[Dovecot] Apple patches 11-14

Mike Abbott michael.abbott at apple.com
Wed Jan 7 03:18:21 EET 2009


> I don't see dtrace-dovecot.h included in it.

dtrace-dovecot.h is generated from dtrace-dovecot.d by the command  
"dtrace -h -s $(DTRACE_SOURCE)" in src/lib/Makefile.am.  See the  
section "BUILDING CODE CONTAINING USDT PROBES" in http://developer.apple.com/documentation/Darwin/Reference/ManPages/man1/dtrace.1.html 
  .  Sorry I should have mentioned that up front.

> Couldn't the DOVECOT_OD_LOOKUP_START() macro be changed

Unfortunately that's out of my control:  those macros are generated.   
But I suppose they could be replaced with an intermediate level of  
macros that are undefined when not wanted, and combine the enabled- 
check and the probe-trigger when wanted, just to remove the in-line  
ifdefs.

>> Patch #12 increases many listen queue sizes
>
> I guess those can be changed. I've really no idea what the tradeoffs
> here are.

Fewer refused connections means happier users on good days, but longer  
wait times when something's wedged.  Too-large queue sizes can lead to  
livelock, according to http://www.hpl.hp.com/personal/Lucy_Cherkasova/papers/ieee-dtoc.ps 
  .  The trick is finding the right compromise.  The queue size of 8  
on the client-facing socket is certainly much too small.

>> Patch #14
> Is it useful to have this code inside Dovecot?

Useful for us....

> -1U would be "more" unlimited. :)


I think we tried that and something else broke.

> Why does it matter which uid/gid is used for dump-capability?

Not sure.


More information about the dovecot mailing list