001/* 002 * #%L 003 * Netarchivesuite - harvester 004 * %% 005 * Copyright (C) 2005 - 2014 The Royal Danish Library, the Danish State and University Library, 006 * the National Library of France and the Austrian National Library. 007 * %% 008 * This program is free software: you can redistribute it and/or modify 009 * it under the terms of the GNU Lesser General Public License as 010 * published by the Free Software Foundation, either version 2.1 of the 011 * License, or (at your option) any later version. 012 * 013 * This program is distributed in the hope that it will be useful, 014 * but WITHOUT ANY WARRANTY; without even the implied warranty of 015 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 016 * GNU General Lesser Public License for more details. 017 * 018 * You should have received a copy of the GNU General Lesser Public 019 * License along with this program. If not, see 020 * <http://www.gnu.org/licenses/lgpl-2.1.html>. 021 * #L% 022 */ 023package dk.netarkivet.harvester.heritrix3; 024 025import java.io.File; 026import java.util.List; 027 028import org.slf4j.Logger; 029import org.slf4j.LoggerFactory; 030 031import dk.netarkivet.common.CommonSettings; 032import dk.netarkivet.common.distribute.ChannelID; 033import dk.netarkivet.common.distribute.JMSConnection; 034import dk.netarkivet.common.distribute.JMSConnectionFactory; 035import dk.netarkivet.common.exceptions.ArgumentNotValid; 036import dk.netarkivet.common.exceptions.IOFailure; 037import dk.netarkivet.common.exceptions.PermissionDenied; 038import dk.netarkivet.common.exceptions.UnknownID; 039import dk.netarkivet.common.lifecycle.PeriodicTaskExecutor; 040import dk.netarkivet.common.utils.ApplicationUtils; 041import dk.netarkivet.common.utils.CleanupIF; 042import dk.netarkivet.common.utils.DomainUtils; 043import dk.netarkivet.common.utils.FileUtils; 044import dk.netarkivet.common.utils.NotificationType; 045import dk.netarkivet.common.utils.NotificationsFactory; 046import dk.netarkivet.common.utils.Settings; 047import dk.netarkivet.common.utils.SystemUtils; 048import dk.netarkivet.common.utils.TimeUtils; 049import dk.netarkivet.harvester.HarvesterSettings; 050import dk.netarkivet.harvester.datamodel.HarvestDefinitionInfo; 051import dk.netarkivet.harvester.datamodel.Job; 052import dk.netarkivet.harvester.datamodel.JobStatus; 053import dk.netarkivet.harvester.distribute.HarvesterChannels; 054import dk.netarkivet.harvester.distribute.HarvesterMessageHandler; 055import dk.netarkivet.harvester.harvesting.distribute.CrawlStatusMessage; 056import dk.netarkivet.harvester.harvesting.distribute.DoOneCrawlMessage; 057import dk.netarkivet.harvester.harvesting.distribute.HarvesterReadyMessage; 058import dk.netarkivet.harvester.harvesting.distribute.HarvesterRegistrationRequest; 059import dk.netarkivet.harvester.harvesting.distribute.HarvesterRegistrationResponse; 060import dk.netarkivet.harvester.harvesting.metadata.MetadataEntry; 061 062/** 063 * This class responds to JMS doOneCrawl messages from the HarvestScheduler and launches a Heritrix crawl with the 064 * received job description. The generated ARC files are uploaded to the bitarchives once a harvest job has been 065 * completed. 066 * <p> 067 * During its operation CrawlStatus messages are sent to the HarvestSchedulerMonitorServer. When starting the actual 068 * harvesting a message is sent with status 'STARTED'. When the harvesting has finished a message is sent with either 069 * status 'DONE' or 'FAILED'. Either a 'DONE' or 'FAILED' message with result should ALWAYS be sent if at all possible, 070 * but only ever one such message per job. 071 * <p> 072 * It is necessary to be able to run the Heritrix harvester on several machines and several processes on each machine. 073 * Each instance of Heritrix is started and monitored by a HarvestControllerServer. 074 * <p> 075 * Initially, all directories under serverdir are scanned for harvestinfo files. If any are found, they are parsed for 076 * information, and all remaining files are attempted uploaded to the bitarchive. It will then send back a 077 * CrawlStatusMessage with status failed. 078 * <p> 079 * A new thread is started for each actual crawl, in which the JMS listener is removed. Threading is required since JMS 080 * will not let the called thread remove the listener that's being handled. 081 * <p> 082 * After a harvestjob has been terminated, either successfully or unsuccessfully, the serverdir is again scanned for 083 * harvestInfo files to attempt upload of files not yet uploaded. Then it begins to listen again after new jobs, if 084 * there is enough room available on the machine. If not, it logs a warning about this, which is also sent as a 085 * notification. 086 */ 087public class HarvestControllerServer extends HarvesterMessageHandler implements CleanupIF { 088 089 /** The logger to use. */ 090 private static final Logger log = LoggerFactory.getLogger(HarvestControllerServer.class); 091 092 /** The unique instance of this class. */ 093 private static HarvestControllerServer instance; 094 095 /** The configured application instance id. @see CommonSettings#APPLICATION_INSTANCE_ID */ 096 private final String applicationInstanceId = Settings.get(CommonSettings.APPLICATION_INSTANCE_ID); 097 098 /** The name of the server this <code>HarvestControllerServer</code> is running on. */ 099 private final String physicalServerName = DomainUtils.reduceHostname(SystemUtils.getLocalHostName()); 100 101 /** Min. space required to start a job. */ 102 private final long minSpaceRequired; 103 104 /** The JMSConnection to use. */ 105 private JMSConnection jmsConnection; 106 107 /** The JMS channel on which to listen for {@link HarvesterRegistrationResponse}s. */ 108 public static final ChannelID HARVEST_CHAN_VALID_RESP_ID = HarvesterChannels 109 .getHarvesterRegistrationResponseChannel(); 110 111 private final PostProcessing postProcessing; 112 113 /** The CHANNEL of jobs processed by this instance. */ 114 private static final String CHANNEL = Settings.get(HarvesterSettings.HARVEST_CONTROLLER_CHANNEL); 115 116 /** Jobs are fetched from this queue. */ 117 private ChannelID jobChannel; 118 119 /** the serverdir, where the harvesting takes place. */ 120 private final File serverDir; 121 122 /** This is true while a doOneCrawl is running. No jobs are accepted while this is running. */ 123 private CrawlStatus status; 124 125 /** 126 * Returns or creates the unique instance of this singleton The server creates an instance of the HarvestController, 127 * uploads arc-files from unfinished harvests, and starts to listen to JMS messages on the incoming jms queues. 128 * @return The instance 129 * @throws PermissionDenied If the serverdir or oldjobsdir can't be created 130 * @throws IOFailure if data from old harvests exist, but contain illegal data 131 */ 132 public static synchronized HarvestControllerServer getInstance() throws IOFailure { 133 if (instance == null) { 134 instance = new HarvestControllerServer(); 135 } 136 return instance; 137 } 138 139 /** 140 * In this constructor, the server creates an instance of the HarvestController, uploads any arc-files from 141 * incomplete harvests. Then it starts listening for new doOneCrawl messages, unless there is no available space. In 142 * that case, it sends a notification to the administrator and pauses. 143 * @throws PermissionDenied If the serverdir or oldjobsdir can't be created. 144 * @throws IOFailure If harvestInfoFile contains invalid data. 145 * @throws UnknownID if the settings file does not specify a valid queue priority. 146 */ 147 private HarvestControllerServer() throws IOFailure { 148 log.info("Starting HarvestControllerServer."); 149 log.info("Bound to harvest channel '{}'", CHANNEL); 150 151 // Make sure serverdir (where active crawl-dirs live) and oldJobsDir 152 // (where old crawl dirs are stored) exist. 153 serverDir = new File(Settings.get(HarvesterSettings.HARVEST_CONTROLLER_SERVERDIR)); 154 ApplicationUtils.dirMustExist(serverDir); 155 log.info("Serverdir: '{}'", serverDir); 156 minSpaceRequired = Settings.getLong(HarvesterSettings.HARVEST_SERVERDIR_MINSPACE); 157 if (minSpaceRequired <= 0L) { 158 log.warn("Wrong setting of minSpaceLeft read from Settings: {}", minSpaceRequired); 159 throw new ArgumentNotValid("Wrong setting of minSpaceLeft read from Settings: " + minSpaceRequired); 160 } 161 log.info("Harvesting requires at least {} bytes free.", minSpaceRequired); 162 163 // Get JMS-connection 164 // Channel THIS_CLIENT is only used for replies to store messages so 165 // do not set as listener here. It is registered in the arcrepository 166 // client. 167 // Channel ANY_xxxPRIORIRY_HACO is used for listening for jobs, and 168 // registered below. 169 170 jmsConnection = JMSConnectionFactory.getInstance(); 171 postProcessing = PostProcessing.getInstance(jmsConnection); 172 log.debug("Obtained JMS connection."); 173 174 status = new CrawlStatus(); 175 176 // If any unprocessed jobs are left on the server, process them now 177 postProcessing.processOldJobs(); 178 179 // Register for listening to harvest channel validity responses 180 JMSConnectionFactory.getInstance().setListener(HARVEST_CHAN_VALID_RESP_ID, this); 181 182 // Ask if the channel this harvester is assigned to is valid 183 jmsConnection.send(new HarvesterRegistrationRequest(HarvestControllerServer.CHANNEL, applicationInstanceId)); 184 log.info("Requested to check the validity of harvest channel '{}'", HarvestControllerServer.CHANNEL); 185 } 186 187 /** 188 * Release all jms connections. Close the Controller 189 */ 190 public synchronized void close() { 191 log.info("Closing HarvestControllerServer."); 192 cleanup(); 193 log.info("Closed down HarvestControllerServer"); 194 } 195 196 /** 197 * Will be called on shutdown. 198 * 199 * @see CleanupIF#cleanup() 200 */ 201 public void cleanup() { 202 if (jmsConnection != null) { 203 jmsConnection.removeListener(HARVEST_CHAN_VALID_RESP_ID, this); 204 if (jobChannel != null) { 205 jmsConnection.removeListener(jobChannel, this); 206 } 207 } 208 // Stop the sending of status messages 209 status.stopSending(); 210 instance = null; 211 } 212 213 @Override 214 public void visit(HarvesterRegistrationResponse msg) { 215 // If we have already started or the message notifies for another channel, resend it. 216 String channelName = msg.getHarvestChannelName(); 217 if (status.isChannelValid() || !CHANNEL.equals(channelName)) { 218 // Controller has already started 219 jmsConnection.resend(msg, msg.getTo()); 220 if (log.isTraceEnabled()) { 221 log.trace("Resending harvest channel validity message for channel '{}'", channelName); 222 } 223 return; 224 } 225 226 if (!msg.isValid()) { 227 log.error("Received message stating that channel '{}' is invalid. Will stop. " 228 + "Probable cause: the channel is not one of the known channels stored in the channels table", channelName); 229 close(); 230 return; 231 } 232 233 log.info("Received message stating that channel '{}' is valid.", channelName); 234 // Environment and connections are now ready for processing of messages 235 jobChannel = HarvesterChannels.getHarvestJobChannelId(channelName, msg.isSnapshot()); 236 237 // Only listen for harvester jobs if enough available space 238 beginListeningIfSpaceAvailable(); 239 240 // Notify the harvest dispatcher that we are ready 241 startAcceptingJobs(); 242 status.startSending(); 243 } 244 245 /** Start listening for crawls, if space available. */ 246 private void beginListeningIfSpaceAvailable() { 247 long availableSpace = FileUtils.getBytesFree(serverDir); 248 if (availableSpace > minSpaceRequired) { 249 log.info("Starts to listen to new jobs on queue '{}'", jobChannel); 250 jmsConnection.setListener(jobChannel, this); 251 } else { 252 String pausedMessage = "Not enough available diskspace. Only " + availableSpace + " bytes available." 253 + " Harvester is paused."; 254 log.error(pausedMessage); 255 NotificationsFactory.getInstance().notify(pausedMessage, NotificationType.ERROR); 256 } 257 } 258 259 /** 260 * Start listening for new crawl requests again. This actually doesn't re-add a listener, but the listener only gets 261 * removed when we're so far committed that we're going to exit at the end. So to start accepting jobs again, we 262 * stop resending messages we get. 263 */ 264 private synchronized void startAcceptingJobs() { 265 // allow this harvestControllerServer to receive messages again 266 status.setRunning(false); 267 } 268 269 /** 270 * Stop accepting more jobs. After this is called, all crawl messages received will be resent to the queue. A bit 271 * further down, we will stop listening altogether, but that requires another thread. 272 */ 273 private synchronized void stopAcceptingJobs() { 274 status.setRunning(true); 275 log.debug("No longer accepting jobs."); 276 } 277 278 /** 279 * Stop listening for new crawl requests. 280 */ 281 private void removeListener() { 282 log.debug("Removing listener on CHANNEL '{}'", jobChannel); 283 jmsConnection.removeListener(jobChannel, this); 284 } 285 286 /** 287 * Checks that we're available to do a crawl, and if so, marks us as unavailable, checks that the job message is 288 * well-formed, and starts the thread that the crawl happens in. If an error occurs starting the crawl, we will 289 * start listening for messages again. 290 * <p> 291 * The sequence of actions involved in a crawl are:</br> 1. If we are already running, resend the job to the queue 292 * and return</br> 2. Check the job for validity</br> 3. Send a CrawlStatus message that crawl has STARTED</br> In a 293 * separate thread:</br> 4. Unregister this HACO as listener</br> 5. Create a new crawldir (based on the JobID and a 294 * timestamp)</br> 6. Write a harvestInfoFile (using JobID and crawldir) and metadata</br> 7. Instantiate a new 295 * HeritrixLauncher</br> 8. Start a crawl</br> 9. Store the generated arc-files and metadata in the known 296 * bit-archives </br>10. _Always_ send CrawlStatus DONE or FAILED</br> 11. Move crawldir into oldJobs dir</br> 297 * 298 * @param msg The crawl job 299 * @throws IOFailure On trouble harvesting, uploading or processing harvestInfo 300 * @throws UnknownID if jobID is null in the message 301 * @throws ArgumentNotValid if the status of the job is not valid - must be SUBMITTED 302 * @throws PermissionDenied if the crawldir can't be created 303 * @see #visit(DoOneCrawlMessage) for more details 304 */ 305 public void visit(DoOneCrawlMessage msg) throws IOFailure, UnknownID, ArgumentNotValid, PermissionDenied { 306 // Only one doOneCrawl at a time. Returning should almost never happen, 307 // since we deregister the listener, but we may receive another message 308 // before the listener is removed. Also, if the job message is 309 // malformed or starting the crawl fails, we re-add the listener. 310 synchronized (this) { 311 if (status.isRunning()) { 312 log.warn( 313 "Received crawl request, but sent it back to queue, as another crawl is already running: '{}'", 314 msg); 315 jmsConnection.resend(msg, jobChannel); 316 try { 317 // Wait a second before listening again, so the message has 318 // a chance of getting snatched by another harvester. 319 Thread.sleep(TimeUtils.SECOND_IN_MILLIS); 320 } catch (InterruptedException e) { 321 // Since we're not waiting for anything critical, we can 322 // ignore this exception. 323 } 324 return; 325 } 326 stopAcceptingJobs(); 327 } 328 329 Thread t = null; 330 331 // This 'try' matches a finally that restores running=false if we don't 332 // start a crawl after all 333 try { 334 final Job job = msg.getJob(); 335 336 // Every job must have an ID or we can do nothing with it, not even 337 // send a proper failure message back. 338 Long jobID = job.getJobID(); 339 if (jobID == null) { 340 log.warn("DoOneCrawlMessage arrived without JobID: '{}'", msg.toString()); 341 throw new UnknownID("DoOneCrawlMessage arrived without JobID"); 342 } 343 344 log.info("Received crawlrequest for job {}: '{}'", jobID, msg); 345 346 // Send message to scheduler that job is started 347 CrawlStatusMessage csmStarted = new CrawlStatusMessage(jobID, JobStatus.STARTED); 348 jmsConnection.send(csmStarted); 349 350 // Jobs should arrive with status "submitted". If this is not the 351 // case, log the error and send a job-failed message back. 352 // HarvestScheduler likes getting a STARTED message before 353 // FAILED, so we oblige it here. 354 if (job.getStatus() != JobStatus.SUBMITTED) { 355 String message = "Message '" + msg.toString() + "' arrived with" + " status " + job.getStatus() 356 + " for job " + jobID + ", should have been STATUS_SUBMITTED"; 357 log.warn(message); 358 sendErrorMessage(jobID, message, message); 359 throw new ArgumentNotValid(message); 360 } 361 362 final List<MetadataEntry> metadataEntries = msg.getMetadata(); 363 364 Thread t1; 365 // Create thread in which harvesting will occur 366 t1 = new HarvesterThread(job, msg.getOrigHarvestInfo(), metadataEntries); 367 // start thread which will remove this listener, harvest, store, and 368 // exit the VM 369 t1.start(); 370 log.info("Started harvester thread for job {}", jobID); 371 // We delay assigning the thread variable until start() has 372 // succeeded. Thus, if start() fails, we will resume accepting 373 // jobs. 374 t = t1; 375 } finally { 376 // If we didn't start a thread for crawling after all, accept more 377 // messages 378 if (t == null) { 379 startAcceptingJobs(); 380 } 381 } 382 // Now return from this method letting the thread do the work. 383 // This is important as it allows us to receive upload-replies from 384 // THIS_CLIENT in the crawl thread. 385 } 386 387 /** 388 * Sends a CrawlStatusMessage for a failed job with the given short message and detailed message. 389 * 390 * @param jobID ID of the job that failed 391 * @param message A short message indicating what went wrong 392 * @param detailedMessage A more detailed message detailing why it went wrong. 393 */ 394 public void sendErrorMessage(long jobID, String message, String detailedMessage) { 395 CrawlStatusMessage csm = new CrawlStatusMessage(jobID, JobStatus.FAILED, null); 396 csm.setHarvestErrors(message); 397 csm.setHarvestErrorDetails(detailedMessage); 398 jmsConnection.send(csm); 399 } 400 401 /** 402 * A Thread class for the actual harvesting. This is required in order to stop listening while we're busy 403 * harvesting, since JMS doesn't allow the called thread to remove the listener that was called. 404 */ 405 private class HarvesterThread extends Thread { 406 407 /** The harvester Job in this thread. */ 408 private final Job job; 409 410 /** Stores documentary information about the harvest. */ 411 private final HarvestDefinitionInfo origHarvestInfo; 412 413 /** The list of metadata associated with this Job. */ 414 private final List<MetadataEntry> metadataEntries; 415 416 /** 417 * Constructor for the HarvesterThread class. 418 * 419 * @param job a harvesting job 420 * @param origHarvestInfo Info about the harvestdefinition that scheduled this job 421 * @param metadataEntries metadata associated with the given job 422 */ 423 public HarvesterThread(Job job, HarvestDefinitionInfo origHarvestInfo, List<MetadataEntry> metadataEntries) { 424 this.job = job; 425 this.origHarvestInfo = origHarvestInfo; 426 this.metadataEntries = metadataEntries; 427 } 428 429 /** 430 * The thread body for the harvester thread. Removes the JMS listener, sets up the files for Heritrix, then 431 * passes control to the HarvestController to perform the actual harvest. 432 * <p> 433 * 434 * @throws PermissionDenied if we cannot create the crawl directory. 435 * @throws IOFailure if there are problems preparing or running the crawl. 436 */ 437 public void run() { 438 try { 439 // We must remove the listener inside a thread, 440 // as JMS doesn't allow us to remove it within the 441 // call it made. 442 removeListener(); 443 444 HarvestJob harvestJob = new HarvestJob(instance); 445 harvestJob.init(job, origHarvestInfo, metadataEntries); 446 Heritrix3Files files = harvestJob.getHeritrix3Files(); 447 448 Throwable crawlException = null; 449 try { 450 harvestJob.runHarvest(); 451 } catch (Throwable e) { 452 String msg = "Error during crawling. The crawl may have been only partially completed."; 453 log.warn(msg, e); 454 crawlException = e; 455 throw new IOFailure(msg, e); 456 } finally { 457 postProcessing.doPostProcessing(files.getCrawlDir(), crawlException); 458 } 459 } catch (Throwable t) { 460 String msg = "Fatal error while operating job '" + job + "'"; 461 log.error(msg, t); 462 NotificationsFactory.getInstance().notify(msg, NotificationType.ERROR, t); 463 } finally { 464 log.info("Ending crawl of job : {}", job.getJobID()); 465 // process serverdir for files not yet uploaded. 466 postProcessing.processOldJobs(); 467 shutdownNowOrContinue(); 468 startAcceptingJobs(); 469 beginListeningIfSpaceAvailable(); 470 } 471 } 472 473 /** 474 * Does the operator want us to shutdown now. TODO In a later implementation, the harvestControllerServer could 475 * be notified over JMX. Now we just look for a "shutdown.txt" file in the HARVEST_CONTROLLER_SERVERDIR 476 */ 477 private void shutdownNowOrContinue() { 478 File shutdownFile = new File(serverDir, "shutdown.txt"); 479 if (shutdownFile.exists()) { 480 log.info("Found shutdown-file in serverdir - " + "shutting down the application"); 481 instance.cleanup(); 482 System.exit(0); 483 } 484 } 485 } 486 487 /** 488 * Used for maintaining the running status of the crawling, is it running or not. Will also take care of notifying 489 * the HarvestJobManager of the status. 490 */ 491 private class CrawlStatus implements Runnable { 492 493 /** The status. */ 494 private Boolean running = false; 495 496 private boolean channelIsValid = false; 497 498 /** Handles the periodic sending of status messages. */ 499 private PeriodicTaskExecutor statusTransmitter; 500 501 private final int SEND_READY_DELAY = Settings.getInt(HarvesterSettings.SEND_READY_DELAY); 502 503 /** 504 * Starts the sending of status messages. 505 */ 506 public void startSending() { 507 this.channelIsValid = true; 508 statusTransmitter = new PeriodicTaskExecutor("HarvesterStatus", this, 0, 509 Settings.getInt(HarvesterSettings.SEND_READY_INTERVAL)); 510 } 511 512 /** 513 * Stops the sending of status messages. 514 */ 515 public void stopSending() { 516 if (statusTransmitter != null) { 517 statusTransmitter.shutdown(); 518 statusTransmitter = null; 519 } 520 } 521 522 /** 523 * Returns <code>true</code> if the a doOneCrawl is running, else <code>false</code>. 524 * @return Whether a crawl running. 525 */ 526 public boolean isRunning() { 527 return running; 528 } 529 530 /** 531 * Use for changing the running state. 532 * @param running The new status 533 */ 534 public void setRunning(boolean running) { 535 this.running = running; 536 } 537 538 /** 539 * @return the channelIsValid 540 */ 541 protected final boolean isChannelValid() { 542 return channelIsValid; 543 } 544 545 @Override 546 public void run() { 547 try { 548 Thread.sleep(SEND_READY_DELAY); 549 } catch (Exception e) { 550 log.error("Unable to sleep", e); 551 } 552 if (!running) { 553 jmsConnection.send(new HarvesterReadyMessage(applicationInstanceId + " on " + physicalServerName, 554 HarvestControllerServer.CHANNEL)); 555 } 556 } 557 558 } 559 560}