1
|
/**
|
2
|
* '$RCSfile$'
|
3
|
* Purpose: Implements a service for managing a Hazelcast cluster member
|
4
|
* Copyright: 2011 Regents of the University of California and the
|
5
|
* National Center for Ecological Analysis and Synthesis
|
6
|
* Authors: Christopher Jones
|
7
|
*
|
8
|
* '$Author: tao $'
|
9
|
* '$Date: 2015-02-12 15:32:36 -0800 (Thu, 12 Feb 2015) $'
|
10
|
* '$Revision: 9116 $'
|
11
|
*
|
12
|
* This program is free software; you can redistribute it and/or modify
|
13
|
* it under the terms of the GNU General Public License as published by
|
14
|
* the Free Software Foundation; either version 2 of the License, or
|
15
|
* (at your option) any later version.
|
16
|
*
|
17
|
* This program is distributed in the hope that it will be useful,
|
18
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
19
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
20
|
* GNU General Public License for more details.
|
21
|
*
|
22
|
* You should have received a copy of the GNU General Public License
|
23
|
* along with this program; if not, write to the Free Software
|
24
|
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
|
25
|
*/
|
26
|
|
27
|
package edu.ucsb.nceas.metacat.dataone.hazelcast;
|
28
|
|
29
|
import java.io.FileNotFoundException;
|
30
|
import java.sql.SQLException;
|
31
|
import java.text.DateFormat;
|
32
|
import java.text.SimpleDateFormat;
|
33
|
import java.util.Calendar;
|
34
|
import java.util.HashSet;
|
35
|
import java.util.Iterator;
|
36
|
import java.util.List;
|
37
|
import java.util.Set;
|
38
|
import java.util.concurrent.ExecutorService;
|
39
|
import java.util.concurrent.Executors;
|
40
|
import java.util.concurrent.locks.Lock;
|
41
|
|
42
|
import org.apache.log4j.Logger;
|
43
|
import org.dataone.service.exceptions.InvalidSystemMetadata;
|
44
|
import org.dataone.service.types.v1.Identifier;
|
45
|
import org.dataone.service.types.v2.SystemMetadata;
|
46
|
|
47
|
import com.hazelcast.config.Config;
|
48
|
import com.hazelcast.config.FileSystemXmlConfig;
|
49
|
import com.hazelcast.core.EntryEvent;
|
50
|
import com.hazelcast.core.EntryListener;
|
51
|
import com.hazelcast.core.Hazelcast;
|
52
|
import com.hazelcast.core.HazelcastInstance;
|
53
|
import com.hazelcast.core.ILock;
|
54
|
import com.hazelcast.core.IMap;
|
55
|
import com.hazelcast.core.ISet;
|
56
|
import com.hazelcast.core.ItemEvent;
|
57
|
import com.hazelcast.core.ItemListener;
|
58
|
import com.hazelcast.core.LifecycleEvent;
|
59
|
import com.hazelcast.core.LifecycleListener;
|
60
|
import com.hazelcast.core.Member;
|
61
|
import com.hazelcast.core.MembershipEvent;
|
62
|
import com.hazelcast.core.MembershipListener;
|
63
|
import com.hazelcast.partition.Partition;
|
64
|
import com.hazelcast.partition.PartitionService;
|
65
|
|
66
|
import edu.ucsb.nceas.metacat.IdentifierManager;
|
67
|
import edu.ucsb.nceas.metacat.McdbDocNotFoundException;
|
68
|
import edu.ucsb.nceas.metacat.common.index.IndexTask;
|
69
|
import edu.ucsb.nceas.metacat.common.index.event.IndexEvent;
|
70
|
import edu.ucsb.nceas.metacat.properties.PropertyService;
|
71
|
import edu.ucsb.nceas.metacat.shared.BaseService;
|
72
|
import edu.ucsb.nceas.metacat.shared.ServiceException;
|
73
|
import edu.ucsb.nceas.metacat.util.DocumentUtil;
|
74
|
import edu.ucsb.nceas.utilities.FileUtil;
|
75
|
import edu.ucsb.nceas.utilities.PropertyNotFoundException;
|
76
|
/**
|
77
|
* The Hazelcast service enables Metacat as a Hazelcast cluster member
|
78
|
*/
|
79
|
public class HazelcastService extends BaseService
|
80
|
implements EntryListener<Identifier, SystemMetadata>, MembershipListener, LifecycleListener, ItemListener<Identifier> {
|
81
|
|
82
|
private static final String MISSING_PID_PREFIX = "missing-";
|
83
|
|
84
|
/* The instance of the logging class */
|
85
|
private static Logger logMetacat = Logger.getLogger(HazelcastService.class);
|
86
|
|
87
|
/* The singleton instance of the hazelcast service */
|
88
|
private static HazelcastService hzService = null;
|
89
|
|
90
|
/* The Hazelcast configuration */
|
91
|
private Config hzConfig;
|
92
|
|
93
|
/* The name of the system metadata map */
|
94
|
private String systemMetadataMap;
|
95
|
|
96
|
/* The Hazelcast distributed system metadata map */
|
97
|
private IMap<Identifier, SystemMetadata> systemMetadata;
|
98
|
|
99
|
/* The name of the identifiers set */
|
100
|
private String identifiersSet;
|
101
|
|
102
|
/* The Hazelcast distributed identifiers set */
|
103
|
private ISet<Identifier> identifiers;
|
104
|
|
105
|
/* The Hazelcast distributed missing identifiers set */
|
106
|
private ISet<Identifier> missingIdentifiers;
|
107
|
|
108
|
/* The Hazelcast distributed index queue */
|
109
|
private String hzIndexQueue;
|
110
|
private IMap<Identifier, IndexTask> indexQueue;
|
111
|
|
112
|
/* The Hazelcast distributed index event map */
|
113
|
private String hzIndexEventMap;
|
114
|
private IMap<Identifier, IndexEvent> indexEventMap;
|
115
|
|
116
|
private HazelcastInstance hzInstance;
|
117
|
|
118
|
/*
|
119
|
* Constructor: Creates an instance of the hazelcast service. Since
|
120
|
* this uses a singleton pattern, use getInstance() to gain the instance.
|
121
|
*/
|
122
|
private HazelcastService() {
|
123
|
|
124
|
super();
|
125
|
_serviceName="HazelcastService";
|
126
|
|
127
|
try {
|
128
|
init();
|
129
|
|
130
|
} catch (ServiceException se) {
|
131
|
logMetacat.error("There was a problem creating the HazelcastService. " +
|
132
|
"The error message was: " + se.getMessage());
|
133
|
|
134
|
}
|
135
|
|
136
|
}
|
137
|
|
138
|
/**
|
139
|
* Get the instance of the HazelcastService that has been instantiated,
|
140
|
* or instantiate one if it has not been already.
|
141
|
*
|
142
|
* @return hazelcastService - The instance of the hazelcast service
|
143
|
*/
|
144
|
public static HazelcastService getInstance(){
|
145
|
|
146
|
if ( hzService == null ) {
|
147
|
|
148
|
hzService = new HazelcastService();
|
149
|
|
150
|
}
|
151
|
return hzService;
|
152
|
}
|
153
|
|
154
|
/**
|
155
|
* Initializes the Hazelcast service
|
156
|
*/
|
157
|
public void init() throws ServiceException {
|
158
|
|
159
|
logMetacat.debug("HazelcastService.init() called.");
|
160
|
|
161
|
String configFileName = null;
|
162
|
try {
|
163
|
configFileName = PropertyService.getProperty("dataone.hazelcast.configFilePath");
|
164
|
hzConfig = new FileSystemXmlConfig(configFileName);
|
165
|
} catch (Exception e) {
|
166
|
configFileName = PropertyService.CONFIG_FILE_DIR + FileUtil.getFS() + "hazelcast.xml";
|
167
|
logMetacat.warn("Custom Hazelcast configuration not defined, using default: " + configFileName);
|
168
|
// make sure we have the config
|
169
|
try {
|
170
|
hzConfig = new FileSystemXmlConfig(configFileName);
|
171
|
} catch (FileNotFoundException e1) {
|
172
|
String msg = e.getMessage();
|
173
|
logMetacat.error(msg);
|
174
|
throw new ServiceException(msg);
|
175
|
}
|
176
|
}
|
177
|
|
178
|
this.hzInstance = Hazelcast.newHazelcastInstance(hzConfig);
|
179
|
|
180
|
logMetacat.debug("Initialized hzInstance");
|
181
|
|
182
|
// Get configuration properties on instantiation
|
183
|
try {
|
184
|
systemMetadataMap =
|
185
|
PropertyService.getProperty("dataone.hazelcast.storageCluster.systemMetadataMap");
|
186
|
identifiersSet = PropertyService.getProperty("dataone.hazelcast.storageCluster.identifiersSet");
|
187
|
|
188
|
// Get a reference to the shared system metadata map as a cluster member
|
189
|
// NOTE: this loads the map from the backing store and can take a long time for large collections
|
190
|
systemMetadata = this.hzInstance.getMap(systemMetadataMap);
|
191
|
|
192
|
logMetacat.debug("Initialized systemMetadata");
|
193
|
|
194
|
// Get a reference to the shared identifiers set as a cluster member
|
195
|
// NOTE: this takes a long time to complete
|
196
|
logMetacat.warn("Retrieving hzIdentifiers from Hazelcast");
|
197
|
identifiers = this.hzInstance.getSet(identifiersSet);
|
198
|
logMetacat.warn("Retrieved hzIdentifiers from Hazelcast");
|
199
|
|
200
|
// for publishing the "PIDs Wanted" list
|
201
|
missingIdentifiers = this.hzInstance.getSet("hzMissingIdentifiersSet");
|
202
|
|
203
|
missingIdentifiers.addItemListener(this, true);
|
204
|
|
205
|
// for index tasks
|
206
|
hzIndexQueue = PropertyService.getProperty("index.hazelcast.indexqueue");
|
207
|
indexQueue = this.hzInstance.getMap(hzIndexQueue);
|
208
|
|
209
|
// for index events (failures)
|
210
|
hzIndexEventMap = PropertyService.getProperty("index.hazelcast.indexeventmap");
|
211
|
indexEventMap = this.hzInstance.getMap(hzIndexEventMap);
|
212
|
|
213
|
// Listen for changes to the system metadata map
|
214
|
systemMetadata.addEntryListener(this, true);
|
215
|
|
216
|
// Listen for members added/removed
|
217
|
hzInstance.getCluster().addMembershipListener(this);
|
218
|
|
219
|
// Listen for lifecycle state changes
|
220
|
hzInstance.getLifecycleService().addLifecycleListener(this);
|
221
|
|
222
|
} catch (PropertyNotFoundException e) {
|
223
|
|
224
|
String msg = "Couldn't find Hazelcast properties for the DataONE clusters. " +
|
225
|
"The error message was: " + e.getMessage();
|
226
|
logMetacat.error(msg);
|
227
|
|
228
|
}
|
229
|
|
230
|
// make sure we have all metadata locally
|
231
|
try {
|
232
|
// synch on restart
|
233
|
resynchInThread();
|
234
|
} catch (Exception e) {
|
235
|
String msg = "Problem resynchronizing system metadata. " + e.getMessage();
|
236
|
logMetacat.error(msg, e);
|
237
|
}
|
238
|
|
239
|
}
|
240
|
|
241
|
/**
|
242
|
* Get the system metadata map
|
243
|
*
|
244
|
* @return systemMetadata - the hazelcast map of system metadata
|
245
|
* @param identifier - the identifier of the object as a string
|
246
|
*/
|
247
|
public IMap<Identifier,SystemMetadata> getSystemMetadataMap() {
|
248
|
return systemMetadata;
|
249
|
}
|
250
|
|
251
|
/**
|
252
|
* Get the identifiers set
|
253
|
* @return identifiers - the set of unique DataONE identifiers in the cluster
|
254
|
*/
|
255
|
public ISet<Identifier> getIdentifiers() {
|
256
|
return identifiers;
|
257
|
|
258
|
}
|
259
|
|
260
|
/**
|
261
|
* Get the index queue
|
262
|
* @return the set of SystemMetadata to be indexed
|
263
|
*/
|
264
|
public IMap<Identifier, IndexTask> getIndexQueue() {
|
265
|
return indexQueue;
|
266
|
}
|
267
|
|
268
|
/**
|
269
|
* Get the index event map
|
270
|
* @return indexEventMap - the hazelcast map of index events
|
271
|
*/
|
272
|
public IMap<Identifier, IndexEvent> getIndexEventMap() {
|
273
|
return indexEventMap;
|
274
|
}
|
275
|
|
276
|
/**
|
277
|
* When Metacat changes the underlying store, we need to refresh the
|
278
|
* in-memory representation of it.
|
279
|
* @param guid
|
280
|
*/
|
281
|
public void refreshSystemMetadataEntry(String guid) {
|
282
|
Identifier identifier = new Identifier();
|
283
|
identifier.setValue(guid);
|
284
|
// force hazelcast to update system metadata in memory from the store
|
285
|
HazelcastService.getInstance().getSystemMetadataMap().evict(identifier);
|
286
|
}
|
287
|
|
288
|
public Lock getLock(String identifier) {
|
289
|
|
290
|
Lock lock = null;
|
291
|
|
292
|
try {
|
293
|
lock = getInstance().getHazelcastInstance().getLock(identifier);
|
294
|
|
295
|
} catch (RuntimeException e) {
|
296
|
logMetacat.info("Couldn't get a lock for identifier " +
|
297
|
identifier + " !!");
|
298
|
}
|
299
|
return lock;
|
300
|
|
301
|
}
|
302
|
|
303
|
/**
|
304
|
* Get the DataONE hazelcast node map
|
305
|
* @return nodes - the hazelcast map of nodes
|
306
|
*/
|
307
|
// public IMap<NodeReference, Node> getNodesMap() {
|
308
|
// return nodes;
|
309
|
// }
|
310
|
|
311
|
/**
|
312
|
* Indicate whether or not this service is refreshable.
|
313
|
*
|
314
|
* @return refreshable - the boolean refreshable status
|
315
|
*/
|
316
|
public boolean refreshable() {
|
317
|
// TODO: Determine the consequences of restarting the Hazelcast instance
|
318
|
// Set this to true if it's okay to drop from the cluster, lose the maps,
|
319
|
// and start back up again
|
320
|
return false;
|
321
|
|
322
|
}
|
323
|
|
324
|
/**
|
325
|
* Stop the HazelcastService. When stopped, the service will no longer
|
326
|
* respond to requests.
|
327
|
*/
|
328
|
public void stop() throws ServiceException {
|
329
|
|
330
|
this.hzInstance.getLifecycleService().shutdown();
|
331
|
|
332
|
}
|
333
|
|
334
|
public HazelcastInstance getHazelcastInstance() {
|
335
|
return this.hzInstance;
|
336
|
|
337
|
}
|
338
|
|
339
|
/**
|
340
|
* Refresh the Hazelcast service by restarting it
|
341
|
*/
|
342
|
@Override
|
343
|
protected void doRefresh() throws ServiceException {
|
344
|
|
345
|
// TODO: verify that the correct config file is still used
|
346
|
this.hzInstance.getLifecycleService().restart();
|
347
|
|
348
|
}
|
349
|
|
350
|
/**
|
351
|
* Implement the EntryListener interface for Hazelcast, reponding to entry
|
352
|
* added events in the hzSystemMetadata map. Evaluate the entry and create
|
353
|
* CNReplicationTasks as appropriate (for DATA, METADATA, RESOURCE)
|
354
|
*
|
355
|
* @param event - The EntryEvent that occurred
|
356
|
*/
|
357
|
@Override
|
358
|
public void entryAdded(EntryEvent<Identifier, SystemMetadata> event) {
|
359
|
|
360
|
logMetacat.info("SystemMetadata entry added event on identifier " +
|
361
|
event.getKey().getValue());
|
362
|
// handle as update - that method will create if necessary
|
363
|
entryUpdated(event);
|
364
|
|
365
|
}
|
366
|
|
367
|
/**
|
368
|
* Implement the EntryListener interface for Hazelcast, reponding to entry
|
369
|
* evicted events in the hzSystemMetadata map. Evaluate the entry and create
|
370
|
* CNReplicationTasks as appropriate (for DATA, METADATA, RESOURCE)
|
371
|
*
|
372
|
* @param event - The EntryEvent that occurred
|
373
|
*/
|
374
|
@Override
|
375
|
public void entryEvicted(EntryEvent<Identifier, SystemMetadata> event) {
|
376
|
|
377
|
logMetacat.info("SystemMetadata entry evicted event on identifier " +
|
378
|
event.getKey().getValue());
|
379
|
|
380
|
// ensure identifiers are listed in the hzIdentifiers set
|
381
|
if ( !identifiers.contains(event.getKey()) ) {
|
382
|
identifiers.add(event.getKey());
|
383
|
}
|
384
|
|
385
|
}
|
386
|
|
387
|
/**
|
388
|
* Implement the EntryListener interface for Hazelcast, reponding to entry
|
389
|
* removed events in the hzSystemMetadata map. Evaluate the entry and create
|
390
|
* CNReplicationTasks as appropriate (for DATA, METADATA, RESOURCE)
|
391
|
*
|
392
|
* @param event - The EntryEvent that occurred
|
393
|
*/
|
394
|
@Override
|
395
|
public void entryRemoved(EntryEvent<Identifier, SystemMetadata> event) {
|
396
|
|
397
|
logMetacat.info("SystemMetadata entry removed event on identifier " +
|
398
|
event.getKey().getValue());
|
399
|
|
400
|
// we typically don't remove objects in Metacat, but can remove System Metadata
|
401
|
IdentifierManager.getInstance().deleteSystemMetadata(event.getValue().getIdentifier().getValue());
|
402
|
|
403
|
// keep the hzIdentifiers set in sync with the systemmetadata table
|
404
|
if ( identifiers.contains(event.getKey()) ) {
|
405
|
identifiers.remove(event.getKey());
|
406
|
|
407
|
}
|
408
|
|
409
|
}
|
410
|
|
411
|
/**
|
412
|
* Implement the EntryListener interface for Hazelcast, reponding to entry
|
413
|
* updated events in the hzSystemMetadata map. Evaluate the entry and create
|
414
|
* CNReplicationTasks as appropriate (for DATA, METADATA, RESOURCE)
|
415
|
*
|
416
|
* @param event - The EntryEvent that occurred
|
417
|
*/
|
418
|
@Override
|
419
|
public void entryUpdated(EntryEvent<Identifier, SystemMetadata> event) {
|
420
|
|
421
|
logMetacat.debug("Entry added/updated to System Metadata map: " + event.getKey().getValue());
|
422
|
PartitionService partitionService = this.hzInstance.getPartitionService();
|
423
|
Partition partition = partitionService.getPartition(event.getKey());
|
424
|
Member ownerMember = partition.getOwner();
|
425
|
SystemMetadata sysmeta = event.getValue();
|
426
|
if (!ownerMember.localMember()) {
|
427
|
if (sysmeta == null) {
|
428
|
logMetacat.warn("No SystemMetadata provided in the event, getting from shared map: " + event.getKey().getValue());
|
429
|
sysmeta = getSystemMetadataMap().get(event.getKey());
|
430
|
if (sysmeta == null) {
|
431
|
// this is a problem
|
432
|
logMetacat.error("Could not find SystemMetadata in shared map for: " + event.getKey().getValue());
|
433
|
// TODO: should probably return at this point since the save will fail
|
434
|
}
|
435
|
}
|
436
|
// need to pull the entry into the local store
|
437
|
saveLocally(event.getValue());
|
438
|
}
|
439
|
|
440
|
// ensure identifiers are listed in the hzIdentifiers set
|
441
|
if (!identifiers.contains(event.getKey())) {
|
442
|
identifiers.add(event.getKey());
|
443
|
}
|
444
|
|
445
|
}
|
446
|
|
447
|
/**
|
448
|
* Save SystemMetadata to local store if needed
|
449
|
* @param sm
|
450
|
*/
|
451
|
private void saveLocally(SystemMetadata sm) {
|
452
|
logMetacat.debug("Saving entry locally: " + sm.getIdentifier().getValue());
|
453
|
try {
|
454
|
|
455
|
IdentifierManager.getInstance().insertOrUpdateSystemMetadata(sm);
|
456
|
|
457
|
} catch (McdbDocNotFoundException e) {
|
458
|
logMetacat.error("Could not save System Metadata to local store.", e);
|
459
|
|
460
|
} catch (SQLException e) {
|
461
|
logMetacat.error("Could not save System Metadata to local store.", e);
|
462
|
|
463
|
} catch (InvalidSystemMetadata e) {
|
464
|
logMetacat.error("Could not save System Metadata to local store.", e);
|
465
|
|
466
|
}
|
467
|
}
|
468
|
|
469
|
/**
|
470
|
* Checks the local backing store for missing SystemMetadata,
|
471
|
* retrieves those entries from the shared map if they exist,
|
472
|
* and saves them locally.
|
473
|
*/
|
474
|
private void synchronizeLocalStore() {
|
475
|
List<String> localIds = IdentifierManager.getInstance().getLocalIdsWithNoSystemMetadata(true, -1);
|
476
|
if (localIds != null) {
|
477
|
logMetacat.debug("Member missing SystemMetadata entries, count = " + localIds.size());
|
478
|
for (String localId: localIds) {
|
479
|
logMetacat.debug("Processing system metadata for localId: " + localId);
|
480
|
try {
|
481
|
String docid = DocumentUtil.getSmartDocId(localId);
|
482
|
int rev = DocumentUtil.getRevisionFromAccessionNumber(localId);
|
483
|
String guid = IdentifierManager.getInstance().getGUID(docid, rev);
|
484
|
logMetacat.debug("Found mapped guid: " + guid);
|
485
|
Identifier pid = new Identifier();
|
486
|
pid.setValue(guid);
|
487
|
SystemMetadata sm = systemMetadata.get(pid);
|
488
|
logMetacat.debug("Found shared system metadata for guid: " + guid);
|
489
|
saveLocally(sm);
|
490
|
logMetacat.debug("Saved shared system metadata locally for guid: " + guid);
|
491
|
} catch (Exception e) {
|
492
|
logMetacat.error("Could not save shared SystemMetadata entry locally, localId: " + localId, e);
|
493
|
}
|
494
|
}
|
495
|
}
|
496
|
}
|
497
|
|
498
|
|
499
|
/**
|
500
|
* Make sure we have a copy of every entry in the shared map.
|
501
|
* We use lazy loading and therefore the CNs may not all be in sync when one
|
502
|
* comes back online after an extended period of being offline
|
503
|
* This method loops through the entries that a FULLY UP-TO-DATE CN has
|
504
|
* and makes sure each one is present on the shared map.
|
505
|
* It is meant to overcome a HZ weakness wherein ownership of a key results in
|
506
|
* null values where the owner does not have a complete backing store.
|
507
|
* This will be an expensive routine and should be run in a background process so that
|
508
|
* the server can continue to service other requests during the synch
|
509
|
* @throws Exception
|
510
|
*/
|
511
|
private void resynchToRemote() {
|
512
|
|
513
|
// the local identifiers not already present in the shared map
|
514
|
Set<Identifier> localIdKeys = loadAllKeys();
|
515
|
|
516
|
// the PIDs missing locally
|
517
|
Set<Identifier> missingIdKeys = new HashSet<Identifier>();
|
518
|
|
519
|
// only contribute PIDs that are not already shared
|
520
|
Iterator<Identifier> idIter = identifiers.iterator();
|
521
|
int processedCount = 0;
|
522
|
while (idIter.hasNext()) {
|
523
|
Identifier pid = idIter.next();
|
524
|
if (localIdKeys.contains(pid)) {
|
525
|
logMetacat.debug("Shared pid is already in local identifier set: " + pid.getValue());
|
526
|
localIdKeys.remove(pid);
|
527
|
} else {
|
528
|
// we don't have this locally, so we should try to get it
|
529
|
missingIdKeys.add(pid);
|
530
|
}
|
531
|
processedCount++;
|
532
|
}
|
533
|
logMetacat.warn("processedCount (identifiers from iterator): " + processedCount);
|
534
|
|
535
|
logMetacat.warn("local pid count not yet shared: " + localIdKeys.size() + ", shared pid count: " + identifiers.size());
|
536
|
|
537
|
//identifiers.addAll(idKeys);
|
538
|
logMetacat.warn("Loading missing local keys into hzIdentifiers");
|
539
|
for (Identifier key: localIdKeys) {
|
540
|
if (!identifiers.contains(key)) {
|
541
|
logMetacat.debug("Adding missing hzIdentifiers key: " + key.getValue());
|
542
|
identifiers.add(key);
|
543
|
}
|
544
|
}
|
545
|
logMetacat.warn("Initialized identifiers with missing local keys");
|
546
|
|
547
|
logMetacat.warn("Processing missing SystemMetadata for missing pid count: " + missingIdKeys.size());
|
548
|
|
549
|
// loop through all the missing PIDs to find any null (missing) SM that needs to be resynched
|
550
|
Iterator<Identifier> missingPids = missingIdKeys.iterator();
|
551
|
while (missingPids.hasNext()) {
|
552
|
Identifier pid = missingPids.next();
|
553
|
// publish that we need this SM entry
|
554
|
logMetacat.debug("Publishing missing pid to wanted list: " + pid.getValue());
|
555
|
missingIdentifiers.add(pid);
|
556
|
}
|
557
|
|
558
|
}
|
559
|
|
560
|
public void resynchInThread() {
|
561
|
logMetacat.debug("launching system metadata resynch in a thread");
|
562
|
ExecutorService executor = Executors.newSingleThreadExecutor();
|
563
|
executor.execute(new Runnable() {
|
564
|
@Override
|
565
|
public void run() {
|
566
|
try {
|
567
|
// this is a push mechanism
|
568
|
DateFormat dateFormat = new SimpleDateFormat("MMM dd, yyyy HH:mm:ss aaa");
|
569
|
System.out.println(dateFormat.format(Calendar.getInstance().getTime())+" Start the hazelcast synchronization");
|
570
|
logMetacat.warn("Start the hazelcast synchronization");
|
571
|
resynchToRemote();
|
572
|
System.out.println(dateFormat.format(Calendar.getInstance().getTime())+" End the hazelcast synchronization");
|
573
|
logMetacat.warn("End the hazelcast synchronization");
|
574
|
} catch (Exception e) {
|
575
|
logMetacat.error("Error in resynchInThread: " + e.getMessage(), e);
|
576
|
}
|
577
|
}
|
578
|
});
|
579
|
executor.shutdown();
|
580
|
}
|
581
|
|
582
|
/**
|
583
|
* When there is missing SystemMetadata on the local member,
|
584
|
* we retrieve it from the shared map and add it to the local
|
585
|
* backing store for safe keeping.
|
586
|
*/
|
587
|
@Override
|
588
|
public void memberAdded(MembershipEvent event) {
|
589
|
Member member = event.getMember();
|
590
|
logMetacat.debug("Member added to cluster: " + member.getInetSocketAddress());
|
591
|
boolean isLocal = member.localMember();
|
592
|
if (isLocal) {
|
593
|
logMetacat.debug("Member islocal: " + member.getInetSocketAddress());
|
594
|
synchronizeLocalStore();
|
595
|
}
|
596
|
}
|
597
|
|
598
|
@Override
|
599
|
public void memberRemoved(MembershipEvent event) {
|
600
|
// TODO Auto-generated method stub
|
601
|
|
602
|
}
|
603
|
|
604
|
/**
|
605
|
* In cases where this cluster is paused, we want to
|
606
|
* check that the local store accurately reflects the shared
|
607
|
* SystemMetadata map
|
608
|
* @param event
|
609
|
*/
|
610
|
@Override
|
611
|
public void stateChanged(LifecycleEvent event) {
|
612
|
logMetacat.debug("HZ LifecycleEvent.state: " + event.getState());
|
613
|
if (event.getState().equals(LifecycleEvent.LifecycleState.RESUMED)) {
|
614
|
logMetacat.debug("HZ LifecycleEvent.state is RESUMED, calling synchronizeLocalStore()");
|
615
|
synchronizeLocalStore();
|
616
|
}
|
617
|
}
|
618
|
|
619
|
/**
|
620
|
* Load all System Metadata keys from the backing store
|
621
|
* @return set of pids
|
622
|
*/
|
623
|
private Set<Identifier> loadAllKeys() {
|
624
|
|
625
|
Set<Identifier> pids = new HashSet<Identifier>();
|
626
|
|
627
|
try {
|
628
|
|
629
|
// ALTERNATIVE 1: this has more overhead than just looking at the GUIDs
|
630
|
// ObjectList ol = IdentifierManager.getInstance().querySystemMetadata(
|
631
|
// null, //startTime,
|
632
|
// null, //endTime,
|
633
|
// null, //objectFormatId,
|
634
|
// false, //replicaStatus,
|
635
|
// 0, //start,
|
636
|
// -1 //count
|
637
|
// );
|
638
|
// for (ObjectInfo o: ol.getObjectInfoList()) {
|
639
|
// Identifier pid = o.getIdentifier();
|
640
|
// if ( !pids.contains(pid) ) {
|
641
|
// pids.add(pid);
|
642
|
// }
|
643
|
// }
|
644
|
|
645
|
// ALTERNATIVE method: look up all the Identifiers from the table
|
646
|
List<String> guids = IdentifierManager.getInstance().getAllSystemMetadataGUIDs();
|
647
|
logMetacat.warn("Local SystemMetadata pid count: " + guids.size());
|
648
|
for (String guid: guids){
|
649
|
Identifier pid = new Identifier();
|
650
|
pid.setValue(guid);
|
651
|
pids.add(pid);
|
652
|
}
|
653
|
|
654
|
} catch (Exception e) {
|
655
|
throw new RuntimeException(e.getMessage(), e);
|
656
|
|
657
|
}
|
658
|
|
659
|
return pids;
|
660
|
}
|
661
|
|
662
|
/**
|
663
|
* Respond to itemAdded events on the hzMissingIdentifiers Set. Uses a
|
664
|
* distributed ILock to try to prevent multiple put calls on hzSystemMetadata
|
665
|
*
|
666
|
* @param pid the identifier of the event
|
667
|
*/
|
668
|
@Override
|
669
|
public void itemAdded(ItemEvent<Identifier> event) {
|
670
|
|
671
|
Identifier pid = (Identifier) event.getItem();
|
672
|
// publish the SM for the pid if we have it locally
|
673
|
logMetacat.debug("Responding to itemAdded for pid: " + pid.getValue());
|
674
|
|
675
|
// lock this event, only if we have a local copy to contribute
|
676
|
ILock lock = null;
|
677
|
try {
|
678
|
// look up the local copy of the SM
|
679
|
SystemMetadata sm = IdentifierManager.getInstance().getSystemMetadata(pid.getValue());
|
680
|
if (sm != null) {
|
681
|
lock = hzInstance.getLock(MISSING_PID_PREFIX + pid.getValue());
|
682
|
|
683
|
if ( lock.tryLock() ) {
|
684
|
// "publish" the system metadata to the shared map since it showed up on the missing queue
|
685
|
logMetacat.debug("Adding SystemMetadata to shared map for pid: " + pid.getValue());
|
686
|
systemMetadata.put(pid, sm);
|
687
|
|
688
|
// remove the entry since we processed it
|
689
|
missingIdentifiers.remove(pid);
|
690
|
|
691
|
} else {
|
692
|
logMetacat.debug(MISSING_PID_PREFIX + pid.getValue() + " was already locked. Skipping.");
|
693
|
}
|
694
|
} else {
|
695
|
// can't help here
|
696
|
logMetacat.warn("Local system metadata not found for pid: " + pid.getValue());
|
697
|
}
|
698
|
} catch (Exception e) {
|
699
|
logMetacat.error("Error looking up missing system metadata for pid: " + pid.getValue());
|
700
|
} finally {
|
701
|
if ( lock != null ) {
|
702
|
lock.unlock();
|
703
|
}
|
704
|
}
|
705
|
}
|
706
|
|
707
|
/**
|
708
|
* Respond to itemRemoved events on the hzMissingIdentifiers Set
|
709
|
*
|
710
|
* @param pid the identifier of the event
|
711
|
*/
|
712
|
@Override
|
713
|
public void itemRemoved(ItemEvent<Identifier> event) {
|
714
|
// do nothing since someone probably handled the wanted PID
|
715
|
|
716
|
}
|
717
|
|
718
|
}
|