1 | /** |
---|
2 | * Licensed to the Apache Software Foundation (ASF) under one |
---|
3 | * or more contributor license agreements. See the NOTICE file |
---|
4 | * distributed with this work for additional information |
---|
5 | * regarding copyright ownership. The ASF licenses this file |
---|
6 | * to you under the Apache License, Version 2.0 (the |
---|
7 | * "License"); you may not use this file except in compliance |
---|
8 | * with the License. You may obtain a copy of the License at |
---|
9 | * |
---|
10 | * http://www.apache.org/licenses/LICENSE-2.0 |
---|
11 | * |
---|
12 | * Unless required by applicable law or agreed to in writing, software |
---|
13 | * distributed under the License is distributed on an "AS IS" BASIS, |
---|
14 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
---|
15 | * See the License for the specific language governing permissions and |
---|
16 | * limitations under the License. |
---|
17 | */ |
---|
18 | package org.apache.hadoop.hdfs.server.namenode; |
---|
19 | |
---|
20 | import junit.framework.TestCase; |
---|
21 | import java.io.*; |
---|
22 | import java.util.Collection; |
---|
23 | import java.util.List; |
---|
24 | import java.util.Iterator; |
---|
25 | import java.util.Random; |
---|
26 | |
---|
27 | import org.apache.hadoop.conf.Configuration; |
---|
28 | import org.apache.hadoop.hdfs.DistributedFileSystem; |
---|
29 | import org.apache.hadoop.hdfs.MiniDFSCluster; |
---|
30 | import org.apache.hadoop.hdfs.protocol.FSConstants.SafeModeAction; |
---|
31 | import org.apache.hadoop.hdfs.server.common.Storage; |
---|
32 | import org.apache.hadoop.hdfs.server.namenode.FSImage.NameNodeFile; |
---|
33 | import org.apache.hadoop.hdfs.server.namenode.SecondaryNameNode.ErrorSimulator; |
---|
34 | import org.apache.hadoop.hdfs.server.common.HdfsConstants.StartupOption; |
---|
35 | import org.apache.hadoop.hdfs.server.common.Storage.StorageDirectory; |
---|
36 | import org.apache.hadoop.hdfs.server.namenode.FSImage.NameNodeDirType; |
---|
37 | import org.apache.hadoop.hdfs.tools.DFSAdmin; |
---|
38 | import org.apache.hadoop.fs.FSDataOutputStream; |
---|
39 | import org.apache.hadoop.fs.FileSystem; |
---|
40 | import org.apache.hadoop.fs.FileUtil; |
---|
41 | import org.apache.hadoop.fs.Path; |
---|
42 | |
---|
43 | /** |
---|
44 | * This class tests the creation and validation of a checkpoint. |
---|
45 | */ |
---|
46 | public class TestCheckpoint extends TestCase { |
---|
47 | static final long seed = 0xDEADBEEFL; |
---|
48 | static final int blockSize = 4096; |
---|
49 | static final int fileSize = 8192; |
---|
50 | static final int numDatanodes = 3; |
---|
51 | short replication = 3; |
---|
52 | |
---|
53 | private void writeFile(FileSystem fileSys, Path name, int repl) |
---|
54 | throws IOException { |
---|
55 | FSDataOutputStream stm = fileSys.create(name, true, |
---|
56 | fileSys.getConf().getInt("io.file.buffer.size", 4096), |
---|
57 | (short)repl, (long)blockSize); |
---|
58 | byte[] buffer = new byte[fileSize]; |
---|
59 | Random rand = new Random(seed); |
---|
60 | rand.nextBytes(buffer); |
---|
61 | stm.write(buffer); |
---|
62 | stm.close(); |
---|
63 | } |
---|
64 | |
---|
65 | |
---|
66 | private void checkFile(FileSystem fileSys, Path name, int repl) |
---|
67 | throws IOException { |
---|
68 | assertTrue(fileSys.exists(name)); |
---|
69 | int replication = fileSys.getFileStatus(name).getReplication(); |
---|
70 | assertEquals("replication for " + name, repl, replication); |
---|
71 | //We should probably test for more of the file properties. |
---|
72 | } |
---|
73 | |
---|
74 | private void cleanupFile(FileSystem fileSys, Path name) |
---|
75 | throws IOException { |
---|
76 | assertTrue(fileSys.exists(name)); |
---|
77 | fileSys.delete(name, true); |
---|
78 | assertTrue(!fileSys.exists(name)); |
---|
79 | } |
---|
80 | |
---|
81 | /** |
---|
82 | * put back the old namedir |
---|
83 | */ |
---|
84 | private void resurrectNameDir(File namedir) |
---|
85 | throws IOException { |
---|
86 | String parentdir = namedir.getParent(); |
---|
87 | String name = namedir.getName(); |
---|
88 | File oldname = new File(parentdir, name + ".old"); |
---|
89 | if (!oldname.renameTo(namedir)) { |
---|
90 | assertTrue(false); |
---|
91 | } |
---|
92 | } |
---|
93 | |
---|
94 | /** |
---|
95 | * remove one namedir |
---|
96 | */ |
---|
97 | private void removeOneNameDir(File namedir) |
---|
98 | throws IOException { |
---|
99 | String parentdir = namedir.getParent(); |
---|
100 | String name = namedir.getName(); |
---|
101 | File newname = new File(parentdir, name + ".old"); |
---|
102 | if (!namedir.renameTo(newname)) { |
---|
103 | assertTrue(false); |
---|
104 | } |
---|
105 | } |
---|
106 | |
---|
107 | /* |
---|
108 | * Verify that namenode does not startup if one namedir is bad. |
---|
109 | */ |
---|
110 | private void testNamedirError(Configuration conf, Collection<File> namedirs) |
---|
111 | throws IOException { |
---|
112 | System.out.println("Starting testNamedirError"); |
---|
113 | MiniDFSCluster cluster = null; |
---|
114 | |
---|
115 | if (namedirs.size() <= 1) { |
---|
116 | return; |
---|
117 | } |
---|
118 | |
---|
119 | // |
---|
120 | // Remove one namedir & Restart cluster. This should fail. |
---|
121 | // |
---|
122 | File first = namedirs.iterator().next(); |
---|
123 | removeOneNameDir(first); |
---|
124 | try { |
---|
125 | cluster = new MiniDFSCluster(conf, 0, false, null); |
---|
126 | cluster.shutdown(); |
---|
127 | assertTrue(false); |
---|
128 | } catch (Throwable t) { |
---|
129 | // no nothing |
---|
130 | } |
---|
131 | resurrectNameDir(first); // put back namedir |
---|
132 | } |
---|
133 | |
---|
134 | /* |
---|
135 | * Simulate namenode crashing after rolling edit log. |
---|
136 | */ |
---|
137 | private void testSecondaryNamenodeError1(Configuration conf) |
---|
138 | throws IOException { |
---|
139 | System.out.println("Starting testSecondaryNamenodeError 1"); |
---|
140 | Path file1 = new Path("checkpointxx.dat"); |
---|
141 | MiniDFSCluster cluster = new MiniDFSCluster(conf, numDatanodes, |
---|
142 | false, null); |
---|
143 | cluster.waitActive(); |
---|
144 | FileSystem fileSys = cluster.getFileSystem(); |
---|
145 | try { |
---|
146 | assertTrue(!fileSys.exists(file1)); |
---|
147 | // |
---|
148 | // Make the checkpoint fail after rolling the edits log. |
---|
149 | // |
---|
150 | SecondaryNameNode secondary = startSecondaryNameNode(conf); |
---|
151 | ErrorSimulator.setErrorSimulation(0); |
---|
152 | |
---|
153 | try { |
---|
154 | secondary.doCheckpoint(); // this should fail |
---|
155 | assertTrue(false); |
---|
156 | } catch (IOException e) { |
---|
157 | } |
---|
158 | ErrorSimulator.clearErrorSimulation(0); |
---|
159 | secondary.shutdown(); |
---|
160 | |
---|
161 | // |
---|
162 | // Create a new file |
---|
163 | // |
---|
164 | writeFile(fileSys, file1, replication); |
---|
165 | checkFile(fileSys, file1, replication); |
---|
166 | } finally { |
---|
167 | fileSys.close(); |
---|
168 | cluster.shutdown(); |
---|
169 | } |
---|
170 | |
---|
171 | // |
---|
172 | // Restart cluster and verify that file exists. |
---|
173 | // Then take another checkpoint to verify that the |
---|
174 | // namenode restart accounted for the rolled edit logs. |
---|
175 | // |
---|
176 | System.out.println("Starting testSecondaryNamenodeError 2"); |
---|
177 | cluster = new MiniDFSCluster(conf, numDatanodes, false, null); |
---|
178 | cluster.waitActive(); |
---|
179 | // Also check that the edits file is empty here |
---|
180 | // and that temporary checkpoint files are gone. |
---|
181 | FSImage image = cluster.getNameNode().getFSImage(); |
---|
182 | for (Iterator<StorageDirectory> it = |
---|
183 | image.dirIterator(NameNodeDirType.IMAGE); it.hasNext();) { |
---|
184 | StorageDirectory sd = it.next(); |
---|
185 | assertFalse(FSImage.getImageFile(sd, NameNodeFile.IMAGE_NEW).exists()); |
---|
186 | } |
---|
187 | for (Iterator<StorageDirectory> it = |
---|
188 | image.dirIterator(NameNodeDirType.EDITS); it.hasNext();) { |
---|
189 | StorageDirectory sd = it.next(); |
---|
190 | assertFalse(image.getEditNewFile(sd).exists()); |
---|
191 | File edits = image.getEditFile(sd); |
---|
192 | assertTrue(edits.exists()); // edits should exist and be empty |
---|
193 | long editsLen = edits.length(); |
---|
194 | assertTrue(editsLen == Integer.SIZE/Byte.SIZE); |
---|
195 | } |
---|
196 | |
---|
197 | fileSys = cluster.getFileSystem(); |
---|
198 | try { |
---|
199 | checkFile(fileSys, file1, replication); |
---|
200 | cleanupFile(fileSys, file1); |
---|
201 | SecondaryNameNode secondary = startSecondaryNameNode(conf); |
---|
202 | secondary.doCheckpoint(); |
---|
203 | secondary.shutdown(); |
---|
204 | } finally { |
---|
205 | fileSys.close(); |
---|
206 | cluster.shutdown(); |
---|
207 | } |
---|
208 | } |
---|
209 | |
---|
210 | /* |
---|
211 | * Simulate a namenode crash after uploading new image |
---|
212 | */ |
---|
213 | private void testSecondaryNamenodeError2(Configuration conf) |
---|
214 | throws IOException { |
---|
215 | System.out.println("Starting testSecondaryNamenodeError 21"); |
---|
216 | Path file1 = new Path("checkpointyy.dat"); |
---|
217 | MiniDFSCluster cluster = new MiniDFSCluster(conf, numDatanodes, |
---|
218 | false, null); |
---|
219 | cluster.waitActive(); |
---|
220 | FileSystem fileSys = cluster.getFileSystem(); |
---|
221 | try { |
---|
222 | assertTrue(!fileSys.exists(file1)); |
---|
223 | // |
---|
224 | // Make the checkpoint fail after uploading the new fsimage. |
---|
225 | // |
---|
226 | SecondaryNameNode secondary = startSecondaryNameNode(conf); |
---|
227 | ErrorSimulator.setErrorSimulation(1); |
---|
228 | |
---|
229 | try { |
---|
230 | secondary.doCheckpoint(); // this should fail |
---|
231 | assertTrue(false); |
---|
232 | } catch (IOException e) { |
---|
233 | } |
---|
234 | ErrorSimulator.clearErrorSimulation(1); |
---|
235 | secondary.shutdown(); |
---|
236 | |
---|
237 | // |
---|
238 | // Create a new file |
---|
239 | // |
---|
240 | writeFile(fileSys, file1, replication); |
---|
241 | checkFile(fileSys, file1, replication); |
---|
242 | } finally { |
---|
243 | fileSys.close(); |
---|
244 | cluster.shutdown(); |
---|
245 | } |
---|
246 | |
---|
247 | // |
---|
248 | // Restart cluster and verify that file exists. |
---|
249 | // Then take another checkpoint to verify that the |
---|
250 | // namenode restart accounted for the rolled edit logs. |
---|
251 | // |
---|
252 | System.out.println("Starting testSecondaryNamenodeError 22"); |
---|
253 | cluster = new MiniDFSCluster(conf, numDatanodes, false, null); |
---|
254 | cluster.waitActive(); |
---|
255 | fileSys = cluster.getFileSystem(); |
---|
256 | try { |
---|
257 | checkFile(fileSys, file1, replication); |
---|
258 | cleanupFile(fileSys, file1); |
---|
259 | SecondaryNameNode secondary = startSecondaryNameNode(conf); |
---|
260 | secondary.doCheckpoint(); |
---|
261 | secondary.shutdown(); |
---|
262 | } finally { |
---|
263 | fileSys.close(); |
---|
264 | cluster.shutdown(); |
---|
265 | } |
---|
266 | } |
---|
267 | |
---|
268 | /* |
---|
269 | * Simulate a secondary namenode crash after rolling the edit log. |
---|
270 | */ |
---|
271 | private void testSecondaryNamenodeError3(Configuration conf) |
---|
272 | throws IOException { |
---|
273 | System.out.println("Starting testSecondaryNamenodeError 31"); |
---|
274 | Path file1 = new Path("checkpointzz.dat"); |
---|
275 | MiniDFSCluster cluster = new MiniDFSCluster(conf, numDatanodes, |
---|
276 | false, null); |
---|
277 | cluster.waitActive(); |
---|
278 | FileSystem fileSys = cluster.getFileSystem(); |
---|
279 | try { |
---|
280 | assertTrue(!fileSys.exists(file1)); |
---|
281 | // |
---|
282 | // Make the checkpoint fail after rolling the edit log. |
---|
283 | // |
---|
284 | SecondaryNameNode secondary = startSecondaryNameNode(conf); |
---|
285 | ErrorSimulator.setErrorSimulation(0); |
---|
286 | |
---|
287 | try { |
---|
288 | secondary.doCheckpoint(); // this should fail |
---|
289 | assertTrue(false); |
---|
290 | } catch (IOException e) { |
---|
291 | } |
---|
292 | ErrorSimulator.clearErrorSimulation(0); |
---|
293 | secondary.shutdown(); // secondary namenode crash! |
---|
294 | |
---|
295 | // start new instance of secondary and verify that |
---|
296 | // a new rollEditLog suceedes inspite of the fact that |
---|
297 | // edits.new already exists. |
---|
298 | // |
---|
299 | secondary = startSecondaryNameNode(conf); |
---|
300 | secondary.doCheckpoint(); // this should work correctly |
---|
301 | secondary.shutdown(); |
---|
302 | |
---|
303 | // |
---|
304 | // Create a new file |
---|
305 | // |
---|
306 | writeFile(fileSys, file1, replication); |
---|
307 | checkFile(fileSys, file1, replication); |
---|
308 | } finally { |
---|
309 | fileSys.close(); |
---|
310 | cluster.shutdown(); |
---|
311 | } |
---|
312 | |
---|
313 | // |
---|
314 | // Restart cluster and verify that file exists. |
---|
315 | // Then take another checkpoint to verify that the |
---|
316 | // namenode restart accounted for the twice-rolled edit logs. |
---|
317 | // |
---|
318 | System.out.println("Starting testSecondaryNamenodeError 32"); |
---|
319 | cluster = new MiniDFSCluster(conf, numDatanodes, false, null); |
---|
320 | cluster.waitActive(); |
---|
321 | fileSys = cluster.getFileSystem(); |
---|
322 | try { |
---|
323 | checkFile(fileSys, file1, replication); |
---|
324 | cleanupFile(fileSys, file1); |
---|
325 | SecondaryNameNode secondary = startSecondaryNameNode(conf); |
---|
326 | secondary.doCheckpoint(); |
---|
327 | secondary.shutdown(); |
---|
328 | } finally { |
---|
329 | fileSys.close(); |
---|
330 | cluster.shutdown(); |
---|
331 | } |
---|
332 | } |
---|
333 | |
---|
334 | /** |
---|
335 | * Simulate a secondary node failure to transfer image |
---|
336 | * back to the name-node. |
---|
337 | * Used to truncate primary fsimage file. |
---|
338 | */ |
---|
339 | void testSecondaryFailsToReturnImage(Configuration conf) |
---|
340 | throws IOException { |
---|
341 | System.out.println("Starting testSecondaryFailsToReturnImage"); |
---|
342 | Path file1 = new Path("checkpointRI.dat"); |
---|
343 | MiniDFSCluster cluster = new MiniDFSCluster(conf, numDatanodes, |
---|
344 | false, null); |
---|
345 | cluster.waitActive(); |
---|
346 | FileSystem fileSys = cluster.getFileSystem(); |
---|
347 | FSImage image = cluster.getNameNode().getFSImage(); |
---|
348 | try { |
---|
349 | assertTrue(!fileSys.exists(file1)); |
---|
350 | StorageDirectory sd = null; |
---|
351 | for (Iterator<StorageDirectory> it = |
---|
352 | image.dirIterator(NameNodeDirType.IMAGE); it.hasNext();) |
---|
353 | sd = it.next(); |
---|
354 | assertTrue(sd != null); |
---|
355 | long fsimageLength = FSImage.getImageFile(sd, NameNodeFile.IMAGE).length(); |
---|
356 | // |
---|
357 | // Make the checkpoint |
---|
358 | // |
---|
359 | SecondaryNameNode secondary = startSecondaryNameNode(conf); |
---|
360 | ErrorSimulator.setErrorSimulation(2); |
---|
361 | |
---|
362 | try { |
---|
363 | secondary.doCheckpoint(); // this should fail |
---|
364 | assertTrue(false); |
---|
365 | } catch (IOException e) { |
---|
366 | System.out.println("testSecondaryFailsToReturnImage: doCheckpoint() " + |
---|
367 | "failed predictably - " + e); |
---|
368 | } |
---|
369 | ErrorSimulator.clearErrorSimulation(2); |
---|
370 | |
---|
371 | // Verify that image file sizes did not change. |
---|
372 | for (Iterator<StorageDirectory> it = |
---|
373 | image.dirIterator(NameNodeDirType.IMAGE); it.hasNext();) { |
---|
374 | assertTrue(FSImage.getImageFile(it.next(), |
---|
375 | NameNodeFile.IMAGE).length() == fsimageLength); |
---|
376 | } |
---|
377 | |
---|
378 | secondary.shutdown(); |
---|
379 | } finally { |
---|
380 | fileSys.close(); |
---|
381 | cluster.shutdown(); |
---|
382 | } |
---|
383 | } |
---|
384 | |
---|
385 | /** |
---|
386 | * Test different startup scenarios. |
---|
387 | * <p><ol> |
---|
388 | * <li> Start of primary name-node in secondary directory must succeed. |
---|
389 | * <li> Start of secondary node when the primary is already running in |
---|
390 | * this directory must fail. |
---|
391 | * <li> Start of primary name-node if secondary node is already running in |
---|
392 | * this directory must fail. |
---|
393 | * <li> Start of two secondary nodes in the same directory must fail. |
---|
394 | * <li> Import of a checkpoint must fail if primary |
---|
395 | * directory contains a valid image. |
---|
396 | * <li> Import of the secondary image directory must succeed if primary |
---|
397 | * directory does not exist. |
---|
398 | * <li> Recover failed checkpoint for secondary node. |
---|
399 | * <li> Complete failed checkpoint for secondary node. |
---|
400 | * </ol> |
---|
401 | */ |
---|
402 | void testStartup(Configuration conf) throws IOException { |
---|
403 | System.out.println("Startup of the name-node in the checkpoint directory."); |
---|
404 | String primaryDirs = conf.get("dfs.name.dir"); |
---|
405 | String primaryEditsDirs = conf.get("dfs.name.edits.dir"); |
---|
406 | String checkpointDirs = conf.get("fs.checkpoint.dir"); |
---|
407 | String checkpointEditsDirs = conf.get("fs.checkpoint.edits.dir"); |
---|
408 | NameNode nn = startNameNode(conf, checkpointDirs, checkpointEditsDirs, |
---|
409 | StartupOption.REGULAR); |
---|
410 | |
---|
411 | // Starting secondary node in the same directory as the primary |
---|
412 | System.out.println("Startup of secondary in the same dir as the primary."); |
---|
413 | SecondaryNameNode secondary = null; |
---|
414 | try { |
---|
415 | secondary = startSecondaryNameNode(conf); |
---|
416 | assertFalse(secondary.getFSImage().isLockSupported(0)); |
---|
417 | secondary.shutdown(); |
---|
418 | } catch (IOException e) { // expected to fail |
---|
419 | assertTrue(secondary == null); |
---|
420 | } |
---|
421 | nn.stop(); nn = null; |
---|
422 | |
---|
423 | // Starting primary node in the same directory as the secondary |
---|
424 | System.out.println("Startup of primary in the same dir as the secondary."); |
---|
425 | // secondary won't start without primary |
---|
426 | nn = startNameNode(conf, primaryDirs, primaryEditsDirs, |
---|
427 | StartupOption.REGULAR); |
---|
428 | boolean succeed = false; |
---|
429 | do { |
---|
430 | try { |
---|
431 | secondary = startSecondaryNameNode(conf); |
---|
432 | succeed = true; |
---|
433 | } catch(IOException ie) { // keep trying |
---|
434 | System.out.println("Try again: " + ie.getLocalizedMessage()); |
---|
435 | } |
---|
436 | } while(!succeed); |
---|
437 | nn.stop(); nn = null; |
---|
438 | try { |
---|
439 | nn = startNameNode(conf, checkpointDirs, checkpointEditsDirs, |
---|
440 | StartupOption.REGULAR); |
---|
441 | assertFalse(nn.getFSImage().isLockSupported(0)); |
---|
442 | nn.stop(); nn = null; |
---|
443 | } catch (IOException e) { // expected to fail |
---|
444 | assertTrue(nn == null); |
---|
445 | } |
---|
446 | |
---|
447 | // Try another secondary in the same directory |
---|
448 | System.out.println("Startup of two secondaries in the same dir."); |
---|
449 | // secondary won't start without primary |
---|
450 | nn = startNameNode(conf, primaryDirs, primaryEditsDirs, |
---|
451 | StartupOption.REGULAR); |
---|
452 | SecondaryNameNode secondary2 = null; |
---|
453 | try { |
---|
454 | secondary2 = startSecondaryNameNode(conf); |
---|
455 | assertFalse(secondary2.getFSImage().isLockSupported(0)); |
---|
456 | secondary2.shutdown(); |
---|
457 | } catch (IOException e) { // expected to fail |
---|
458 | assertTrue(secondary2 == null); |
---|
459 | } |
---|
460 | nn.stop(); nn = null; |
---|
461 | secondary.shutdown(); |
---|
462 | |
---|
463 | // Import a checkpoint with existing primary image. |
---|
464 | System.out.println("Import a checkpoint with existing primary image."); |
---|
465 | try { |
---|
466 | nn = startNameNode(conf, primaryDirs, primaryEditsDirs, |
---|
467 | StartupOption.IMPORT); |
---|
468 | assertTrue(false); |
---|
469 | } catch (IOException e) { // expected to fail |
---|
470 | assertTrue(nn == null); |
---|
471 | } |
---|
472 | |
---|
473 | // Remove current image and import a checkpoint. |
---|
474 | System.out.println("Import a checkpoint with existing primary image."); |
---|
475 | List<File> nameDirs = (List<File>)FSNamesystem.getNamespaceDirs(conf); |
---|
476 | List<File> nameEditsDirs = (List<File>)FSNamesystem. |
---|
477 | getNamespaceEditsDirs(conf); |
---|
478 | long fsimageLength = new File(new File(nameDirs.get(0), "current"), |
---|
479 | NameNodeFile.IMAGE.getName()).length(); |
---|
480 | for(File dir : nameDirs) { |
---|
481 | if(dir.exists()) |
---|
482 | if(!(FileUtil.fullyDelete(dir))) |
---|
483 | throw new IOException("Cannot remove directory: " + dir); |
---|
484 | if (!dir.mkdirs()) |
---|
485 | throw new IOException("Cannot create directory " + dir); |
---|
486 | } |
---|
487 | |
---|
488 | for(File dir : nameEditsDirs) { |
---|
489 | if(dir.exists()) |
---|
490 | if(!(FileUtil.fullyDelete(dir))) |
---|
491 | throw new IOException("Cannot remove directory: " + dir); |
---|
492 | if (!dir.mkdirs()) |
---|
493 | throw new IOException("Cannot create directory " + dir); |
---|
494 | } |
---|
495 | |
---|
496 | nn = startNameNode(conf, primaryDirs, primaryEditsDirs, |
---|
497 | StartupOption.IMPORT); |
---|
498 | // Verify that image file sizes did not change. |
---|
499 | FSImage image = nn.getFSImage(); |
---|
500 | for (Iterator<StorageDirectory> it = |
---|
501 | image.dirIterator(NameNodeDirType.IMAGE); it.hasNext();) { |
---|
502 | assertTrue(FSImage.getImageFile(it.next(), |
---|
503 | NameNodeFile.IMAGE).length() == fsimageLength); |
---|
504 | } |
---|
505 | nn.stop(); |
---|
506 | |
---|
507 | // recover failed checkpoint |
---|
508 | nn = startNameNode(conf, primaryDirs, primaryEditsDirs, |
---|
509 | StartupOption.REGULAR); |
---|
510 | Collection<File> secondaryDirs = FSImage.getCheckpointDirs(conf, null); |
---|
511 | for(File dir : secondaryDirs) { |
---|
512 | Storage.rename(new File(dir, "current"), |
---|
513 | new File(dir, "lastcheckpoint.tmp")); |
---|
514 | } |
---|
515 | secondary = startSecondaryNameNode(conf); |
---|
516 | secondary.shutdown(); |
---|
517 | for(File dir : secondaryDirs) { |
---|
518 | assertTrue(new File(dir, "current").exists()); |
---|
519 | assertFalse(new File(dir, "lastcheckpoint.tmp").exists()); |
---|
520 | } |
---|
521 | |
---|
522 | // complete failed checkpoint |
---|
523 | for(File dir : secondaryDirs) { |
---|
524 | Storage.rename(new File(dir, "previous.checkpoint"), |
---|
525 | new File(dir, "lastcheckpoint.tmp")); |
---|
526 | } |
---|
527 | secondary = startSecondaryNameNode(conf); |
---|
528 | secondary.shutdown(); |
---|
529 | for(File dir : secondaryDirs) { |
---|
530 | assertTrue(new File(dir, "current").exists()); |
---|
531 | assertTrue(new File(dir, "previous.checkpoint").exists()); |
---|
532 | assertFalse(new File(dir, "lastcheckpoint.tmp").exists()); |
---|
533 | } |
---|
534 | nn.stop(); nn = null; |
---|
535 | |
---|
536 | // Check that everything starts ok now. |
---|
537 | MiniDFSCluster cluster = new MiniDFSCluster(conf, numDatanodes, false, null); |
---|
538 | cluster.waitActive(); |
---|
539 | cluster.shutdown(); |
---|
540 | } |
---|
541 | |
---|
542 | NameNode startNameNode( Configuration conf, |
---|
543 | String imageDirs, |
---|
544 | String editsDirs, |
---|
545 | StartupOption start) throws IOException { |
---|
546 | conf.set("fs.default.name", "hdfs://localhost:0"); |
---|
547 | conf.set("dfs.http.address", "0.0.0.0:0"); |
---|
548 | conf.set("dfs.name.dir", imageDirs); |
---|
549 | conf.set("dfs.name.edits.dir", editsDirs); |
---|
550 | String[] args = new String[]{start.getName()}; |
---|
551 | NameNode nn = NameNode.createNameNode(args, conf); |
---|
552 | assertTrue(nn.isInSafeMode()); |
---|
553 | return nn; |
---|
554 | } |
---|
555 | |
---|
556 | SecondaryNameNode startSecondaryNameNode(Configuration conf |
---|
557 | ) throws IOException { |
---|
558 | conf.set("dfs.secondary.http.address", "0.0.0.0:0"); |
---|
559 | return new SecondaryNameNode(conf); |
---|
560 | } |
---|
561 | |
---|
562 | /** |
---|
563 | * Tests checkpoint in HDFS. |
---|
564 | */ |
---|
565 | public void testCheckpoint() throws IOException { |
---|
566 | Path file1 = new Path("checkpoint.dat"); |
---|
567 | Path file2 = new Path("checkpoint2.dat"); |
---|
568 | Collection<File> namedirs = null; |
---|
569 | |
---|
570 | Configuration conf = new Configuration(); |
---|
571 | conf.set("dfs.secondary.http.address", "0.0.0.0:0"); |
---|
572 | replication = (short)conf.getInt("dfs.replication", 3); |
---|
573 | MiniDFSCluster cluster = new MiniDFSCluster(conf, numDatanodes, true, null); |
---|
574 | cluster.waitActive(); |
---|
575 | FileSystem fileSys = cluster.getFileSystem(); |
---|
576 | |
---|
577 | try { |
---|
578 | // |
---|
579 | // verify that 'format' really blew away all pre-existing files |
---|
580 | // |
---|
581 | assertTrue(!fileSys.exists(file1)); |
---|
582 | assertTrue(!fileSys.exists(file2)); |
---|
583 | namedirs = cluster.getNameDirs(); |
---|
584 | |
---|
585 | // |
---|
586 | // Create file1 |
---|
587 | // |
---|
588 | writeFile(fileSys, file1, replication); |
---|
589 | checkFile(fileSys, file1, replication); |
---|
590 | |
---|
591 | // |
---|
592 | // Take a checkpoint |
---|
593 | // |
---|
594 | SecondaryNameNode secondary = startSecondaryNameNode(conf); |
---|
595 | ErrorSimulator.initializeErrorSimulationEvent(3); |
---|
596 | secondary.doCheckpoint(); |
---|
597 | secondary.shutdown(); |
---|
598 | } finally { |
---|
599 | fileSys.close(); |
---|
600 | cluster.shutdown(); |
---|
601 | } |
---|
602 | |
---|
603 | // |
---|
604 | // Restart cluster and verify that file1 still exist. |
---|
605 | // |
---|
606 | cluster = new MiniDFSCluster(conf, numDatanodes, false, null); |
---|
607 | cluster.waitActive(); |
---|
608 | fileSys = cluster.getFileSystem(); |
---|
609 | try { |
---|
610 | // check that file1 still exists |
---|
611 | checkFile(fileSys, file1, replication); |
---|
612 | cleanupFile(fileSys, file1); |
---|
613 | |
---|
614 | // create new file file2 |
---|
615 | writeFile(fileSys, file2, replication); |
---|
616 | checkFile(fileSys, file2, replication); |
---|
617 | |
---|
618 | // |
---|
619 | // Take a checkpoint |
---|
620 | // |
---|
621 | SecondaryNameNode secondary = startSecondaryNameNode(conf); |
---|
622 | secondary.doCheckpoint(); |
---|
623 | secondary.shutdown(); |
---|
624 | } finally { |
---|
625 | fileSys.close(); |
---|
626 | cluster.shutdown(); |
---|
627 | } |
---|
628 | |
---|
629 | // |
---|
630 | // Restart cluster and verify that file2 exists and |
---|
631 | // file1 does not exist. |
---|
632 | // |
---|
633 | cluster = new MiniDFSCluster(conf, numDatanodes, false, null); |
---|
634 | cluster.waitActive(); |
---|
635 | fileSys = cluster.getFileSystem(); |
---|
636 | |
---|
637 | assertTrue(!fileSys.exists(file1)); |
---|
638 | |
---|
639 | try { |
---|
640 | // verify that file2 exists |
---|
641 | checkFile(fileSys, file2, replication); |
---|
642 | } finally { |
---|
643 | fileSys.close(); |
---|
644 | cluster.shutdown(); |
---|
645 | } |
---|
646 | |
---|
647 | // file2 is left behind. |
---|
648 | |
---|
649 | testSecondaryNamenodeError1(conf); |
---|
650 | testSecondaryNamenodeError2(conf); |
---|
651 | testSecondaryNamenodeError3(conf); |
---|
652 | testNamedirError(conf, namedirs); |
---|
653 | testSecondaryFailsToReturnImage(conf); |
---|
654 | testStartup(conf); |
---|
655 | } |
---|
656 | |
---|
657 | /** |
---|
658 | * Tests save namepsace. |
---|
659 | */ |
---|
660 | public void testSaveNamespace() throws IOException { |
---|
661 | MiniDFSCluster cluster = null; |
---|
662 | DistributedFileSystem fs = null; |
---|
663 | try { |
---|
664 | Configuration conf = new Configuration(); |
---|
665 | cluster = new MiniDFSCluster(conf, numDatanodes, false, null); |
---|
666 | cluster.waitActive(); |
---|
667 | fs = (DistributedFileSystem)(cluster.getFileSystem()); |
---|
668 | |
---|
669 | // Saving image without safe mode should fail |
---|
670 | DFSAdmin admin = new DFSAdmin(conf); |
---|
671 | String[] args = new String[]{"-saveNamespace"}; |
---|
672 | try { |
---|
673 | admin.run(args); |
---|
674 | } catch(IOException eIO) { |
---|
675 | assertTrue(eIO.getLocalizedMessage().contains("Safe mode should be turned ON")); |
---|
676 | } catch(Exception e) { |
---|
677 | throw new IOException(e); |
---|
678 | } |
---|
679 | // create new file |
---|
680 | Path file = new Path("namespace.dat"); |
---|
681 | writeFile(fs, file, replication); |
---|
682 | checkFile(fs, file, replication); |
---|
683 | // verify that the edits file is NOT empty |
---|
684 | Collection<File> editsDirs = cluster.getNameEditsDirs(); |
---|
685 | for(File ed : editsDirs) { |
---|
686 | assertTrue(new File(ed, "current/edits").length() > Integer.SIZE/Byte.SIZE); |
---|
687 | } |
---|
688 | |
---|
689 | // Saving image in safe mode should succeed |
---|
690 | fs.setSafeMode(SafeModeAction.SAFEMODE_ENTER); |
---|
691 | try { |
---|
692 | admin.run(args); |
---|
693 | } catch(Exception e) { |
---|
694 | throw new IOException(e); |
---|
695 | } |
---|
696 | // verify that the edits file is empty |
---|
697 | for(File ed : editsDirs) { |
---|
698 | assertTrue(new File(ed, "current/edits").length() == Integer.SIZE/Byte.SIZE); |
---|
699 | } |
---|
700 | |
---|
701 | // restart cluster and verify file exists |
---|
702 | cluster.shutdown(); |
---|
703 | cluster = null; |
---|
704 | |
---|
705 | cluster = new MiniDFSCluster(conf, numDatanodes, false, null); |
---|
706 | cluster.waitActive(); |
---|
707 | fs = (DistributedFileSystem)(cluster.getFileSystem()); |
---|
708 | checkFile(fs, file, replication); |
---|
709 | } finally { |
---|
710 | if(fs != null) fs.close(); |
---|
711 | if(cluster!= null) cluster.shutdown(); |
---|
712 | } |
---|
713 | } |
---|
714 | } |
---|