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