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.hbase.master.snapshot;
19
20 import java.io.FileNotFoundException;
21 import java.io.IOException;
22 import java.util.Collection;
23 import java.util.HashMap;
24 import java.util.HashSet;
25 import java.util.List;
26 import java.util.Map;
27 import java.util.Set;
28 import java.util.Timer;
29 import java.util.TimerTask;
30 import java.util.concurrent.locks.Lock;
31
32 import com.google.common.annotations.VisibleForTesting;
33 import com.google.common.collect.Lists;
34 import org.apache.commons.logging.Log;
35 import org.apache.commons.logging.LogFactory;
36 import org.apache.hadoop.hbase.classification.InterfaceAudience;
37 import org.apache.hadoop.hbase.classification.InterfaceStability;
38 import org.apache.hadoop.conf.Configuration;
39 import org.apache.hadoop.fs.FileStatus;
40 import org.apache.hadoop.fs.FileSystem;
41 import org.apache.hadoop.fs.Path;
42 import org.apache.hadoop.hbase.Stoppable;
43 import org.apache.hadoop.hbase.snapshot.CorruptedSnapshotException;
44 import org.apache.hadoop.hbase.snapshot.SnapshotDescriptionUtils;
45 import org.apache.hadoop.hbase.util.FSUtils;
46
47 /**
48 * Intelligently keep track of all the files for all the snapshots.
49 * <p>
50 * A cache of files is kept to avoid querying the {@link FileSystem} frequently. If there is a cache
51 * miss the directory modification time is used to ensure that we don't rescan directories that we
52 * already have in cache. We only check the modification times of the snapshot directories
53 * (/hbase/.snapshot/[snapshot_name]) to determine if the files need to be loaded into the cache.
54 * <p>
55 * New snapshots will be added to the cache and deleted snapshots will be removed when we refresh
56 * the cache. If the files underneath a snapshot directory are changed, but not the snapshot itself,
57 * we will ignore updates to that snapshot's files.
58 * <p>
59 * This is sufficient because each snapshot has its own directory and is added via an atomic rename
60 * <i>once</i>, when the snapshot is created. We don't need to worry about the data in the snapshot
61 * being run.
62 * <p>
63 * Further, the cache is periodically refreshed ensure that files in snapshots that were deleted are
64 * also removed from the cache.
65 * <p>
66 * A {@link SnapshotFileCache.SnapshotFileInspector} must be passed when creating <tt>this</tt> to
67 * allow extraction of files under /hbase/.snapshot/[snapshot name] directory, for each snapshot.
68 * This allows you to only cache files under, for instance, all the logs in the .logs directory or
69 * all the files under all the regions.
70 * <p>
71 * <tt>this</tt> also considers all running snapshots (those under /hbase/.snapshot/.tmp) as valid
72 * snapshots and will attempt to cache files from those snapshots as well.
73 * <p>
74 * Queries about a given file are thread-safe with respect to multiple queries and cache refreshes.
75 */
76 @InterfaceAudience.Private
77 @InterfaceStability.Evolving
78 public class SnapshotFileCache implements Stoppable {
79 interface SnapshotFileInspector {
80 /**
81 * Returns a collection of file names needed by the snapshot.
82 * @param snapshotDir {@link Path} to the snapshot directory to scan.
83 * @return the collection of file names needed by the snapshot.
84 */
85 Collection<String> filesUnderSnapshot(final Path snapshotDir) throws IOException;
86 }
87
88 private static final Log LOG = LogFactory.getLog(SnapshotFileCache.class);
89 private volatile boolean stop = false;
90 private final FileSystem fs;
91 private final SnapshotFileInspector fileInspector;
92 private final Path snapshotDir;
93 private final Set<String> cache = new HashSet<String>();
94 /**
95 * This is a helper map of information about the snapshot directories so we don't need to rescan
96 * them if they haven't changed since the last time we looked.
97 */
98 private final Map<String, SnapshotDirectoryInfo> snapshots =
99 new HashMap<String, SnapshotDirectoryInfo>();
100 private final Timer refreshTimer;
101
102 private long lastModifiedTime = Long.MIN_VALUE;
103
104 /**
105 * Create a snapshot file cache for all snapshots under the specified [root]/.snapshot on the
106 * filesystem.
107 * <p>
108 * Immediately loads the file cache.
109 * @param conf to extract the configured {@link FileSystem} where the snapshots are stored and
110 * hbase root directory
111 * @param cacheRefreshPeriod frequency (ms) with which the cache should be refreshed
112 * @param refreshThreadName name of the cache refresh thread
113 * @param inspectSnapshotFiles Filter to apply to each snapshot to extract the files.
114 * @throws IOException if the {@link FileSystem} or root directory cannot be loaded
115 */
116 public SnapshotFileCache(Configuration conf, long cacheRefreshPeriod, String refreshThreadName,
117 SnapshotFileInspector inspectSnapshotFiles) throws IOException {
118 this(FSUtils.getCurrentFileSystem(conf), FSUtils.getRootDir(conf), 0, cacheRefreshPeriod,
119 refreshThreadName, inspectSnapshotFiles);
120 }
121
122 /**
123 * Create a snapshot file cache for all snapshots under the specified [root]/.snapshot on the
124 * filesystem
125 * @param fs {@link FileSystem} where the snapshots are stored
126 * @param rootDir hbase root directory
127 * @param cacheRefreshPeriod period (ms) with which the cache should be refreshed
128 * @param cacheRefreshDelay amount of time to wait for the cache to be refreshed
129 * @param refreshThreadName name of the cache refresh thread
130 * @param inspectSnapshotFiles Filter to apply to each snapshot to extract the files.
131 */
132 public SnapshotFileCache(FileSystem fs, Path rootDir, long cacheRefreshPeriod,
133 long cacheRefreshDelay, String refreshThreadName, SnapshotFileInspector inspectSnapshotFiles) {
134 this.fs = fs;
135 this.fileInspector = inspectSnapshotFiles;
136 this.snapshotDir = SnapshotDescriptionUtils.getSnapshotsDir(rootDir);
137 // periodically refresh the file cache to make sure we aren't superfluously saving files.
138 this.refreshTimer = new Timer(refreshThreadName, true);
139 this.refreshTimer.scheduleAtFixedRate(new RefreshCacheTask(), cacheRefreshDelay,
140 cacheRefreshPeriod);
141 }
142
143 /**
144 * Trigger a cache refresh, even if its before the next cache refresh. Does not affect pending
145 * cache refreshes.
146 * <p>
147 * Blocks until the cache is refreshed.
148 * <p>
149 * Exposed for TESTING.
150 */
151 public void triggerCacheRefreshForTesting() {
152 try {
153 SnapshotFileCache.this.refreshCache();
154 } catch (IOException e) {
155 LOG.warn("Failed to refresh snapshot hfile cache!", e);
156 }
157 LOG.debug("Current cache:" + cache);
158 }
159
160 /**
161 * Check to see if any of the passed file names is contained in any of the snapshots.
162 * First checks an in-memory cache of the files to keep. If its not in the cache, then the cache
163 * is refreshed and the cache checked again for that file.
164 * This ensures that we never return files that exist.
165 * <p>
166 * Note this may lead to periodic false positives for the file being referenced. Periodically, the
167 * cache is refreshed even if there are no requests to ensure that the false negatives get removed
168 * eventually. For instance, suppose you have a file in the snapshot and it gets loaded into the
169 * cache. Then at some point later that snapshot is deleted. If the cache has not been refreshed
170 * at that point, cache will still think the file system contains that file and return
171 * <tt>true</tt>, even if it is no longer present (false positive). However, if the file never was
172 * on the filesystem, we will never find it and always return <tt>false</tt>.
173 * @param files file to check, NOTE: Relies that files are loaded from hdfs before method
174 * is called (NOT LAZY)
175 * @return <tt>unReferencedFiles</tt> the collection of files that do not have snapshot references
176 * @throws IOException if there is an unexpected error reaching the filesystem.
177 */
178 // XXX this is inefficient to synchronize on the method, when what we really need to guard against
179 // is an illegal access to the cache. Really we could do a mutex-guarded pointer swap on the
180 // cache, but that seems overkill at the moment and isn't necessarily a bottleneck.
181 public synchronized Iterable<FileStatus> getUnreferencedFiles(Iterable<FileStatus> files,
182 SnapshotManager snapshotManager) throws IOException {
183 List<FileStatus> unReferencedFiles = Lists.newArrayList();
184 List<String> snapshotsInProgress = null;
185 boolean refreshed = false;
186 Lock lock = null;
187 if (snapshotManager != null) {
188 lock = snapshotManager.getTakingSnapshotLock().writeLock();
189 }
190 if (lock == null || lock.tryLock()) {
191 try {
192 if (snapshotManager != null && snapshotManager.isTakingAnySnapshot()) {
193 LOG.warn("Not checking unreferenced files since snapshot is running, it will "
194 + "skip to clean the HFiles this time");
195 return unReferencedFiles;
196 }
197 for (FileStatus file : files) {
198 String fileName = file.getPath().getName();
199 if (!refreshed && !cache.contains(fileName)) {
200 refreshCache();
201 refreshed = true;
202 }
203 if (cache.contains(fileName)) {
204 continue;
205 }
206 if (snapshotsInProgress == null) {
207 snapshotsInProgress = getSnapshotsInProgress();
208 }
209 if (snapshotsInProgress.contains(fileName)) {
210 continue;
211 }
212 unReferencedFiles.add(file);
213 }
214 } finally {
215 if (lock != null) {
216 lock.unlock();
217 }
218 }
219 }
220 return unReferencedFiles;
221 }
222
223 private synchronized void refreshCache() throws IOException {
224 long lastTimestamp = Long.MAX_VALUE;
225 boolean hasChanges = false;
226
227 // get the status of the snapshots directory and check if it is has changes
228 try {
229 FileStatus dirStatus = fs.getFileStatus(snapshotDir);
230 lastTimestamp = dirStatus.getModificationTime();
231 hasChanges |= (lastTimestamp >= lastModifiedTime);
232 } catch (FileNotFoundException e) {
233 if (this.cache.size() > 0) {
234 LOG.error("Snapshot directory: " + snapshotDir + " doesn't exist");
235 }
236 return;
237 }
238
239 // get the status of the snapshots temporary directory and check if it has changes
240 // The top-level directory timestamp is not updated, so we have to check the inner-level.
241 try {
242 Path snapshotTmpDir = new Path(snapshotDir, SnapshotDescriptionUtils.SNAPSHOT_TMP_DIR_NAME);
243 FileStatus tempDirStatus = fs.getFileStatus(snapshotTmpDir);
244 lastTimestamp = Math.min(lastTimestamp, tempDirStatus.getModificationTime());
245 hasChanges |= (lastTimestamp >= lastModifiedTime);
246 if (!hasChanges) {
247 FileStatus[] tmpSnapshots = FSUtils.listStatus(fs, snapshotDir);
248 if (tmpSnapshots != null) {
249 for (FileStatus dirStatus: tmpSnapshots) {
250 lastTimestamp = Math.min(lastTimestamp, dirStatus.getModificationTime());
251 }
252 hasChanges |= (lastTimestamp >= lastModifiedTime);
253 }
254 }
255 } catch (FileNotFoundException e) {
256 // Nothing todo, if the tmp dir is empty
257 }
258
259 // if the snapshot directory wasn't modified since we last check, we are done
260 if (!hasChanges) {
261 return;
262 }
263
264 // directory was modified, so we need to reload our cache
265 // there could be a slight race here where we miss the cache, check the directory modification
266 // time, then someone updates the directory, causing us to not scan the directory again.
267 // However, snapshot directories are only created once, so this isn't an issue.
268
269 // 1. update the modified time
270 this.lastModifiedTime = lastTimestamp;
271
272 // 2.clear the cache
273 this.cache.clear();
274 Map<String, SnapshotDirectoryInfo> known = new HashMap<String, SnapshotDirectoryInfo>();
275
276 // 3. check each of the snapshot directories
277 FileStatus[] snapshots = FSUtils.listStatus(fs, snapshotDir);
278 if (snapshots == null) {
279 // remove all the remembered snapshots because we don't have any left
280 if (LOG.isDebugEnabled() && this.snapshots.size() > 0) {
281 LOG.debug("No snapshots on-disk, cache empty");
282 }
283 this.snapshots.clear();
284 return;
285 }
286
287 // 3.1 iterate through the on-disk snapshots
288 for (FileStatus snapshot : snapshots) {
289 String name = snapshot.getPath().getName();
290 // its not the tmp dir,
291 if (!name.equals(SnapshotDescriptionUtils.SNAPSHOT_TMP_DIR_NAME)) {
292 SnapshotDirectoryInfo files = this.snapshots.remove(name);
293 // 3.1.1 if we don't know about the snapshot or its been modified, we need to update the
294 // files the latter could occur where I create a snapshot, then delete it, and then make a
295 // new snapshot with the same name. We will need to update the cache the information from
296 // that new snapshot, even though it has the same name as the files referenced have
297 // probably changed.
298 if (files == null || files.hasBeenModified(snapshot.getModificationTime())) {
299 // get all files for the snapshot and create a new info
300 Collection<String> storedFiles = fileInspector.filesUnderSnapshot(snapshot.getPath());
301 files = new SnapshotDirectoryInfo(snapshot.getModificationTime(), storedFiles);
302 }
303 // 3.2 add all the files to cache
304 this.cache.addAll(files.getFiles());
305 known.put(name, files);
306 }
307 }
308
309 // 4. set the snapshots we are tracking
310 this.snapshots.clear();
311 this.snapshots.putAll(known);
312 }
313
314 @VisibleForTesting
315 List<String> getSnapshotsInProgress() throws IOException {
316 List<String> snapshotInProgress = Lists.newArrayList();
317 // only add those files to the cache, but not to the known snapshots
318 Path snapshotTmpDir = new Path(snapshotDir, SnapshotDescriptionUtils.SNAPSHOT_TMP_DIR_NAME);
319 // only add those files to the cache, but not to the known snapshots
320 FileStatus[] running = FSUtils.listStatus(fs, snapshotTmpDir);
321 if (running != null) {
322 for (FileStatus run : running) {
323 try {
324 snapshotInProgress.addAll(fileInspector.filesUnderSnapshot(run.getPath()));
325 } catch (CorruptedSnapshotException e) {
326 // See HBASE-16464
327 if (e.getCause() instanceof FileNotFoundException) {
328 // If the snapshot is corrupt, we will delete it
329 fs.delete(run.getPath(), true);
330 LOG.warn("delete the " + run.getPath() + " due to exception:", e.getCause());
331 } else {
332 throw e;
333 }
334 }
335 }
336 }
337 return snapshotInProgress;
338 }
339
340 /**
341 * Simple helper task that just periodically attempts to refresh the cache
342 */
343 public class RefreshCacheTask extends TimerTask {
344 @Override
345 public void run() {
346 try {
347 SnapshotFileCache.this.refreshCache();
348 } catch (IOException e) {
349 LOG.warn("Failed to refresh snapshot hfile cache!", e);
350 }
351 }
352 }
353
354 @Override
355 public void stop(String why) {
356 if (!this.stop) {
357 this.stop = true;
358 this.refreshTimer.cancel();
359 }
360
361 }
362
363 @Override
364 public boolean isStopped() {
365 return this.stop;
366 }
367
368 /**
369 * Information about a snapshot directory
370 */
371 private static class SnapshotDirectoryInfo {
372 long lastModified;
373 Collection<String> files;
374
375 public SnapshotDirectoryInfo(long mtime, Collection<String> files) {
376 this.lastModified = mtime;
377 this.files = files;
378 }
379
380 /**
381 * @return the hfiles in the snapshot when <tt>this</tt> was made.
382 */
383 public Collection<String> getFiles() {
384 return this.files;
385 }
386
387 /**
388 * Check if the snapshot directory has been modified
389 * @param mtime current modification time of the directory
390 * @return <tt>true</tt> if it the modification time of the directory is newer time when we
391 * created <tt>this</tt>
392 */
393 public boolean hasBeenModified(long mtime) {
394 return this.lastModified < mtime;
395 }
396 }
397 }