You are viewing a plain text version of this content. The canonical link for it is here.
Posted to hdfs-dev@hadoop.apache.org by "Jay Hankinson (JIRA)" <ji...@apache.org> on 2018/05/01 17:41:01 UTC
[jira] [Created] (HDFS-13518) hdfsHSync()/hdfsFileClose() doesn't
release lease with libhdfs on Azure ADL
Jay Hankinson created HDFS-13518:
------------------------------------
Summary: hdfsHSync()/hdfsFileClose() doesn't release lease with libhdfs on Azure ADL
Key: HDFS-13518
URL: https://issues.apache.org/jira/browse/HDFS-13518
Project: Hadoop HDFS
Issue Type: Bug
Components: libhdfs
Affects Versions: 2.7.3
Environment: HDInsight in Azure:
actian@hn0-vector:~$ hadoop version
Hadoop 2.7.3.2.6.2.25-1
Subversion git@github.com:hortonworks/hadoop.git -r 1ceeb58bb3bb5904df0cbb7983389bcaf2ffd0b6
Compiled by jenkins on 2017-11-29T15:28Z
Compiled with protoc 2.5.0
From source with checksum 90b73c4c185645c1f47b61f942230
This command was run using /usr/hdp/2.6.2.25-1/hadoop/hadoop-common-2.7.3.2.6.2.25-1.jar
actian@hn0-vector:~$ cat /etc/lsb-release
DISTRIB_ID=Ubuntu
DISTRIB_RELEASE=16.04
DISTRIB_CODENAME=xenial
DISTRIB_DESCRIPTION="Ubuntu 16.04.4 LTS"
Reporter: Jay Hankinson
If hdfsFileClose() is called on a file opened for write/append directly after hdfsHSync(), without further data being written to the file, the file doesn't close and the lease isn't released. In the ADL
output trace, with hdfsHSync() immediately before hdfsFileClose() I see:
18/04/26 23:31:11 DEBUG store.HttpTransport: HTTPRequest,Succeeded,cReqId:607c6051-2ef3-4e57-9b20-caea70f22a31.0,lat:43,err:,Reqlen:14,Resplen:0,token_ns:3700,sReqId:40b2a8df-c9b5-4eff-b84e-c1346f4fdb88,path:/tmp/testfile.txt,qp:op=APPEND&append=true&syncFlag=METADATA&filesessionid=76b19d5b-f3ad-4400-a742-2f7996450859&leaseid=76b19d5b-f3ad-4400-a742-2f7996450859&offset=28&api-version=2016-11-01
but nothing else. I also get an error if I try to re-open the file:
18/04/26 23:37:28 DEBUG store.HttpTransport: HTTPRequest,Failed,cReqId:f216b9d6-3e35-4166-92cf-a7c4e2389dcd.0,lat:9,err:HTTP400(IllegalArgumentException),Reqlen:0,Resplen:357,token_ns:13001,sReqId:4b9e1e11-3f61-46a6-b461-3e5b180266c5,path:/tmp/testfile.txt,qp:op=APPEND&append=true&syncFlag=DATA&filesessionid=3197c37c-356d-491a-a902-aa19b6d8a374&leaseid=3197c37c-356d-491a-a902-aa19b6d8a374&api-version=2016-11-01
hdfsOpenFile(/tmp/testfile.txt): FileSystem#append((Lorg/apache/hadoop/fs/Path;)Lorg/apache/hadoop/fs/FSDataOutputStream;) error:
java.io.IOException: APPEND failed with error 0x83090a16 (Failed to perform the requested operation because the file is currently open in write mode by another user or process.). [4b9e1e11-3f61-46a6-b461-3e5b180266c5][2018-04-26T16:37:28.6549201-07:00]
at com.microsoft.azure.datalake.store.ADLStoreClient.getRemoteException(ADLStoreClient.java:1142)
at com.microsoft.azure.datalake.store.ADLStoreClient.getExceptionFromResponse(ADLStoreClient.java:1106)
at com.microsoft.azure.datalake.store.ADLStoreClient.getAppendStream(ADLStoreClient.java:294)
at org.apache.hadoop.fs.adl.AdlFileSystem.append(AdlFileSystem.java:398)
at org.apache.hadoop.fs.FileSystem.append(FileSystem.java:1187)
If hdfsWrite()/flush() is called after hdfsHSync() and before hdfsFileClose() I see the APPEND() with syncFlag=CLOSE as expected.
18/04/26 23:37:53 DEBUG store.HttpTransport: HTTPRequest,Succeeded,cReqId:ecd19d36-b535-4dd5-8c49-1588a2a60947.0,lat:124,err:,Reqlen:14,Resplen:0,token_ns:4700,sReqId:4a6a2c94-ab31-47b2-aae8-1e4250974d27,path:/tmp/testfile.txt,qp:op=APPEND&append=true&syncFlag=METADATA&filesessionid=21adf26e-3208-40fc-803a-1c18e1914090&leaseid=21adf26e-3208-40fc-803a-1c18e1914090&offset=0&api-version=2016-11-01
18/04/26 23:37:53 DEBUG store.HttpTransport: HTTPRequest,Succeeded,cReqId:08708059-6396-4af9-a4da-d7d60c44041b.0,lat:43,err:,Reqlen:14,Resplen:0,token_ns:3900,sReqId:a4b5f419-f87f-4bed-99ed-8ad1d24c9ab7,path:/tmp/testfile.txt,qp:op=APPEND&append=true&syncFlag=CLOSE&filesessionid=21adf26e-3208-40fc-803a-1c18e1914090&leaseid=21adf26e-3208-40fc-803a-1c18e1914090&offset=14&api-version=2016-11-01
Appending to the file is then prevented for the next 8 or so minutes until the lease expires. libhdfs repro case attached.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-dev-unsubscribe@hadoop.apache.org
For additional commands, e-mail: hdfs-dev-help@hadoop.apache.org