[
https://issues.apache.org/jira/browse/HDFS-12217?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16106143#comment-16106143
]
Yongjun Zhang edited comment on HDFS-12217 at 7/29/17 3:22 PM:
---------------------------------------------------------------
Hi [~manojg],
{quote}
LeaseManager#getINodeWithLeases already logs the warning message with full
exception and stack trace.
{quote}
The logging you mentioned above happens at the server side and is about the
cause of the failure. What I hope to see is the exception chain at client side
when SnapshotException is thrown, so user can see immediately what caused the
exception in one place. Collecting the server side log to match to the client
symptom is a support effort which I hope we can avoid if possible. Another
thing is, server side log can be rolled too.
Currently SnapshotException is:
{code}
public class SnapshotException extends IOException {
private static final long serialVersionUID = 1L;
public SnapshotException(final String message) {
super(message);
}
public SnapshotException(final Throwable cause) {
super(cause);
}
}
{code}
I examined all the places that call {{SnapshotException(final String
message)}}, none of them is called inside exception catch block. So the patch
here is the first time that we are throwing SnapshotException inside an
exception catch block. Interestingly, I don't see any places using
{{SnapshotException(final Throwable cause)}}. So it seems introducing a method
{{public SnapshotException(String message, Throwable cause)}} and use it in
this patch will not introduce inconsistency because it will be the first use of
it.
That's what I preferred. However, if you like to create a new jira to add the
new API, and change the single use which is introduced by the patch here, It's
ok with me too.
Thanks.
was (Author: yzhangal):
Hi [~manojg],
{quote}
LeaseManager#getINodeWithLeases already logs the warning message with full
exception and stack trace.
{quote}
The logging you mentioned above happens at the server side and is about the
cause of the failure. What I hope to see is the exception chain at client side
when SnapshotException is thrown, so user can see immediately what caused the
exception in one place. Collecting the server side log to match to the client
symptom is a support effort which I hope we can avoid if possible.
Currently SnapshotException is:
{code}
public class SnapshotException extends IOException {
private static final long serialVersionUID = 1L;
public SnapshotException(final String message) {
super(message);
}
public SnapshotException(final Throwable cause) {
super(cause);
}
}
{code}
I examined all the places that call {{SnapshotException(final String
message)}}, none of them is called inside exception catch block. So the patch
here is the first time that we are throwing SnapshotException inside an
exception catch block. Interestingly, I don't see any places using
{{SnapshotException(final Throwable cause)}}. So it seems introducing a method
{{public SnapshotException(String message, Throwable cause)}} and use it in
this patch will not introduce inconsistency because it will be the first use of
it.
That's what I preferred. However, if you like to create a new jira to add the
new API, and change the single use which is introduced by the patch here, It's
ok with me too.
Thanks.
> HDFS snapshots doesn't capture all open files when one of the open files is
> deleted
> -----------------------------------------------------------------------------------
>
> Key: HDFS-12217
> URL: https://issues.apache.org/jira/browse/HDFS-12217
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: snapshots
> Affects Versions: 3.0.0-alpha1
> Reporter: Manoj Govindassamy
> Assignee: Manoj Govindassamy
> Attachments: HDFS-12217.01.patch, HDFS-12217.02.patch,
> HDFS-12217.03.patch
>
>
> With the fix for HDFS-11402, HDFS Snapshots can additionally capture all the
> open files. Just like all other files, these open files in the snapshots will
> remain immutable. But, sometimes it is found that snapshots fail to capture
> all the open files in the system.
> Under the following conditions, LeaseManager will fail to find INode
> corresponding to an active lease
> * a file is opened for writing (LeaseManager allots a lease), and
> * the same file is deleted while it is still open for writing and having
> active lease, and
> * the same file is not referenced in any other Snapshots/Trash
> {{INode[] LeaseManager#getINodesWithLease()}} can thus return null for few
> leases there by causing the caller to trip over and not return all the open
> files needed by the snapshot manager.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
