hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Aaron T. Myers (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDFS-1974) HA: Introduce active and standby states to the namenode
Date Tue, 09 Aug 2011 02:14:28 GMT

    [ https://issues.apache.org/jira/browse/HDFS-1974?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13081371#comment-13081371
] 

Aaron T. Myers commented on HDFS-1974:
--------------------------------------

Thanks for addressing my comments, Suresh. Comments inline.

bq. How do you handle operations related to checkpointing and journaling that namenode in
it's various states/roles cannot satisfy?

Good point. StandbyException is perhaps a poor name, or should perhaps be a subclass of UnsupportedActionException.
Should we perhaps move UnsupportedActionException to some common package? It's certainly not
HA- or even HDFS-specific.

bq. Further we need to think if StandbyException should be at IPC layer? Is this not Service
related exception?

The layer which performs client failover should be able to determine whether or not an IPC
failed because the call was made to the wrong node in a set of HA servers which provide a
service. That is why I introduced StandbyException at the IPC layer.

bq. Setting a state involves multiple steps - that existing current state and entering a new
state. standbyToActive and activeToStandby are interface methods.

Right, but I'm under the impression that {{exitState}} and {{enterState}} are the interface
methods which should be override by the state subclasses to perform the steps of state transition.
The implementations you've provided for {{standbyToActive}} and {{activeToStandby}} are either
just no-ops or aliases for {{setState(some state)}}, and therefore seem unnecessary.

bq. exitState does the cleanup required for exiting a state, with in the state. I have made
a change to set the state of NN before enterState.

I don't see where this change was made in the latest patch.

A few more comments:

# Why aren't you adding {{checkOperation}} calls in all the read operations of the NN?
# {{ActiveState.checkOperation}} doesn't actually check if the op is of type {{JOURNAL}},
although its comment claims to allow all operations other than {{JOURNAL}}.

> HA: Introduce active and standby states to the namenode
> -------------------------------------------------------
>
>                 Key: HDFS-1974
>                 URL: https://issues.apache.org/jira/browse/HDFS-1974
>             Project: Hadoop HDFS
>          Issue Type: Sub-task
>          Components: name-node
>            Reporter: Suresh Srinivas
>            Assignee: Suresh Srinivas
>         Attachments: HDFS-1974.1.patch, HDFS-1974.2.patch, HDFS-1974.3.patch, HDFS-1974.4.patch,
HDFS-1974.patch
>
>
> Currently namenode supports active, secondary and backup roles. To support namenode high
availability, active and standby states are needed. Note that this is different from the existing
notion of namenode role, where a namenode cannot transition from one role to the other.

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

Mime
View raw message