#!/usr/bin/python # -*- coding: utf-8 -*- #--------------------------------------------------------------------------------# # This file is part of the PALM model system. # # PALM is free software: you can redistribute it and/or modify it under the terms # of the GNU General Public License as published by the Free Software Foundation, # either version 3 of the License, or (at your option) any later version. # # PALM is distributed in the hope that it will be useful, but WITHOUT ANY # WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR # A PARTICULAR PURPOSE. See the GNU General Public License for more details. # # You should have received a copy of the GNU General Public License along with # PALM. If not, see . # # Copyright 1997-2018 Leibniz Universitaet Hannover #--------------------------------------------------------------------------------# # # Current revisions: # ----------------- # # # Former revisions: # ----------------- # $Id: palm_wd 2718 2018-01-02 08:49:38Z schwenkel $ # Corrected "Former revisions" section # # 2696 2017-12-14 17:12:51Z kanani # Change in file header (GPL part) # # 2416 2017-09-06 14:28:14Z maronga # Adapted for palmrun # # 1754 2016-02-22 13:50:22Z maronga # # 1753 2016-02-22 13:49:49Z maronga # Bugfix: use of global variables is required after updating configuration file # # 1751 2016-02-15 07:44:16Z maronga # Bugfixes: runs on multiple hosts caused crash of the watchdog. Progress bar # occasionally showed wrong progress. # New: Hosts can be switched on/off and update_frequency can be modified via # File->Options. Security check for cancelation of jobs. # # 1618 2015-07-13 06:52:15Z maronga # Added steering via configuration file .wd.config, to be place in the local # palm directory. Watchdog save files are automatically created upon first start. # # 1613 2015-07-08 14:53:29Z maronga # Bugfix: tooltip for queuing name did not show up on first update. # New: added contect menu for showing the parameter file and the run control # output # # 1611 2015-07-07 12:23:22Z maronga # Initial revision # # # Description: # ------------ # PALM watchdog client for monitoring batch jobs on a variety of hosts specified # by the user # # Instructions: # ------------- # 1) Modify the header section of palm_wd # 2) Move .wd.olddata and .wd.newdata to your palm directory # (e.g. /home/user/current_version/.wd.newdata etc.) # 3) Modify a copy of palm_wdd for each host to be monitored and move it to the # respective hosts # 4) Start the client either from mrungui or from shell by "nohup palm_wd&" # To do: # ------ # 1) Add "Options", "Help" and "Manual" # 2) Move user settings to a configuration file #------------------------------------------------------------------------------! import ConfigParser import datetime import os from PyQt4 import QtGui, QtCore, uic from PyQt4.QtCore import pyqtSlot,SIGNAL,SLOT import shutil import subprocess as sub import sys import time # Determine PALM directories try: devnull = open(os.devnull, 'w') out = sub.check_output("echo $PALM_BIN", shell=True, stderr=sub.STDOUT) palm_bin = out.rstrip() palm_dir = out.split("palm")[0] + "palm/" + out.split("palm")[1].split("/")[1] out = None except: print "Error. $PALM_BIN is not set." raise SystemExit # Read configuration file # First check if the configuration file exists if ( os.path.exists(palm_dir + '/.wd.config') == False ): print "Error. No configuration file .wd.config found in " + palm_dir raise SystemExit config = ConfigParser.RawConfigParser() config.read(palm_dir + '/.wd.config') description = [] hostname = [] username = [] for i in range(0,len(config.sections())): description_tmp = config.sections()[i] if ( description_tmp != 'Settings' ): if ( config.get(description_tmp, 'enabled') == 'true' ): description.append(description_tmp) hostname.append(config.get(description_tmp, 'hostname')) username.append(config.get(description_tmp, 'username')) else: update_frequency = int(config.get(description_tmp, 'update_frequency'))*60000 # Check if .wd.olddata and .wd.newdata exist. Otherwise create the files if ( os.path.exists(palm_dir + '/.wd.olddata') == False ): print "No .wd.olddata found. Creating..." file1 = open(palm_dir + '/.wd.olddata', 'a') file1.close() if ( os.path.exists(palm_dir + '/.wd.newdata') == False ): print "No .wd.newdata found. Creating..." file1 = open(palm_dir + '/.wd.newdata', 'a') file1.close() # Dummy variables for the jobname and the progressbars jobname = "" timestamp = "" pbars = [] job_data_list = [] # Message box for showing RUN_CONTROL output class MessageBoxScroll(QtGui.QMessageBox): def __init__(self): QtGui.QMessageBox.__init__(self) self.setSizeGripEnabled(True) def event(self, e): result = QtGui.QMessageBox.event(self, e) self.setMinimumHeight(100) self.setMaximumHeight(16777215) self.setMinimumWidth(400) self.setMaximumWidth(16777215) self.setSizePolicy(QtGui.QSizePolicy.Expanding, QtGui.QSizePolicy.Expanding) textEdit = self.findChild(QtGui.QTextEdit) if textEdit != None : textEdit.setMinimumHeight(800) textEdit.setMaximumHeight(16777215) textEdit.setMinimumWidth(1000) textEdit.setMaximumWidth(16777215) textEdit.setSizePolicy(QtGui.QSizePolicy.Expanding, QtGui.QSizePolicy.Expanding) f = QtGui.QFont('not_a_font') # f.setStyleHint(f.TypeWriter, f.PreferDefault) textEdit.setFont(f) return result # Message box for showing RUN_CONTROL output class OptionBox(QtGui.QDialog): def __init__(self): super(OptionBox, self).__init__() uic.loadUi(palm_bin + '/palm_wd_files/wdoptions.ui', self) hostname = [] self.checkbox = [] j = -1 ypos = 0 self.vbox = QtGui.QVBoxLayout(self) self.vbox.setSpacing(0) self.vbox.setMargin(0) config.read(palm_dir + '/.wd.config') for i in range(0,len(config.sections())): description_tmp = config.sections()[i] if ( description_tmp != 'Settings' ): description.append(description_tmp) hostname.append(config.get(description_tmp, 'hostname')) j = j + 1 self.checkbox.append(j) self.checkbox[j] = QtGui.QCheckBox(description_tmp, self) ypos = ypos + 20 self.checkbox[j].move(10,0) if ( config.get(description_tmp, 'enabled') == 'true' ): self.checkbox[j].toggle() self.vbox.addWidget(self.checkbox[j]) else: self.update_spin.setValue(int(config.get(description_tmp, 'update_frequency'))) self.hostBox.setLayout(self.vbox) self.hostBox.setGeometry(0, 0, 272, 50+ypos) self.settingBox.setGeometry(281, 0, 214, 50+ypos) self.ok.setGeometry(385, 50+ypos, 111, 24) self.cancel.setGeometry(270, 50+ypos, 111, 24) self.setGeometry(0, 0, 500, 80+ypos) self.show() return # save updtes to file def SaveAndClose(self): for i in range(0,len(self.checkbox)): if ( self.checkbox[i].checkState() == 0 ): config.set(str(self.checkbox[i].text()),'enabled','false') else: config.set(str(self.checkbox[i].text()),'enabled','true') config.set('Settings','update_frequency',self.update_spin.text()) with open(palm_dir + '/.wd.config', 'wb') as configfile: config.write(configfile) self.close() return # MainWindow class class Watchdog(QtGui.QMainWindow): def __init__(self): super(Watchdog, self).__init__() self.InitUi() # Initialized MainWindow UI def InitUi(self): # Load predefined mainwindow uic.loadUi(palm_bin + '/palm_wd_files/wd.ui', self) # Resize columns and set number of rows to zero. Column 6 is hidden and # contains the remote jobname (e.g. hannover.174610) self.table.setColumnWidth(0,230) self.table.setColumnWidth(1,80) self.table.setColumnWidth(2,50) self.table.setColumnWidth(3,80) self.table.setColumnWidth(4,180) self.table.setColumnWidth(5,115) self.table.setColumnHidden(6, True) self.table.setRowCount(0) # Display MainWindow self.show() QtGui.QApplication.processEvents() # Start refresh timer. On timeout perform update self.timer = QtCore.QTimer(self) self.timer.timeout.connect(self.Refresh) self.timer.setSingleShot(False) self.timer.start(update_frequency) # The timetimer counts the time since last update self.timetimer= QtCore.QElapsedTimer() self.timetimer.start() # The labeltimer induces the update of the remaining time in the UI self.labeltimer = QtCore.QTimer(self) self.labeltimer.timeout.connect(self.UpdateLabel) self.labeltimer.setSingleShot(False) # Update in the UI will be performed at each 1/10 of the update interval self.labeltimer.start(update_frequency/10) self.label.setText("Next update in " + str(update_frequency/1000/60) + "min") # Update the list now self.Refresh() # Add a custom context menu def OpenContextMenu(self, position): menu = QtGui.QMenu() # "Show details" -> fetch details for the selected job detailsjobAction = menu.addAction('Show job details') detailsjobAction.setStatusTip('Display detailed info for job') detailsjobAction.triggered.connect(self.ShowDetails) # "Show parameter file" -> show the contents of PARIN parinAction = menu.addAction('Show parameter file') parinAction.setStatusTip('Display the parameter file of the job (e.g. PARIN / _p3d)') parinAction.triggered.connect(self.ShowPARIN) rcAction = menu.addAction('Show run control file') rcAction.setStatusTip('Display the current run control file (e.g. RUN_CONTROL / _rc)') rcAction.triggered.connect(self.ShowRC) # "Cancel job" -> remove job from queuing system canceljobAction = menu.addAction('Cancel job') canceljobAction.setStatusTip('Remove job from queueing system') canceljobAction.triggered.connect(self.CancelJob) # "Force stop" -> force termination of job stopjobAction = menu.addAction('Force stop') stopjobAction.setStatusTip('Terminate job properly') stopjobAction.triggered.connect(self.DoStop) # "Force restart" -> force rermination and restart of job restartjobAction = menu.addAction('Force restart') restartjobAction.setStatusTip('Terminate job properly and initiate restart') restartjobAction.triggered.connect(self.DoRestart) # "Remove from list" -> remove a completed job from the list removefromlistAction = menu.addAction('Remove from list') removefromlistAction.setStatusTip('Remove finished job') removefromlistAction.triggered.connect(lambda: self.RemoveFromList(-1)) # Activate/deactive contect menu items based on the status of the runs state = self.table.item(self.table.currentRow(),3).text() if (state == "Canceled" or state == "Completed" or state == "Terminated"): detailsjobAction.setEnabled(False) canceljobAction.setEnabled(False) removefromlistAction.setEnabled(True) else: detailsjobAction.setEnabled(True) canceljobAction.setEnabled(True) removefromlistAction.setEnabled(False) if ( state == "Running" ): stopjobAction.setEnabled(True) restartjobAction.setEnabled(True) parinAction.setEnabled(True) rcAction.setEnabled(True) else: stopjobAction.setEnabled(False) restartjobAction.setEnabled(False) parinAction.setEnabled(False) rcAction.setEnabled(False) # Show the context menu action = menu.exec_(self.table.mapToGlobal(position)) # Collecting watchdog data and display in the UI def Refresh(self): # Use global variables global pbars global update_frequency # Deactivate timer processes until update is finished # QtGui.QApplication.processEvents() updates the UI whenever needed self.labeltimer.stop() self.label.setText("Updating...") QtGui.QApplication.processEvents() self.setEnabled(False) QtGui.QApplication.processEvents() # Get current timestamp timestamp = datetime.datetime.now().strftime("%d/%m/%Y %H:%M") # Open an empty file for the watchdog data file = open(palm_dir + '/.wd.newdata', 'w') # Set all required variables to their initial values. Sorting must be # disabled. self.table.setRowCount(0) self.table.setSortingEnabled(False) i = -1 job_data_list = [] pbars = [] # Scan all hosts in the variable hostname. For each host perform the # update loop for h in range(0,len(hostname)): # Perform ssh command host = username[h] + "@" + hostname[h] ssh = sub.Popen(["ssh", "%s" % host, "/sw/tools/python/2.7.6/generic/bin/python palm_wdd queue " + username[h]], shell=False, stdout=sub.PIPE, stderr=sub.PIPE) result = ssh.stdout.readlines() result = filter(None, result) # Delete all job data job_data_tmp = [] job_data = [] job_data_n = [] job_progress = 0 # In case of errors display error message if ( len(result) < 1 ): error = ssh.stderr.readlines() if ( error != [] ): notify = QtGui.QMessageBox.warning(self,'Collect data',"Error. An error occured during read of job data for user " + username[h] +" for host " + description[h] + ".\n\nError message:\n" + ''.join(error)) # No error -> save job data else: # Successively write to job_data for j in range(0,len(result)): job_data_tmp.append(j) job_data_tmp[j] = result[j].split(" ") job_data_tmp[j] = filter(None, job_data_tmp[j]) job_data.append(j) job_data[j] = [job_data_tmp[j][0], description[h], job_data_tmp[j][2], job_data_tmp[j][3], int(float(job_data_tmp[j][4])*100), job_data_tmp[j][5], job_data_tmp[j][1], job_data_tmp[j][6].rstrip()] job_data_n.append(j) job_data_n[j] = job_data_tmp[j][1] del(result) # Now read the data from the last update from file. These data are # already in the prefered format file2 = open(palm_dir + '/.wd.olddata', 'r') result = file2.readlines() file2.close() job_data_old = [] job_data_old_n = [] k = -1 for j in range(0,len(result)): if ( result[j].split()[1] == description[h] ): k = k + 1 job_data_old.append(k) job_data_old[k] = result[j].split(" ") job_data_old[k] = filter(None, job_data_old[k]) job_data_old[k] = [line.rstrip('\n') for line in job_data_old[k]] job_data_old_n.append(k) job_data_old_n[k] = job_data_old[k][6] # Merge the job_data and job_data_old to find completed, new and # still queued jobs if ( len(job_data_n) > 0 and len(job_data_old_n) > 0 ): jobs_full = list(set(job_data_old_n) | set(job_data_n)) jobs_known = list(set(job_data_old_n) & set(job_data_n)) jobs_complete = set(job_data_old_n) - set(job_data_n) jobs_new = set(job_data_n) - set(job_data_old_n) elif ( len(job_data_n) > 0 ): jobs_full = job_data_n jobs_known = [] jobs_new = job_data_n jobs_complete = [] elif ( len(job_data_old_n) > 0 ): jobs_full = job_data_old_n jobs_known = [] jobs_new = [] jobs_complete = job_data_old_n else: jobs_full = [] jobs_known = [] jobs_new = [] jobs_complete = [] # Display all known jobs to_display = [list(job_data_n).index(item) for item in list(jobs_known) if list(jobs_known) and list(job_data_n)] for j in range(0,len(to_display)): i = i + 1 self.table.insertRow(i) item = QtGui.QTableWidgetItem(job_data[to_display[j]][0]) item.setToolTip("Queuing name: " + job_data[to_display[j]][6]) self.table.setItem(i, 0, item) self.table.setItem(i, 1, QtGui.QTableWidgetItem(job_data[to_display[j]][1])) self.table.setItem(i, 2, QtGui.QTableWidgetItem(job_data[to_display[j]][2])) self.table.setItem(i, 3, QtGui.QTableWidgetItem(job_data[to_display[j]][3])) item = QtGui.QTableWidgetItem(job_data[to_display[j]][5]) item.setToolTip("Estimated job start: " + job_data[to_display[j]][7]) self.table.setItem(i, 5, item) self.table.setItem(i, 6, QtGui.QTableWidgetItem(job_data[to_display[j]][6])) self.table.item(i,2).setTextAlignment(QtCore.Qt.AlignRight) self.table.item(i,5).setTextAlignment(QtCore.Qt.AlignRight) pbars.append(i) pbars[i] = QtGui.QProgressBar(self) pbars[i].setValue(int(job_data[to_display[j]][4])) # Apply a color depending on the status of the job if ( job_data[to_display[j]][3] == "Running" ): self.table.setCellWidget(i,4,pbars[i]) self.table.item(i,0).setBackground(QtGui.QColor(255, 251, 168)) self.table.item(i,1).setBackground(QtGui.QColor(255, 251, 168)) self.table.item(i,2).setBackground(QtGui.QColor(255, 251, 168)) self.table.item(i,3).setBackground(QtGui.QColor(255, 251, 168)) self.table.item(i,5).setBackground(QtGui.QColor(255, 251, 168)) self.table.item(i,6).setBackground(QtGui.QColor(255, 251, 168)) else: pbars[j].setEnabled(False) self.table.setCellWidget(i,4,pbars[i]) self.table.item(i,0).setBackground(QtGui.QColor(255, 112, 118)) self.table.item(i,1).setBackground(QtGui.QColor(255, 112, 118)) self.table.item(i,2).setBackground(QtGui.QColor(255, 112, 118)) self.table.item(i,3).setBackground(QtGui.QColor(255, 112, 118)) self.table.item(i,5).setBackground(QtGui.QColor(255, 112, 118)) self.table.item(i,6).setBackground(QtGui.QColor(255, 112, 118)) # Save the job data to file job_data_list.append(i) job_data_list[i] = job_data[to_display[j]][0] printstring = str(job_data[to_display[j]][0]) + " " + \ str(job_data[to_display[j]][1]) + " " + \ str(job_data[to_display[j]][2]) + " " + \ str(job_data[to_display[j]][3]) + " " + \ str(job_data[to_display[j]][4]) + " " + \ str(job_data[to_display[j]][5]) + " " + \ str(job_data[to_display[j]][6]) + " " + \ str(job_data[to_display[j]][7]) + "\n" file.write(printstring) # Display all new jobs to_display = [list(job_data_n).index(item) for item in list(jobs_new) if list(jobs_new) and list(job_data_n)] for j in range(0,len(to_display)): i = i + 1 self.table.insertRow(i) item = QtGui.QTableWidgetItem(job_data[to_display[j]][0]) item.setToolTip("Queuing name: " + job_data[to_display[j]][6]) self.table.setItem(i, 0, item) self.table.setItem(i, 1, QtGui.QTableWidgetItem(job_data[to_display[j]][1])) self.table.setItem(i, 2, QtGui.QTableWidgetItem(job_data[to_display[j]][2])) self.table.setItem(i, 3, QtGui.QTableWidgetItem(job_data[to_display[j]][3])) item = QtGui.QTableWidgetItem(job_data[to_display[j]][5]) item.setToolTip("Estimated job start: " + job_data[to_display[j]][7]) self.table.setItem(i, 5, item) self.table.setItem(i, 6, QtGui.QTableWidgetItem(job_data[to_display[j]][6])) self.table.item(i,2).setTextAlignment(QtCore.Qt.AlignRight) self.table.item(i,5).setTextAlignment(QtCore.Qt.AlignRight) pbars.append(i) pbars[i] = QtGui.QProgressBar(self) pbars[i].setValue(int(job_data[to_display[j]][4])) # Apply a color depending on the status of the job if ( job_data[to_display[j]][3] == "Running" ): self.table.setCellWidget(i,4,pbars[i]) self.table.item(i,0).setBackground(QtGui.QColor(255, 251, 168)) self.table.item(i,1).setBackground(QtGui.QColor(255, 251, 168)) self.table.item(i,2).setBackground(QtGui.QColor(255, 251, 168)) self.table.item(i,3).setBackground(QtGui.QColor(255, 251, 168)) self.table.item(i,5).setBackground(QtGui.QColor(255, 251, 168)) self.table.item(i,6).setBackground(QtGui.QColor(255, 251, 168)) else: pbars[j].setEnabled(False) self.table.setCellWidget(i,4,pbars[i]) self.table.item(i,0).setBackground(QtGui.QColor(255, 112, 118)) self.table.item(i,1).setBackground(QtGui.QColor(255, 112, 118)) self.table.item(i,2).setBackground(QtGui.QColor(255, 112, 118)) self.table.item(i,3).setBackground(QtGui.QColor(255, 112, 118)) self.table.item(i,5).setBackground(QtGui.QColor(255, 112, 118)) self.table.item(i,6).setBackground(QtGui.QColor(255, 112, 118)) # Save job data to file job_data_list.append(i) job_data_list[i] = job_data[to_display[j]][0] printstring = str(job_data[to_display[j]][0]) + " " + \ str(job_data[to_display[j]][1]) + " " + \ str(job_data[to_display[j]][2]) + " " + \ str(job_data[to_display[j]][3]) + " " + \ str(job_data[to_display[j]][4]) + " " + \ str(job_data[to_display[j]][5]) + " " + \ str(job_data[to_display[j]][6]) + " " + \ str(job_data[to_display[j]][7]) + "\n" file.write(printstring) # Display all completed/canceled/aborted jobs. The watchdog cannot # distinguish why the job has been finished to_display = [list(job_data_old_n).index(item) for item in list(jobs_complete) if list(jobs_complete) and list(job_data_old_n)] for j in range(0,len(to_display)): i = i + 1 self.table.insertRow(i) item = QtGui.QTableWidgetItem(job_data_old[to_display[j]][0]) item.setToolTip("Queuing name: " + job_data_old[to_display[j]][6]) self.table.setItem(i, 0, item) self.table.setItem(i, 1, QtGui.QTableWidgetItem(job_data_old[to_display[j]][1])) self.table.setItem(i, 2, QtGui.QTableWidgetItem(job_data_old[to_display[j]][2])) self.table.setItem(i, 3, QtGui.QTableWidgetItem(job_data_old[to_display[j]][3])) pbars.append(i) pbars[j] = QtGui.QProgressBar(self) pbars[j].setValue(int(job_data_old[to_display[j]][4])) pbars[j].setEnabled(False) self.table.setCellWidget(i,4,pbars[j]) item = QtGui.QTableWidgetItem(job_data_old[to_display[j]][5]) item.setToolTip("Estimated job start: " + job_data_old[to_display[j]][7]) self.table.setItem(i, 5, item) self.table.setItem(i, 6, QtGui.QTableWidgetItem(job_data_old[to_display[j]][6])) self.table.item(i,2).setTextAlignment(QtCore.Qt.AlignRight) self.table.item(i,5).setTextAlignment(QtCore.Qt.AlignRight) self.table.setItem(i, 3, QtGui.QTableWidgetItem("Completed")) # Apply a color depending on the status of the job self.table.item(i,0).setBackground(QtGui.QColor(172, 252, 175)) self.table.item(i,1).setBackground(QtGui.QColor(172, 252, 175)) self.table.item(i,2).setBackground(QtGui.QColor(172, 252, 175)) self.table.item(i,3).setBackground(QtGui.QColor(172, 252, 175)) self.table.item(i,5).setBackground(QtGui.QColor(172, 252, 175)) self.table.item(i,6).setBackground(QtGui.QColor(172, 252, 175)) # Save job data to file job_data_list.append(i) job_data_list[i] = job_data_old[to_display[j]][0] printstring = str(job_data_old[to_display[j]][0]) + " " + \ str(job_data_old[to_display[j]][1]) + " " + \ str(job_data_old[to_display[j]][2]) + " " + \ str(job_data_old[to_display[j]][3]) + " " + \ str(job_data_old[to_display[j]][4]) + " " + \ str(job_data_old[to_display[j]][5]) + " " + \ str(job_data_old[to_display[j]][6]) + " " + \ str(job_data_old[to_display[j]][7]) + "\n" file.write(printstring) del(jobs_complete) del(jobs_new) del(result) del(job_data_old) del(job_data_old_n) file.close() # Update is complete, sorting can thus be re-enabled self.table.setSortingEnabled(True) # Update the logfile if ( os.path.isfile(palm_dir + '/.wd.newdata') ): shutil.copy(palm_dir + '/.wd.newdata',palm_dir + '/.wd.olddata') # Re-enable timer processes self.setEnabled(True) self.label2.setText("Last update: " + timestamp) self.label.setText("Update complete.") self.timetimer.start() self.timer.start(update_frequency) self.labeltimer.start(update_frequency/10) self.UpdateLabel() QtGui.QApplication.processEvents() # Canceling selected job from table def CancelJob(self): # Return internal jobname jobname = self.table.item(self.table.currentRow(),6).text() jobrealname = self.table.item(self.table.currentRow(),0).text() # Check description of job in order to login on the correct host descr = self.table.item(self.table.currentRow(),1).text() querybox = QtGui.QMessageBox() querybox.setText("Attention!\nYou are trying to cancel a job. It will not be able to terminate properly.") querybox.setInformativeText("Do you really want to cancel " + jobname + " on host " + descr + "?") querybox.setStandardButtons(QtGui.QMessageBox.Yes | QtGui.QMessageBox.No) querybox.setDefaultButton(QtGui.QMessageBox.No) returnvalue = querybox.exec_() if ( returnvalue == QtGui.QMessageBox.Yes ): for h in range(0,len(description)): if ( descr == description[h] ): host = username[h] + "@" + hostname[h] ssh = sub.Popen(["ssh", "%s" % host, "/sw/tools/python/2.7.6/generic/bin/python palm_wdd cancel " + jobname], shell=False, stdout=sub.PIPE, stderr=sub.PIPE) result = ssh.stdout.readlines() result = filter(None, result) # In case of error display a warning message if ( len(result) == 0 ): error = ssh.stderr.readlines() notify = QtGui.QMessageBox.warning(self,'Cancel job',"Error. Could not cancel job " + jobname + ".\n\nError message:\n" + ''.join(error)) # Otherwise inform the user and mark the job in the table else: self.table.setItem(self.table.currentRow(),3,QtGui.QTableWidgetItem("Canceled")) notify = QtGui.QMessageBox.information(self,'Cancel job',"Job" + jobname + " canceled!\n\nServer message:\n" + ''.join(result)) # Show detailed information on job. For documentation see above def ShowDetails(self): jobname = self.table.item(self.table.currentRow(),6).text() descr = self.table.item(self.table.currentRow(),1).text() for h in range(0,len(description)): if ( descr == description[h] ): host = username[h] + "@" + hostname[h] ssh = sub.Popen(["ssh", "%s" % host, "/sw/tools/python/2.7.6/generic/bin/python palm_wdd check " + jobname], shell=False, stdout=sub.PIPE, stderr=sub.PIPE) result = ssh.stdout.readlines() result = filter(None, result) if ( len(result) == 0 ): error = ssh.stderr.readlines() notify = QtGui.QMessageBox.warning(self,'Show job details',"Error. Could not fetch job details for " + jobname + ".\n\nError message:\n" + ''.join(error)) else: notify = QtGui.QMessageBox.information(self,'Job details',"Details for job " + jobname + ":\n\n" + ''.join(result)) # Perform a forced stop on the job def DoStop(self): # Return internal jobname jobname = self.table.item(self.table.currentRow(),6).text() jobrealname = self.table.item(self.table.currentRow(),0).text() # Check description of job in order to login on the correct host descr = self.table.item(self.table.currentRow(),1).text() for h in range(0,len(description)): if ( descr == description[h] ): host = username[h] + "@" + hostname[h] user = username[h] ssh = sub.Popen(["ssh", "%s" % host, "/sw/tools/python/2.7.6/generic/bin/python palm_wdd stop " + jobrealname], shell=False, stdout=sub.PIPE, stderr=sub.PIPE) result = ssh.stdout.readlines() result = filter(None, result) # In case of error display a warning message if ( len(result) == 0 ): error = ssh.stderr.readlines() notify = QtGui.QMessageBox.warning(self,'Proper termination of job',"Error. Could not stop job " + jobname + ".\n\nError message:\n" + ''.join(error)) # Otherwise inform the user and mark the job in the table else: self.table.setItem(self.table.currentRow(),3,QtGui.QTableWidgetItem("Terminated")) notify = QtGui.QMessageBox.information(self,'Terminate job',"Termination of job " + jobname + " was initiated!\n\nServer message:\n" + ''.join(result)) # Perform a forced stop on the job def DoRestart(self): # Return internal jobname jobname = self.table.item(self.table.currentRow(),6).text() jobrealname = self.table.item(self.table.currentRow(),0).text() # Check description of job in order to login on the correct host descr = self.table.item(self.table.currentRow(),1).text() for h in range(0,len(description)): if ( descr == description[h] ): host = username[h] + "@" + hostname[h] user = username[h] ssh = sub.Popen(["ssh", "%s" % host, "/sw/tools/python/2.7.6/generic/bin/python palm_wdd restart " + jobrealname], shell=False, stdout=sub.PIPE, stderr=sub.PIPE) result = ssh.stdout.readlines() result = filter(None, result) # In case of error display a warning message if ( len(result) == 0 ): error = ssh.stderr.readlines() notify = QtGui.QMessageBox.warning(self,'Proper termination of job',"Error. Could not stop job " + jobname + ".\n\nError message:\n" + ''.join(error)) # Otherwise inform the user and mark the job in the table else: self.table.setItem(self.table.currentRow(),3,QtGui.QTableWidgetItem("Terminated")) notify = QtGui.QMessageBox.information(self,'Terminate job for restart',"Restart for job" + jobname + " was initiated!\n\nServer message:\n" + ''.join(result)) # Read the PARIN file of the job def ShowPARIN(self): # Return internal jobname jobname = self.table.item(self.table.currentRow(),6).text() jobrealname = self.table.item(self.table.currentRow(),0).text() # Check description of job in order to login on the correct host descr = self.table.item(self.table.currentRow(),1).text() for h in range(0,len(description)): if ( descr == description[h] ): host = username[h] + "@" + hostname[h] user = username[h] ssh = sub.Popen(["ssh", "%s" % host, "/sw/tools/python/2.7.6/generic/bin/python palm_wdd parin " + jobrealname], shell=False, stdout=sub.PIPE, stderr=sub.PIPE) result = ssh.stdout.readlines() result = filter(None, result) # In case of error display a warning message if ( len(result) == 0 ): error = ssh.stderr.readlines() notify = QtGui.QMessageBox.warning(self,'Showing parameter file',"Error. Could not fetch parameter file for job " + jobrealname + " (" + jobname + ").\n\nError message:\n" + ''.join(error)) # Otherwise inform the user and mark the job in the table else: mb = MessageBoxScroll() mb.setText("Parameter file for job: " + jobrealname + "") mb.setDetailedText(''.join(result)) mb.exec_() # Read the PARIN file of the job def ShowRC(self): # Return internal jobname and real job name jobname = self.table.item(self.table.currentRow(),6).text() jobrealname = self.table.item(self.table.currentRow(),0).text() # Check description of job in order to login on the correct host descr = self.table.item(self.table.currentRow(),1).text() for h in range(0,len(description)): if ( descr == description[h] ): host = username[h] + "@" + hostname[h] user = username[h] ssh = sub.Popen(["ssh", "%s" % host, "/sw/tools/python/2.7.6/generic/bin/python palm_wdd rc " + jobrealname], shell=False, stdout=sub.PIPE, stderr=sub.PIPE) result = ssh.stdout.readlines() result = filter(None, result) # In case of error display a warning message if ( len(result) == 0 ): error = ssh.stderr.readlines() notify = QtGui.QMessageBox.warning(self,'Showing run control',"Error. Could not fetch run control file for job " + jobrealname + "(" + jobname + ").\n\nError message:\n" + ''.join(error)) # Otherwise inform the user and mark the job in the table else: mb = MessageBoxScroll() lastline = result[len(result)-2].split()[2] mb.setText("Simulated time for job " + jobrealname + " is currently: " + lastline) mb.setDetailedText(''.join(result)) mb.exec_() # Remove a job from list - removes the current row from the table and from # save file def RemoveFromList(self, row): if ( row == -1 ): row = self.table.currentRow() # Read data from save file job_to_delete = self.table.item(row,6).text() self.table.removeRow(row) file = open(palm_dir + '/.wd.olddata', 'r') result = file.readlines() result = filter(None, result) file.close() file = open(palm_dir + '/.wd.olddata', 'w') job_data_old = [] if ( len(result) == 0 ): notify = QtGui.QMessageBox.warning(self,'Read from .wd.olddata',"Error message:\n\nCould not read from file. I/O error.") else: # Save data in array job_data_old for j in range(0,len(result)): job_data_old.append(j) job_data_old[j] = result[j].split(" ") job_data_old[j] = filter(None, job_data_old[j]) job_data_old[j] = [line.rstrip('\n') for line in job_data_old[j]] # Check if line j is the selected job, if not -> save to file if ( job_data_old[j][6] != job_to_delete ): printstring = str(job_data_old[j][0]) + " " + \ str(job_data_old[j][1]) + " " + \ str(job_data_old[j][2]) + " " + \ str(job_data_old[j][3]) + " " + \ str(job_data_old[j][4]) + " " + \ str(job_data_old[j][5]) + " " + \ str(job_data_old[j][6]) + " " + \ str(job_data_old[j][7]) + "\n" file.write(printstring) file.close() # Remove all completed jobs from list def ClearList(self): num_of_lines = self.table.rowCount() # Delete all lines with completed/canceled jobs from list. The counter # must decrease as the line numbers would be messed up otherwise for j in range(num_of_lines-1,-1,-1): state = self.table.item(j,3).text() if ( state == "Completed" or state == "Canceled" or state == "Terminated"): self.RemoveFromList(j) # Update the label def UpdateLabel(self): remaining_time = (update_frequency - self.timetimer.elapsed()) / 1000 / 60 self.label.setText("Next update in " + str(remaining_time) + " min") # Enter Options menu def Options(self): # disable mainwindow self.setEnabled(False) # show Options Dialog opt = OptionBox() opt.exec_() self.UpdateHosts() self.setEnabled(True) # Update settings def UpdateHosts(self): global update_frequency global description global hostname global username description = [] hostname = [] username = [] for i in range(0,len(config.sections())): description_tmp = config.sections()[i] if ( description_tmp != 'Settings' ): if ( config.get(description_tmp, 'enabled') == 'true' ): description.append(description_tmp) hostname.append(config.get(description_tmp, 'hostname')) username.append(config.get(description_tmp, 'username')) else: update_frequency = int(config.get(description_tmp, 'update_frequency'))*60000 # Main loop def main(): app = QtGui.QApplication(sys.argv) res = Watchdog() sys.exit(app.exec_()) if __name__ == '__main__': main()