[SAtalk] kernel panics and the .spamassassin/bayes_ stuff

[Alex J . Avriette]

There was recently a construction accident here which caused our MX to be offline for a week. We did not have a backup MX, and as such, there were about 10,000 messages waiting when we came back online. The MX chewed and chewed and chewed, but eventually OpenBSD's otherwise fairly stable kernel panicked. I suppose it can be forgiven for this, as the load was around 30 as it processed many emails, each one launching its own sa process, which then tried to read the disk, became io blocked, and so on.

Eventually, it got through everything, and we're back online and reading messages in realtime. What bothers me is that the .spamassassin/bayes_ stuff (we are using qmail and virtual domains) became corrupted.

I've been feeding sa-learn all the spams I see now for a few months, and I'm very unhappy that there is simply no method for recovering those files. If you google around for "Cannot open bayes_path," you encounter several people responding on mailing lists that the only way to recover from this error is to throw out the directory and start anew.

So I moved the files to the side, and everything was okay.

Being both a professional unix administrator and a perl programmer, I can express my strong displeasure with this situation, on the grounds that I would not deploy software that did this. If you had to reinstall Oracle every time the server crashed, we'd all be using xml or other flatfile formats. As such, I'm inclined to say that sa is "not quite ready for prime time."

What can be done to make this more robust?

Thanks,
alex

--
[EMAIL PROTECTED]
Alex J. Avriette, Unix Systems Gladiator
"Shut down some of the bullshit the government is spending money on and use it to buy all the Microsoft stock. If we did that, we could ... just bronze Gates, turn him into a statue, and stick him in front of the Commerce Department." - Scott McNealy