HDFS-15210. EC : File write hanged when DN is shutdown by admin command. Contributed by Surendra Singh Lilhore.

(cherry picked from commit db6252b6c3)
(cherry picked from commit 7856af2cc65377352db5ebc93a778f373f71d215)
This commit is contained in:
Surendra Singh Lilhore 2020-04-29 10:58:35 +05:30 committed by Wei-Chiu Chuang
parent 6863a5bb8a
commit 929411acce
2 changed files with 17 additions and 1 deletions

View File

@ -282,6 +282,7 @@ public class DFSStripedOutputStream extends DFSOutputStream
private ExecutorService flushAllExecutor;
private CompletionService<Void> flushAllExecutorCompletionService;
private int blockGroupIndex;
private long datanodeRestartTimeout;
/** Construct a new output stream for creating a file. */
DFSStripedOutputStream(DFSClient dfsClient, String src, HdfsFileStatus stat,
@ -321,6 +322,7 @@ public class DFSStripedOutputStream extends DFSOutputStream
streamers.add(streamer);
}
currentPackets = new DFSPacket[streamers.size()];
datanodeRestartTimeout = dfsClient.getConf().getDatanodeRestartTimeout();
setCurrentStreamer(0);
}
@ -621,6 +623,11 @@ public class DFSStripedOutputStream extends DFSOutputStream
"streamer: " + streamer);
streamer.setExternalError();
healthySet.add(streamer);
} else if (!streamer.streamerClosed()
&& streamer.getErrorState().hasDatanodeError()
&& streamer.getErrorState().doWaitForRestart()) {
healthySet.add(streamer);
failedStreamers.remove(streamer);
}
}
return healthySet;
@ -685,6 +692,14 @@ public class DFSStripedOutputStream extends DFSOutputStream
for (int i = 0; i < numAllBlocks; i++) {
coordinator.offerStreamerUpdateResult(i, newFailed.size() == 0);
}
//wait for get notify to failed stream
if (newFailed.size() != 0) {
try {
Thread.sleep(datanodeRestartTimeout);
} catch (InterruptedException e) {
// Do nothing
}
}
}
}

View File

@ -143,7 +143,8 @@ public class StripedDataStreamer extends DataStreamer {
// set up the pipeline again with the remaining nodes. when a striped
// data streamer comes here, it must be in external error state.
assert getErrorState().hasExternalError();
assert getErrorState().hasExternalError()
|| getErrorState().doWaitForRestart();
success = createBlockOutputStream(nodes, nodeStorageTypes,
nodeStorageIDs, newGS, true);