private static HRegionServer setDrainingServer(final HRegionServer hrs) throws KeeperException {
LOG.info(
"Making "
+ hrs.getServerName()
+ " the draining server; "
+ "it has "
+ hrs.getNumberOfOnlineRegions()
+ " online regions");
ZooKeeperWatcher zkw = hrs.getZooKeeper();
String hrsDrainingZnode = ZKUtil.joinZNode(zkw.drainingZNode, hrs.getServerName().toString());
ZKUtil.createWithParents(zkw, hrsDrainingZnode);
return hrs;
}
/**
* Test adding server to draining servers and then move regions off it. Make sure that no regions
* are moved back to the draining server.
*
* @throws IOException
* @throws KeeperException
*/
@Test // (timeout=30000)
public void testDrainingServerOffloading() throws Exception {
// I need master in the below.
HMaster master = TEST_UTIL.getMiniHBaseCluster().getMaster();
HRegionInfo hriToMoveBack = null;
// Set first server as draining server.
HRegionServer drainingServer =
setDrainingServer(TEST_UTIL.getMiniHBaseCluster().getRegionServer(0));
try {
final int regionsOnDrainingServer = drainingServer.getNumberOfOnlineRegions();
Assert.assertTrue(regionsOnDrainingServer > 0);
List<HRegionInfo> hris = ProtobufUtil.getOnlineRegions(drainingServer);
for (HRegionInfo hri : hris) {
// Pass null and AssignmentManager will chose a random server BUT it
// should exclude draining servers.
master.moveRegion(
null, RequestConverter.buildMoveRegionRequest(hri.getEncodedNameAsBytes(), null));
// Save off region to move back.
hriToMoveBack = hri;
}
// Wait for regions to come back on line again.
waitForAllRegionsOnline();
Assert.assertEquals(0, drainingServer.getNumberOfOnlineRegions());
} finally {
unsetDrainingServer(drainingServer);
}
// Now we've unset the draining server, we should be able to move a region
// to what was the draining server.
master.moveRegion(
null,
RequestConverter.buildMoveRegionRequest(
hriToMoveBack.getEncodedNameAsBytes(),
Bytes.toBytes(drainingServer.getServerName().toString())));
// Wait for regions to come back on line again.
waitForAllRegionsOnline();
Assert.assertEquals(1, drainingServer.getNumberOfOnlineRegions());
}
/**
* Test that draining servers are ignored even after killing regionserver(s). Verify that the
* draining server is not given any of the dead servers regions.
*
* @throws KeeperException
* @throws IOException
*/
@Test(timeout = 30000)
public void testDrainingServerWithAbort() throws KeeperException, Exception {
HMaster master = TEST_UTIL.getHBaseCluster().getMaster();
waitForAllRegionsOnline();
final long regionCount = TEST_UTIL.getMiniHBaseCluster().countServedRegions();
// Let's get a copy of the regions today.
Collection<HRegion> regions = new ArrayList<HRegion>();
for (int i = 0; i < NB_SLAVES; i++) {
HRegionServer hrs = TEST_UTIL.getMiniHBaseCluster().getRegionServer(i);
regions.addAll(hrs.getCopyOfOnlineRegionsSortedBySize().values());
}
// Choose the draining server
HRegionServer drainingServer = TEST_UTIL.getMiniHBaseCluster().getRegionServer(0);
final int regionsOnDrainingServer = drainingServer.getNumberOfOnlineRegions();
Assert.assertTrue(regionsOnDrainingServer > 0);
ServerManager sm = master.getServerManager();
Collection<HRegion> regionsBefore =
drainingServer.getCopyOfOnlineRegionsSortedBySize().values();
LOG.info("Regions of drained server are: " + regionsBefore);
try {
// Add first server to draining servers up in zk.
setDrainingServer(drainingServer);
// wait for the master to receive and manage the event
while (sm.createDestinationServersList().contains(drainingServer.getServerName())) {
Thread.sleep(1);
}
LOG.info("The available servers are: " + sm.createDestinationServersList());
Assert.assertEquals(
"Nothing should have happened here.",
regionsOnDrainingServer,
drainingServer.getNumberOfOnlineRegions());
Assert.assertFalse(
"We should not have regions in transition here. List is: "
+ master.getAssignmentManager().getRegionStates().getRegionsInTransition(),
master.getAssignmentManager().getRegionStates().isRegionsInTransition());
// Kill a few regionservers.
for (int aborted = 0; aborted <= 2; aborted++) {
HRegionServer hrs = TEST_UTIL.getMiniHBaseCluster().getRegionServer(aborted + 1);
hrs.abort("Aborting");
}
// Wait for regions to come back online again.
waitForAllRegionsOnline();
Collection<HRegion> regionsAfter =
drainingServer.getCopyOfOnlineRegionsSortedBySize().values();
LOG.info("Regions of drained server are: " + regionsAfter);
Assert.assertEquals(
"Test conditions are not met: regions were" + " created/deleted during the test. ",
regionCount,
TEST_UTIL.getMiniHBaseCluster().countServedRegions());
// Assert the draining server still has the same regions.
StringBuilder result = new StringBuilder();
for (HRegion r : regionsAfter) {
if (!regionsBefore.contains(r)) {
result.append(r).append(" was added after the drain");
if (regions.contains(r)) {
result.append("(existing region");
} else {
result.append("(new region)");
}
result.append("; ");
}
}
for (HRegion r : regionsBefore) {
if (!regionsAfter.contains(r)) {
result.append(r).append(" was removed after the drain; ");
}
}
Assert.assertTrue("Errors are: " + result.toString(), result.length() == 0);
} finally {
unsetDrainingServer(drainingServer);
}
}
