[Chandler-dev] Re: 0.14 Hub Update

[Jared Rhine]

> It seems that something environmental is causing the Dojo build to  
> hang (apparently) indefinitely during its "string interning" phase,  
> in which HTML files are packed into Javascript strings. My current  
> best guess is that this is a result of using the JRockit VM for the  
> build...

> 2) Install a new VM, specifically the Sun JVM, which is well  
> supported (this is currently the solution we're using on the lab  
> instances).

Could you describe more?  Pretty sure lab has been and is currently  
running Jrockit.  0.14 builds on lab but not Hub?  If so, it's  
something a bit more than just Jrockit.  It might be external to the  
JVM, or 64-bit ness.  I would at least like to see a bump in the  
JRockit version first.

> 3) Discontinue our current practice of producing a Hub-local  
> Javascript JAR, instead pulling from the OSAF Maven repos.
>
> In the past I believe we've avoided option 3 due to confusion  
> surrounding the deployment of of Javascript JARs to our Maven repo,  
> but with the smaller corps of active developers (and specifically  
> developers pushing JS JARs to our repo) I'm more comfortable with  
> doing that now. As a result, this is the option I'm leaning toward.

We've avoided pulling from OSAF maven repos not because of errors, but  
because it's bad production practice (IMO) to deploy open source  
software you can't build yourself.  It can make it difficult to deploy/ 
revert mildly-forked service instances, and it's just sketchy to be  
running binary blobs that only someone else can produce (not so much  
for security, but for repeatability and isolation from upstream).   
It's merely a happy coincidence that building from source has avoiding  
various and ongoing problems with the OSAF repositories.

> I remain somewhat concerned that our build/release process is  
> somewhat prone to errors like the ones that made us originally  
> choose to do the Dojo builds locally on Hub, and I'd like to follow  
> up by proposing some infrastructure research and work to rectify this.

I support ongoing work to make Cosmo build reliably.  It's not really  
accurate to say we'd be using OSAF-provided blobs if there had never  
been any problems.  It's been great to be able to build Cosmo "from  
source" from any revision in svn regardless of the state of Dojo  
deployments.

Overall, I don't feel good about either changing the VM (remember we  
switched away from the Sun VM originally because of real-world  
production problems, including full crashes, though that was with  
Jackrabbit), or switching to OSAF-built Dojo.

 From the perils of being "out"/unavailable, I won't throw up a hard  
block, but I'd be pleased to see more characterization first.  What  
environments *do* produce working Dojo builds?  What happens on a more  
recent 64-bit JRockit?  What happens on a 32-bit JRockit on Hub?

-- Jared