Ceki, I think you are correct, doing it at the filter level seems awkward and more complex than it needs to be. I think that the appender method is going to work out the best. Thanks for the advice and the info. -Eric Ceki Gulcu wrote:
Hello Eric,
You could place the logic in a TurboFilter. TurboFilters are set at the LoggerContext (i.e. global) level and they intervene before all other components in the processing pipeline.
If my memory serves me correctly, creating a LoggingEvent takes about 200 nanoseconds which is pretty fast. However, for a disabled logging event logback can reach a decision (not to log) in about 10 nanoseconds.
But since you wish to buffer the logging events, they must be created. You can't buffer what does not exist. Moreover, TurboFilters are not supposed to buffer events. They could, but it's not a use case we've considered nor currently support.
The current architecture is linear at the beginning and branches out at the level of appenders. In ASCII art, this looks like
/-->ap0-->F00-->F01->F0n /-->ap1-->F10-->... TB0->TB1->...->TBm->LLFilter->--|-->ap2-->F20->... \-->ap... \-->apX->FX0->...
What you are suggesting resembbles
/-->apJ-->... /.... /-->apJ+1-->.. / /-->apJ+...--> / /-->apJ+j--> / TB0--------------->TB1-->....-->LLFilter-->....unchanged
Where TB stands for TurboFilter LLFilter stands for Logger Level filter which is the main filter in logback ap stands for appender
As I understand it, the advantage of "sectional logging," as we have been discussing, is to be silent if things go well, unless an error occurs, in which case we dump detailed logging information. Is this how you see things or am entirely I missing the point?
Doing sectional logging at the appender level would be the simplest way to proceed. You could do it at other points in the processing pipeline but you'd need to code a lot more on your own and frankly I don't see what you would gain. However, I might be wrong.
Eric Faden wrote:
Cool thanks. I got the idea then. Is there any way to do something like that in the filter level or before it got to the appender? Also how inefficient would it be to create the LoggingEvent? My idea would be to only put this appender on certain classes if for instance there is an exception we are trying to track down. It seems that based on the code creating the LoggingEvent isn't very expensive since the string isn't built until it is actually appended. So setting the Debug level higher and having the Appender handle it wouldn't incur that much of a performance penalty. Right?
-Eric