Here is the skeleton of proposed change to HbaseObjectWritable.readObject():
if(Writable.class.isAssignableFrom(instanceClass)){
if(WritableWithoutDefaultCtor.class.isAssignableFrom(instanceClass)){
Method method = instanceClass.getMethod("newInstance",
DataInput.class);
try {
Writable writable = method.invoke(null, in);
} catch (Exception e) {
...
}
}
}
On Fri, Jul 22, 2011 at 10:48 PM, Ted Yu <[email protected]> wrote:
> The scope of changes is bigger than I thought initially.
>
> The introduction of this method:
> public static Writable newInstance(DataInput dataInput)
> would probably lead to an abstract class which is called, e.g.
> WritableWithoutDefaultCtor.
>
> if(Writable.class.isAssignableFrom(instanceClass)){
>
> if(WritableWithoutDefaultCtor.class.isAssignableFrom(instanceClass)){
>
> HTD would be the first beneficiary.
>
>
> On Fri, Jul 22, 2011 at 10:03 PM, Stack <[email protected]> wrote:
>
>> You are fixing it for HTD only? Not all Writables?
>> St>Ack
>>
>> On Fri, Jul 22, 2011 at 9:50 PM, Ted Yu <[email protected]> wrote:
>> > Looks like I need to enhance the following code in HbaseObjectWritable
>> (line
>> > 575):
>> > Writable writable = WritableFactories.newInstance(instanceClass,
>> > conf);
>> >
>> > If HTableDescriptor has the following method where T is
>> HTableDescriptor:
>> > public <T> T newInstance(DataInput dataInput)
>> >
>> > HbaseObjectWritable.readObject() would be able to create an HTD instance
>> > without relying on its default ctor.
>> >
>> > On Fri, Jul 22, 2011 at 9:31 PM, Stack <[email protected]> wrote:
>> >
>> >> Try your proposal. I don't think it will work. It'll fail over in
>> >> RPC where expectation is that there is a null/default constructor on
>> >> every Writable.
>> >> St.Ack
>> >>
>> >> On Fri, Jul 22, 2011 at 9:24 PM, Ted Yu <[email protected]> wrote:
>> >> > My proposal below would make default HTableDescriptor ctor package
>> >> private.
>> >> >
>> >> > If there is no object, I will log a new JIRA.
>> >> >
>> >> > On Fri, Jul 22, 2011 at 4:27 PM, Ted Yu (JIRA) <[email protected]>
>> wrote:
>> >> >
>> >> >>
>> >> >> [
>> >> >>
>> >>
>> https://issues.apache.org/jira/browse/HBASE-4127?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13069840#comment-13069840
>> >> ]
>> >> >>
>> >> >> Ted Yu commented on HBASE-4127:
>> >> >> -------------------------------
>> >> >>
>> >> >> +1 on the patch.
>> >> >>
>> >> >> Long term, we should hide this ctor, HTableDescriptor().
>> >> >> Its existence is to support the following pattern:
>> >> >> {code}
>> >> >> hTableDescriptor = new HTableDescriptor();
>> >> >> hTableDescriptor.readFields(fsDataInputStream);
>> >> >> {code}
>> >> >> I think we should replace this ctor with the following:
>> >> >> {code}
>> >> >> public HTableDescriptor getTableDescriptor(DataInputStream
>> >> dataInputStream)
>> >> >> {code}
>> >> >>
>> >> >> > HBaseAdmin : Don't modify table's name away
>> >> >> > -------------------------------------------
>> >> >> >
>> >> >> > Key: HBASE-4127
>> >> >> > URL:
>> https://issues.apache.org/jira/browse/HBASE-4127
>> >> >> > Project: HBase
>> >> >> > Issue Type: Bug
>> >> >> > Components: client, master
>> >> >> > Affects Versions: 0.92.0
>> >> >> > Reporter: Nicolas Spiegelberg
>> >> >> > Assignee: Nicolas Spiegelberg
>> >> >> > Priority: Blocker
>> >> >> > Attachments: HBASE-4127.patch
>> >> >> >
>> >> >> >
>> >> >> > One of the developers was using the default constructor for
>> >> >> HTableDescriptor, which is sadly a bad constructor that should never
>> be
>> >> >> used. It made the tablename empty in META & caused an ERROR cycle as
>> >> region
>> >> >> onlining kept failing. We should have never let this happen. Don't
>> do
>> >> table
>> >> >> modifications if the HTableDescriptor name doesn't match the table
>> name
>> >> >> passed in.
>> >> >>
>> >> >> --
>> >> >> This message is automatically generated by JIRA.
>> >> >> For more information on JIRA, see:
>> >> http://www.atlassian.com/software/jira
>> >> >>
>> >> >>
>> >> >>
>> >> >
>> >>
>> >
>>
>
>
