[
https://issues.apache.org/jira/browse/HADOOP-3628?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12627663#action_12627663
]
Steve Loughran commented on HADOOP-3628:
----------------------------------------
Konstantin Shvachko has proposed that instead of throwing an exception on a
health failure, different error text should be returned. His concern is that
the cost of constructing and marshalling an exception can be high, and should
not be used if the exception is likely to be returned regularly.
This is an interesting thought. The arguments in favour of an exception are
1.For individual nodes, failure is the unexpected state. Normally all should be
well.
2.Much of the cost of creating an exception is the cost of creating a stack
trace; this can be useful for later diagnostics.
3.By returning different exceptions for different problems, callers can
diagnose and act on different failures. It is much harder for programs to act
on simple strings.
4.Exceptions are going to be raised if the far end is unreachable, so the
caller needs to be prepared for those exceptions, and know that any exception
raised on a call is a sign of a failure.
However, there are some good arguments in favour of returning a structure
response instead.
1.Stack traces are less useful when they are just code inside the health check.
2.Unmarshalling exceptions reliably requires the caller to have a set of
exception classes and versions in their JVM. With nested exceptions, that
implies the entire exception list needs to be unmarshallable.
3.The aggregate health checks of the clusters themselves will inevitably
include failed nodes. Should an aggregate health check include those node
failures in a report that says "overall, we are healthy, here are the nodes
that are not"
The alternative to sending exceptions back on a ping() would be to return a
NodeHealth structure that included node name, IPAddress, service type and a
list of what was wrong with the node, as well as an aggregate "live/not live"
response. The list of what was wrong could include standard constant values for
machine interpretation, as well as human-readable messages.
What do others think?
> Add a lifecycle interface for Hadoop components: namenodes, job clients, etc.
> -----------------------------------------------------------------------------
>
> Key: HADOOP-3628
> URL: https://issues.apache.org/jira/browse/HADOOP-3628
> Project: Hadoop Core
> Issue Type: Improvement
> Components: dfs, mapred
> Affects Versions: 0.19.0
> Reporter: Steve Loughran
> Assignee: Steve Loughran
> Attachments: AbstractHadoopComponent.java, hadoop-3628.patch,
> hadoop-3628.patch, hadoop-3628.patch, hadoop-3628.patch, hadoop-3628.patch,
> hadoop-3628.patch, hadoop-3628.patch, hadoop-3628.patch, hadoop-3628.patch,
> hadoop-3628.patch, hadoop-3628.patch
>
>
> I'd like to propose we have a standard interface for hadoop components, the
> things that get started or stopped when you bring up a namenode. currently,
> some of these classes have a stop() or shutdown() method, with no standard
> name/interface, but no way of seeing if they are live, checking their health
> of shutting them down reliably. Indeed, there is a tendency for the spawned
> threads to not want to die; to require the entire process to be killed to
> stop the workers.
> Having a standard interface would make it easier for
> * management tools to manage the different things
> * monitoring the state of things
> * subclassing
> The latter is interesting as right now TaskTracker and JobTracker start up
> threads in their constructor; that's very dangerous as subclasses may have
> their methods called before they are full initialised. Adding this interface
> would be the right time to clean up the startup process so that subclassing
> is less risky.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
