big-data/hadoop
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

HADOOP-9459. ActiveStandbyElector can join election even before Service HEALTHY, and results in null data at ActiveBreadCrumb. Contributed by Vinay and Todd Lipcon.

Browse Source 
git-svn-id: https://svn.apache.org/repos/asf/hadoop/common/trunk@1482227 13f79535-47bb-0310-9956-ffa450edef68
This commit is contained in:
Todd Lipcon 2013-05-14 06:43:47 +00:00
parent febf951220
commit 8a4ff65b3c
3 changed files with 34 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
hadoop-common-project/hadoop-common/CHANGES.txt
View File

@ -710,6 +710,10 @@ Release 2.0.5-beta - UNRELEASED
HADOOP-9485. No default value in the code for
hadoop.rpc.socket.factory.class.default. (Colin Patrick McCabe via atm)
HADOOP-9459. ActiveStandbyElector can join election even before
Service HEALTHY, and results in null data at ActiveBreadCrumb.
(Vinay and todd via todd)
Release 2.0.4-alpha - 2013-04-25
INCOMPATIBLE CHANGES

14
hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/ha/ActiveStandbyElector.java
View File

@ -643,6 +643,8 @@ private void monitorActiveStatus() {
}
private void joinElectionInternal() {
Preconditions.checkState(appData != null,
"trying to join election without any app data");
if (zkClient == null) {
if (!reEstablishSession()) {
fatalError("Failed to reEstablish connection with ZooKeeper");
@ -669,8 +671,14 @@ private void reJoinElection(int sleepTime) {
try {
terminateConnection();
sleepFor(sleepTime);
joinElectionInternal();
// Should not join election even before the SERVICE is reported
// as HEALTHY from ZKFC monitoring.
if (appData != null) {
joinElectionInternal();
} else {
LOG.info("Not joining election since service has not yet been " +
"reported as healthy.");
}
} finally {
sessionReestablishLockForTests.unlock();
}
@ -798,6 +806,8 @@ private boolean becomeActive() {
*/
private void writeBreadCrumbNode(Stat oldBreadcrumbStat)
throws KeeperException, InterruptedException {
Preconditions.checkState(appData != null, "no appdata");
LOG.info("Writing znode " + zkBreadCrumbPath +
" to indicate that the local node is the most recent active...");
if (oldBreadcrumbStat == null) {

18
hadoop-common-project/hadoop-common/src/test/java/org/apache/hadoop/ha/TestActiveStandbyElector.java
View File

@ -721,4 +721,22 @@ public void testWithoutZKServer() throws Exception {
GenericTestUtils.assertExceptionContains( "ConnectionLoss", ke);
}
}
/**
* joinElection(..) should happen only after SERVICE_HEALTHY.
*/
@Test
public void testBecomeActiveBeforeServiceHealthy() throws Exception {
mockNoPriorActive();
WatchedEvent mockEvent = Mockito.mock(WatchedEvent.class);
Mockito.when(mockEvent.getType()).thenReturn(Event.EventType.None);
// session expired should enter safe mode
// But for first time, before the SERVICE_HEALTY i.e. appData is set,
// should not enter the election.
Mockito.when(mockEvent.getState()).thenReturn(Event.KeeperState.Expired);
elector.processWatchEvent(mockZK, mockEvent);
// joinElection should not be called.
Mockito.verify(mockZK, Mockito.times(0)).create(ZK_LOCK_NAME, null,
Ids.OPEN_ACL_UNSAFE, CreateMode.EPHEMERAL, elector, mockZK);
}
}