kafka-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "John Roesler" <vvcep...@apache.org>
Subject Re: [VOTE] KIP-441: Smooth Scaling Out for Kafka Streams
Date Mon, 04 May 2020 02:09:57 GMT
Hi Guozhang,

Ah, good question. Yes, the assignor will always now try to achieve a perfect balance. This
was also the proposed default in the KIP before. The config would have allowed users to relax
the search for perfection.

This is actually one of our motivations now to remove it. We feel it’s simpler to reason
about the behavior of the system if you know it’s always going to produce a balanced assignment.


Thanks,
John 

On Sun, May 3, 2020, at 19:03, Guozhang Wang wrote:
> Hello John / Sophie:
> 
> With this config removed, would the assignor always tries to to achieve the
> "perfect balance" (of course, it may be a sub-optimal local plateau) or
> there's an internal hard-coded factor to still retain some satisfying
> threshold?
> 
> Guozhang
> 
> On Sun, May 3, 2020 at 9:23 AM John Roesler <vvcephei@apache.org> wrote:
> 
> > Hi Matthias,
> >
> > We originally proposed that config to allow us to skip migrating tasks if
> > the current balance is “good enough”. But during implementation, we became
> > concerned that supporting this option increased code complexity, and it’s
> > also an extra concept for users to have to learn.
> >
> > To keep the new balancing system simpler both internally and externally,
> > we’d like to drop it from the API for now, with the idea of adding it later
> > if needed.
> >
> > Does that seem reasonable?
> >
> > Thanks,
> > John
> >
> > On Fri, May 1, 2020, at 14:18, Matthias J. Sax wrote:
> > > Can you elaborate why to remove it?
> > >
> > > On 5/1/20 11:29 AM, Sophie Blee-Goldman wrote:
> > > > Hey all,
> > > >
> > > > We'd like to make a slight modification to the proposal in this KIP and
> > > > remove
> > > > the *balance.factor* config. We will update the KIP accordingly.
> > Please let
> > > > us know
> > > > if you have any concerns.
> > > >
> > > > Cheers,
> > > > Sophie
> > > >
> > > > On Wed, Jan 15, 2020 at 12:48 PM John Roesler <vvcephei@apache.org>
> > wrote:
> > > >
> > > >> Hello all,
> > > >>
> > > >> After a long hiatus, I've just realized that I'm now able to upgrade
> > my
> > > >> non-binding support to a binding +1 for KIP-441.
> > > >>
> > > >> This brings the vote tally to:
> > > >> 3 binding +1s: Guozhang, Bill, and myself
> > > >> 3 non-binding +1s: Bruno, Vinoth, and Sophie
> > > >>
> > > >> Since the vote has been open for at least 72 hours, the KIP is
> > accepted.
> > > >>
> > > >> Thanks all,
> > > >> -John
> > > >>
> > > >>
> > > >>
> > > >> On Mon, Oct 28, 2019 at 21:02 PM John Roesler <john@confluent.io>
> > wrote:
> > > >>> Hey all,
> > > >>>
> > > >>> Now that the 2.4 release storm is over, I'd like to bump this
vote
> > > >> thread.
> > > >>>
> > > >>> Currently, we have two binding +1s (Guozhang and Bill), and four
> > > >>> non-binding ones (Bruno, Vinoth, Sophie, and myself), and no vetoes.
> > > >>>
> > > >>> Thanks,
> > > >>> -John
> > > >>>
> > > >>> On Thu, Sep 12, 2019 at 12:54 PM Bill Bejeck <bbejeck@gmail.com>
> > wrote:
> > > >>>>
> > > >>>> +1 (binding)
> > > >>>>
> > > >>>> On Thu, Sep 12, 2019 at 1:53 PM Sophie Blee-Goldman <
> > > >> sophie@confluent.io> wrote:
> > > >>>>
> > > >>>>> +1 (non-binding)
> > > >>>>>
> > > >>>>> On Wed, Sep 11, 2019 at 11:38 AM Vinoth Chandar <
> > > >> vchandar@confluent.io> wrote:
> > > >>>>>
> > > >>>>>> +1 (non-binding).
> > > >>>>>>
> > > >>>>>> On Fri, Sep 6, 2019 at 12:46 AM Bruno Cadonna <bruno@confluent.io
> > >
> > > >> wrote:
> > > >>>>>>
> > > >>>>>>> +1 (non-binding)
> > > >>>>>>>
> > > >>>>>>> On Fri, Sep 6, 2019 at 12:32 AM Guozhang Wang
<
> > > >> wangguoz@gmail.com> wrote:
> > > >>>>>>>>
> > > >>>>>>>> +1 (binding).
> > > >>>>>>>>
> > > >>>>>>>> On Thu, Sep 5, 2019 at 2:47 PM John Roesler
<john@confluent.io>
> > > >> wrote:
> > > >>>>>>>>
> > > >>>>>>>>> Hello, all,
> > > >>>>>>>>>
> > > >>>>>>>>> After a great discussion, I'd like to
open voting on KIP-441,
> > > >>>>>>>>> to avoid long restore times in Streams
after rebalancing.
> > > >>>>>>>>> Please cast your votes!
> > > >>>>>>>>>
> > > >>>>>>>>>
> > > >>>>>>>>>
> > > >>>>>>>>>
> > > >>>>>>>>>
> > > >>>>>>>>>
> > > >>
> > https://cwiki.apache.org/confluence/display/KAFKA/KIP-441:+Smooth+Scaling+Out+for+Kafka+Streams
> > > >>>>>>>>>
> > > >>>>>>>>> Thanks,
> > > >>>>>>>>> -John
> > > >>>>>>>>>
> > > >>>>>>>>
> > > >>>>>>>>
> > > >>>>>>>> --
> > > >>>>>>>> -- Guozhang
> > > >>>>>>>
> > > >>>>>>
> > > >>>>>
> > > >>>
> > > >>
> > > >
> > >
> > >
> > > Attachments:
> > > * signature.asc
> >
> 
> 
> -- 
> -- Guozhang
>

Mime
View raw message