Open Stack

On Tue, Dec 22, 2015 at 03:36:02PM +0000, Dougal Matthews wrote:
>    Hi all,
> 
>    This topic came up in the 2015-12-15 meeting[1], and again briefly today.
>    After working with the code that came out of the deployment library
>    spec[2] I
>    had some concerns with how we are storing the templates.
> 
>    Simply put, when we are dealing with 100+ files from
>    tripleo-heat-templates
>    how can we ensure consistency in Swift without any atomicity or
>    transactions.
>    I think this is best explained with a couple of examples.
> 
>    Â - When we create a new deployment plan (upload all the templates to
>    swift)
>    Â Â  how do we handle the case where there is an error? For example, if we
>    are
>    Â Â  uploading 10 files - what do we do if the 5th one fails for some
>    reason?
>    Â Â  There is a patch to do a manual rollback[3], but I have concerns
>    about
>    Â Â  doing this in Python. If Swift is completely inaccessible for a short
>    Â Â  period the rollback wont work either.

How does using a different data store help fix this error path?

Regardless of the Swift/DB/Git choice, you have to detect something failed,
and mark the plan as in a failed state.

>    Â - When deploying to Heat, we need to download all the YAML files from
>    Swift.
>    Â Â  This can take a couple of seconds. What happens if somebody starts to
>    Â Â  upload a new version of the plan in the middle? We could end up
>    trying to
>    Â Â  deploy half old and half new files. We wouldn't have a consistent
>    view of
>    Â Â  the database.

If this happens, the API design is wrong - we should have a plan reference
a unique identifier, including a version (e.g through swift versioned
objects, git tags/branches or whatever), then on deployment heat should
download exactly one version of those artefacts (e.g via a swift tempurl or
a git URL, or whatever).

FYI heatclient already supports downloading template objects directly from
swift, and heat/heatclient already support downloading from http URLs, so
all we have to do AFAICS is generate a URL pointing to a consistent
version of the templates.

>    We had a few suggestions in the meeting:
> 
>    Â - Add a locking mechanism. I would be concerned about deadlocks or
>    having to
>    Â Â  lock for the full duration of a deploy.

I think we need to create a new version of a plan every time it's modified,
then on deployment the concern around locking is much reduced, or even
eliminated?

>    Â - Store the files in a tarball (so we only deal with one file). I think
>    we
>    Â Â  could still hit issues with multiple operations unless we guarantee
>    that
>    Â Â  only one instance of the API is ever running.
> 
>    I think these could potentially be made to work, but at this point are we
>    using the best tool for the job?
> 
>    For alternatives, I see a can think of a couple of options:
> 
>    - Use a database, like we did for Tuskar and most OpenStack API's do.

Storing template files in a database with Tuskar was a mistake - we should
*not* do that again IMO - it was a very opaque interface and caused a lot
of confusion in use.

Storing code (in this case yaml files) in some sort of versioned repository
is *such* a solved problem - it'll be total wheel-reinventing if we
implement template revision control inside a bespoke DB IMHO.

I think swift is probably a workable solution (using versioning and
tempurl's), but we should really consider making the template store
pluggable, as having the API backed by an (operator visible) git repo is
likely to be a nicer and more familiar solution.

The question around validation data I'm less clear on - we already store
discovery data in swift, which seems to work OK - how is the validation
data different, in that it warrants a bespoke DB data store?

Steve