HDFS-10477. Stop decommission a rack of DataNodes caused NameNode fail over to standby. Contributed by yunjiong zhao and Wei-Chiu Chuang.
This commit is contained in:
parent
3b0c5016b2
commit
be488b6070
|
@ -4235,21 +4235,41 @@ public class BlockManager implements BlockStatsMXBean {
|
|||
if (!isPopulatingReplQueues()) {
|
||||
return;
|
||||
}
|
||||
final Iterator<BlockInfo> it = srcNode.getBlockIterator();
|
||||
|
||||
int numExtraRedundancy = 0;
|
||||
while(it.hasNext()) {
|
||||
final BlockInfo block = it.next();
|
||||
if (block.isDeleted()) {
|
||||
//Orphan block, will be handled eventually, skip
|
||||
for (DatanodeStorageInfo datanodeStorageInfo : srcNode.getStorageInfos()) {
|
||||
// the namesystem lock is released between iterations. Make sure the
|
||||
// storage is not removed before continuing.
|
||||
if (srcNode.getStorageInfo(datanodeStorageInfo.getStorageID()) == null) {
|
||||
continue;
|
||||
}
|
||||
int expectedReplication = this.getExpectedRedundancyNum(block);
|
||||
NumberReplicas num = countNodes(block);
|
||||
if (shouldProcessExtraRedundancy(num, expectedReplication)) {
|
||||
// extra redundancy block
|
||||
processExtraRedundancyBlock(block, (short) expectedReplication, null,
|
||||
null);
|
||||
numExtraRedundancy++;
|
||||
final Iterator<BlockInfo> it = datanodeStorageInfo.getBlockIterator();
|
||||
while(it.hasNext()) {
|
||||
final BlockInfo block = it.next();
|
||||
if (block.isDeleted()) {
|
||||
//Orphan block, will be handled eventually, skip
|
||||
continue;
|
||||
}
|
||||
int expectedReplication = this.getExpectedRedundancyNum(block);
|
||||
NumberReplicas num = countNodes(block);
|
||||
if (shouldProcessExtraRedundancy(num, expectedReplication)) {
|
||||
// extra redundancy block
|
||||
processExtraRedundancyBlock(block, (short) expectedReplication, null,
|
||||
null);
|
||||
numExtraRedundancy++;
|
||||
}
|
||||
}
|
||||
// When called by tests like TestDefaultBlockPlacementPolicy.
|
||||
// testPlacementWithLocalRackNodesDecommissioned, it is not protected by
|
||||
// lock, only when called by DatanodeManager.refreshNodes have writeLock
|
||||
if (namesystem.hasWriteLock()) {
|
||||
namesystem.writeUnlock();
|
||||
try {
|
||||
Thread.sleep(1);
|
||||
} catch (InterruptedException e) {
|
||||
Thread.currentThread().interrupt();
|
||||
}
|
||||
namesystem.writeLock();
|
||||
}
|
||||
}
|
||||
LOG.info("Invalidated {} extra redundancy blocks on {} after "
|
||||
|
|
Loading…
Reference in New Issue