View Javadoc

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.regionserver.snapshot;
19  
20  import java.io.IOException;
21  import java.util.ArrayList;
22  import java.util.Collection;
23  import java.util.List;
24  import java.util.concurrent.Callable;
25  import java.util.concurrent.ExecutionException;
26  import java.util.concurrent.ExecutorCompletionService;
27  import java.util.concurrent.Future;
28  import java.util.concurrent.LinkedBlockingQueue;
29  import java.util.concurrent.ThreadPoolExecutor;
30  import java.util.concurrent.TimeUnit;
31  
32  import org.apache.commons.logging.Log;
33  import org.apache.commons.logging.LogFactory;
34  import org.apache.hadoop.classification.InterfaceAudience;
35  import org.apache.hadoop.classification.InterfaceStability;
36  import org.apache.hadoop.conf.Configuration;
37  import org.apache.hadoop.hbase.DaemonThreadFactory;
38  import org.apache.hadoop.hbase.errorhandling.ForeignException;
39  import org.apache.hadoop.hbase.errorhandling.ForeignExceptionDispatcher;
40  import org.apache.hadoop.hbase.master.snapshot.MasterSnapshotVerifier;
41  import org.apache.hadoop.hbase.master.snapshot.SnapshotManager;
42  import org.apache.hadoop.hbase.procedure.ProcedureMember;
43  import org.apache.hadoop.hbase.procedure.ProcedureMemberRpcs;
44  import org.apache.hadoop.hbase.procedure.Subprocedure;
45  import org.apache.hadoop.hbase.procedure.SubprocedureFactory;
46  import org.apache.hadoop.hbase.procedure.ZKProcedureMemberRpcs;
47  import org.apache.hadoop.hbase.protobuf.generated.HBaseProtos.SnapshotDescription;
48  import org.apache.hadoop.hbase.regionserver.HRegion;
49  import org.apache.hadoop.hbase.regionserver.HRegionServer;
50  import org.apache.hadoop.hbase.regionserver.RegionServerServices;
51  import org.apache.hadoop.hbase.snapshot.SnapshotCreationException;
52  import org.apache.hadoop.hbase.util.Bytes;
53  import org.apache.hadoop.hbase.zookeeper.ZooKeeperWatcher;
54  import org.apache.zookeeper.KeeperException;
55  
56  import com.google.protobuf.InvalidProtocolBufferException;
57  
58  /**
59   * This manager class handles the work dealing with snapshots for a {@link HRegionServer}.
60   * <p>
61   * This provides the mechanism necessary to kick off a online snapshot specific
62   * {@link Subprocedure} that is responsible for the regions being served by this region server.
63   * If any failures occur with the subprocedure, the RegionSeverSnapshotManager's subprocedure
64   * handler, {@link ProcedureMember}, notifies the master's ProcedureCoordinator to abort all
65   * others.
66   * <p>
67   * On startup, requires {@link #start()} to be called.
68   * <p>
69   * On shutdown, requires {@link #stop(boolean)} to be called
70   */
71  @InterfaceAudience.Private
72  @InterfaceStability.Unstable
73  public class RegionServerSnapshotManager {
74    private static final Log LOG = LogFactory.getLog(RegionServerSnapshotManager.class);
75  
76    /** Maximum number of snapshot region tasks that can run concurrently */
77    private static final String CONCURENT_SNAPSHOT_TASKS_KEY = "hbase.snapshot.region.concurrentTasks";
78    private static final int DEFAULT_CONCURRENT_SNAPSHOT_TASKS = 3;
79  
80    /** Conf key for number of request threads to start snapshots on regionservers */
81    public static final String SNAPSHOT_REQUEST_THREADS_KEY = "hbase.snapshot.region.pool.threads";
82    /** # of threads for snapshotting regions on the rs. */
83    public static final int SNAPSHOT_REQUEST_THREADS_DEFAULT = 10;
84  
85    /** Conf key for max time to keep threads in snapshot request pool waiting */
86    public static final String SNAPSHOT_TIMEOUT_MILLIS_KEY = "hbase.snapshot.region.timeout";
87    /** Keep threads alive in request pool for max of 60 seconds */
88    public static final long SNAPSHOT_TIMEOUT_MILLIS_DEFAULT = 60000;
89  
90    /** Conf key for millis between checks to see if snapshot completed or if there are errors*/
91    public static final String SNAPSHOT_REQUEST_WAKE_MILLIS_KEY = "hbase.snapshot.region.wakefrequency";
92    /** Default amount of time to check for errors while regions finish snapshotting */
93    private static final long SNAPSHOT_REQUEST_WAKE_MILLIS_DEFAULT = 500;
94  
95    private final RegionServerServices rss;
96    private final ProcedureMemberRpcs memberRpcs;
97    private final ProcedureMember member;
98  
99    /**
100    * Exposed for testing.
101    * @param conf HBase configuration.
102    * @param parent parent running the snapshot handler
103    * @param memberRpc use specified memberRpc instance
104    * @param procMember use specified ProcedureMember
105    */
106    RegionServerSnapshotManager(Configuration conf, HRegionServer parent,
107       ProcedureMemberRpcs memberRpc, ProcedureMember procMember) {
108     this.rss = parent;
109     this.memberRpcs = memberRpc;
110     this.member = procMember;
111   }
112 
113   /**
114    * Create a default snapshot handler - uses a zookeeper based member controller.
115    * @param rss region server running the handler
116    * @throws KeeperException if the zookeeper cluster cannot be reached
117    */
118   public RegionServerSnapshotManager(RegionServerServices rss)
119       throws KeeperException {
120     this.rss = rss;
121     ZooKeeperWatcher zkw = rss.getZooKeeper();
122     this.memberRpcs = new ZKProcedureMemberRpcs(zkw,
123         SnapshotManager.ONLINE_SNAPSHOT_CONTROLLER_DESCRIPTION);
124 
125     // read in the snapshot request configuration properties
126     Configuration conf = rss.getConfiguration();
127     long wakeMillis = conf.getLong(SNAPSHOT_REQUEST_WAKE_MILLIS_KEY, SNAPSHOT_REQUEST_WAKE_MILLIS_DEFAULT);
128     long keepAlive = conf.getLong(SNAPSHOT_TIMEOUT_MILLIS_KEY, SNAPSHOT_TIMEOUT_MILLIS_DEFAULT);
129     int opThreads = conf.getInt(SNAPSHOT_REQUEST_THREADS_KEY, SNAPSHOT_REQUEST_THREADS_DEFAULT);
130 
131     // create the actual snapshot procedure member
132     ThreadPoolExecutor pool = ProcedureMember.defaultPool(wakeMillis, keepAlive, opThreads, 
133       rss.getServerName().toString());
134     this.member = new ProcedureMember(memberRpcs, pool, new SnapshotSubprocedureBuilder());
135   }
136 
137   /**
138    * Start accepting snapshot requests.
139    */
140   public void start() {
141     LOG.debug("Start Snapshot Manager " + rss.getServerName().toString());
142     this.memberRpcs.start(rss.getServerName().toString(), member);
143   }
144 
145   /**
146    * Close <tt>this</tt> and all running snapshot tasks
147    * @param force forcefully stop all running tasks
148    * @throws IOException
149    */
150   public void stop(boolean force) throws IOException {
151     String mode = force ? "abruptly" : "gracefully";
152     LOG.info("Stopping RegionServerSnapshotManager " + mode + ".");
153 
154     try {
155       this.member.close();
156     } finally {
157       this.memberRpcs.close();
158     }
159   }
160 
161   /**
162    * If in a running state, creates the specified subprocedure for handling an online snapshot.
163    *
164    * Because this gets the local list of regions to snapshot and not the set the master had,
165    * there is a possibility of a race where regions may be missed.  This detected by the master in
166    * the snapshot verification step.
167    *
168    * @param snapshot
169    * @return Subprocedure to submit to the ProcedureMemeber.
170    */
171   public Subprocedure buildSubprocedure(SnapshotDescription snapshot) {
172 
173     // don't run a snapshot if the parent is stop(ping)
174     if (rss.isStopping() || rss.isStopped()) {
175       throw new IllegalStateException("Can't start snapshot on RS: " + rss.getServerName()
176           + ", because stopping/stopped!");
177     }
178 
179     // check to see if this server is hosting any regions for the snapshots
180     // check to see if we have regions for the snapshot
181     List<HRegion> involvedRegions;
182     try {
183       involvedRegions = getRegionsToSnapshot(snapshot);
184     } catch (IOException e1) {
185       throw new IllegalStateException("Failed to figure out if we should handle a snapshot - "
186           + "something has gone awry with the online regions.", e1);
187     }
188 
189     // We need to run the subprocedure even if we have no relevant regions.  The coordinator
190     // expects participation in the procedure and without sending message the snapshot attempt
191     // will hang and fail.
192 
193     LOG.debug("Launching subprocedure for snapshot " + snapshot.getName() + " from table "
194         + snapshot.getTable());
195     ForeignExceptionDispatcher exnDispatcher = new ForeignExceptionDispatcher();
196     Configuration conf = rss.getConfiguration();
197     long timeoutMillis = conf.getLong(SNAPSHOT_TIMEOUT_MILLIS_KEY,
198         SNAPSHOT_TIMEOUT_MILLIS_DEFAULT);
199     long wakeMillis = conf.getLong(SNAPSHOT_REQUEST_WAKE_MILLIS_KEY,
200         SNAPSHOT_REQUEST_WAKE_MILLIS_DEFAULT);
201 
202     switch (snapshot.getType()) {
203     case FLUSH:
204       SnapshotSubprocedurePool taskManager =
205         new SnapshotSubprocedurePool(rss.getServerName().toString(), conf);
206       return new FlushSnapshotSubprocedure(member, exnDispatcher, wakeMillis,
207           timeoutMillis, involvedRegions, snapshot, taskManager);
208     default:
209       throw new UnsupportedOperationException("Unrecognized snapshot type:" + snapshot.getType());
210     }
211   }
212 
213   /**
214    * Determine if the snapshot should be handled on this server
215    *
216    * NOTE: This is racy -- the master expects a list of regionservers.
217    * This means if a region moves somewhere between the calls we'll miss some regions.
218    * For example, a region move during a snapshot could result in a region to be skipped or done
219    * twice.  This is manageable because the {@link MasterSnapshotVerifier} will double check the
220    * region lists after the online portion of the snapshot completes and will explicitly fail the
221    * snapshot.
222    *
223    * @param snapshot
224    * @return the list of online regions. Empty list is returned if no regions are responsible for
225    *         the given snapshot.
226    * @throws IOException
227    */
228   private List<HRegion> getRegionsToSnapshot(SnapshotDescription snapshot) throws IOException {
229     byte[] table = Bytes.toBytes(snapshot.getTable());
230     return rss.getOnlineRegions(table);
231   }
232 
233   /**
234    * Build the actual snapshot runner that will do all the 'hard' work
235    */
236   public class SnapshotSubprocedureBuilder implements SubprocedureFactory {
237 
238     @Override
239     public Subprocedure buildSubprocedure(String name, byte[] data) {
240       try {
241         // unwrap the snapshot information
242         SnapshotDescription snapshot = SnapshotDescription.parseFrom(data);
243         return RegionServerSnapshotManager.this.buildSubprocedure(snapshot);
244       } catch (InvalidProtocolBufferException e) {
245         throw new IllegalArgumentException("Could not read snapshot information from request.");
246       }
247     }
248 
249   }
250 
251   /**
252    * We use the SnapshotSubprocedurePool, a class specific thread pool instead of
253    * {@link org.apache.hadoop.hbase.executor.ExecutorService}.
254    *
255    * It uses a {@link java.util.concurrent.ExecutorCompletionService} which provides queuing of
256    * completed tasks which lets us efficiently cancel pending tasks upon the earliest operation
257    * failures.
258    *
259    * HBase's ExecutorService (different from {@link java.util.concurrent.ExecutorService}) isn't
260    * really built for coordinated tasks where multiple threads as part of one larger task.  In
261    * RS's the HBase Executor services are only used for open and close and not other threadpooled
262    * operations such as compactions and replication  sinks.
263    */
264   static class SnapshotSubprocedurePool {
265     private final ExecutorCompletionService<Void> taskPool;
266     private final ThreadPoolExecutor executor;
267     private volatile boolean stopped;
268     private final List<Future<Void>> futures = new ArrayList<Future<Void>>();
269     private final String name;
270 
271     SnapshotSubprocedurePool(String name, Configuration conf) {
272       // configure the executor service
273       long keepAlive = conf.getLong(
274         RegionServerSnapshotManager.SNAPSHOT_TIMEOUT_MILLIS_KEY,
275         RegionServerSnapshotManager.SNAPSHOT_TIMEOUT_MILLIS_DEFAULT);
276       int threads = conf.getInt(CONCURENT_SNAPSHOT_TASKS_KEY, DEFAULT_CONCURRENT_SNAPSHOT_TASKS);
277       this.name = name;
278       executor = new ThreadPoolExecutor(1, threads, keepAlive, TimeUnit.MILLISECONDS,
279           new LinkedBlockingQueue<Runnable>(), new DaemonThreadFactory("rs("
280               + name + ")-snapshot-pool"));
281       taskPool = new ExecutorCompletionService<Void>(executor);
282     }
283 
284     boolean hasTasks() {
285       return futures.size() != 0;
286     }
287 
288     /**
289      * Submit a task to the pool.
290      *
291      * NOTE: all must be submitted before you can safely {@link #waitForOutstandingTasks()}. This
292      * version does not support issuing tasks from multiple concurrent table snapshots requests.
293      */
294     void submitTask(final Callable<Void> task) {
295       Future<Void> f = this.taskPool.submit(task);
296       futures.add(f);
297     }
298 
299     /**
300      * Wait for all of the currently outstanding tasks submitted via {@link #submitTask(Callable)}.
301      * This *must* be called after all tasks are submitted via submitTask.
302      *
303      * @return <tt>true</tt> on success, <tt>false</tt> otherwise
304      * @throws InterruptedException
305      * @throws SnapshotCreationException if the snapshot failed while we were waiting
306      */
307     boolean waitForOutstandingTasks() throws ForeignException, InterruptedException {
308       LOG.debug("Waiting for local region snapshots to finish.");
309 
310       int sz = futures.size();
311       try {
312         // Using the completion service to process the futures that finish first first.
313         for (int i = 0; i < sz; i++) {
314           Future<Void> f = taskPool.take();
315           f.get();
316           if (!futures.remove(f)) {
317             LOG.warn("unexpected future" + f);
318           }
319           LOG.debug("Completed " + (i+1) + "/" + sz +  " local region snapshots.");
320         }
321         LOG.debug("Completed " + sz +  " local region snapshots.");
322         return true;
323       } catch (InterruptedException e) {
324         LOG.warn("Got InterruptedException in SnapshotSubprocedurePool", e);
325         if (!stopped) {
326           Thread.currentThread().interrupt();
327           throw new ForeignException("SnapshotSubprocedurePool", e);
328         }
329         // we are stopped so we can just exit.
330       } catch (ExecutionException e) {
331         if (e.getCause() instanceof ForeignException) {
332           LOG.warn("Rethrowing ForeignException from SnapshotSubprocedurePool", e);
333           throw (ForeignException)e.getCause();
334         }
335         LOG.warn("Got Exception in SnapshotSubprocedurePool", e);
336         throw new ForeignException(name, e.getCause());
337       } finally {
338         cancelTasks();
339       }
340       return false;
341     }
342 
343     /**
344      * This attempts to cancel out all pending and in progress tasks (interruptions issues)
345      * @throws InterruptedException
346      */
347     void cancelTasks() throws InterruptedException {
348       Collection<Future<Void>> tasks = futures;
349       LOG.debug("cancelling " + tasks.size() + " tasks for snapshot " + name);
350       for (Future<Void> f: tasks) {
351         // TODO Ideally we'd interrupt hbase threads when we cancel.  However it seems that there
352         // are places in the HBase code where row/region locks are taken and not released in a
353         // finally block.  Thus we cancel without interrupting.  Cancellations will be slower to
354         // complete but we won't suffer from unreleased locks due to poor code discipline.
355         f.cancel(false);
356       }
357 
358       // evict remaining tasks and futures from taskPool.
359       LOG.debug(taskPool);
360       while (!futures.isEmpty()) {
361         // block to remove cancelled futures;
362         LOG.warn("Removing cancelled elements from taskPool");
363         futures.remove(taskPool.take());
364       }
365       stop();
366     }
367 
368     /**
369      * Abruptly shutdown the thread pool.  Call when exiting a region server.
370      */
371     void stop() {
372       if (this.stopped) return;
373 
374       this.stopped = true;
375       this.executor.shutdownNow();
376     }
377   }
378 }