Re: [Chicken-users] egg announcement: remote-repl

[Elf]

On Tue, 19 Aug 2008, Jörg F. Wittenberger wrote:

> Am Montag, den 18.08.2008, 09:19 -0700 schrieb Elf:
> > On Mon, 18 Aug 2008, Jörg F. Wittenberger wrote:
<snip>
> > as a sidenote, i cannot think of any legitimate reason to change timeouts
> > dynamically.
>
> So you probably never had to debug a larger amount of communicating
> machines?  I consider it legitimate to test for behavioral changes of
> the whole system, when timing changes and/or node become uncooperative.
> Apart from wild guesses I can't even see another way to find out good
> values.

ive had to debug a ridiculous amount of HA machines.  ive helped optimise 
the physical layout on the bloody disk to improve seektimes over clusters
that were simultaneously parsing insane amounts of data while being queried
at tens of thousands of times per second, and needed to generate consistent
constraints over the data simultaneously.  i have never had an application 
improve its characteristics from dynamic timeout changes.  the closest
reasonable things that i HAVE needed to do are 1) figure out what the system
was doing beforehand, so i could find sane timeouts and quanta, and 2)
very rarely, have the machine repeat a loop before it 'should', to give it 
an extra chance.  its not a particularly difficult set of algos to work 
out... its essentially a modification of starvation prevention algos. 
changing the timeouts while the system is running can have any number of 
bad effects, if you have carefully timed and optimised machines.

> > every repl session is in its own thread.  as tcp timeouts are parameters,
> > if you want to change the timeouts for that particular thread, all you have
> > to do is:
>
> That's what I have been able to do bevor.  That's not the aim.  I need
> to change all timeouts for all thread which are started after the
> change.

then all you have to do is use the environments egg to make or copy the 
base env before starting the server, and define an evaler to eval certain
expressions in that environment rather than the local.  if you would read 
the docs, it might save us both a lot of time, as this is explained, and 
examples are given.

-elf