[cosmo-dev] 0.9 post mortem

[Bobby Rullo]

Allow me to summarize:

Ok.

DOH!

DOH!

As for #1, it may not always be possible/practical to revert schema  
(though I bet it almost always is) but it's something we should talk  
about for every release.


On Nov 15, 2007, at 3:05 PM, Randy Letness wrote:

> Bobby Rullo wrote:
>> Here's some thoughts I've had after the 0.9.0 Revert/Data Loss  
>> Fiasco of 07
>>
>> 1) We should have DB Revert scripts. This doesn't mean a script  
>> which reverts to the backup. What I am talking about is a script  
>> which
>>    i) Makes a post-update backup of the database
>>    ii) Migrates the database backwards to the old (previous  
>> release) schema
>
> This would be nice.  In fact the schema changes in 0.9 were small  
> enough that we could have done a revert with a single update  
> statement.  But in past migrations a revert script would have been  
> more complicated, and this is something that would have to be tested  
> too.  I should have thought of the revert option during the fiasco  
> but I had to leave at the exact moment stuff started to blow up.  It  
> would have been nice to have had more time but I didn't know when  
> the update was going to happen.
>
>> Obviously this is only for releases which have a schema change. I'm  
>> not saying that we have to have this for every database we support  
>> - just for hub. Maybe it's not even the Cosmo team's job, maybe  
>> it's Hub (working with Cosmo team of course)
>>
>> 2) I don't think we should have any release dependent on roving  
>> "SNAPSHOT" - it makes bugs difficult to reproduce, since you  
>> potentially get a different build every day. When we test an RC, we  
>> should pin to that day's SNAPSHOT.
>
> +1 We ran into this before with ical4j...should have learned our  
> lesson..doh!
>
>> 3) I'm not sure to what extent we are doing this, but we need to  
>> make sure to test on a cosmo instance with migrated data. This  
>> testing may or may not include the normal windmill stuff, but it  
>> should have added performance stuff, so we know if things are way  
>> out of wack.
>
> +1  Although in this case, because we were using SNAPSHOT  
> dependencies, each test may have been using different code...doh!
>
> -Randy
> _______________________________________________
> cosmo-dev mailing list
> cosmo-dev at lists.osafoundation.org
> http://lists.osafoundation.org/mailman/listinfo/cosmo-dev