[ntp:hackers] Patches to reinstate rank variable and add end
keyword.
Danny Mayer
mayer at gis.net
Sat Aug 6 18:52:56 UTC 2005
Paul Vixie wrote:
> # > I don't care about how to implement an asynchronous resolver and forking
> # > is fine with me. That's the way the code used to do it and somebody took
> # > it out.
> #
> # Dave, that's not true. That code is still there.
>
> then forget my note about popen(), plz, and reenable the old fork()-based
> asynchronous resolver, and we'll proceed as though this problem was solved
> (rather than worked around).
>
> # The current proposal is to port the ISC's BIND 8 eventlib code (though I
> # think it originated from somewhere else, Paul can tell you for sure) and
> # use that to handle the asynchonous resolver issues.
>
> that's not the current proposal. the current proposal is to rip select()
> out of ntpd's mainline and convert all of ntpd to use ISC eventlib so that
> ntpd's asynchrony can be portable. at which point it's a trivial matter
> for me to come up with an asynchronous resolver that fits that framework.
> (an asynchronous resolver needs multiple file descriptors, plus the ability
> to open tcp sockets and wait for them to complete, plus multiple timers;
> ISC eventlib provides all of this, portably; ntpd's mainline provides all
> of this, non-portably.)
>
> and ISC eventlib originated with me, not from anywhere else. there's a huge
> API debt owed by eventlib to MIT-X11-Xt and DEC-SRC-Modula3, but no source
> code except what i wrote originally, and what bob halley rewrote, and
> various bug fixes over the years. here's a copy of the freebsd version of
> the man page for this library. i'll repeat my earlier offer to convert ntpd
> to this asynchrony framework, assuming that poul henning-kamp will help me.
> i'll also repeat my earlier offer to release an asynchronous resolver that
> lives within this framework, if ntpd is successfully converted to this
> framework.
>
In my opinion we should do it. It would resolve (no pun intended) a lot
of problems we currently have in NTP. The only issue that I would bring
up is the fact that this code uses timers and ntp fiddles with the
clock. Would this cause any problems with the eventlib code?
Danny
More information about the hackers
mailing list