The box it was hosted on was shut down (possibly uncleanly), and when it came back up, mongod refused to start. I ran mongod --repair and ended up with a running system, but the data was mangled. There were parts of records inside other records and a large amount of data that I just couldn't find. Fortunately, most of it was backed up elsewhere.
I've been told that maybe some other version is more reliable. I've been told I need to configure it properly by hand before I use it. I don't care. This was a trivial single-machine installation and there is one thing any sane database absolutely must do: avoid completely losing all the data that has been stored in it.
What version was this? 2.0 (which has been out almost a year) has journaling enabled by default which almost always provides complete single server durability for these situations.
Any database can suffer corruption if you shut down the server uncleanly - it depends on the state of the filesystem when it comes back up. That's why you have replication to protect against failure of a single node.
Having run Mongo in production for over 3 years, I've never lost any data despite numerous failure scenarios (including unclean shutdown).
I believe it was 1.8. I know it wasn't a high-reliability configuration, and it could be that an unclean shutdown was to blame. I do not claim to know objectively how reliable MongoDB is.
I am, however very hesitant to rely on it for anything now, much as I would be hesitant to touch a machine that previously gave me a severe electric shock even though its creator and other users assure me it's safe now.
I can tell you now that it's extremely reliable - we store a lot of data in our mongodb - several million documents a day. As mentioned, it would be because journalling wasn't enabled which would mean possible corruption on non-clean shutdown. That's history.
That makes me willing to consider subjecting a test environment to abuse to see if it loses data so as to consider it for production use, but only if I had a compelling use case for it.
I don't understand how writing migrations takes "weeks (if not months)". I can see "hours (if not days)". But, you still have a schema and migrations, they just exist in the application layer instead without a clear upgrade path.
I did need a schemaless store for a project I was prototyping, but then I didn't see why I had to install MongoDB just to get a simple schemaless store. So, I wrote Goatfish:
The box it was hosted on was shut down (possibly uncleanly), and when it came back up, mongod refused to start. I ran mongod --repair and ended up with a running system, but the data was mangled. There were parts of records inside other records and a large amount of data that I just couldn't find. Fortunately, most of it was backed up elsewhere.
I've been told that maybe some other version is more reliable. I've been told I need to configure it properly by hand before I use it. I don't care. This was a trivial single-machine installation and there is one thing any sane database absolutely must do: avoid completely losing all the data that has been stored in it.