Does FATE equate to a transaction at the Mutation level?

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

Does FATE equate to a transaction at the Mutation level?

David Medinets
I went to a talk about Foundation DB the other day. They said that
Foundation DB was the only NoSQL tool with transactions. But then I
thought, does FATE serve as a transaction boundary ... at least for
Mutations?
Reply | Threaded
Open this post in threaded view
|

Re: Does FATE equate to a transaction at the Mutation level?

John Vines
No, FATE is for ensuring all steps of an action get completed, not
atomicity. I think of FATE as system level eventual consistency.

For example, when we create a table, we put in !METADATA entries and put
some records in Zookeeper. If the master dies midway through, we're then
left with a table in a half-state that must be manually cleaned up. FATE
provides a mechanism for ensuring that all parts of a multi-phase system
process get completed, but it doesn't do anything for ensuring that the
entirety of a complex operation is completed before it gets noticed.

However, it is possible to jimmy it into the current system via combination
of custom iterators and embedded transaction IDs in the Key. But we
currently provide nothing of the sort.

John

On Sun, Jul 1, 2012 at 12:01 PM, David Medinets <[hidden email]>wrote:

> I went to a talk about Foundation DB the other day. They said that
> Foundation DB was the only NoSQL tool with transactions. But then I
> thought, does FATE serve as a transaction boundary ... at least for
> Mutations?
>
Reply | Threaded
Open this post in threaded view
|

Re: Does FATE equate to a transaction at the Mutation level?

Adam Fuchs
In reply to this post by David Medinets
FATE is really designed to provide low frequency atomic operations across
distributed subcomponents components, rather than the high-speed
transactions across distributed partitions that Foundation DB supports.
Performance in terms of transactions per second is limited with FATE, and
certainly doesn't scale linearly as the cluster grows.

Adam


On Sun, Jul 1, 2012 at 12:00 PM, David Medinets <[hidden email]>wrote:

> I went to a talk about Foundation DB the other day. They said that
> Foundation DB was the only NoSQL tool with transactions. But then I
> thought, does FATE serve as a transaction boundary ... at least for
> Mutations?
>
Reply | Threaded
Open this post in threaded view
|

Re: Does FATE equate to a transaction at the Mutation level?

David Medinets
<sigh> Which your presentation explained basically on the next slide
... I should have read more.

On Sun, Jul 1, 2012 at 12:09 PM, Adam Fuchs <[hidden email]> wrote:

> FATE is really designed to provide low frequency atomic operations across
> distributed subcomponents components, rather than the high-speed
> transactions across distributed partitions that Foundation DB supports.
> Performance in terms of transactions per second is limited with FATE, and
> certainly doesn't scale linearly as the cluster grows.
>
> Adam
>
>
> On Sun, Jul 1, 2012 at 12:00 PM, David Medinets <[hidden email]>wrote:
>
>> I went to a talk about Foundation DB the other day. They said that
>> Foundation DB was the only NoSQL tool with transactions. But then I
>> thought, does FATE serve as a transaction boundary ... at least for
>> Mutations?
>>
Reply | Threaded
Open this post in threaded view
|

Re: Does FATE equate to a transaction at the Mutation level?

Billie J Rinaldi
Just to be clear, Accumulo provides transactions within individual Mutations, but it doesn't use FATE to do so.

Billie


----- Original Message -----

> From: "David Medinets" <[hidden email]>
> To: [hidden email]
> Sent: Sunday, July 1, 2012 12:11:59 PM
> Subject: Re: Does FATE equate to a transaction at the Mutation level?
> <sigh> Which your presentation explained basically on the next slide
> ... I should have read more.
>
> On Sun, Jul 1, 2012 at 12:09 PM, Adam Fuchs <[hidden email]> wrote:
> > FATE is really designed to provide low frequency atomic operations
> > across
> > distributed subcomponents components, rather than the high-speed
> > transactions across distributed partitions that Foundation DB
> > supports.
> > Performance in terms of transactions per second is limited with
> > FATE, and
> > certainly doesn't scale linearly as the cluster grows.
> >
> > Adam
> >
> >
> > On Sun, Jul 1, 2012 at 12:00 PM, David Medinets
> > <[hidden email]>wrote:
> >
> >> I went to a talk about Foundation DB the other day. They said that
> >> Foundation DB was the only NoSQL tool with transactions. But then I
> >> thought, does FATE serve as a transaction boundary ... at least for
> >> Mutations?
> >>