| [1611] | 1 | #!/usr/bin/python |
|---|
| 2 | # -*- coding: utf-8 -*- |
|---|
| 3 | #--------------------------------------------------------------------------------# |
|---|
| 4 | # This file is part of PALM. |
|---|
| 5 | # |
|---|
| 6 | # PALM is free software: you can redistribute it and/or modify it under the terms |
|---|
| 7 | # of the GNU General Public License as published by the Free Software Foundation, |
|---|
| 8 | # either version 3 of the License, or (at your option) any later version. |
|---|
| 9 | # |
|---|
| 10 | # PALM is distributed in the hope that it will be useful, but WITHOUT ANY |
|---|
| 11 | # WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR |
|---|
| 12 | # A PARTICULAR PURPOSE. See the GNU General Public License for more details. |
|---|
| 13 | # |
|---|
| 14 | # You should have received a copy of the GNU General Public License along with |
|---|
| 15 | # PALM. If not, see <http://www.gnu.org/licenses/>. |
|---|
| 16 | # |
|---|
| 17 | # Copyright 1997-2015 Leibniz Universitaet Hannover |
|---|
| 18 | #--------------------------------------------------------------------------------# |
|---|
| 19 | # |
|---|
| 20 | # Current revisions: |
|---|
| 21 | # ----------------- |
|---|
| 22 | # |
|---|
| [1612] | 23 | # |
|---|
| [1611] | 24 | # Former revisions: |
|---|
| 25 | # ----------------- |
|---|
| 26 | # $Id: palm_wd 1612 2015-07-07 12:25:21Z maronga $ |
|---|
| 27 | # |
|---|
| [1612] | 28 | # 1611 2015-07-07 12:23:22Z maronga |
|---|
| 29 | # Initial revision |
|---|
| 30 | # |
|---|
| [1611] | 31 | # |
|---|
| 32 | # Description: |
|---|
| 33 | # ------------ |
|---|
| 34 | # PALM watchdog client for monitoring batch jobs on a variety of hosts specified |
|---|
| 35 | # by the user |
|---|
| 36 | # |
|---|
| 37 | # Instructions: |
|---|
| 38 | # ------------- |
|---|
| 39 | # 1) Modify the header section of palm_wd |
|---|
| 40 | # 2) Move .wd.olddata and .wd.newdata to your palm directory |
|---|
| 41 | # (e.g. /home/user/current_version/.wd.newdata etc.) |
|---|
| 42 | # 3) Modify a copy of palm_wdd for each host to be monitored and move it to the |
|---|
| 43 | # respective hosts |
|---|
| 44 | # 4) Start the client either from mrungui or from shell by "nohup palm_wd&" |
|---|
| 45 | |
|---|
| 46 | # To do: |
|---|
| 47 | # ------ |
|---|
| 48 | # 1) Add "Options", "Help" and "Manual" |
|---|
| 49 | # 2) Move user settings to a configuration file |
|---|
| 50 | #------------------------------------------------------------------------------! |
|---|
| 51 | |
|---|
| 52 | import sys |
|---|
| 53 | import subprocess as sub |
|---|
| 54 | from PyQt4 import QtGui, QtCore, uic |
|---|
| 55 | from PyQt4.QtCore import pyqtSlot,SIGNAL,SLOT |
|---|
| 56 | import time |
|---|
| 57 | import datetime |
|---|
| 58 | import shutil |
|---|
| 59 | import os |
|---|
| 60 | |
|---|
| 61 | # START OF HEADER |
|---|
| 62 | # Each list must contain the login host, the username and a description |
|---|
| 63 | hostname = ["hlogin.hlrn.de", "blogin.hlrn.de"] |
|---|
| 64 | username = ["nikmaron" , "nikmaron" ] |
|---|
| 65 | description = ["Hannover" , "Berlin" ] |
|---|
| 66 | |
|---|
| 67 | update_frequency = 600*1000 |
|---|
| 68 | |
|---|
| 69 | # END OF HEADER |
|---|
| 70 | |
|---|
| 71 | # Determine PALM directories |
|---|
| 72 | try: |
|---|
| 73 | devnull = open(os.devnull, 'w') |
|---|
| 74 | out = sub.check_output("echo $PALM_BIN", shell=True, stderr=sub.STDOUT) |
|---|
| 75 | palm_bin = out.rstrip() |
|---|
| 76 | palm_dir = out.split("palm")[0] + "palm/" + out.split("palm")[1].split("/")[1] |
|---|
| 77 | out = None |
|---|
| 78 | except: |
|---|
| 79 | print "Error. $PALM_BIN is not set." |
|---|
| 80 | raise SystemExit |
|---|
| 81 | |
|---|
| 82 | |
|---|
| 83 | # Dummy variables for the jobname and the progressbars |
|---|
| 84 | jobname = "" |
|---|
| 85 | timestamp = "" |
|---|
| 86 | pbars = [] |
|---|
| 87 | job_data_list = [] |
|---|
| 88 | |
|---|
| 89 | # MainWindow class |
|---|
| 90 | class Watchdog(QtGui.QMainWindow): |
|---|
| 91 | |
|---|
| 92 | def __init__(self): |
|---|
| 93 | super(Watchdog, self).__init__() |
|---|
| 94 | |
|---|
| 95 | self.InitUi() |
|---|
| 96 | |
|---|
| 97 | |
|---|
| 98 | # Initialized MainWindow UI |
|---|
| 99 | def InitUi(self): |
|---|
| 100 | |
|---|
| 101 | # Load predefined mainwindow |
|---|
| 102 | uic.loadUi(palm_bin + '/palm_wd_files/wd.ui', self) |
|---|
| 103 | |
|---|
| 104 | # Resize columns and set number of rows to zero. Column 6 is hidden and |
|---|
| 105 | # contains the remote jobname (e.g. hannover.174610) |
|---|
| 106 | self.table.setColumnWidth(0,230) |
|---|
| 107 | self.table.setColumnWidth(1,80) |
|---|
| 108 | self.table.setColumnWidth(2,50) |
|---|
| 109 | self.table.setColumnWidth(3,80) |
|---|
| 110 | self.table.setColumnWidth(4,180) |
|---|
| 111 | self.table.setColumnWidth(5,115) |
|---|
| 112 | self.table.setColumnHidden(6, True) |
|---|
| 113 | self.table.setRowCount(0) |
|---|
| 114 | |
|---|
| 115 | # Display MainWindow |
|---|
| 116 | self.show() |
|---|
| 117 | QtGui.QApplication.processEvents() |
|---|
| 118 | |
|---|
| 119 | # Start refresh timer. On timeout perform update |
|---|
| 120 | self.timer = QtCore.QTimer(self) |
|---|
| 121 | self.timer.timeout.connect(self.Refresh) |
|---|
| 122 | self.timer.setSingleShot(False) |
|---|
| 123 | self.timer.start(update_frequency) |
|---|
| 124 | |
|---|
| 125 | # The timetimer counts the time since last update |
|---|
| 126 | self.timetimer= QtCore.QElapsedTimer() |
|---|
| 127 | self.timetimer.start() |
|---|
| 128 | |
|---|
| 129 | # The labeltimer induces the update of the remaining time in the UI |
|---|
| 130 | self.labeltimer = QtCore.QTimer(self) |
|---|
| 131 | self.labeltimer.timeout.connect(self.UpdateLabel) |
|---|
| 132 | self.labeltimer.setSingleShot(False) |
|---|
| 133 | |
|---|
| 134 | # Update in the UI will be performed at each 1/10 of the update interval |
|---|
| 135 | self.labeltimer.start(update_frequency/10) |
|---|
| 136 | self.label.setText("Next update in " + str(update_frequency/1000/60) + "min") |
|---|
| 137 | |
|---|
| 138 | # Update the list now |
|---|
| 139 | self.Refresh() |
|---|
| 140 | |
|---|
| 141 | |
|---|
| 142 | # Add a custom context menu |
|---|
| 143 | def OpenContextMenu(self, position): |
|---|
| 144 | menu = QtGui.QMenu() |
|---|
| 145 | |
|---|
| 146 | # "Show details" -> fetch details for the selected job |
|---|
| 147 | detailsjobAction = menu.addAction('Show details') |
|---|
| 148 | detailsjobAction.setStatusTip('Display detailed info for job') |
|---|
| 149 | detailsjobAction.triggered.connect(self.ShowDetails) |
|---|
| 150 | |
|---|
| 151 | # "Cancel job" -> remove job from queuing system |
|---|
| 152 | canceljobAction = menu.addAction('Cancel job') |
|---|
| 153 | canceljobAction.setStatusTip('Remove job from queueing system') |
|---|
| 154 | canceljobAction.triggered.connect(self.CancelJob) |
|---|
| 155 | |
|---|
| 156 | # "Force stop" -> remove job from queuing system |
|---|
| 157 | stopjobAction = menu.addAction('Force stop') |
|---|
| 158 | stopjobAction.setStatusTip('Terminate job properly') |
|---|
| 159 | stopjobAction.triggered.connect(self.DoStop) |
|---|
| 160 | |
|---|
| 161 | # "Force restart" -> remove job from queuing system |
|---|
| 162 | restartjobAction = menu.addAction('Force restart') |
|---|
| 163 | restartjobAction.setStatusTip('Terminate job properly and initiate restart') |
|---|
| 164 | restartjobAction.triggered.connect(self.DoRestart) |
|---|
| 165 | |
|---|
| 166 | # "Remove from list" -> remove a completed job from the list |
|---|
| 167 | removefromlistAction = menu.addAction('Remove from list') |
|---|
| 168 | removefromlistAction.setStatusTip('Remove finished job') |
|---|
| 169 | removefromlistAction.triggered.connect(lambda: self.RemoveFromList(-1)) |
|---|
| 170 | |
|---|
| 171 | |
|---|
| 172 | # Activate/deactive contect menu items based on the status of the runs |
|---|
| 173 | state = self.table.item(self.table.currentRow(),3).text() |
|---|
| 174 | if (state == "Canceled" or state == "Completed" or state == "Terminated"): |
|---|
| 175 | detailsjobAction.setEnabled(False) |
|---|
| 176 | canceljobAction.setEnabled(False) |
|---|
| 177 | removefromlistAction.setEnabled(True) |
|---|
| 178 | else: |
|---|
| 179 | detailsjobAction.setEnabled(True) |
|---|
| 180 | canceljobAction.setEnabled(True) |
|---|
| 181 | removefromlistAction.setEnabled(False) |
|---|
| 182 | |
|---|
| 183 | if ( state == "Running" ): |
|---|
| 184 | stopjobAction.setEnabled(True) |
|---|
| 185 | restartjobAction.setEnabled(True) |
|---|
| 186 | else: |
|---|
| 187 | stopjobAction.setEnabled(False) |
|---|
| 188 | restartjobAction.setEnabled(False) |
|---|
| 189 | |
|---|
| 190 | # Show the context menu |
|---|
| 191 | action = menu.exec_(self.table.mapToGlobal(position)) |
|---|
| 192 | |
|---|
| 193 | |
|---|
| 194 | # Collecting watchdog data and display in the UI |
|---|
| 195 | def Refresh(self): |
|---|
| 196 | |
|---|
| 197 | # Use global variables |
|---|
| 198 | global jobs_save |
|---|
| 199 | global pbars |
|---|
| 200 | |
|---|
| 201 | # Deactivate timer processes until update is finished |
|---|
| 202 | # QtGui.QApplication.processEvents() updates the UI whenever needed |
|---|
| 203 | self.labeltimer.stop() |
|---|
| 204 | self.label.setText("Updating...") |
|---|
| 205 | QtGui.QApplication.processEvents() |
|---|
| 206 | self.setEnabled(False) |
|---|
| 207 | QtGui.QApplication.processEvents() |
|---|
| 208 | |
|---|
| 209 | # Get current timestamp |
|---|
| 210 | timestamp = datetime.datetime.now().strftime("%d/%m/%Y %H:%M") |
|---|
| 211 | |
|---|
| 212 | # Open an empty file for the watchdog data |
|---|
| 213 | file = open(palm_dir + '/.wd.newdata', 'w') |
|---|
| 214 | |
|---|
| 215 | |
|---|
| 216 | # Set all required variables to their initial values. Sorting must be |
|---|
| 217 | # disabled. |
|---|
| 218 | self.table.setRowCount(0) |
|---|
| 219 | self.table.setSortingEnabled(False) |
|---|
| 220 | i = -1 |
|---|
| 221 | job_data_list = [] |
|---|
| 222 | pbars = [] |
|---|
| 223 | |
|---|
| 224 | # Scan all hosts in the variable hostname. For each host perform the |
|---|
| 225 | # update loop |
|---|
| 226 | for h in range(0,len(hostname)): |
|---|
| 227 | |
|---|
| 228 | # Perform ssh command |
|---|
| 229 | host = username[h] + "@" + hostname[h] |
|---|
| 230 | ssh = sub.Popen(["ssh", "%s" % host, "/sw/tools/python/2.7.6/generic/bin/python palm_wdd queue " + username[h]], |
|---|
| 231 | shell=False, |
|---|
| 232 | stdout=sub.PIPE, |
|---|
| 233 | stderr=sub.PIPE) |
|---|
| 234 | result = ssh.stdout.readlines() |
|---|
| 235 | result = filter(None, result) |
|---|
| 236 | |
|---|
| 237 | |
|---|
| 238 | # Delete all job data |
|---|
| 239 | job_data_tmp = [] |
|---|
| 240 | job_data = [] |
|---|
| 241 | job_data_n = [] |
|---|
| 242 | job_data_old = [] |
|---|
| 243 | job_data_old_n = [] |
|---|
| 244 | job_progress = 0 |
|---|
| 245 | |
|---|
| 246 | # In case of errors display error message |
|---|
| 247 | if ( len(result) < 1 ): |
|---|
| 248 | error = ssh.stderr.readlines() |
|---|
| 249 | if ( error != [] ): |
|---|
| 250 | 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)) |
|---|
| 251 | |
|---|
| 252 | # No error -> save job data |
|---|
| 253 | else: |
|---|
| 254 | |
|---|
| 255 | # Successively write to job_data |
|---|
| 256 | for j in range(0,len(result)): |
|---|
| 257 | |
|---|
| 258 | job_data_tmp.append(j) |
|---|
| 259 | job_data_tmp[j] = result[j].split(" ") |
|---|
| 260 | job_data_tmp[j] = filter(None, job_data_tmp[j]) |
|---|
| 261 | |
|---|
| 262 | job_data.append(j) |
|---|
| 263 | job_data[j] = [job_data_tmp[j][0], description[h], job_data_tmp[j][2], |
|---|
| 264 | job_data_tmp[j][3], int(float(job_data_tmp[j][4])*100), job_data_tmp[j][5], |
|---|
| 265 | job_data_tmp[j][1], job_data_tmp[j][6].rstrip()] |
|---|
| 266 | job_data_n.append(j) |
|---|
| 267 | job_data_n[j] = job_data_tmp[j][1] |
|---|
| 268 | |
|---|
| 269 | del(result) |
|---|
| 270 | |
|---|
| 271 | |
|---|
| 272 | |
|---|
| 273 | # Now read the data from the last update from file. These data are |
|---|
| 274 | # already in the prefered format |
|---|
| 275 | file2 = open(palm_dir + '/.wd.olddata', 'r') |
|---|
| 276 | result = file2.readlines() |
|---|
| 277 | file2.close() |
|---|
| 278 | |
|---|
| 279 | for j in range(0,len(result)): |
|---|
| 280 | if ( result[j].split()[1] == description[h] ): |
|---|
| 281 | job_data_old.append(j) |
|---|
| 282 | job_data_old[j] = result[j].split(" ") |
|---|
| 283 | job_data_old[j] = filter(None, job_data_old[j]) |
|---|
| 284 | job_data_old[j] = [line.rstrip('\n') for line in job_data_old[j]] |
|---|
| 285 | |
|---|
| 286 | job_data_old_n.append(j) |
|---|
| 287 | job_data_old_n[j] = job_data_old[j][6] |
|---|
| 288 | |
|---|
| 289 | # Merge the job_data and job_data_old to find completed, new and |
|---|
| 290 | # still queued jobs |
|---|
| 291 | if ( len(job_data_n) > 0 and len(job_data_old_n) > 0 ): |
|---|
| 292 | jobs_full = list(set(job_data_old_n) | set(job_data_n)) |
|---|
| 293 | jobs_known = list(set(job_data_old_n) & set(job_data_n)) |
|---|
| 294 | jobs_complete = set(job_data_old_n) - set(job_data_n) |
|---|
| 295 | jobs_new = set(job_data_n) - set(job_data_old_n) |
|---|
| 296 | elif ( len(job_data_n) > 0 ): |
|---|
| 297 | jobs_full = job_data_n |
|---|
| 298 | jobs_known = [] |
|---|
| 299 | jobs_new = job_data_n |
|---|
| 300 | jobs_complete = [] |
|---|
| 301 | elif ( len(job_data_old_n) > 0 ): |
|---|
| 302 | jobs_full = job_data_old_n |
|---|
| 303 | jobs_known = [] |
|---|
| 304 | jobs_new = [] |
|---|
| 305 | jobs_complete = job_data_old_n |
|---|
| 306 | else: |
|---|
| 307 | jobs_full = [] |
|---|
| 308 | jobs_known = [] |
|---|
| 309 | jobs_new = [] |
|---|
| 310 | jobs_complete = [] |
|---|
| 311 | |
|---|
| 312 | # Display all known jobs |
|---|
| 313 | to_display = [list(job_data_n).index(item) for item in list(jobs_known) if list(jobs_known) and list(job_data_n)] |
|---|
| 314 | for j in range(0,len(to_display)): |
|---|
| 315 | i = i + 1 |
|---|
| 316 | self.table.insertRow(i) |
|---|
| 317 | item = QtGui.QTableWidgetItem(job_data[to_display[j]][0]) |
|---|
| 318 | item.setToolTip("Queuing name: " + job_data[to_display[j]][6]) |
|---|
| 319 | self.table.setItem(i, 0, item) |
|---|
| 320 | self.table.setItem(i, 1, QtGui.QTableWidgetItem(job_data[to_display[j]][1])) |
|---|
| 321 | self.table.setItem(i, 2, QtGui.QTableWidgetItem(job_data[to_display[j]][2])) |
|---|
| 322 | self.table.setItem(i, 3, QtGui.QTableWidgetItem(job_data[to_display[j]][3])) |
|---|
| 323 | item = QtGui.QTableWidgetItem(job_data[to_display[j]][5]) |
|---|
| 324 | item.setToolTip("Estimated job start: " + job_data[to_display[j]][7]) |
|---|
| 325 | self.table.setItem(i, 5, item) |
|---|
| 326 | self.table.setItem(i, 6, QtGui.QTableWidgetItem(job_data[to_display[j]][6])) |
|---|
| 327 | self.table.item(i,2).setTextAlignment(QtCore.Qt.AlignRight) |
|---|
| 328 | self.table.item(i,5).setTextAlignment(QtCore.Qt.AlignRight) |
|---|
| 329 | pbars.append(i) |
|---|
| 330 | pbars[i] = QtGui.QProgressBar(self) |
|---|
| 331 | pbars[i].setValue(int(job_data[to_display[j]][4])) |
|---|
| 332 | |
|---|
| 333 | # Apply a color depending on the status of the job |
|---|
| 334 | if ( job_data[to_display[j]][3] == "Running" ): |
|---|
| 335 | self.table.setCellWidget(i,4,pbars[j]) |
|---|
| 336 | self.table.item(i,0).setBackground(QtGui.QColor(255, 251, 168)) |
|---|
| 337 | self.table.item(i,1).setBackground(QtGui.QColor(255, 251, 168)) |
|---|
| 338 | self.table.item(i,2).setBackground(QtGui.QColor(255, 251, 168)) |
|---|
| 339 | self.table.item(i,3).setBackground(QtGui.QColor(255, 251, 168)) |
|---|
| 340 | self.table.item(i,5).setBackground(QtGui.QColor(255, 251, 168)) |
|---|
| 341 | self.table.item(i,6).setBackground(QtGui.QColor(255, 251, 168)) |
|---|
| 342 | else: |
|---|
| 343 | pbars[j].setEnabled(False) |
|---|
| 344 | self.table.setCellWidget(i,4,pbars[j]) |
|---|
| 345 | self.table.item(i,0).setBackground(QtGui.QColor(255, 112, 118)) |
|---|
| 346 | self.table.item(i,1).setBackground(QtGui.QColor(255, 112, 118)) |
|---|
| 347 | self.table.item(i,2).setBackground(QtGui.QColor(255, 112, 118)) |
|---|
| 348 | self.table.item(i,3).setBackground(QtGui.QColor(255, 112, 118)) |
|---|
| 349 | self.table.item(i,5).setBackground(QtGui.QColor(255, 112, 118)) |
|---|
| 350 | self.table.item(i,6).setBackground(QtGui.QColor(255, 112, 118)) |
|---|
| 351 | |
|---|
| 352 | # Save the job data to file |
|---|
| 353 | job_data_list.append(i) |
|---|
| 354 | job_data_list[i] = job_data[to_display[j]][0] |
|---|
| 355 | |
|---|
| 356 | printstring = str(job_data[to_display[j]][0]) + " " + \ |
|---|
| 357 | str(job_data[to_display[j]][1]) + " " + \ |
|---|
| 358 | str(job_data[to_display[j]][2]) + " " + \ |
|---|
| 359 | str(job_data[to_display[j]][3]) + " " + \ |
|---|
| 360 | str(job_data[to_display[j]][4]) + " " + \ |
|---|
| 361 | str(job_data[to_display[j]][5]) + " " + \ |
|---|
| 362 | str(job_data[to_display[j]][6]) + " " + \ |
|---|
| 363 | str(job_data[to_display[j]][7]) + "\n" |
|---|
| 364 | file.write(printstring) |
|---|
| 365 | |
|---|
| 366 | |
|---|
| 367 | # Display all new jobs |
|---|
| 368 | to_display = [list(job_data_n).index(item) for item in list(jobs_new) if list(jobs_new) and list(job_data_n)] |
|---|
| 369 | for j in range(0,len(to_display)): |
|---|
| 370 | i = i + 1 |
|---|
| 371 | self.table.insertRow(i) |
|---|
| 372 | item = QtGui.QTableWidgetItem(job_data[to_display[j]][0]) |
|---|
| 373 | item.setToolTip("Queuing name: " + job_data[to_display[j]][6]) |
|---|
| 374 | self.table.setItem(i, 0, item) |
|---|
| 375 | self.table.setItem(i, 0, QtGui.QTableWidgetItem(job_data[to_display[j]][0])) |
|---|
| 376 | self.table.setItem(i, 1, QtGui.QTableWidgetItem(job_data[to_display[j]][1])) |
|---|
| 377 | self.table.setItem(i, 2, QtGui.QTableWidgetItem(job_data[to_display[j]][2])) |
|---|
| 378 | self.table.setItem(i, 3, QtGui.QTableWidgetItem(job_data[to_display[j]][3])) |
|---|
| 379 | item = QtGui.QTableWidgetItem(job_data[to_display[j]][5]) |
|---|
| 380 | item.setToolTip("Estimated job start: " + job_data[to_display[j]][7]) |
|---|
| 381 | self.table.setItem(i, 5, item) |
|---|
| 382 | self.table.setItem(i, 6, QtGui.QTableWidgetItem(job_data[to_display[j]][6])) |
|---|
| 383 | self.table.item(i,2).setTextAlignment(QtCore.Qt.AlignRight) |
|---|
| 384 | self.table.item(i,5).setTextAlignment(QtCore.Qt.AlignRight) |
|---|
| 385 | |
|---|
| 386 | pbars.append(i) |
|---|
| 387 | pbars[i] = QtGui.QProgressBar(self) |
|---|
| 388 | pbars[i].setValue(int(job_data[to_display[j]][4])) |
|---|
| 389 | |
|---|
| 390 | # Apply a color depending on the status of the job |
|---|
| 391 | if ( job_data[to_display[j]][3] == "Running" ): |
|---|
| 392 | self.table.setCellWidget(i,4,pbars[i]) |
|---|
| 393 | self.table.item(i,0).setBackground(QtGui.QColor(255, 251, 168)) |
|---|
| 394 | self.table.item(i,1).setBackground(QtGui.QColor(255, 251, 168)) |
|---|
| 395 | self.table.item(i,2).setBackground(QtGui.QColor(255, 251, 168)) |
|---|
| 396 | self.table.item(i,3).setBackground(QtGui.QColor(255, 251, 168)) |
|---|
| 397 | self.table.item(i,5).setBackground(QtGui.QColor(255, 251, 168)) |
|---|
| 398 | self.table.item(i,6).setBackground(QtGui.QColor(255, 251, 168)) |
|---|
| 399 | else: |
|---|
| 400 | pbars[j].setEnabled(False) |
|---|
| 401 | self.table.setCellWidget(i,4,pbars[i]) |
|---|
| 402 | self.table.item(i,0).setBackground(QtGui.QColor(255, 112, 118)) |
|---|
| 403 | self.table.item(i,1).setBackground(QtGui.QColor(255, 112, 118)) |
|---|
| 404 | self.table.item(i,2).setBackground(QtGui.QColor(255, 112, 118)) |
|---|
| 405 | self.table.item(i,3).setBackground(QtGui.QColor(255, 112, 118)) |
|---|
| 406 | self.table.item(i,5).setBackground(QtGui.QColor(255, 112, 118)) |
|---|
| 407 | self.table.item(i,6).setBackground(QtGui.QColor(255, 112, 118)) |
|---|
| 408 | |
|---|
| 409 | # Save job data to file |
|---|
| 410 | job_data_list.append(i) |
|---|
| 411 | job_data_list[i] = job_data[to_display[j]][0] |
|---|
| 412 | |
|---|
| 413 | printstring = str(job_data[to_display[j]][0]) + " " + \ |
|---|
| 414 | str(job_data[to_display[j]][1]) + " " + \ |
|---|
| 415 | str(job_data[to_display[j]][2]) + " " + \ |
|---|
| 416 | str(job_data[to_display[j]][3]) + " " + \ |
|---|
| 417 | str(job_data[to_display[j]][4]) + " " + \ |
|---|
| 418 | str(job_data[to_display[j]][5]) + " " + \ |
|---|
| 419 | str(job_data[to_display[j]][6]) + " " + \ |
|---|
| 420 | str(job_data[to_display[j]][7]) + "\n" |
|---|
| 421 | file.write(printstring) |
|---|
| 422 | |
|---|
| 423 | |
|---|
| 424 | # Display all completed/canceled/aborted jobs. The watchdog cannot |
|---|
| 425 | # distinguish why the job has been finished |
|---|
| 426 | 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)] |
|---|
| 427 | for j in range(0,len(to_display)): |
|---|
| 428 | i = i + 1 |
|---|
| 429 | self.table.insertRow(i) |
|---|
| 430 | item = QtGui.QTableWidgetItem(job_data_old[to_display[j]][0]) |
|---|
| 431 | item.setToolTip("Queuing name: " + job_data_old[to_display[j]][6]) |
|---|
| 432 | self.table.setItem(i, 0, item) |
|---|
| 433 | self.table.setItem(i, 1, QtGui.QTableWidgetItem(job_data_old[to_display[j]][1])) |
|---|
| 434 | self.table.setItem(i, 2, QtGui.QTableWidgetItem(job_data_old[to_display[j]][2])) |
|---|
| 435 | self.table.setItem(i, 3, QtGui.QTableWidgetItem(job_data_old[to_display[j]][3])) |
|---|
| 436 | pbars.append(i) |
|---|
| 437 | pbars[j] = QtGui.QProgressBar(self) |
|---|
| 438 | pbars[j].setValue(int(job_data_old[to_display[j]][4])) |
|---|
| 439 | pbars[j].setEnabled(False) |
|---|
| 440 | self.table.setCellWidget(i,4,pbars[j]) |
|---|
| 441 | item = QtGui.QTableWidgetItem(job_data_old[to_display[j]][5]) |
|---|
| 442 | item.setToolTip("Estimated job start: " + job_data_old[to_display[j]][7]) |
|---|
| 443 | self.table.setItem(i, 5, item) |
|---|
| 444 | self.table.setItem(i, 6, QtGui.QTableWidgetItem(job_data_old[to_display[j]][6])) |
|---|
| 445 | self.table.item(i,2).setTextAlignment(QtCore.Qt.AlignRight) |
|---|
| 446 | self.table.item(i,5).setTextAlignment(QtCore.Qt.AlignRight) |
|---|
| 447 | self.table.setItem(i, 3, QtGui.QTableWidgetItem("Completed")) |
|---|
| 448 | |
|---|
| 449 | # Apply a color depending on the status of the job |
|---|
| 450 | self.table.item(i,0).setBackground(QtGui.QColor(172, 252, 175)) |
|---|
| 451 | self.table.item(i,1).setBackground(QtGui.QColor(172, 252, 175)) |
|---|
| 452 | self.table.item(i,2).setBackground(QtGui.QColor(172, 252, 175)) |
|---|
| 453 | self.table.item(i,3).setBackground(QtGui.QColor(172, 252, 175)) |
|---|
| 454 | self.table.item(i,5).setBackground(QtGui.QColor(172, 252, 175)) |
|---|
| 455 | self.table.item(i,6).setBackground(QtGui.QColor(172, 252, 175)) |
|---|
| 456 | |
|---|
| 457 | |
|---|
| 458 | # Save job data to file |
|---|
| 459 | job_data_list.append(i) |
|---|
| 460 | job_data_list[i] = job_data_old[to_display[j]][0] |
|---|
| 461 | |
|---|
| 462 | printstring = str(job_data_old[to_display[j]][0]) + " " + \ |
|---|
| 463 | str(job_data_old[to_display[j]][1]) + " " + \ |
|---|
| 464 | str(job_data_old[to_display[j]][2]) + " " + \ |
|---|
| 465 | str(job_data_old[to_display[j]][3]) + " " + \ |
|---|
| 466 | str(job_data_old[to_display[j]][4]) + " " + \ |
|---|
| 467 | str(job_data_old[to_display[j]][5]) + " " + \ |
|---|
| 468 | str(job_data_old[to_display[j]][6]) + " " + \ |
|---|
| 469 | str(job_data_old[to_display[j]][7]) + "\n" |
|---|
| 470 | file.write(printstring) |
|---|
| 471 | |
|---|
| 472 | del(jobs_complete) |
|---|
| 473 | del(jobs_new) |
|---|
| 474 | del(result) |
|---|
| 475 | |
|---|
| 476 | file.close() |
|---|
| 477 | |
|---|
| 478 | |
|---|
| 479 | # Update is complete, sorting can thus be re-enabled |
|---|
| 480 | self.table.setSortingEnabled(True) |
|---|
| 481 | |
|---|
| 482 | # Update the logfile |
|---|
| 483 | if ( os.path.isfile(palm_dir + '/.wd.newdata') ): |
|---|
| 484 | shutil.copy(palm_dir + '/.wd.newdata',palm_dir + '/.wd.olddata') |
|---|
| 485 | |
|---|
| 486 | # Re-enable timer processes |
|---|
| 487 | self.setEnabled(True) |
|---|
| 488 | self.label2.setText("Last update: " + timestamp) |
|---|
| 489 | self.label.setText("Update complete.") |
|---|
| 490 | self.timetimer.restart() |
|---|
| 491 | self.timer.start() |
|---|
| 492 | self.labeltimer.start() |
|---|
| 493 | QtGui.QApplication.processEvents() |
|---|
| 494 | |
|---|
| 495 | # Canceling selected job from table |
|---|
| 496 | def CancelJob(self): |
|---|
| 497 | |
|---|
| 498 | # Return internal jobname |
|---|
| 499 | jobname = self.table.item(self.table.currentRow(),6).text() |
|---|
| 500 | |
|---|
| 501 | # Check description of job in order to login on the correct host |
|---|
| 502 | descr = self.table.item(self.table.currentRow(),1).text() |
|---|
| 503 | for h in range(0,len(description)): |
|---|
| 504 | if ( descr == description[h] ): |
|---|
| 505 | host = username[h] + "@" + hostname[h] |
|---|
| 506 | |
|---|
| 507 | ssh = sub.Popen(["ssh", "%s" % host, "/sw/tools/python/2.7.6/generic/bin/python palm_wdd cancel " + jobname], |
|---|
| 508 | shell=False, |
|---|
| 509 | stdout=sub.PIPE, |
|---|
| 510 | stderr=sub.PIPE) |
|---|
| 511 | result = ssh.stdout.readlines() |
|---|
| 512 | result = filter(None, result) |
|---|
| 513 | |
|---|
| 514 | # In case of error display a warning message |
|---|
| 515 | if ( len(result) == 0 ): |
|---|
| 516 | error = ssh.stderr.readlines() |
|---|
| 517 | notify = QtGui.QMessageBox.warning(self,'Cancel job',"Error. Could not cancel job " + jobname + ".\n\nError message:\n" + ''.join(error)) |
|---|
| 518 | |
|---|
| 519 | # Otherwise inform the user and mark the job in the table |
|---|
| 520 | else: |
|---|
| 521 | self.table.setItem(self.table.currentRow(),3,QtGui.QTableWidgetItem("Canceled")) |
|---|
| 522 | notify = QtGui.QMessageBox.information(self,'Cancel job',"Job" + jobname + " canceled!\n\nServer message:\n" + ''.join(result)) |
|---|
| 523 | |
|---|
| 524 | |
|---|
| 525 | # Show detailed information on job. For documentation see above |
|---|
| 526 | def ShowDetails(self): |
|---|
| 527 | |
|---|
| 528 | jobname = self.table.item(self.table.currentRow(),6).text() |
|---|
| 529 | descr = self.table.item(self.table.currentRow(),1).text() |
|---|
| 530 | for h in range(0,len(description)): |
|---|
| 531 | if ( descr == description[h] ): |
|---|
| 532 | host = username[h] + "@" + hostname[h] |
|---|
| 533 | |
|---|
| 534 | ssh = sub.Popen(["ssh", "%s" % host, "/sw/tools/python/2.7.6/generic/bin/python palm_wdd check " + jobname], |
|---|
| 535 | shell=False, |
|---|
| 536 | stdout=sub.PIPE, |
|---|
| 537 | stderr=sub.PIPE) |
|---|
| 538 | result = ssh.stdout.readlines() |
|---|
| 539 | result = filter(None, result) |
|---|
| 540 | |
|---|
| 541 | if ( len(result) == 0 ): |
|---|
| 542 | error = ssh.stderr.readlines() |
|---|
| 543 | notify = QtGui.QMessageBox.warning(self,'Show job details',"Error. Could not fetch job details for " + jobname + ".\n\nError message:\n" + ''.join(error)) |
|---|
| 544 | else: |
|---|
| 545 | notify = QtGui.QMessageBox.information(self,'Job details',"Details for job " + jobname + ":\n\n" + ''.join(result)) |
|---|
| 546 | |
|---|
| 547 | |
|---|
| 548 | # Perform a forced stop on the job |
|---|
| 549 | def DoStop(self): |
|---|
| 550 | |
|---|
| 551 | # Return internal jobname |
|---|
| 552 | jobname = self.table.item(self.table.currentRow(),6).text() |
|---|
| 553 | |
|---|
| 554 | # Check description of job in order to login on the correct host |
|---|
| 555 | descr = self.table.item(self.table.currentRow(),1).text() |
|---|
| 556 | for h in range(0,len(description)): |
|---|
| 557 | if ( descr == description[h] ): |
|---|
| 558 | host = username[h] + "@" + hostname[h] |
|---|
| 559 | user = username[h] |
|---|
| 560 | |
|---|
| 561 | ssh = sub.Popen(["ssh", "%s" % host, "/sw/tools/python/2.7.6/generic/bin/python palm_wdd stop " + user + " " + jobname], |
|---|
| 562 | shell=False, |
|---|
| 563 | stdout=sub.PIPE, |
|---|
| 564 | stderr=sub.PIPE) |
|---|
| 565 | result = ssh.stdout.readlines() |
|---|
| 566 | result = filter(None, result) |
|---|
| 567 | |
|---|
| 568 | # In case of error display a warning message |
|---|
| 569 | if ( len(result) == 0 ): |
|---|
| 570 | error = ssh.stderr.readlines() |
|---|
| 571 | notify = QtGui.QMessageBox.warning(self,'Proper termination of job',"Error. Could not stop job " + jobname + ".\n\nError message:\n" + ''.join(error)) |
|---|
| 572 | |
|---|
| 573 | # Otherwise inform the user and mark the job in the table |
|---|
| 574 | else: |
|---|
| 575 | self.table.setItem(self.table.currentRow(),3,QtGui.QTableWidgetItem("Terminated")) |
|---|
| 576 | notify = QtGui.QMessageBox.information(self,'Terminatie job',"Termination of job " + jobname + " was initiated!\n\nServer message:\n" + ''.join(result)) |
|---|
| 577 | |
|---|
| 578 | |
|---|
| 579 | # Perform a forced stop on the job |
|---|
| 580 | def DoRestart(self): |
|---|
| 581 | |
|---|
| 582 | # Return internal jobname |
|---|
| 583 | jobname = self.table.item(self.table.currentRow(),6).text() |
|---|
| 584 | |
|---|
| 585 | # Check description of job in order to login on the correct host |
|---|
| 586 | descr = self.table.item(self.table.currentRow(),1).text() |
|---|
| 587 | for h in range(0,len(description)): |
|---|
| 588 | if ( descr == description[h] ): |
|---|
| 589 | host = username[h] + "@" + hostname[h] |
|---|
| 590 | user = username[h] |
|---|
| 591 | |
|---|
| 592 | ssh = sub.Popen(["ssh", "%s" % host, "/sw/tools/python/2.7.6/generic/bin/python palm_wdd restart " + user + " " + jobname], |
|---|
| 593 | shell=False, |
|---|
| 594 | stdout=sub.PIPE, |
|---|
| 595 | stderr=sub.PIPE) |
|---|
| 596 | result = ssh.stdout.readlines() |
|---|
| 597 | result = filter(None, result) |
|---|
| 598 | |
|---|
| 599 | # In case of error display a warning message |
|---|
| 600 | if ( len(result) == 0 ): |
|---|
| 601 | error = ssh.stderr.readlines() |
|---|
| 602 | notify = QtGui.QMessageBox.warning(self,'Proper termination of job',"Error. Could not stop job " + jobname + ".\n\nError message:\n" + ''.join(error)) |
|---|
| 603 | |
|---|
| 604 | # Otherwise inform the user and mark the job in the table |
|---|
| 605 | else: |
|---|
| 606 | self.table.setItem(self.table.currentRow(),3,QtGui.QTableWidgetItem("Terminated")) |
|---|
| 607 | notify = QtGui.QMessageBox.information(self,'Terminate job for restart',"Restart for job" + jobname + " was initiated!\n\nServer message:\n" + ''.join(result)) |
|---|
| 608 | |
|---|
| 609 | |
|---|
| 610 | # Remove a job from list - removes the current row from the table and from |
|---|
| 611 | # save file |
|---|
| 612 | def RemoveFromList(self, row): |
|---|
| 613 | if ( row == -1 ): |
|---|
| 614 | row = self.table.currentRow() |
|---|
| 615 | |
|---|
| 616 | # Read data from save file |
|---|
| 617 | job_to_delete = self.table.item(row,6).text() |
|---|
| 618 | self.table.removeRow(row) |
|---|
| 619 | file = open(palm_dir + '/.wd.olddata', 'r') |
|---|
| 620 | result = file.readlines() |
|---|
| 621 | result = filter(None, result) |
|---|
| 622 | file.close() |
|---|
| 623 | file = open(palm_dir + '/.wd.olddata', 'w') |
|---|
| 624 | |
|---|
| 625 | job_data_old = [] |
|---|
| 626 | |
|---|
| 627 | if ( len(result) == 0 ): |
|---|
| 628 | notify = QtGui.QMessageBox.warning(self,'Read from .wd.olddata',"Error message:\n\nCould not read from file. I/O error.") |
|---|
| 629 | else: |
|---|
| 630 | # Save data in array job_data_old |
|---|
| 631 | for j in range(0,len(result)): |
|---|
| 632 | job_data_old.append(j) |
|---|
| 633 | job_data_old[j] = result[j].split(" ") |
|---|
| 634 | job_data_old[j] = filter(None, job_data_old[j]) |
|---|
| 635 | job_data_old[j] = [line.rstrip('\n') for line in job_data_old[j]] |
|---|
| 636 | |
|---|
| 637 | # Check if line j is the selected job, if not -> save to file |
|---|
| 638 | if ( job_data_old[j][6] != job_to_delete ): |
|---|
| 639 | printstring = str(job_data_old[j][0]) + " " + \ |
|---|
| 640 | str(job_data_old[j][1]) + " " + \ |
|---|
| 641 | str(job_data_old[j][2]) + " " + \ |
|---|
| 642 | str(job_data_old[j][3]) + " " + \ |
|---|
| 643 | str(job_data_old[j][4]) + " " + \ |
|---|
| 644 | str(job_data_old[j][5]) + " " + \ |
|---|
| 645 | str(job_data_old[j][6]) + " " + \ |
|---|
| 646 | str(job_data_old[j][7]) + "\n" |
|---|
| 647 | file.write(printstring) |
|---|
| 648 | |
|---|
| 649 | file.close() |
|---|
| 650 | |
|---|
| 651 | # Remove all completed jobs from list |
|---|
| 652 | def ClearList(self): |
|---|
| 653 | |
|---|
| 654 | num_of_lines = self.table.rowCount() |
|---|
| 655 | |
|---|
| 656 | # Delete all lines with completed/canceled jobs from list. The counter |
|---|
| 657 | # must decrease as the line numbers would be messed up otherwise |
|---|
| 658 | for j in range(num_of_lines-1,-1,-1): |
|---|
| 659 | state = self.table.item(j,3).text() |
|---|
| 660 | if ( state == "Completed" or state == "Canceled" or state == "Terminated"): |
|---|
| 661 | self.RemoveFromList(j) |
|---|
| 662 | |
|---|
| 663 | # Update the label |
|---|
| 664 | def UpdateLabel(self): |
|---|
| 665 | remaining_time = (update_frequency - self.timetimer.elapsed()) / 1000 / 60 |
|---|
| 666 | self.label.setText("Next update in " + str(remaining_time) + " min") |
|---|
| 667 | |
|---|
| 668 | |
|---|
| 669 | # Main loop |
|---|
| 670 | def main(): |
|---|
| 671 | |
|---|
| 672 | app = QtGui.QApplication(sys.argv) |
|---|
| 673 | res = Watchdog() |
|---|
| 674 | sys.exit(app.exec_()) |
|---|
| 675 | |
|---|
| 676 | |
|---|
| 677 | if __name__ == '__main__': |
|---|
| 678 | main() |
|---|
