[
https://issues.apache.org/jira/browse/HDFS-17564?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17863226#comment-17863226
]
ASF GitHub Bot commented on HDFS-17564:
---------------------------------------
Hexiaoqiao merged PR #6911:
URL: https://github.com/apache/hadoop/pull/6911
> EC: Fix the issue of inaccurate metrics when decommission mark busy DN
> ----------------------------------------------------------------------
>
> Key: HDFS-17564
> URL: https://issues.apache.org/jira/browse/HDFS-17564
> Project: Hadoop HDFS
> Issue Type: Bug
> Reporter: Haiyang Hu
> Assignee: Haiyang Hu
> Priority: Major
> Labels: pull-request-available
>
> If DataNode is marked as busy and contains many EC blocks, when running
> decommission DataNode, when execute ErasureCodingWork#addTaskToDatanode, here
> will no replication work will be generated for ecBlocksToBeReplicated, but
> related metrics (such as DatanodeDescriptor#currApproxBlocksScheduled,
> pendingReconstruction and needReconstruction) will still updated.
> *Specific codeļ¼*
> BlockManager#scheduleReconstruction -> BlockManager#chooseSourceDatanodes
> [2628~2650]
> If DataNode is marked as busy and contains many EC blocks here will not add
> to srcNodes.
> .
> {code:java}
> @VisibleForTesting
> DatanodeDescriptor[] chooseSourceDatanodes(BlockInfo block,
> List<DatanodeDescriptor> containingNodes,
> List<DatanodeStorageInfo> nodesContainingLiveReplicas,
> NumberReplicas numReplicas, List<Byte> liveBlockIndices,
> List<Byte> liveBusyBlockIndices, List<Byte> excludeReconstructed, int
> priority) {
> containingNodes.clear();
> nodesContainingLiveReplicas.clear();
> List<DatanodeDescriptor> srcNodes = new ArrayList<>();
> ...
> for (DatanodeStorageInfo storage : blocksMap.getStorages(block)) {
> final DatanodeDescriptor node = getDatanodeDescriptorFromStorage(storage);
> final StoredReplicaState state = checkReplicaOnStorage(numReplicas, block,
> storage, corruptReplicas.getNodes(block), false);
> ...
> // for EC here need to make sure the numReplicas replicates state correct
> // because in the scheduleReconstruction it need the numReplicas to check
> // whether need to reconstruct the ec internal block
> byte blockIndex = -1;
> if (isStriped) {
> blockIndex = ((BlockInfoStriped) block)
> .getStorageBlockIndex(storage);
> countLiveAndDecommissioningReplicas(numReplicas, state,
> liveBitSet, decommissioningBitSet, blockIndex);
> }
> if (priority != LowRedundancyBlocks.QUEUE_HIGHEST_PRIORITY
> && (!node.isDecommissionInProgress() && !node.isEnteringMaintenance())
> && node.getNumberOfBlocksToBeReplicated() +
> node.getNumberOfBlocksToBeErasureCoded() >= maxReplicationStreams) {
> if (isStriped && (state == StoredReplicaState.LIVE
> || state == StoredReplicaState.DECOMMISSIONING)) {
> liveBusyBlockIndices.add(blockIndex);
> //HDFS-16566 ExcludeReconstructed won't be reconstructed.
> excludeReconstructed.add(blockIndex);
> }
> continue; // already reached replication limit
> }
> if (node.getNumberOfBlocksToBeReplicated() +
> node.getNumberOfBlocksToBeErasureCoded() >=
> replicationStreamsHardLimit) {
> if (isStriped && (state == StoredReplicaState.LIVE
> || state == StoredReplicaState.DECOMMISSIONING)) {
> liveBusyBlockIndices.add(blockIndex);
> //HDFS-16566 ExcludeReconstructed won't be reconstructed.
> excludeReconstructed.add(blockIndex);
> }
> continue;
> }
> if(isStriped || srcNodes.isEmpty()) {
> srcNodes.add(node);
> if (isStriped) {
> liveBlockIndices.add(blockIndex);
> }
> continue;
> }
> ...
> {code}
> ErasureCodingWork#addTaskToDatanode[149~157]
> {code:java}
> @Override
> void addTaskToDatanode(NumberReplicas numberReplicas) {
> final DatanodeStorageInfo[] targets = getTargets();
> assert targets.length > 0;
> BlockInfoStriped stripedBlk = (BlockInfoStriped) getBlock();
> ...
> } else if ((numberReplicas.decommissioning() > 0 ||
> numberReplicas.liveEnteringMaintenanceReplicas() > 0) &&
> hasAllInternalBlocks()) {
> List<Integer> leavingServiceSources = findLeavingServiceSources();
> // decommissioningSources.size() should be >= targets.length
> // if the leavingServiceSources size is 0, here will not to
> createReplicationWork
> final int num = Math.min(leavingServiceSources.size(), targets.length);
> for (int i = 0; i < num; i++) {
> createReplicationWork(leavingServiceSources.get(i), targets[i]);
> }
> ...
> }
> // Since there is no decommission busy datanode in srcNodes, here return the
> set size of srcIndices as 0.
> private List<Integer> findLeavingServiceSources() {
> // Mark the block in normal node.
> BlockInfoStriped block = (BlockInfoStriped)getBlock();
> BitSet bitSet = new BitSet(block.getRealTotalBlockNum());
> for (int i = 0; i < getSrcNodes().length; i++) {
> if (getSrcNodes()[i].isInService()) {
> bitSet.set(liveBlockIndices[i]);
> }
> }
> // If the block is on the node which is decommissioning or
> // entering_maintenance, and it doesn't exist on other normal nodes,
> // we just add the node into source list.
> List<Integer> srcIndices = new ArrayList<>();
> for (int i = 0; i < getSrcNodes().length; i++) {
> if ((getSrcNodes()[i].isDecommissionInProgress() ||
> (getSrcNodes()[i].isEnteringMaintenance() &&
> getSrcNodes()[i].isAlive())) &&
> !bitSet.get(liveBlockIndices[i])) {
> srcIndices.add(i);
> }
> }
> return srcIndices;
> }
> {code}
> so we need to fix this logic to avoid inaccurate metrics.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
