Extreme apprehension about RavenDB for production usage.

[Chris Marisic]

The application I've been developing that is using a Raven backend is
slated to goto production in the coming weeks. However I'm really
starting to face extreme apprehension about using Raven. It seems like
every week there's a trade in of 1 set of severe problems for another
set of severe but different problems.

And by severe, I truly mean severe. For how awesome working with Raven
is I can handle just about any minor bug but the 2 things that are
completely inflexible is persistence itself, and indexing. As long as
I can depend on my documents being accurately saved, and that I can
accurately query them everything else pales in comparison to those.
Yet issues with indexing seems to occur far too frequently like that
they stop rebuilding as it is currently doing now.

How much work left is there to be done against indexes? Because at
some point there really needs to be a line drawn and only allowing
changes to index operations to be included in major version releases
that are extensively regression tested.

[Ayende Rahien]

Chris,

The index rebuilding issues has been fixed. It took a long time to resolve them because on the nature of the problem (only show up after reboot).

There are three things left to be done for indexing:

* change the default index mode to NotAnalyzed

* make indexing multi threaded (actually, this is done, but we need to get all the tests working here).

* modify the reduce operation to use the etag approach rather than the task approach.

Aside from the index rebuilding, I don't think that we had major issues that weren't fixed in a very short amount of time.

[Daniel Steigerwald]

Chris, I have two projects going to production soon, so I share your
pain with bugs. I thought about Raven production readiness deeply. I
have to say, I was decided to switch out two times. But I didn't. Here
is my reasons.

1. I never ever lost any data.
2. Each time I had a severe problem, Ayende took personal
responsibility in 24h, and solved bugs in couple of days. He is also
reachable by Skype.
3. Raven is pretty new stuff, which is invaluable. We can change, or
influence how things work.
4. There are several other smart people involved.
5. Last but not least, similarity with http://normproject.org/ gives
us ability to make relative fast migration ;)

The recently increased bug rate is caused the more attention the Raven
gets.

[Matt]

I share your concern but Im sticking with it now, major issues for me
were smuggler being broken (lesson: always test your backups), flawed
transactions support - in terms of the semantics we'd expect, and many
issues near the start (datetime issues, index issues, boolean
issues, .net 3.5 client being fcked etc.. etc..). But I am a week or
two away from deployment and relatively happy with the stability now,
things are working smoothly finally.

Biggest lesson learnt in this - is not to jump on the bandwagon (as
often that's all it is), I reckon at the time I started using Raven
there were less than two dozen real users, however of course if you
looked at the alt.net kids (and most of them are just sheep now) and
the twitter you'd easily conclude this had massive community adoption
- not the case, i think the alt.net kids just want to twit #ravendb.
In hindsight, if I could roll back time I'd have started with
something more proven and reliable, rather then dogfeeding Raven, I'd
have gone with CouchDB. If Im upmost honest I was probably neglient in
my contractual obligations in selecting RavenDB (the perceived coolest
and .net adoption factor blinded me), and as a result I've absorbed
many days working around ravendb issues to get things back ontime.

Where things are now, I say we got around 50 users properly
dogfeeding this (ayende, might say more but unless they've been
reporting bugs or gripes here I doubt they doing anything but ScottGu
style examples), I'd like to see more delays between releases (or
certain releases being hailed as stable), the regressions we've
endured in the past releases are just not tolerable going forward (I
think ayende realises this)..