incubator-general mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Steve Loughran <>
Subject Re: Slider Proposal
Date Mon, 14 Apr 2014 22:22:05 GMT
On 14 April 2014 04:43, Andreas Neumann <> wrote:

> I'd like to comment on what has been said about Twill.
> >> Twill: handles all the AM logic for running new code packaged as a JAR
> with
> >> an executor method
> The goal of Twill is much broader than to support writing new code. Its
> goal is to ease the development and deployment of any distributed service
> in a cluster. Its impact is most significantly for the development of new
> applications but here are some other aspects:
It should please you to know that for my Berlin Buzzwords talks I'm going
to call out Twill as where you should be starting coding any new YARN app
-stay at the djikstra and knuth layer and avoid going near the "Lamport

>    - Twill can run existing Java applications, even if they were not
>    developed against Twill APIs and do not make use of Twill's advanced
>    features, After all the existing main() method of Java programs bears a
> lot
>    of resemblance to a runnable. For example, we have successfully run
> Presto
>    on Twill, and the same should be possible for other distributed services
>    written in Java (and non-Java language support is on the horizon for
> Twill
>    as well).
>    - Twill does provide very useful features that most distributed apps
>    need (such as service discovery, lifecycle management, elastic scaling,
>    etc.). Even if the first step may be to port an existing applications
>    without modification, it is very much possible that these applications -
>    over time - make use of Twill features.
>    - Twill is not limited to YARN, in fact its APIs are completely
>    YARN-independent, and in addition to a YARN-based implementation, there
> can
>    well be others - for example, it is not far--fetched to think of a
>    MesosTwillRunner.
> In this sense, Twill is more generic than Slider, which is optimized
> specifically for a few existing distributed services. At this time, Slider
> is probably more optimized for running - say - HBase or Accumulo than
> Twill. Yet most likely Twill will eventually have all the required
> features,
> and wouldn't it be better if things like ZK-bindings were done in a widely
> used and common way? Twill has the potential to lead us there.
I'd love all YARN collaborate on service registry stuff -I've been playing
with curator service discovery which is a good starting point, but I see
where I need to go beyond it. And for it to work, we do need as much
commonality as we can

If we can build something driven by the needs of different YARN apps, we
can go to the YARN project itself and say "This is what works for us"

I've just uploaded to Slideshare some slides on Slider, with bits of what I
want from a registry in there

NOTICE: This message is intended for the use of the individual or entity to 
which it is addressed and may contain information that is confidential, 
privileged and exempt from disclosure under applicable law. If the reader 
of this message is not the intended recipient, you are hereby notified that 
any printing, copying, dissemination, distribution, disclosure or 
forwarding of this communication is strictly prohibited. If you have 
received this communication in error, please contact the sender immediately 
and delete it from your system. Thank You.

  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message