I was wondering if there is any mechanism implemented to detect WALEdits that haven't been entirely written to the wal. Say that a transaction (using the terminology in the header of the WALEdit class) has three edits, but only two get written to the wal. How do we detect that it is a corrupt record when reading?

Thanks!
-Flavio  

>From javadoc:

 * In the new world, all the edits for a given transaction are written
 * out as a single record, for example:
 *
 *   <logseq#-for-entire-txn>:<WALEdit-for-entire-txn>

The keyvalues for the three edits would be stored in one record.

  public void readFields(DataInput in) throws IOException {

If only two got written, the above method would throw IOException.

Cheers

On Mon, Oct 8, 2012 at 12:54 PM, Flavio Junqueira <[EMAIL PROTECTED]> wrote:

> I was wondering if there is any mechanism implemented to detect WALEdits
> that haven't been entirely written to the wal. Say that a transaction
> (using the terminology in the header of the WALEdit class) has three edits,
> but only two get written to the wal. How do we detect that it is a corrupt
> record when reading?
>
> Thanks!
> -Flavio

Thanks, Ted. It detects it by writing and reading back the number of elements. Does it detect if the last one has been partially written?

-Flavio

On Oct 8, 2012, at 11:10 PM, Ted Yu wrote:

> From javadoc:
>
> * In the new world, all the edits for a given transaction are written
> * out as a single record, for example:
> *
> *   <logseq#-for-entire-txn>:<WALEdit-for-entire-txn>
>
> The keyvalues for the three edits would be stored in one record.
>
>  public void readFields(DataInput in) throws IOException {
>
> If only two got written, the above method would throw IOException.
>
> Cheers
>
> On Mon, Oct 8, 2012 at 12:54 PM, Flavio Junqueira <[EMAIL PROTECTED]> wrote:
>
>> I was wondering if there is any mechanism implemented to detect WALEdits
>> that haven't been entirely written to the wal. Say that a transaction
>> (using the terminology in the header of the WALEdit class) has three edits,
>> but only two get written to the wal. How do we detect that it is a corrupt
>> record when reading?
>>
>> Thanks!
>> -Flavio

I would expect IOException to be thrown if the last one was partially
written.

Cheers

On Mon, Oct 8, 2012 at 2:34 PM, Flavio Junqueira <[EMAIL PROTECTED]> wrote:

> Thanks, Ted. It detects it by writing and reading back the number of
> elements. Does it detect if the last one has been partially written?
>
> -Flavio
>
> On Oct 8, 2012, at 11:10 PM, Ted Yu wrote:
>
> > From javadoc:
> >
> > * In the new world, all the edits for a given transaction are written
> > * out as a single record, for example:
> > *
> > *   <logseq#-for-entire-txn>:<WALEdit-for-entire-txn>
> >
> > The keyvalues for the three edits would be stored in one record.
> >
> >  public void readFields(DataInput in) throws IOException {
> >
> > If only two got written, the above method would throw IOException.
> >
> > Cheers
> >
> > On Mon, Oct 8, 2012 at 12:54 PM, Flavio Junqueira <[EMAIL PROTECTED]>
> wrote:
> >
> >> I was wondering if there is any mechanism implemented to detect WALEdits
> >> that haven't been entirely written to the wal. Say that a transaction
> >> (using the terminology in the header of the WALEdit class) has three
> edits,
> >> but only two get written to the wal. How do we detect that it is a
> corrupt
> >> record when reading?
> >>
> >> Thanks!
> >> -Flavio
>
>