[ACCUMULO-1685] bench testing shows that the NN loses the WAL

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Fix Version/s: 1.6.0
• Component/s: tserver

Environment

Hadoop 1.0.4, single node dev't system

• Assignee: Eric C. Newton
• Reporter: Eric C. Newton
• Created: 04/Sep/13 16:53
• Updated: 11/Sep/13 12:49
• Resolved: 05/Sep/13 15:27

Description

Doing bench testing; I build accumulo:

$ mvn -Pnative package -DskipTests

I go into the assembly area and configure and run accumulo

$ cd assemble/target/accumulo-1.6.0-SNAPSHOT-dev/accumulo-1.6.0-SNAPSHOT
$ cp ~/conf/* conf
$ hadoop fs -rmr /accumulo
Moved to trash: hdfs://somehost:9000/accumulo
$ ( echo test ; echo Y ; echo secret ; echo secret ) | ./bin/accumulo init
$ 2013-09-04 12:23:51,558 [util.Initialize] INFO : Hadoop Filesystem is hdfs://somehost:9000
2013-09-04 12:23:51,559 [util.Initialize] INFO : Accumulo data dirs are [hdfs://somehost:9000/accumulo]
2013-09-04 12:23:51,559 [util.Initialize] INFO : Zookeeper server is localhost:2181
2013-09-04 12:23:51,559 [util.Initialize] INFO : Checking if Zookeeper is available. If this hangs, then you need to make sure zookeeper is running
Instance name : test
Instance name "test" exists. Delete existing entry from zookeeper? [Y/N] : Y
Enter initial password for root (this may not be applicable for your security setup): ******
Confirm initial password for root: ******
$ ./bin/start-all.sh 
Starting monitor on localhost
Starting tablet servers .... done
Starting tablet server on localhost
2013-09-04 12:26:24,545 [server.Accumulo] INFO : Attempting to talk to zookeeper
2013-09-04 12:26:24,675 [server.Accumulo] INFO : Zookeeper connected and initialized, attemping to talk to HDFS
2013-09-04 12:26:24,679 [server.Accumulo] INFO : Connected to HDFS
Starting master on localhost
Starting garbage collector on localhost
Starting tracer on localhost

Next, create a table

$ ./bin/accumulo shell -u root -p secret
2013-09-04 12:27:01,628 [shell.Shell] WARN : Specifying a raw password is deprecated.

Shell - Apache Accumulo Interactive Shell
- 
- version: 1.6.0-SNAPSHOT
- instance name: test
- instance id: 1967c1ec-cc0f-439b-b4da-4029debd16e3
- 
- type 'help' for a list of available commands
- 
root@test> createtable t
root@test t> 

Then I checked the tserver log for the write-ahead log created for this update to the root table:

$ fgrep -a /wal/ logs/tserver_*.debug.log
2013-09-04 12:26:27,130 [log.DfsLogger] DEBUG: Got new write-ahead log: localhost+9997/hdfs://rd6ul-14706v.tycho.ncsc.mil:9000/accumulo/wal/localhost+9997/1dd2727f-1de9-417b-a5a2-e56f7d8020a9
2013-09-04 12:26:58,264 [tabletserver.Tablet] DEBUG: Logs for memory compacted: !!R<< localhost+9997/hdfs://somehost:9000/accumulo/wal/localhost+9997/1dd2727f-1de9-417b-a5a2-e56f7d8020a9

Now, let's check for the file:

$ hadoop fs -ls hdfs://somehost:9000/accumulo/wal/localhost+9997/1dd2727f-1de9-417b-a5a2-e56f7d8020a9
ls: Cannot access hdfs://somehost:9000/accumulo/wal/localhost+9997/1dd2727f-1de9-417b-a5a2-e56f7d8020a9: No such file or directory.

What?

Check the NN logs:

$ fgrep 1dd2727f /some/log/dir/hadoop-ecnewt2-local-namenode-somehost.log 
2013-09-04 12:26:27,075 INFO org.apache.hadoop.hdfs.StateChange: BLOCK* NameSystem.allocateBlock: /accumulo/wal/localhost+9997/1dd2727f-1de9-417b-a5a2-e56f7d8020a9. blk_-6011963215434912690_971163
2013-09-04 12:26:27,113 INFO org.apache.hadoop.hdfs.StateChange: BLOCK* NameSystem.fsync: file /accumulo/wal/localhost+9997/1dd2727f-1de9-417b-a5a2-e56f7d8020a9 for DFSClient_-787226921

So, the NN seems to be making the file, but it's not there when we go to look!

Here's my hdfs-site.xml file:

<?xml version="1.0"?>
<?xml-stylesheet type="text/xsl" href="configuration.xsl"?>

<!-- Put site-specific property overrides in this file. -->

<configuration>
  <property>
      <name>dfs.replication</name>
      <value>1</value>
  </property>
  <property>
      <name>dfs.name.dir</name>
      <value>/local/ecn/data/hadoop/nn</value>
  </property>
  <property>
      <name>dfs.data.dir</name>
      <value>/disk01/data/hadoop/dn,/disk02/data/hadoop/dn,/disk03/data/hadoop/dn</value>
  </property>
  <property>
      <name>dfs.support.append</name>
      <value>true</value>
  </property>
  <property>
      <name>dfs.data.synconclose</name>
      <value>true</value>
  </property>
</configuration>

I have written an integration test that I dumped into RestartIT.java, but that doesn't seem to fail in same way.

Issue Links

• is related to: ACCUMULO-1684 WAL encryption beaks WAL
• is related to: ACCUMULO-118 accumulo could work across HDFS instances, which would help it to scale past a single namenode