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.IOException;
21 import java.util.Collection;
22 import java.util.HashMap;
23 import java.util.HashSet;
24 import java.util.List;
25 import java.util.Map;
26 import java.util.Set;
27 import java.util.Timer;
28 import java.util.TimerTask;
29 import java.util.concurrent.locks.Lock;
30
31 import com.google.common.collect.Lists;
32 import org.apache.commons.logging.Log;
33 import org.apache.commons.logging.LogFactory;
34 import org.apache.hadoop.fs.PathFilter;
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.SnapshotDescriptionUtils;
43 import org.apache.hadoop.hbase.util.ArrayUtils;
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 /**
102 * Create a snapshot file cache for all snapshots under the specified [root]/.snapshot on the
103 * filesystem.
104 * <p>
105 * Immediately loads the file cache.
106 * @param conf to extract the configured {@link FileSystem} where the snapshots are stored and
107 * hbase root directory
108 * @param cacheRefreshPeriod frequency (ms) with which the cache should be refreshed
109 * @param refreshThreadName name of the cache refresh thread
110 * @param inspectSnapshotFiles Filter to apply to each snapshot to extract the files.
111 * @throws IOException if the {@link FileSystem} or root directory cannot be loaded
112 */
113 public SnapshotFileCache(Configuration conf, long cacheRefreshPeriod, String refreshThreadName,
114 SnapshotFileInspector inspectSnapshotFiles) throws IOException {
115 this(FSUtils.getCurrentFileSystem(conf), FSUtils.getRootDir(conf), 0, cacheRefreshPeriod,
116 refreshThreadName, inspectSnapshotFiles);
117 }
118
119 /**
120 * Create a snapshot file cache for all snapshots under the specified [root]/.snapshot on the
121 * filesystem
122 * @param fs {@link FileSystem} where the snapshots are stored
123 * @param rootDir hbase root directory
124 * @param cacheRefreshPeriod period (ms) with which the cache should be refreshed
125 * @param cacheRefreshDelay amount of time to wait for the cache to be refreshed
126 * @param refreshThreadName name of the cache refresh thread
127 * @param inspectSnapshotFiles Filter to apply to each snapshot to extract the files.
128 */
129 public SnapshotFileCache(FileSystem fs, Path rootDir, long cacheRefreshPeriod,
130 long cacheRefreshDelay, String refreshThreadName,
131 SnapshotFileInspector inspectSnapshotFiles) {
132 this.fs = fs;
133 this.fileInspector = inspectSnapshotFiles;
134 this.snapshotDir = SnapshotDescriptionUtils.getSnapshotsDir(rootDir);
135 // periodically refresh the file cache to make sure we aren't superfluously saving files.
136 this.refreshTimer = new Timer(refreshThreadName, true);
137 this.refreshTimer.scheduleAtFixedRate(new RefreshCacheTask(), cacheRefreshDelay,
138 cacheRefreshPeriod);
139 }
140
141 /**
142 * Trigger a cache refresh, even if its before the next cache refresh. Does not affect pending
143 * cache refreshes.
144 * <p/>
145 * Blocks until the cache is refreshed.
146 * <p/>
147 * Exposed for TESTING.
148 */
149 public synchronized void triggerCacheRefreshForTesting() {
150 try {
151 refreshCache();
152 } catch (IOException e) {
153 LOG.warn("Failed to refresh snapshot hfile cache!", e);
154 }
155 LOG.debug("Current cache:" + cache);
156 }
157
158 /**
159 * Check to see if any of the passed file names is contained in any of the snapshots. First checks
160 * an in-memory cache of the files to keep. If its not in the cache, then the cache is refreshed
161 * and the cache checked again for that file. This ensures that we never return files that exist.
162 * <p>
163 * Note this may lead to periodic false positives for the file being referenced. Periodically, the
164 * cache is refreshed even if there are no requests to ensure that the false negatives get removed
165 * eventually. For instance, suppose you have a file in the snapshot and it gets loaded into the
166 * cache. Then at some point later that snapshot is deleted. If the cache has not been refreshed
167 * at that point, cache will still think the file system contains that file and return
168 * <tt>true</tt>, even if it is no longer present (false positive). However, if the file never was
169 * on the filesystem, we will never find it and always return <tt>false</tt>.
170 * @param files file to check, NOTE: Relies that files are loaded from hdfs before method is
171 * called (NOT LAZY)
172 * @return <tt>unReferencedFiles</tt> the collection of files that do not have snapshot references
173 * @throws IOException if there is an unexpected error reaching the filesystem.
174 */
175 // XXX this is inefficient to synchronize on the method, when what we really need to guard against
176 // is an illegal access to the cache. Really we could do a mutex-guarded pointer swap on the
177 // cache, but that seems overkill at the moment and isn't necessarily a bottleneck.
178 public synchronized Iterable<FileStatus> getUnreferencedFiles(Iterable<FileStatus> files,
179 final SnapshotManager snapshotManager) throws IOException {
180 List<FileStatus> unReferencedFiles = Lists.newArrayList();
181 boolean refreshed = false;
182 Lock lock = null;
183 if (snapshotManager != null) {
184 lock = snapshotManager.getTakingSnapshotLock().writeLock();
185 }
186 if (lock == null || lock.tryLock()) {
187 try {
188 if (snapshotManager != null && snapshotManager.isTakingAnySnapshot()) {
189 LOG.warn("Not checking unreferenced files since snapshot is running, it will " +
190 "skip to clean the HFiles this time");
191 return unReferencedFiles;
192 }
193 for (FileStatus file : files) {
194 String fileName = file.getPath().getName();
195 if (!refreshed && !cache.contains(fileName)) {
196 refreshCache();
197 refreshed = true;
198 }
199 if (cache.contains(fileName)) {
200 continue;
201 }
202 unReferencedFiles.add(file);
203 }
204 } finally {
205 if (lock != null) {
206 lock.unlock();
207 }
208 }
209 }
210 return unReferencedFiles;
211 }
212
213 private void refreshCache() throws IOException {
214 // just list the snapshot directory directly, do not check the modification time for the root
215 // snapshot directory, as some file system implementations do not modify the parent directory's
216 // modTime when there are new sub items, for example, S3.
217 FileStatus[] snapshotDirs = FSUtils.listStatus(fs, snapshotDir, new PathFilter() {
218 @Override
219 public boolean accept(Path path) {
220 return !path.getName().equals(SnapshotDescriptionUtils.SNAPSHOT_TMP_DIR_NAME);
221 }
222 });
223 // clear the cache, as in the below code, either we will also clear the snapshots, or we will
224 // refill the file name cache again.
225 this.cache.clear();
226 if (ArrayUtils.isEmpty(snapshotDirs)) {
227 // remove all the remembered snapshots because we don't have any left
228 if (LOG.isDebugEnabled() && this.snapshots.size() > 0) {
229 LOG.debug("No snapshots on-disk, clear cache");
230 }
231 this.snapshots.clear();
232 return;
233 }
234
235 // iterate over all the cached snapshots and see if we need to update some, it is not an
236 // expensive operation if we do not reload the manifest of snapshots.
237 Map<String, SnapshotDirectoryInfo> newSnapshots = new HashMap<>();
238 for (FileStatus snapshotDir : snapshotDirs) {
239 String name = snapshotDir.getPath().getName();
240 SnapshotDirectoryInfo files = this.snapshots.remove(name);
241 // if we don't know about the snapshot or its been modified, we need to update the
242 // files the latter could occur where I create a snapshot, then delete it, and then make a
243 // new snapshot with the same name. We will need to update the cache the information from
244 // that new snapshot, even though it has the same name as the files referenced have
245 // probably changed.
246 if (files == null || files.hasBeenModified(snapshotDir.getModificationTime())) {
247 Collection<String> storedFiles = fileInspector.filesUnderSnapshot(snapshotDir.getPath());
248 files = new SnapshotDirectoryInfo(snapshotDir.getModificationTime(), storedFiles);
249 }
250 // add all the files to cache
251 this.cache.addAll(files.getFiles());
252 newSnapshots.put(name, files);
253 }
254 // set the snapshots we are tracking
255 this.snapshots.clear();
256 this.snapshots.putAll(newSnapshots);
257 }
258
259 /**
260 * Simple helper task that just periodically attempts to refresh the cache
261 */
262 public class RefreshCacheTask extends TimerTask {
263 @Override
264 public void run() {
265 synchronized (SnapshotFileCache.this) {
266 try {
267 SnapshotFileCache.this.refreshCache();
268 } catch (IOException e) {
269 LOG.warn("Failed to refresh snapshot hfile cache!", e);
270 // clear all the cached entries if we meet an error
271 cache.clear();
272 snapshots.clear();
273 }
274 }
275
276 }
277 }
278
279 @Override
280 public void stop(String why) {
281 if (!this.stop) {
282 this.stop = true;
283 this.refreshTimer.cancel();
284 }
285
286 }
287
288 @Override
289 public boolean isStopped() {
290 return this.stop;
291 }
292
293 /**
294 * Information about a snapshot directory
295 */
296 private static class SnapshotDirectoryInfo {
297 long lastModified;
298 Collection<String> files;
299
300 public SnapshotDirectoryInfo(long mtime, Collection<String> files) {
301 this.lastModified = mtime;
302 this.files = files;
303 }
304
305 /**
306 * @return the hfiles in the snapshot when <tt>this</tt> was made.
307 */
308 public Collection<String> getFiles() {
309 return this.files;
310 }
311
312 /**
313 * Check if the snapshot directory has been modified
314 * @param mtime current modification time of the directory
315 * @return <tt>true</tt> if it the modification time of the directory is newer time when we
316 * created <tt>this</tt>
317 */
318 public boolean hasBeenModified(long mtime) {
319 return this.lastModified < mtime;
320 }
321 }
322 }