[Sugar-devel] new journal and datastore maintainer

[Tomeu Vizoso]

On Thu, Sep 17, 2009 at 16:51, Martin Langhoff
<martin.langhoff at gmail.com> wrote:
> On Thu, Sep 17, 2009 at 4:27 PM, Gary C Martin <gary at garycmartin.com> wrote:
>> FWIW, your three major points were fixed way back by Tomeu's 0.84 data-store
>> re-write to make for robustness. Actually I'm fairly sure it was these
>> kind'a reports that prompted Tomeu to make that fine effort. Prior to this
>> (0.82) I had seen occasional datastore 'loss', sometime recoverable,
>> sometimes not (while stress testing 3 XOs here). But since 0.84 I've not
>> seen any failures.
>
> Somewhat in agreement. Clearly, there have been huge efforts -- and
> huge thanks go to Tomeu. That is the good side.
>
> But a rewrite is always a risky business. Got to channel jwz here and
> link to this oldie: http://jwz.livejournal.com/154529.html
>
> A rewrite means we have to approach the new DS as a new piece of
> software (or library or service or component). May fix some bugs, may
> add others -- new code new bugs. Hasn't seen the field. And for
> example, cannot be easily backported for a trial run in a deployment,
> and/or to be incorporated into my "papercuts" release.

I'm not sure if Gary said that we can forget about the DS problems in
0.82 because we rewrote the DS in 0.84. I think he was noting that the
feedback we got on 0.82 doesn't apply necessarily to the current code
base because it changed drastically. Actually, if it was redesigned
was for addressing the issues you mention.

>> I guess we're waiting on the OLPC timeline for an official XO signed release
>> with 0.84, before deployments would be comfortable moving forward with the
>> solution. 0.84 is just about 6 months old now, so happy half Birthday! :-)
>
> You are right there. But I would like to note that any widely deployed
> software has a very long "release/deploy" cycle. This happens on OLPC
> deployments, and will happen on all large deployments.
>
> The DS is a very important component. When we see dataloss bugs, we
> need a patch for the deployed version, and a "rescue my data" utility.
>
> "It is rewritten in the next release" is -- in practice -- not very
> useful. There is no fix for the users in any reasonable timeframe. And
> in fact, we don't know how that rewrite will work in the field.
>
> This is not to disparage Tomeu's work. It is positive, and necessary.
> But when we think about users and deployments, there is an aspect that
> is actually more important: getting a fix and a 'repair' path out to
> users, in a way deployments can use.
>
> The current situation with this specific issue is understandable,
> 8.2.x was of course released at a time of turmoil for us all, and the
> DS rewrite was done shortly after.
>
> So I am strictly looking forward -- and wondering if we can help these
> users of 0.82 now. And how we'll handle similar issues when 0.84 is
> deployed and if dataloss bugs are reported.

I also wonder about that ;) Any deployer wants to talk about actual
problems, possible solutions, and how a possible fix could be
deployed?

Regards,

Tomeu

-- 
«Sugar Labs is anyone who participates in improving and using Sugar.
What Sugar Labs does is determined by the participants.» - David
Farning