Thanks guys for the details and update.
I'm updating my WriteFn and Coder accordingly.
Regards
JB
On 06/20/2017 05:52 PM, Lukasz Cwik wrote:
Filed https://issues.apache.org/jira/browse/BEAM-2482 for updating
CodedValueMutationDetector
On Tue, Jun 20, 2017 at 8:29 AM, Kenneth Knowles
wrote:
Filed https://issues.apache.org/jira/browse/BEAM-2482 for updating
CodedValueMutationDetector
On Tue, Jun 20, 2017 at 8:29 AM, Kenneth Knowles
wrote:
> Replacing full encoding with structural value is a good way to provide an
> opportunity for a fast past. File a starter JIRA?
>
> The equals che
Replacing full encoding with structural value is a good way to provide an
opportunity for a fast past. File a starter JIRA?
The equals check should be retained since it will sometimes be even faster,
and structural value falls back to full encoding.
On Tue, Jun 20, 2017 at 8:19 AM, Lukasz Cwik
w
I think the mutation detector could be updated to use the coder's
structural value and the coder could then provide a structural value which
wraps the message and does the equality comparison however it chooses.
https://github.com/apache/beam/blob/01b3f87f977d44eac23eb5488074bbc638858a9d/sdks/java/
Either Java object equality or its coder needs to be deterministic for that
check to hold.
On Tue, Jun 20, 2017 at 7:49 AM, Reuven Lax
wrote:
> Him
>
> That is only a fast path. If equals returns false, it then encodes the
> values to a byte array and checks the byte array for equality. So as lo
Him
That is only a fast path. If equals returns false, it then encodes the
values to a byte array and checks the byte array for equality. So as long
as you havev a correct coder, this should work.
On Tue, Jun 20, 2017 at 2:06 AM, Jean-Baptiste Onofré
wrote:
> Hi Kenn,
>
> I checked in MutationD
Hi Kenn,
I checked in MutationDetectors, and we use the CodedValueMutationDetector(T
value, Coder coder).
To verify mutation, we use the verifyUnmodified() method calling
verifyUnmodifiedThrowingCheckedExceptions().
In the verifyUnmodifiedThrowingCheckedExceptions() method, basically, we do
Thanks Kenn,
let me investigate a bit if I can do something on the coder side.
Regards
JB
On 06/19/2017 07:57 AM, Kenneth Knowles wrote:
Last I checked, equals() was used only as a shortcut. If the two are not
equals() then their encoded forms should be checked. If neither equals()
nor the cod
Last I checked, equals() was used only as a shortcut. If the two are not
equals() then their encoded forms should be checked. If neither equals()
nor the coder can work for this, you will have a bad time.
On Sun, Jun 18, 2017 at 10:14 PM, Jean-Baptiste Onofré
wrote:
> Hi team,
>
> The direct run