Charles B wrote:
Thanks for the plug for logterse, Chris :-)
You're welcome. The existing logging infrastructure didn't do it for me ;-)
Are we perhaps in need of a two-channel logging strategy where all debug/error messages go to stderr and stdout just logs a [terse :-)] one-line summary of the transaction?
I don't think this is particularly necessary. In a production environment, pretty much all you want to emit to the log is a summary of each transaction. A "transaction log object" if you will. Everything else is either server errors (high priority) which you want to stand out, or debug (low priority) and normally not emitted.
If that was the case then the logterse functionality could be brought closer in to the core. I imagine some folks who don't use multilog might find it a pain to have to extract the logterse lines from the rest of the info in the log output, so this would help them.
My logs normally consist exclusively of a few terse startup diagnostics, reams and reams of logterse records, and the rare error.
It isn't the most elegant thing in the world, but I don't have problems plucking out the logterse lines and feeding them into PostgreSQL. Or rather, the complexity isn't in plucking the log lines out, it's getting so damn many of them into PostgreSQL fast enough and keeping the DBMS structured so that expiration of old records doesn't take weeks.
