[
https://issues.apache.org/jira/browse/HADOOP-4044?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12637158#action_12637158
]
Sanjay Radia commented on HADOOP-4044:
--------------------------------------
If throwing a UnresolvedPathException is considered clean style inside the
namenode then it should be clean for RPC interface.
We need to be consistent here. It is either clean style or it s not.
In my opnion, throwing the UnresolvedPathException across the rpc interface is
clean -- it is one of many internal interfaces.
Joshua Bloch in "Effective Java" (item 40) argues that exceptions can be used
for recoverable conditions. The processing the symbolic link is
a recoverable condition. Later in the same item he argues that such exceptions
can provide information to allow the recovery (in this case the remaining
pathname to resolve).
Indeed the link that Doug gave argues in favour of using declared exceptions
for recoverable conditions
[ http://192.220.96.201/essays/java-style/exceptions.html]:
_"My rule of thumb for whether I make an exception declared or undeclared is
whether localized recovery from the exception being thrown is sensible. If the
calling method (or one of its recent callers) of the code is the right place to
handle a given failure type, I represent that failure with a declared
exception. If, on the other hand, if the failure case is best handled by a
global handler which catches all the exceptions for a given component of a
program and treats them all as failures of that subsystem, I use undeclared
exceptions"._
In the case of symbolic links, localized recovery by following the symbolic
link is possible, and hence using an exception is acceptable.
Here the author is distinguishing between declared and undeclared -- but the
point is that declared exceptions are okay for recoverable conditions.
If java did not have the distinction for declared vs undeclared and exceptions
were only for errors, then Doug's argument would have been valid.
Dhruba made the right choice in throwing the exception inside the name node -
otherwise all the interfaces and code would have become unclean.
The same style should have continued for the RPC operation.
The RPC interfaces in the patch have been changed as follows.
void create(...) become FSLink create(..)
and boolean setReplcation() becomes FSLinkBoolean setReplication(..)
... similar for many other methods.
The original interfaces were intuitive -- the input parameters and the return
values are obvious for each method.
The newer ones have this FSlinkXXX return value that, IMHO, is not clean.
They will get worse with the union overloading overtime if we add new reasons
to recover on the client side.
The use of exceptions would have kept the interface clean (just as was done
inside the name node code).
> Create symbolic links in HDFS
> -----------------------------
>
> Key: HADOOP-4044
> URL: https://issues.apache.org/jira/browse/HADOOP-4044
> Project: Hadoop Core
> Issue Type: New Feature
> Components: dfs
> Reporter: dhruba borthakur
> Assignee: dhruba borthakur
> Attachments: symLink1.patch, symLink1.patch, symLink4.patch,
> symLink5.patch, symLink6.patch, symLink8.patch, symLink9.patch
>
>
> HDFS should support symbolic links. A symbolic link is a special type of file
> that contains a reference to another file or directory in the form of an
> absolute or relative path and that affects pathname resolution. Programs
> which read or write to files named by a symbolic link will behave as if
> operating directly on the target file. However, archiving utilities can
> handle symbolic links specially and manipulate them directly.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
