> This may be a bug and not a feature request.  I was reading through
> the tenjin.py to make sure that you were handling gaememcache
> invalidation.  I do not understand why you are marking the templates
> with a timestamp from time.time() when it is created instead of with
> getmtime().  I believe this opens up a scenario where old templates
> can be served.

> GAE allows you to have multiple live versions.  So you can have a v1
> and v2 or dev and production.  Live versions share the same datastore
> and memcache though.  So it is possible you could have this scenario
> at a slow point in the middle of the night:

> 12:00 Update v1 (production) with new templates
> 12:05 Update v2 (development) with new templates
> 12:08 User connects to v1 enviornment and creates & memcaches template
> with timestamp 12:08
> 12:10 Make v2 the live (production) enviornemnt
> 12:15 User connects to v2 (now production) and gets from memcache v1
> template marked 12:08 timestamp
>       BUG: Should be serving v2 template from disk with 12:05
> timestamp

> Because you can shift back to older versions (like in case of
> discovering an error) not only should you be using the mtime to
> compare, but I believe you should be doing an == compare, not just

> >=.  Continuing the earlier example:

> 12:20 Memcache somehow expires v1 template (not likely, but it can
> happen)
> 12:25 User cocnnects to v2 on new process, creates and memcaches v2
> template with timestamp 12:05
> 12:30 Admin notices problem in v2 application and switches back to v1
> 12:35 User connects and gets memcached v2 template timestamp 12:05
>       BUG: Should be serving v1 template with timestamp 12:00

> Thanks for your consideration,
> Steve