* * * * *
                                        
                             Those deployment blues
                                        
My department at The Corporation had a deployment this morning (2:00 am).
These deployments don't happen that often (the last one happened in January
of this year; last year we had a total of four deployments) but usually there
are no problems afterwards.

This time we weren't so lucky.

It wasn't a problem with our code, but with a vendor our customer, The
Monopolistic Phone Company, uses. The vendor in question wasn't sending some
critical information we were sending back to The Monopolistic Phone Company.
We didn't notice this initially since our testing just happened to use the
other vendor The Monopolistic Phone Comapny uses. So while it technically
wasn't our problem, getting that particular vendor to even look at a problem,
much less solve it, is a multi-month and multi-money problem, practically it
is our problem.

The base problem is that one vendor who shall rename nameless  is supposed to
forward all SIP (Session Initiation Protocol) headers that start with a
common prefix, but they have a limit to the number of non-standard SIP
headers they'll forward and we've exceeded said limit. Apparently, a new
feature we added, plus moving some existing data to its own header, bumped
the number of headers past this limit. The fix was easy (just put the
existing data we moved back in the old header while keeping it in the new
header) but there was a bit of concern about installing it into production.

You see, because our customer is The Monopolistic Phone Company, and they
have regulartory issues with respect to reliability to contend with, there's
a whole process involved with deployment. Just for starters, we have to give
them a 10-business day notice of any changes, which they can veto …

Oh, and have I mentioned the very scary SLA (Service Level Agreement)s we
have with them? Where vast amounts of money start flowing to The Monopolistic
Phone Company for violations of said SLAs? So you can see why it takes a
significant amount of time to get deployed, and why we have so few.

Fortunately, we're given a number of emergency deployments we can use and
thus, we used one of them today.

All told, from initial bug fix to re-deployment took a total of three hours.
That is the fastest deployment I've seen of our department's code.


Email author at sean@conman.org