Graduating Gelly in 0.10

classic Classic list List threaded Threaded
5 messages Options
Reply | Threaded
Open this post in threaded view
|

Graduating Gelly in 0.10

Vasiliki Kalavri
Hello squirrels,

since it was already mentioned in the other thread, I'm starting this
discussion to see what the community thinks about graduating Gelly in the
upcoming 0.10 release.

In my opinion, in order to do that we have to look into the following
issues:

1. FLINK-2663 <https://issues.apache.org/jira/browse/FLINK-2663>: I've
already opened a PR and will merge if no objections.

2. FLINK-2561 <https://issues.apache.org/jira/browse/FLINK-2561>, FLINK-2785
<https://issues.apache.org/jira/browse/FLINK-2785>: These are about syncing
the Scala and Java APIs. There is one method missing, documentation, and
examples. I'm working on these and should have a PR tomorrow or Thursday.

3. FLINK-2443 <https://issues.apache.org/jira/browse/FLINK-2443>: This is a
bug in the CompactingHashTable that might affect Gelly iterations. The
problem with this one is that it's hard to reproduce and debug. I ran
GSAConnectedComponents with the same dataset mentioned in the JIRA on EC2
and couldn't reproduce the bug. However, I did ran into (probably) the same
bug with some other delta iteration code of mine (not Gelly), but I had no
luck debugging it. You can see the discussion here
<https://mail-archives.apache.org/mod_mbox/flink-dev/201509.mbox/%3CCAJZ2dcW5JK3khtndSu4ixPfe5WgHOj-%3DiCJaUN3pdhK1DY_6ZQ%40mail.gmail.com%3E>.
The good news is that Gelly iteration methods have a
"setSolutionSetUnmanaged" option, which avoids this.

From my side, I don't see any blocking issues or major functionality
missing (after 1-2 are addressed).

My only concern is FLINK-2443. What do you think we should do about it?

Thank you!

Cheers,
Vasia.
Reply | Threaded
Open this post in threaded view
|

Re: Graduating Gelly in 0.10

Kostas Tzoumas-2
Does the workaround work, or can FLINK-2443 still appear?

If we have a workaround, we can schedule a fix to FLINK-2443 for 0.10.1

On Tue, Sep 29, 2015 at 9:59 PM, Vasiliki Kalavri <[hidden email]
> wrote:

> Hello squirrels,
>
> since it was already mentioned in the other thread, I'm starting this
> discussion to see what the community thinks about graduating Gelly in the
> upcoming 0.10 release.
>
> In my opinion, in order to do that we have to look into the following
> issues:
>
> 1. FLINK-2663 <https://issues.apache.org/jira/browse/FLINK-2663>: I've
> already opened a PR and will merge if no objections.
>
> 2. FLINK-2561 <https://issues.apache.org/jira/browse/FLINK-2561>,
> FLINK-2785
> <https://issues.apache.org/jira/browse/FLINK-2785>: These are about
> syncing
> the Scala and Java APIs. There is one method missing, documentation, and
> examples. I'm working on these and should have a PR tomorrow or Thursday.
>
> 3. FLINK-2443 <https://issues.apache.org/jira/browse/FLINK-2443>: This is
> a
> bug in the CompactingHashTable that might affect Gelly iterations. The
> problem with this one is that it's hard to reproduce and debug. I ran
> GSAConnectedComponents with the same dataset mentioned in the JIRA on EC2
> and couldn't reproduce the bug. However, I did ran into (probably) the same
> bug with some other delta iteration code of mine (not Gelly), but I had no
> luck debugging it. You can see the discussion here
> <
> https://mail-archives.apache.org/mod_mbox/flink-dev/201509.mbox/%3CCAJZ2dcW5JK3khtndSu4ixPfe5WgHOj-%3DiCJaUN3pdhK1DY_6ZQ%40mail.gmail.com%3E
> >.
> The good news is that Gelly iteration methods have a
> "setSolutionSetUnmanaged" option, which avoids this.
>
> From my side, I don't see any blocking issues or major functionality
> missing (after 1-2 are addressed).
>
> My only concern is FLINK-2443. What do you think we should do about it?
>
> Thank you!
>
> Cheers,
> Vasia.
>
Reply | Threaded
Open this post in threaded view
|

Re: Graduating Gelly in 0.10

Ufuk Celebi-2

> On 30 Sep 2015, at 10:06, Kostas Tzoumas <[hidden email]> wrote:
>
> Does the workaround work, or can FLINK-2443 still appear?
>
> If we have a workaround, we can schedule a fix to FLINK-2443 for 0.10.1

+1 Sounds like a good plan to me.

– Ufuk
Reply | Threaded
Open this post in threaded view
|

Re: Graduating Gelly in 0.10

Vasiliki Kalavri
If by workaround you mean setting the solutionSetUnmanaged option, yes that
works, because the CompactingHashTable is skipped.

-V.

On 30 September 2015 at 10:44, Ufuk Celebi <[hidden email]> wrote:

>
> > On 30 Sep 2015, at 10:06, Kostas Tzoumas <[hidden email]> wrote:
> >
> > Does the workaround work, or can FLINK-2443 still appear?
> >
> > If we have a workaround, we can schedule a fix to FLINK-2443 for 0.10.1
>
> +1 Sounds like a good plan to me.
>
> – Ufuk
Reply | Threaded
Open this post in threaded view
|

Re: Graduating Gelly in 0.10

Kostas Tzoumas-2
Yes, exactly. Then, I do not think FLINK-2443 is a blocker for graduating

On Wed, Sep 30, 2015 at 11:09 AM, Vasiliki Kalavri <
[hidden email]> wrote:

> If by workaround you mean setting the solutionSetUnmanaged option, yes that
> works, because the CompactingHashTable is skipped.
>
> -V.
>
> On 30 September 2015 at 10:44, Ufuk Celebi <[hidden email]> wrote:
>
> >
> > > On 30 Sep 2015, at 10:06, Kostas Tzoumas <[hidden email]> wrote:
> > >
> > > Does the workaround work, or can FLINK-2443 still appear?
> > >
> > > If we have a workaround, we can schedule a fix to FLINK-2443 for 0.10.1
> >
> > +1 Sounds like a good plan to me.
> >
> > – Ufuk
>