Moving udd away from sqlite

[Jonathan Lange]

On 21 June 2012 12:23, John Arbash Meinel <john at arbash-meinel.com> wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> ...
>
>> We haven't had any movement on this for a few days.
>>
>> To sum up, this is my understanding of the situation. 1. We need to
>> switch our udd to postgres 2. We need to make code changes to do
>> so 3. We want to integrate these changes into trunk to avoid a
>> fork 4. We cannot because udd core developers consider it too
>> dangerous to do so, and the costs of production experimentation too
>> high 5. We therefore must demonstrate that it's not too dangerous
>> 6. We do not know how to do so 7. Thus, to proceed, we need a
>> finite, clear, actionable, cheap plan about to demonstrate the
>> safety of this change or to reduce the cost/risk of changes to
>> production to an acceptable level 8. We need this from you. Until
>> we receive it, we are blocked on this task
>>
>> I hope this helps. I really, really look forward to hearing a plan
>> from the udd maintainers on how we can do this.
>>
>> jml
>>
>
> So my proposal was to update udd to use the Storm code, with some work
> to make it less likely to deadlock. (Specifically (2) from James's
> Original Post.)
>
> If this "doesn't work" to fix the deadlock issues, we rollback and
> re-evaluate. Which may still mean going whole-hog to Postgres, but if
> it does work, we've taken out a lot of the fear, and have been able to
> do it in incremental steps, rather than all at once.
>

We'd definitely be in favour of this. Do we need to co-ordinate
timings with you, or can we just fire off the MP and let you handle
it, the deploy & rollback?

>From our end, it would probably be best if you could do the rollout
with the Storm changes when James Westby is around, since he has the
most experience debugging these issues.

jml