[
https://issues.apache.org/jira/browse/HDFS-16912?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17707209#comment-17707209
]
jiangchunyang commented on HDFS-16912:
--------------------------------------
Through investigation, we found that when ibr reports the heartbeat information
to nn, it needs to obtain the global lock of ReentrantReadWriteLock, and when
our cluster cleaning script cleans up the files in the .trash directory, it
will go through a long process, resulting in the lock not being released. The
reporting thread can only wait for the lock to be released, resulting in a
timeout after all retries.
Our solution is to expand the datanode node horizontally. Because the cluster
was migrated before, it may be caused by insufficient storage node dn. At
present, such problems rarely occur, but they still exist.
> Block is COMMITTED but not COMPLETE(numNodes= 0 < minimum = 1)
> ---------------------------------------------------------------
>
> Key: HDFS-16912
> URL: https://issues.apache.org/jira/browse/HDFS-16912
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: block placement
> Affects Versions: 3.3.1
> Environment: hadoop:3.3.1
> Reporter: jiangchunyang
> Priority: Major
> Attachments: image-2023-02-09-16-36-42-818.png
>
>
> We use hdfs federation mode: ns1, ns2. The table data is written under
> dc-hdfs. But we designate a specific library under a specific ns according to
> the business division.
> Use parquetWriter to write data to the staging temporary file directory of
> each table under a specific ns, but when the Writer is closed, an exception
> will be reported, which will trigger our operation to restore the file lease,
> but when the file is found to be restoring the lease An exception will be
> reported:
> It looks like dn and nn have temporarily lost communication, and this doesn't
> happen with every write.
> {code:java}
> java.lang.IllegalArgumentException: java.net.UnknownHostException: ns1hdfs
> at
> org.apache.hadoop.security.SecurityUtil.buildTokenService(SecurityUtil.java:445)
> at
> org.apache.hadoop.hdfs.NameNodeProxiesClient.createProxyWithClientProtocol(NameNodeProxiesClient.java:140)
> at org.apache.hadoop.hdfs.DFSClient.<init>(DFSClient.java:357)
> at org.apache.hadoop.hdfs.DFSClient.<init>(DFSClient.java:291)
> at
> org.apache.hadoop.hdfs.DistributedFileSystem.initialize(DistributedFileSystem.java:181)
> at
> com.onething.dc.flink.parquet.sink.template.ProcessParquetSinkTemplate.recoverLease(ProcessParquetSinkTemplate.java:180)
> at
> com.onething.dc.flink.parquet.sink.template.ProcessParquetSinkTemplate.close(ProcessParquetSinkTemplate.java:208)
> at
> com.onething.dc.flink.parquet.sink.ProcessParquetSink$1.processElement(ProcessParquetSink.java:118)
> at
> com.onething.dc.flink.parquet.sink.ProcessParquetSink$1.processElement(ProcessParquetSink.java:35)
> at
> org.apache.flink.streaming.api.operators.ProcessOperator.processElement(ProcessOperator.java:66)
> at
> org.apache.flink.streaming.runtime.tasks.OneInputStreamTask$StreamTaskNetworkOutput.emitRecord(OneInputStreamTask.java:205)
> at
> org.apache.flink.streaming.runtime.io.AbstractStreamTaskNetworkInput.processElement(AbstractStreamTaskNetworkInput.java:134)
> at
> org.apache.flink.streaming.runtime.io.AbstractStreamTaskNetworkInput.emitNext(AbstractStreamTaskNetworkInput.java:105)
> at
> org.apache.flink.streaming.runtime.io.StreamOneInputProcessor.processInput(StreamOneInputProcessor.java:66)
> at
> org.apache.flink.streaming.runtime.tasks.StreamTask.processInput(StreamTask.java:419)
> at
> org.apache.flink.streaming.runtime.tasks.mailbox.MailboxProcessor.runMailboxLoop(MailboxProcessor.java:204)
> at
> org.apache.flink.streaming.runtime.tasks.StreamTask.runMailboxLoop(StreamTask.java:661)
> at
> org.apache.flink.streaming.runtime.tasks.StreamTask.invoke(StreamTask.java:623)
> at org.apache.flink.runtime.taskmanager.Task.doRun(Task.java:776)
> at org.apache.flink.runtime.taskmanager.Task.run(Task.java:563)
> at java.lang.Thread.run(Thread.java:748)
> Suppressed: java.lang.IllegalArgumentException:
> java.net.UnknownHostException: ns1hdfs
> at
> org.apache.hadoop.security.SecurityUtil.buildTokenService(SecurityUtil.java:445)
> at
> org.apache.hadoop.hdfs.NameNodeProxiesClient.createProxyWithClientProtocol(NameNodeProxiesClient.java:140)
> at org.apache.hadoop.hdfs.DFSClient.<init>(DFSClient.java:357)
> at org.apache.hadoop.hdfs.DFSClient.<init>(DFSClient.java:291)
> at
> org.apache.hadoop.hdfs.DistributedFileSystem.initialize(DistributedFileSystem.java:181)
> at
> com.onething.dc.flink.parquet.sink.template.ProcessParquetSinkTemplate.recoverLease(ProcessParquetSinkTemplate.java:180)
> at
> com.onething.dc.flink.parquet.sink.template.ProcessParquetSinkTemplate.close(ProcessParquetSinkTemplate.java:208)
> at
> org.apache.flink.api.common.functions.util.FunctionUtils.closeFunction(FunctionUtils.java:41)
> at
> org.apache.flink.streaming.api.operators.AbstractUdfStreamOperator.dispose(AbstractUdfStreamOperator.java:117)
> at
> org.apache.flink.streaming.runtime.tasks.StreamTask.disposeAllOperators(StreamTask.java:837)
> at
> org.apache.flink.streaming.runtime.tasks.StreamTask.runAndSuppressThrowable(StreamTask.java:816)
> at
> org.apache.flink.streaming.runtime.tasks.StreamTask.cleanUpInvoke(StreamTask.java:733)
> at
> org.apache.flink.streaming.runtime.tasks.StreamTask.invoke(StreamTask.java:637)
> ... 3 more
> Caused by: java.net.UnknownHostException: ns1hdfs
> ... 16 more
> Caused by: java.net.UnknownHostException: ns1hdfs
> ... 21 more {code}
> I can't see what the problem is, but when I checked the log of the namenode,
> I found that the block status of the file could not change from COMMITTED to
> COMPLETE. The reason is that dn needs to report ibr to namenode when closing
> the file. And it will close after receiving ack confirmation. However, dn
> failed to report ibr, which made it impossible to close the file. And it will
> retry every time the report fails, and the waiting time of the client is
> doubled in turn: 400ms, 800ms, 1600ms, 3200ms, 6400ms. These retries can be
> seen in the log of the namenode.
>
> {code:java}
> 2023-02-09 10:00:10,638 INFO hdfs.StateChange
> (FSDirWriteFileOp.java:logAllocatedBlock(802)) - BLOCK* allocate
> blk_1092451654_18751000, replicas=10.146.144.69:1019, 10.146.80.45:1019 for
> /user/hive/warehouse/dc_orig.db/o_dc_http_topic_metric/_flume_parquet_insert_staging/bdyun-suzhou-tw04a0032.su.baidu.internal_1675908007110_628122711
> 2023-02-09 10:00:11,072 INFO namenode.FSNamesystem
> (FSNamesystem.java:checkBlocksComplete(3151)) - BLOCK*
> blk_1092451654_18751000 is COMMITTED but not COMPLETE(numNodes= 0 < minimum
> = 1) in file
> /user/hive/warehouse/dc_orig.db/o_dc_http_topic_metric/_flume_parquet_insert_staging/bdyun-suzhou-tw04a0032.su.baidu.internal_1675908007110_628122711
> 2023-02-09 10:00:11,474 INFO namenode.FSNamesystem
> (FSNamesystem.java:checkBlocksComplete(3151)) - BLOCK*
> blk_1092451654_18751000 is COMMITTED but not COMPLETE(numNodes= 0 < minimum
> = 1) in file
> /user/hive/warehouse/dc_orig.db/o_dc_http_topic_metric/_flume_parquet_insert_staging/bdyun-suzhou-tw04a0032.su.baidu.internal_1675908007110_628122711
> 2023-02-09 10:00:12,285 INFO namenode.FSNamesystem
> (FSNamesystem.java:checkBlocksComplete(3151)) - BLOCK*
> blk_1092451654_18751000 is COMMITTED but not COMPLETE(numNodes= 0 < minimum
> = 1) in file
> /user/hive/warehouse/dc_orig.db/o_dc_http_topic_metric/_flume_parquet_insert_staging/bdyun-suzhou-tw04a0032.su.baidu.internal_1675908007110_628122711
> 2023-02-09 10:00:13,887 INFO namenode.FSNamesystem
> (FSNamesystem.java:checkBlocksComplete(3151)) - BLOCK*
> blk_1092451654_18751000 is COMMITTED but not COMPLETE(numNodes= 0 < minimum
> = 1) in file
> /user/hive/warehouse/dc_orig.db/o_dc_http_topic_metric/_flume_parquet_insert_staging/bdyun-suzhou-tw04a0032.su.baidu.internal_1675908007110_628122711
> 2023-02-09 10:00:17,089 INFO namenode.FSNamesystem
> (FSNamesystem.java:checkBlocksComplete(3151)) - BLOCK*
> blk_1092451654_18751000 is COMMITTED but not COMPLETE(numNodes= 0 < minimum
> = 1) in file
> /user/hive/warehouse/dc_orig.db/o_dc_http_topic_metric/_flume_parquet_insert_staging/bdyun-suzhou-tw04a0032.su.baidu.internal_1675908007110_628122711
> 2023-02-09 10:00:23,490 INFO namenode.FSNamesystem
> (FSNamesystem.java:checkBlocksComplete(3151)) - BLOCK*
> blk_1092451654_18751000 is COMMITTED but not COMPLETE(numNodes= 0 < minimum
> = 1) in file
> /user/hive/warehouse/dc_orig.db/o_dc_http_topic_metric/_flume_parquet_insert_staging/bdyun-suzhou-tw04a0032.su.baidu.internal_1675908007110_628122711{code}
>
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
