On Wed, Oct 28, 2015 at 6:57 AM, Robert Haas <robertmh...@gmail.com> wrote: > On Wed, Oct 28, 2015 at 1:51 AM, Tom Lane <t...@sss.pgh.pa.us> wrote: >> Craig Ringer <cr...@2ndquadrant.com> writes: >>> I think it'd be helpful to define some level of policy about what the >>> debug levels are intended for, so there's some guidance on what level >>> to emit messages on rather than playing "pick a number". >> >> +1 ... I doubt anyone has ever looked at that in a holistic way. > > Well, I don't know that I could give really specific guidance on each > individual level, but what I think is pretty clear is that messages > which tell you about something that's likely to happen very frequently > should only appear if you have really cranked the logging up to the > maximum. So a message that fires every time you touch a data block > had better be DEBUG5, but a message that fires at most once per > checkpoint cycle can afford to be DEBUG1. Qualitatively, I think that > higher debug level (DEBUG1, DEBUG2) should focus on telling you about > things that are potentially interesting and maybe a little bit > unusual, while tracing messages that report things which are entirely > routine should use lower debug levels (DEBUG4, DEBUG5).
I figured it would go something like this: DEBUG1 once or a few times per statement/autovac/checkpoint DEBUG2 several times per statement/autovac/checkpoint (like once per joined relation in the planner) DEBUG3 once or a few times per row/block DEBUG4 several times per row/block DEBUG5 memory allocation. > > I agree with Craig that the transaction system is interesting and > important, but it's not the *only* interesting and important subsystem > we have ... and any log level below DEBUG2 is basically useless right > now; the stuff you want is almost guaranteed to be lost in the noise. The main use I have for any of the DEBUG levels above 1 is to step through them with vim in a specific source file to see what someone felt was important to log when they were debugging that file in the past. Then I change the ones I like to LOG or NOTICE and recompile to get a targeted verbose logging build. This is not to disparage their usefulness, it is hard to imagine me getting into a new part of the code without the ability to use these archeological markers. They are often more useful than the /* comments */ . Cheers, Jeff -- Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-hackers
