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.procedure;
19  
20  import java.io.IOException;
21  import java.util.Collection;
22  import java.util.HashSet;
23  import java.util.List;
24  import java.util.Set;
25  import java.util.concurrent.ConcurrentMap;
26  import java.util.concurrent.ExecutorService;
27  import java.util.concurrent.Future;
28  import java.util.concurrent.RejectedExecutionException;
29  import java.util.concurrent.SynchronousQueue;
30  import java.util.concurrent.ThreadPoolExecutor;
31  import java.util.concurrent.TimeUnit;
32  
33  import org.apache.commons.logging.Log;
34  import org.apache.commons.logging.LogFactory;
35  import org.apache.hadoop.classification.InterfaceAudience;
36  import org.apache.hadoop.classification.InterfaceStability;
37  import org.apache.hadoop.hbase.DaemonThreadFactory;
38  import org.apache.hadoop.hbase.errorhandling.ForeignException;
39  import org.apache.hadoop.hbase.errorhandling.ForeignExceptionDispatcher;
40  
41  import com.google.common.collect.MapMaker;
42  
43  /**
44   * This is the master side of a distributed complex procedure execution.
45   * <p>
46   * The {@link Procedure} is generic and subclassing or customization shouldn't be
47   * necessary -- any customization should happen just in {@link Subprocedure}s.
48   */
49  @InterfaceAudience.Public
50  @InterfaceStability.Evolving
51  public class ProcedureCoordinator {
52    private static final Log LOG = LogFactory.getLog(ProcedureCoordinator.class);
53  
54    final static long KEEP_ALIVE_MILLIS_DEFAULT = 5000;
55    final static long TIMEOUT_MILLIS_DEFAULT = 60000;
56    final static long WAKE_MILLIS_DEFAULT = 500;
57  
58    private final ProcedureCoordinatorRpcs rpcs;
59    private final ExecutorService pool;
60    private final long wakeTimeMillis;
61    private final long timeoutMillis;
62  
63    // Running procedure table.  Maps procedure name to running procedure reference
64    private final ConcurrentMap<String, Procedure> procedures =
65        new MapMaker().concurrencyLevel(4).weakValues().makeMap();
66  
67    /**
68     * Create and start a ProcedureCoordinator.
69     *
70     * The rpc object registers the ProcedureCoordinator and starts any threads in this
71     * constructor.
72     *
73     * @param rpcs
74     * @param pool Used for executing procedures.
75     */
76    public ProcedureCoordinator(ProcedureCoordinatorRpcs rpcs, ThreadPoolExecutor pool) {
77      this(rpcs, pool, TIMEOUT_MILLIS_DEFAULT, WAKE_MILLIS_DEFAULT);
78    }
79  
80    /**
81     * Create and start a ProcedureCoordinator.
82     *
83     * The rpc object registers the ProcedureCoordinator and starts any threads in
84     * this constructor.
85     *
86     * @param rpcs
87     * @param pool Used for executing procedures.
88     * @param timeoutMillis
89     */
90    public ProcedureCoordinator(ProcedureCoordinatorRpcs rpcs, ThreadPoolExecutor pool,
91        long timeoutMillis, long wakeTimeMillis) {
92      this.timeoutMillis = timeoutMillis;
93      this.wakeTimeMillis = wakeTimeMillis;
94      this.rpcs = rpcs;
95      this.pool = pool;
96      this.rpcs.start(this);
97    }
98  
99    /**
100    * Default thread pool for the procedure
101    *
102    * @param coordName
103    * @param opThreads the maximum number of threads to allow in the pool
104    */
105   public static ThreadPoolExecutor defaultPool(String coordName, int opThreads) {
106     return defaultPool(coordName, opThreads, KEEP_ALIVE_MILLIS_DEFAULT);
107   }
108 
109   /**
110    * Default thread pool for the procedure
111    *
112    * @param coordName
113    * @param opThreads the maximum number of threads to allow in the pool
114    * @param keepAliveMillis the maximum time (ms) that excess idle threads will wait for new tasks
115    */
116   public static ThreadPoolExecutor defaultPool(String coordName, int opThreads,
117       long keepAliveMillis) {
118     return new ThreadPoolExecutor(1, opThreads, keepAliveMillis, TimeUnit.MILLISECONDS,
119         new SynchronousQueue<Runnable>(),
120         new DaemonThreadFactory("(" + coordName + ")-proc-coordinator-pool"));
121   }
122 
123   /**
124    * Shutdown the thread pools and release rpc resources
125    * @throws IOException
126    */
127   public void close() throws IOException {
128     // have to use shutdown now to break any latch waiting
129     pool.shutdownNow();
130     rpcs.close();
131   }
132 
133   /**
134    * Submit an procedure to kick off its dependent subprocedures.
135    * @param proc Procedure to execute
136    * @return <tt>true</tt> if the procedure was started correctly, <tt>false</tt> if the
137    *         procedure or any subprocedures could not be started.  Failure could be due to
138    *         submitting a procedure multiple times (or one with the same name), or some sort
139    *         of IO problem.  On errors, the procedure's monitor holds a reference to the exception
140    *         that caused the failure.
141    */
142   boolean submitProcedure(Procedure proc) {
143     // if the submitted procedure was null, then we don't want to run it
144     if (proc == null) {
145       return false;
146     }
147     String procName = proc.getName();
148 
149     // make sure we aren't already running a procedure of that name
150     synchronized (procedures) {
151       Procedure oldProc = procedures.get(procName);
152       if (oldProc != null) {
153         // procedures are always eventually completed on both successful and failed execution
154         if (oldProc.completedLatch.getCount() != 0) {
155           LOG.warn("Procedure " + procName + " currently running.  Rejecting new request");
156           return false;
157         }
158         LOG.debug("Procedure " + procName + " was in running list but was completed.  Accepting new attempt.");
159         procedures.remove(procName);
160       }
161     }
162 
163     // kick off the procedure's execution in a separate thread
164     Future<Void> f = null;
165     try {
166       synchronized (procedures) {
167         f = this.pool.submit(proc);
168         // if everything got started properly, we can add it known running procedures
169         this.procedures.put(procName, proc);
170       }
171       return true;
172     } catch (RejectedExecutionException e) {
173       LOG.warn("Procedure " + procName + " rejected by execution pool.  Propagating error and " +
174           "cancelling operation.", e);
175       // the thread pool is full and we can't run the procedure
176       proc.receive(new ForeignException(procName, e));
177 
178       // cancel procedure proactively
179       if (f != null) {
180         f.cancel(true);
181       }
182     }
183     return false;
184   }
185 
186   /**
187    * The connection to the rest of the procedure group (members and coordinator) has been
188    * broken/lost/failed. This should fail any interested procedures, but not attempt to notify other
189    * members since we cannot reach them anymore.
190    * @param message description of the error
191    * @param cause the actual cause of the failure
192    */
193   void rpcConnectionFailure(final String message, final IOException cause) {
194     Collection<Procedure> toNotify = procedures.values();
195 
196     for (Procedure proc : toNotify) {
197       if (proc == null) {
198         continue;
199       }
200       // notify the elements, if they aren't null
201       proc.receive(new ForeignException(proc.getName(), cause));
202     }
203   }
204 
205   /**
206    * Abort the procedure with the given name
207    * @param procName name of the procedure to abort
208    * @param reason serialized information about the abort
209    */
210   public void abortProcedure(String procName, ForeignException reason) {
211     // if we know about the Procedure, notify it
212     synchronized(procedures) {
213       Procedure proc = procedures.get(procName);
214       if (proc == null) {
215         return;
216       }
217       proc.receive(reason);
218     }
219   }
220 
221   /**
222    * Exposed for hooking with unit tests.
223    * @param procName
224    * @param procArgs
225    * @param expectedMembers
226    * @return
227    */
228   Procedure createProcedure(ForeignExceptionDispatcher fed, String procName, byte[] procArgs,
229       List<String> expectedMembers) {
230     // build the procedure
231     return new Procedure(this, fed, wakeTimeMillis, timeoutMillis,
232         procName, procArgs, expectedMembers);
233   }
234 
235   /**
236    * Kick off the named procedure
237    * @param procName name of the procedure to start
238    * @param procArgs arguments for the procedure
239    * @param expectedMembers expected members to start
240    * @return handle to the running procedure, if it was started correctly, <tt>null</tt> otherwise
241    * @throws RejectedExecutionException if there are no more available threads to run the procedure
242    */
243   public Procedure startProcedure(ForeignExceptionDispatcher fed, String procName, byte[] procArgs,
244       List<String> expectedMembers) throws RejectedExecutionException {
245     Procedure proc = createProcedure(fed, procName, procArgs, expectedMembers);
246     if (!this.submitProcedure(proc)) {
247       LOG.error("Failed to submit procedure '" + procName + "'");
248       return null;
249     }
250     return proc;
251   }
252 
253   /**
254    * Notification that the procedure had the specified member acquired its part of the barrier
255    * via {@link Subprocedure#acquireBarrier()}.
256    * @param procName name of the procedure that acquired
257    * @param member name of the member that acquired
258    */
259   void memberAcquiredBarrier(String procName, final String member) {
260     Procedure proc = procedures.get(procName);
261     if (proc != null) {
262       proc.barrierAcquiredByMember(member);
263     }
264   }
265 
266   /**
267    * Notification that the procedure had another member finished executing its in-barrier subproc
268    * via {@link Subprocedure#insideBarrier()}.
269    * @param procName name of the subprocedure that finished
270    * @param member name of the member that executed and released its barrier
271    */
272   void memberFinishedBarrier(String procName, final String member) {
273     Procedure proc = procedures.get(procName);
274     if (proc != null) {
275       proc.barrierReleasedByMember(member);
276     }
277   }
278 
279   /**
280    * @return the rpcs implementation for all current procedures
281    */
282   ProcedureCoordinatorRpcs getRpcs() {
283     return rpcs;
284   }
285 
286   /**
287    * Returns the procedure.  This Procedure is a live instance so should not be modified but can
288    * be inspected.
289    * @param name Name of the procedure
290    * @return Procedure or null if not present any more
291    */
292   public Procedure getProcedure(String name) {
293     return procedures.get(name);
294   }
295 
296   /**
297    * @return Return set of all procedure names.
298    */
299   public Set<String> getProcedureNames() {
300     return new HashSet<String>(procedures.keySet());
301   }
302 }