Thread: [Bigdata-commit] SF.net SVN: bigdata:[7614] branches/MGC_1_3_0

SourceForge Headquarters 225 Broadway Suite 1600 San Diego, CA 92101 +1 (858) 422-6466

Revision: 7614
          http://bigdata.svn.sourceforge.net/bigdata/?rev=7614&view=rev
Author:   martyncutcher
Date:     2013-12-09 08:18:21 +0000 (Mon, 09 Dec 2013)
Log Message:
-----------
Add service specific methods for doPipelineRemove/doWithdrawVote.. etc to allow the forcing of zookeeper state changes by "controlling" services, where previously global zookeeper state changes were only triggered by the effected service.  This has been introduced specifically to support the forced removal of a service as a result of a sudden death or wire pull.

Modified Paths:
--------------
    branches/MGC_1_3_0/bigdata/src/java/com/bigdata/quorum/AbstractQuorum.java
    branches/MGC_1_3_0/bigdata/src/java/com/bigdata/quorum/QuorumActor.java
    branches/MGC_1_3_0/bigdata/src/test/com/bigdata/quorum/MockQuorumFixture.java
    branches/MGC_1_3_0/bigdata-jini/src/java/com/bigdata/quorum/zk/ZKQuorumImpl.java

Modified: branches/MGC_1_3_0/bigdata/src/java/com/bigdata/quorum/AbstractQuorum.java
===================================================================

--- branches/MGC_1_3_0/bigdata/src/java/com/bigdata/quorum/AbstractQuorum.java	2013-12-09 08:13:26 UTC (rev 7613)
+++ branches/MGC_1_3_0/bigdata/src/java/com/bigdata/quorum/AbstractQuorum.java	2013-12-09 08:18:21 UTC (rev 7614)
@@ -2289,16 +2289,36 @@
 
         abstract protected void doMemberAdd();
 
-        abstract protected void doMemberRemove();
+        final protected void doMemberRemove() {
+            doMemberRemove(serviceId);
+        }
 
+        abstract protected void doMemberRemove(UUID serviceId);
+
         abstract protected void doCastVote(long lastCommitTime);
 
-        abstract protected void doWithdrawVote();
+        final protected void doWithdrawVote() {
+            doWithdrawVote(serviceId);
+        }
 
+        abstract protected void doWithdrawVote(UUID serviceId);
+
         abstract protected void doPipelineAdd();
 
-        abstract protected void doPipelineRemove();
+        final protected void doPipelineRemove() {
+            doPipelineRemove(serviceId);
+        }
 
+        abstract protected void doPipelineRemove(UUID serviceId);
+
+        @Override
+        public void forceRemoveService(final UUID psid) {
+            doMemberRemove(psid);
+            doWithdrawVote(psid);
+            doPipelineRemove(psid);
+            doServiceLeave(psid);
+        }
+
         /**
          * Invoked when our client will become the leader to (a) reorganize the
          * write pipeline such that our client is the first service in the write
@@ -2396,8 +2416,12 @@
         
         abstract protected void doServiceJoin();
 
-        abstract protected void doServiceLeave();
+        final protected void doServiceLeave() {
+            doServiceLeave(serviceId);
+        }
 
+        abstract protected void doServiceLeave(UUID serviceId);
+
         abstract protected void doSetToken(long newToken);
 
 //        abstract protected void doSetLastValidToken(long newToken);

Modified: branches/MGC_1_3_0/bigdata/src/java/com/bigdata/quorum/QuorumActor.java
===================================================================
--- branches/MGC_1_3_0/bigdata/src/java/com/bigdata/quorum/QuorumActor.java	2013-12-09 08:13:26 UTC (rev 7613)
+++ branches/MGC_1_3_0/bigdata/src/java/com/bigdata/quorum/QuorumActor.java	2013-12-09 08:18:21 UTC (rev 7614)
@@ -205,4 +205,21 @@
      */
     void clearToken();
 
+    /**
+     * Remove the service from the quorum. This should be called when a problem
+     * with the service is reported to the quorum leader, for example as a
+     * result of a failed RMI request or failed socket level write replication.
+     * Such errors arise either from network connectivity or service death.
+     * These problems will generally be cured, but the heatbeat timeout to cure
+     * the problem can cause write replication to block. This method may be used
+     * to force the timely reordering of the pipeline in order to work around
+     * the replication problem. This is not a permenant disabling of the service
+     * - the service may be restarted or may recover and reenter the quorum at
+     * any time.
+     * 
+     * @param serviceId
+     *            The UUID of the service to be removed.
+     */
+    public void forceRemoveService(UUID serviceId);
+
 }

Modified: branches/MGC_1_3_0/bigdata/src/test/com/bigdata/quorum/MockQuorumFixture.java
===================================================================
--- branches/MGC_1_3_0/bigdata/src/test/com/bigdata/quorum/MockQuorumFixture.java	2013-12-09 08:13:26 UTC (rev 7613)
+++ branches/MGC_1_3_0/bigdata/src/test/com/bigdata/quorum/MockQuorumFixture.java	2013-12-09 08:18:21 UTC (rev 7614)
@@ -846,32 +846,20 @@
                 fixture.memberAdd(serviceId);
             }
 
-            protected void doMemberRemove() {
-                fixture.memberRemove(serviceId);
-            }
-
             protected void doCastVote(final long lastCommitTime) {
                 fixture.castVote(serviceId, lastCommitTime);
             }
 
-            protected void doWithdrawVote() {
-                fixture.withdrawVote(serviceId);
-            }
-
             protected void doPipelineAdd() {
                 fixture.pipelineAdd(serviceId);
             }
 
-            protected void doPipelineRemove() {
-                fixture.pipelineRemove(serviceId);
-            }
-
             protected void doServiceJoin() {
                 fixture.serviceJoin(serviceId);
             }
 
-            protected void doServiceLeave() {
-                fixture.serviceLeave(serviceId);
+            protected void doServiceLeave(final UUID service) {
+                fixture.serviceLeave(service);
             }
 
             protected void doSetToken(final long newToken) {
@@ -890,6 +878,21 @@
                 fixture.clearToken();
             }
 
+			@Override
+			protected void doMemberRemove(UUID service) {
+				fixture.memberRemove(service);
+			}
+
+			@Override
+			protected void doWithdrawVote(UUID service) {
+				fixture.withdrawVote(service);
+			}
+
+			@Override
+			protected void doPipelineRemove(UUID service) {
+				fixture.pipelineRemove(service);
+			}
+
 //            /**
 //             * {@inheritDoc}
 //             * <p>

Modified: branches/MGC_1_3_0/bigdata-jini/src/java/com/bigdata/quorum/zk/ZKQuorumImpl.java
===================================================================
--- branches/MGC_1_3_0/bigdata-jini/src/java/com/bigdata/quorum/zk/ZKQuorumImpl.java	2013-12-09 08:13:26 UTC (rev 7613)
+++ branches/MGC_1_3_0/bigdata-jini/src/java/com/bigdata/quorum/zk/ZKQuorumImpl.java	2013-12-09 08:18:21 UTC (rev 7614)
@@ -327,7 +327,7 @@
         }
 
         @Override
-        protected void doMemberRemove() {
+        protected void doMemberRemove(final UUID service) {
             // get a valid zookeeper connection object.
             final ZooKeeper zk;
             try {
@@ -340,7 +340,7 @@
             try {
                 zk.delete(logicalServiceId + "/" + QUORUM + "/"
                                 + QUORUM_MEMBER + "/" + QUORUM_MEMBER_PREFIX
-                                + serviceIdStr, -1/* anyVersion */);
+                                + service.toString(), -1/* anyVersion */);
             } catch (NoNodeException e) {
                 // ignore.
             } catch (KeeperException e) {
@@ -414,7 +414,7 @@
         }
 
         @Override
-        protected void doPipelineRemove() {
+        protected void doPipelineRemove(final UUID service) {
             // get a valid zookeeper connection object.
             final ZooKeeper zk;
             try {
@@ -446,7 +446,7 @@
                         }
                         final QuorumServiceState state = (QuorumServiceState) SerializerUtil
                                 .deserialize(b);
-                        if (serviceId.equals(state.serviceUUID())) {
+                        if (service.equals(state.serviceUUID())) {
                             zk.delete(zpath + "/" + s, -1/* anyVersion */);
                             return;
                         }
@@ -636,7 +636,7 @@
          * handles a concurrent delete by a simple retry loop.
          */
         @Override
-        protected void doWithdrawVote() {
+        protected void doWithdrawVote(final UUID service) {
             // zpath for votes.
             final String votesZPath = getVotesZPath();
             if (log.isInfoEnabled())
@@ -724,7 +724,7 @@
                         Thread.currentThread().interrupt();
                         return;
                     }
-                    if (serviceId.equals(state.serviceUUID())) {
+                    if (service.equals(state.serviceUUID())) {
                         // found our vote.
                         try {
                             // delete our vote.
@@ -761,7 +761,7 @@
                             }
                             // done.
                             if (log.isInfoEnabled())
-                                log.info("withdrawn: serviceId=" + serviceIdStr
+                                log.info("withdrawn: serviceId=" + service.toString()
                                         + ", lastCommitTime=" + lastCommitTime);
                             return;
                         } catch (NoNodeException e) {
@@ -836,7 +836,7 @@
         }
 
         @Override
-        protected void doServiceLeave() {
+        protected void doServiceLeave(final UUID service) {
             // get a valid zookeeper connection object.
             final ZooKeeper zk;
             try {
@@ -871,7 +871,7 @@
                         }
                         final QuorumServiceState state = (QuorumServiceState) SerializerUtil
                                 .deserialize(b);
-                        if (serviceId.equals(state.serviceUUID())) {
+                        if (service.equals(state.serviceUUID())) {
                             // Found this service.
                             zk.delete(zpath + "/" + s, -1/* anyVersion */);
                             return;

This was sent by the SourceForge.net collaborative development platform, the world's largest Open Source development site.





Thread: [Bigdata-commit] SF.net SVN: bigdata:[7614] branches/MGC_1_3_0

Fast, scalable, robust graph database platform

bigdata-commit