hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Mingliang Liu (JIRA)" <j...@apache.org>
Subject [jira] [Comment Edited] (HDFS-11094) Send back HAState along with NamespaceInfo during a versionRequest as an optional parameter
Date Tue, 29 Nov 2016 02:18:58 GMT

    [ https://issues.apache.org/jira/browse/HDFS-11094?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15703890#comment-15703890
] 

Mingliang Liu edited comment on HDFS-11094 at 11/29/16 2:18 AM:
----------------------------------------------------------------

Hi [~ebadger], thanks for updating the patch. Sorry for returning late from holiday. The patch
looks good to me overall. I have two thoughts for your consideration.

# I discussed with [~arpitagarwal] offline and he suggested us use the same logic in {{updateActorStatesFromHeartbeat}}
to update the active NN {{bpServiceToActive}}, which has dealt with several cases carefully.
Moreover, if we are updating {{bpServiceToActive}} we should likely also update {{lastActiveClaimTxId}}.
To achieve this, I think we can pass {{NNHAStatusHeartbeatProto}} instead of {{HAServiceStateProto}}
in {{NamespaceInfoProto}}.
# For the unit test, can we set a very large heartbeat interval in configuration, and check
the active NN is not null after {{cluster.waitForActive()}}? Mocked tests are useful as well
and can be kept. Another idea is to drop heartbeat request against a spied HeartbeatManager.


was (Author: liuml07):
Hi [~ebadger], thanks for updating the patch. Sorry for returning late from holiday. The patch
looks good to me overall. I have two thoughts for your consideration.

# I discussed with [~arpitagarwal] and he suggest us consider using the same logic in {{updateActorStatesFromHeartbeat}}
to update the active NN {{bpServiceToActive}}, which has dealt with several cases carefully.
If we are updating {{bpServiceToActive}} we should likely also update {{lastActiveClaimTxId}}.

> Send back HAState along with NamespaceInfo during a versionRequest as an optional parameter
> -------------------------------------------------------------------------------------------
>
>                 Key: HDFS-11094
>                 URL: https://issues.apache.org/jira/browse/HDFS-11094
>             Project: Hadoop HDFS
>          Issue Type: Bug
>          Components: datanode
>            Reporter: Eric Badger
>            Assignee: Eric Badger
>         Attachments: HDFS-11094.001.patch, HDFS-11094.002.patch, HDFS-11094.003.patch,
HDFS-11094.004.patch, HDFS-11094.005.patch, HDFS-11094.006.patch, HDFS-11094.007.patch, HDFS-11094.008.patch,
HDFS-11094.009.patch
>
>
> The datanode should know which NN is active when it is connecting/registering to the
NN. Currently, it only figures this out during its first (and subsequent) heartbeat(s) and
so there is a period of time where the datanode is alive and registered, but can't actually
do anything because it doesn't know which NN is active. A byproduct of this is that the MiniDFSCluster
will become active before it knows what NN is active, which can lead to NPEs when calling
getActiveNN(). 



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-help@hadoop.apache.org


Mime
View raw message