empty line)
1232 boards_delimiter = [''] * len(l_read[0])
1233 # Make the list of boards, by splitting with the delimiter
1234 l_boards = [list(y) for x, y in itertools.groupby(l_read,
1235 lambda z: z == boards_delimiter) if not x]
1237 # loop over the csv lists of lines and get the rows, columns and jobs
1239 for input_board in l_boards:
1241 board_name = input_board[0][0]
1244 columns = input_board[0][1:]
1248 for line in input_board[1:]:
1250 for i, square in enumerate(line[1:]):
1253 days = square.split(DAYS_SEPARATOR)
1254 days = [int(day) for day in days]
1258 job = (row, columns[i])
1261 d_boards[board_name] = {"rows" : rows,
1262 "columns" : columns,
1265 self.d_input_boards = d_boards
1267 def update_xml_files(self, l_jobs):
1268 '''Write all the xml files with updated information about the jobs
1270 :param l_jobs List: the list of jobs that run today
1272 for xml_file in [self.xml_global_file] + list(
1273 self.d_xml_board_files.values()):
1274 self.update_xml_file(l_jobs, xml_file)
1277 self.write_xml_files()
1279 def update_xml_file(self, l_jobs, xml_file):
1280 '''update information about the jobs for the file xml_file
1282 :param l_jobs List: the list of jobs that run today
1283 :param xml_file xmlManager.XmlLogFile: the xml instance to update
1286 xml_node_jobs = xml_file.xmlroot.find('jobs')
1287 # Update the job names and status node
1289 # Find the node corresponding to the job and delete it
1290 # in order to recreate it
1291 for xmljob in xml_node_jobs.findall('job'):
1292 if xmljob.attrib['name'] == job.name:
1293 xml_node_jobs.remove(xmljob)
1297 T0 = time.strftime('%Y-%m-%d %H:%M:%S',
1298 time.localtime(job._T0))
1301 Tf = time.strftime('%Y-%m-%d %H:%M:%S',
1302 time.localtime(job._Tf))
1304 # recreate the job node
1305 xmlj = src.xmlManager.add_simple_node(xml_node_jobs,
1307 attrib={"name" : job.name})
1308 src.xmlManager.add_simple_node(xmlj, "machine", job.machine.name)
1309 src.xmlManager.add_simple_node(xmlj, "host", job.machine.host)
1310 src.xmlManager.add_simple_node(xmlj, "port", str(job.machine.port))
1311 src.xmlManager.add_simple_node(xmlj, "user", job.machine.user)
1312 src.xmlManager.add_simple_node(xmlj, "sat_path",
1313 job.machine.sat_path)
1314 src.xmlManager.add_simple_node(xmlj, "application", job.application)
1315 src.xmlManager.add_simple_node(xmlj, "distribution",
1316 job.machine.distribution)
1317 src.xmlManager.add_simple_node(xmlj, "board", job.board)
1318 src.xmlManager.add_simple_node(xmlj, "timeout", str(job.timeout))
1319 src.xmlManager.add_simple_node(xmlj, "commands",
1320 " ; ".join(job.commands))
1321 src.xmlManager.add_simple_node(xmlj, "state", job.get_status())
1322 src.xmlManager.add_simple_node(xmlj, "begin", T0)
1323 src.xmlManager.add_simple_node(xmlj, "end", Tf)
1324 src.xmlManager.add_simple_node(xmlj, "out",
1325 src.printcolors.cleancolor(job.out))
1326 src.xmlManager.add_simple_node(xmlj, "err",
1327 src.printcolors.cleancolor(job.err))
1328 src.xmlManager.add_simple_node(xmlj, "res", str(job.res_job))
1329 if len(job.remote_log_files) > 0:
1330 src.xmlManager.add_simple_node(xmlj,
1331 "remote_log_file_path",
1332 job.remote_log_files[0])
1334 src.xmlManager.add_simple_node(xmlj,
1335 "remote_log_file_path",
1338 xmlafter = src.xmlManager.add_simple_node(xmlj, "after", job.after)
1339 # get the job father
1340 if job.after is not None:
1343 if jb.name == job.after:
1346 if (job_father is not None and
1347 len(job_father.remote_log_files) > 0):
1348 link = job_father.remote_log_files[0]
1351 src.xmlManager.append_node_attrib(xmlafter, {"link" : link})
1353 # Verify that the job is to be done today regarding the input csv
1355 if job.board and job.board in self.d_input_boards.keys():
1357 for dist, appli in self.d_input_boards[job.board]["jobs"]:
1358 if (job.machine.distribution == dist
1359 and job.application == appli):
1361 src.xmlManager.add_simple_node(xmlj,
1366 src.xmlManager.add_simple_node(xmlj,
1372 xml_node_infos = xml_file.xmlroot.find('infos')
1373 src.xmlManager.append_node_attrib(xml_node_infos,
1375 datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S")})
1379 def last_update(self, finish_status = "finished"):
1380 '''update information about the jobs for the file xml_file
1382 :param l_jobs List: the list of jobs that run today
1383 :param xml_file xmlManager.XmlLogFile: the xml instance to update
1385 for xml_file in [self.xml_global_file] + list(self.d_xml_board_files.values()):
1386 xml_node_infos = xml_file.xmlroot.find('infos')
1387 src.xmlManager.append_node_attrib(xml_node_infos,
1388 attrib={"JobsCommandStatus" : finish_status})
1390 self.write_xml_files()
1392 def write_xml_files(self):
1393 ''' Write the xml files
1395 self.xml_global_file.write_tree(STYLESHEET_GLOBAL)
1396 for xml_file in self.d_xml_board_files.values():
1397 xml_file.write_tree(STYLESHEET_BOARD)
1400 # Describes the command
1402 return _("The jobs command launches maintenances that are described"
1403 " in the dedicated jobs configuration file.")
1407 def run(args, runner, logger):
1409 (options, args) = parser.parse_args(args)
1411 l_cfg_dir = runner.cfg.PATHS.JOBPATH
1413 # list option : display all the available config files
1415 for cfg_dir in l_cfg_dir:
1416 if not options.no_label:
1417 logger.write("------ %s\n" %
1418 src.printcolors.printcHeader(cfg_dir))
1420 for f in sorted(os.listdir(cfg_dir)):
1421 if not f.endswith('.pyconf'):
1424 logger.write("%s\n" % cfilename)
1427 # Make sure the jobs_config option has been called
1428 if not options.jobs_cfg:
1429 message = _("The option --jobs_config is required\n")
1430 src.printcolors.printcError(message)
1433 # Find the file in the directories
1435 for cfg_dir in l_cfg_dir:
1436 file_jobs_cfg = os.path.join(cfg_dir, options.jobs_cfg)
1437 if not file_jobs_cfg.endswith('.pyconf'):
1438 file_jobs_cfg += '.pyconf'
1440 if not os.path.exists(file_jobs_cfg):
1447 msg = _("The file configuration %(name_file)s was not found."
1448 "\nUse the --list option to get the possible files.")
1449 src.printcolors.printcError(msg)
1453 (_("Platform"), runner.cfg.VARS.dist),
1454 (_("File containing the jobs configuration"), file_jobs_cfg)
1456 src.print_info(logger, info)
1458 # Read the config that is in the file
1459 config_jobs = src.read_config_from_a_file(file_jobs_cfg)
1460 if options.only_jobs:
1461 l_jb = src.pyconf.Sequence()
1462 for jb in config_jobs.jobs:
1463 if jb.name in options.only_jobs:
1465 "Adding a job that was given in only_jobs option parameters")
1466 config_jobs.jobs = l_jb
1469 today_jobs = Jobs(runner,
1473 # SSH connection to all machines
1474 today_jobs.ssh_connection_all_machines()
1475 if options.test_connection:
1480 # Copy the stylesheets in the log directory
1481 log_dir = runner.cfg.USER.log_dir
1482 xsl_dir = os.path.join(runner.cfg.VARS.srcDir, 'xsl')
1484 files_to_copy.append(os.path.join(xsl_dir, STYLESHEET_GLOBAL))
1485 files_to_copy.append(os.path.join(xsl_dir, STYLESHEET_BOARD))
1486 files_to_copy.append(os.path.join(xsl_dir, "running.gif"))
1487 for file_path in files_to_copy:
1488 shutil.copy2(file_path, log_dir)
1490 # Instanciate the Gui in order to produce the xml files that contain all
1492 gui = Gui(runner.cfg.USER.log_dir,
1494 today_jobs.ljobs_not_today,
1495 file_boards = options.input_boards)
1497 # Display the list of the xml files
1498 logger.write(src.printcolors.printcInfo(("Here is the list of published"
1500 logger.write("%s\n" % gui.xml_global_file.logFile, 4)
1501 for board in gui.d_xml_board_files.keys():
1502 file_path = gui.d_xml_board_files[board].logFile
1503 file_name = os.path.basename(file_path)
1504 logger.write("%s\n" % file_path, 4)
1505 logger.add_link(file_name, "board", 0, board)
1507 logger.write("\n", 4)
1509 today_jobs.gui = gui
1513 # Run all the jobs contained in config_jobs
1514 today_jobs.run_jobs()
1515 except KeyboardInterrupt:
1517 logger.write("\n\n%s\n\n" %
1518 (src.printcolors.printcWarning(_("Forced interruption"))), 1)
1521 msg = _("Killing the running jobs and trying"
1522 " to get the corresponding logs\n")
1523 logger.write(src.printcolors.printcWarning(msg))
1525 # find the potential not finished jobs and kill them
1526 for jb in today_jobs.ljobs:
1527 if not jb.has_finished():
1529 jb.kill_remote_process()
1530 except Exception as e:
1531 msg = _("Failed to kill job %s: %s\n" % (jb.name, e))
1532 logger.write(src.printcolors.printcWarning(msg))
1535 today_jobs.gui.last_update(_("Forced interruption"))
1538 today_jobs.gui.last_update()
1539 # Output the results
1540 today_jobs.write_all_results()