[Bro-Dev] New logging architecture

[Robin Sommer]

On Mon, Jul 04, 2011 at 19:16 -0400, you wrote:

> That makes sense.  See note about message types, however; after reading  
> Jon's comments, I'm starting to think that Rotate / Init / etc. should  
> be handled by the client exclusively.

Not quite sure what you mean be handled exlusively. They'd still need
to be triggered by the LogManager so we would need messages to send
the action over to then be performed by the client. So what I'd
picture is essentially one message per method in LogWriter.

One reason that the manager needs to trigger them is that there are
parameters to pass (like path, etc. for Init). There're more though:
some can be triggered explicitly from the script-layer at any time
(e.g., Flush()); and others need to be executed synchronously across
log files (rotation: all files should rotate at the same time as
otherwise one gets inconsistent timestamps).

> No.  I was referring more to the LogWriter::Write -- when dealing with  
> something inproc, it doesn't make sense to take the entire LogVal ** and  
> encapsulate it into a 0mq message if we can just pass the LogVal pointer  
> directly instead.

Got it. Yes, that makes sense.

Robin

-- 
Robin Sommer * Phone +1 (510) 722-6541 * robin at icir.org
ICSI/LBNL    * Fax   +1 (510) 666-2956 *   www.icir.org