Details
-
Type:
Bug
-
Status: Open
-
Priority:
Major
-
Resolution: Unresolved
-
Affects Version/s: CDH4.3.0
-
Fix Version/s: None
-
Component/s: MapReduce
-
Labels:None
-
Environment:Deployment: Blades on SAN
DN blades each have a 133 GB local disk on the blade with 2 TB SAN as storage attached.
Description
Version: Hadoop 2.0.0-cdh4.3.0
First I mounted mapred.local.dir (/opt/var/mapred/local/1) on local disk. The disk soon got filled up.
After I softlink /opt/var/mapred/local/1 to SAN, everything ran fine for a day. Later, somehow the softlink was removed and Task Tracker process created a directory again as /opt/var/mapred/local/1 on local disk again by itself.
Quesiton:
1. What is disk quota requirement for mapred.local.dir?
2. How often mapred.local.dir get a cleanup?
2. Does Task Tracker automatically switch mapred.local.dir to local disk?
FYI - local disk is used now. I did create softlink 1 -> /opt/var/hadoop/dfs/dn/1/mapred.tmp, but the soft link was removed.
$ pwd
/opt/var/mapred/local/1
$ ls -al
total 20
drwxrwxr-x. 5 mapred hadoop 4096 Sep 15 23:02 .
drwxrwxr-x. 3 mapred hadoop 4096 Sep 15 05:48 ..
drwxr-xr-x. 5 mapred mapred 4096 Sep 16 04:39 taskTracker
drwxr-xr-x. 3 mapred mapred 4096 Sep 15 23:02 ttprivate
drwxr-xr-x. 20 mapred mapred 4096 Sep 16 05:49 userlogs
SAN location for mapred.local.dir (softlinked)
$ pwd
/opt/var/hadoop/dfs/dn/1/mapred.tmp
$ ls -al
total 28
drwxr-xr-x. 7 mapred hadoop 4096 Sep 13 23:44 .
drwxrwxr-x. 5 hdfs hdfs 4096 Sep 13 23:12 ..
drwxr-xr-x. 3 mapred mapred 4096 Sep 14 22:53 taskTracker
drwxr-xr-x. 2 mapred mapred 4096 Sep 13 23:44 toBeDeleted
drwxr-xr-x. 2 mapred mapred 4096 Sep 13 23:44 tt_log_tmp
drwx------. 3 mapred mapred 4096 Sep 14 22:53 ttprivate
drwxr-xr-x. 9 mapred mapred 4096 Sep 14 23:04 userlogs