|
Nigel Daley made changes - 11/Feb/09 08:42 PM
Robert Chansler made changes - 12/Feb/09 11:34 PM
Jim,
btw, this is assigned to me just to provide a patch. The decision is not mine. I will let others reply to the above valid concern. This particular item turns off append() at the interface without affecting the implementation or normal behavior of sync().
A related item, Hadoop-5225, disables some of the recovery code associated with sync in the case of data node failure. The use of either of these facilities in 0.19.0 is dangerous to the integrity of the file system. The goal is to provide 0.19.1 with the same file system integrity as 0.18.3. Further development of append and sync recovery can then continue. patch for review. The only real change is that DistributedFileSystem.append() throws UnsupportedOperationException. Rest of the changes are for disabling tests that depend on HDFS append.
Raghu Angadi made changes - 13/Feb/09 08:20 PM
'ant test-contrib' passes as well.
Raghu Angadi made changes - 13/Feb/09 09:54 PM
'ant test-libhdfs' passes as well. ant test-patch : [exec] +1 overall.
[exec]
[exec] +1 @author. The patch does not contain any @author tags.
[exec]
[exec] +1 tests included. The patch appears to include 9 new or modified tests.
[exec]
[exec] +1 javadoc. The javadoc tool did not generate any warning messages.
[exec]
[exec] +1 javac. The applied patch does not increase the total number of javac compiler warnings.
[exec]
[exec] +1 findbugs. The patch does not introduce any new Findbugs warnings.
Looks like TestStickyBit uses append(). Something tells me it should have failed.
correct. But TestStickyBit is not in 0.19.
I just committed committed this to 0.19 (only).
Raghu Angadi made changes - 13/Feb/09 11:55 PM
Hi folks, please pardon me for re-opening this issue. I am wondering if there is a better way to disable "appends" for 0.19.1 We coudld set the config parameter dfs.support.append to false in hadoop-defaults.xml. This means that a default installation of hadoop will not support "appends". However, for people that are testing their apps (not in production), it gives them an easy way to test their apps. There could be big-and-large warning message associated with the dfs.support.append parameter warning people to the fact that there could be data corruption if they set that flag to true.
I am re-opening this issue because I am not sure whether this approach was considered. If it was considered and was shot down, please let me know why and u can close this issue.
dhruba borthakur made changes - 14/Feb/09 05:37 AM
Making the config parameter dfs.support.append to "false" by default might allow Jim Kellerman to continue testing his Hbase code with 0.19.1 (till these get resolved in later releases).
I think I shouldn't have been assigned this jira in the first place.
Raghu Angadi made changes - 14/Feb/09 05:02 PM
Hi Raghu, I uploaded a patch where the only change is to set "dfs.support.append" to false. Does this look ok to you?
dhruba borthakur made changes - 17/Feb/09 03:09 PM
Dhruba, folks who want to test with append can use 0.19.0.
The uploaded patch creates a dfs.support.append config entry but doesn't seem to use it in the append method to affect it's behavior. What am I missing?
I think this should be a new Jira that could be fixed in 0.19.2. If you agree, please close this and open a new Jira. Nigel, thanks for the suggestion. I am closing this one.
dhruba borthakur made changes - 18/Feb/09 05:42 AM
> 'ant test-libhdfs' passes as well.
my bad. I had reverted the patch by mistake before running 'ant test-libhdfs'. So the patch I attached is incomplete since it does not remove the libhdfs append test. Please disregard it. libhdfs.patch : disables append tests in libhdfs. The command ' ant -Dcompile.c++=true -Dlibhdfs=true clean compile test-libhdfs ' fails with current 0.19 and passes with this patch applied. I will commit this correction soon today.
Raghu Angadi made changes - 19/Feb/09 07:10 PM
I just committed libhdfs.patch to 0.19 (only).
Nigel Daley made changes - 24/Feb/09 11:37 PM
dhruba borthakur made changes - 26/Feb/09 05:43 AM
Owen O'Malley made changes - 08/Jul/09 04:43 PM
Owen O'Malley made changes - 08/Jul/09 05:05 PM
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
The testing I have done to date indicates that sync() does, in fact, work. See
HBASE-1155