[
https://issues.apache.org/jira/browse/HADOOP-3153?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12587135#action_12587135
]
Hemanth Yamijala commented on HADOOP-3153:
------------------------------------------
Basic fix seem right.
Some comments:
- If the $HOME/.hod directory does not exist, (and so the clusters.state file
does not exist), allocate is not working.
- I think we can make some small changes to the error messages. Currently, in
the error scenario, for allocate we say:
Invalid Permissions on clusters' data file :
/home/yhemanth/.hod/clusters.state. It has to be a readable file.
Instead, we can say:
Requested operation cannot be performed. Cannot read
/home/yhemanth/.hod/clusters.state: Permission denied.
This is somewhat similar to what Unix errors report.
Likewise, for deallocate, when you don't have write permissions, we say:
Invalid Permissions on clusters' data file :
/home/yhemanth/.hod/clusters.state. It has to be a writable file. info and list
operations might show old cluster data.
Instead, we can say:
Cannot update /home/yhemanth/.hod/clusters.state: Permission denied.\n Cluster
is deallocated, but info and list operations might show incorrect information.
- In the except block of the last part of the allocate command, the
deallocation is still not happening, right ? It should be, as per your comment ?
Minor points:
- For the checkStateFile, I think we can pass in some const strings instead of
the hardcoded numbers. In fact, it would be great if we can reuse os.R_OK etc.
Can we not OR os.R_OK and os.W_OK to accomodate mode 3 ?
- Likewise the error strings, if possible should be only at one place. That
way, it will be easier to change.
Test cases:
- Move setupConf to a library type function.. maybe in a util class in the same
module.
- One more test case I can think of is what happens if the cluster state
directory does not have write and execute permissions. This is actually the
scenario that we have fixed
> [HOD] Hod should deallocate cluster if there's a problem in writing
> information to the state file
> -------------------------------------------------------------------------------------------------
>
> Key: HADOOP-3153
> URL: https://issues.apache.org/jira/browse/HADOOP-3153
> Project: Hadoop Core
> Issue Type: Bug
> Components: contrib/hod
> Affects Versions: 0.16.0
> Reporter: Hemanth Yamijala
> Assignee: Vinod Kumar Vavilapalli
> Fix For: 0.17.0
>
> Attachments: HADOOP-3153
>
>
> Consider a scenario where hod runs allocate successfully, but isn't able to
> save teh allocated information to the clusters.state file. In such a case, it
> gets an error and exits. But the cluster remains allocated, and unfortunately
> the user cannot deallocate the cluster now unless he knows the cluster
> directory.
> It is better if HOD can deallocate the cluster in such an error condition.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
