On Wed, Sep 24, 2008 at 03:43:43PM +0200, Michael Hanselmann wrote:
>
> Get rid of ssconf and convert to configuration instead.
> ---
Hmm... Patch LGTM, but I have more doubts now that I realize that for
each ssh connection and such we open and parse the whole config file...
I did a very quick test:
# python -m timeit -s 'import simplejson' 'data = open("config.data").read()'
'x = simplejson.loads(data)'
10 loops, best of 3: 30.8 msec per loop
# python -m timeit -s 'import simplejson' 'data =
open("ssconf_cluster_name").read()'
10000 loops, best of 3: 32.8 usec per loop
Which means that, roughly, on an average-sized config file (and this is
a small cluster), we have a three orders of magnitude penalty in reading
this. 30 miliseconds is not slow, but also not insigifiant.
I wonder if in the future the nodes will be able to re-use the ssconf
files, as written when the config is uploaded. What do you think?
iustin
