|
On Wed, May 23, 2001 at 09:23:09AM +0200, Roberto Nibali wrote:
> > What does LVSM offer?
> >
> > - Level 4 switching with LVS
> > - High Availabity with heartbeat
> > - Service Level monitoring with mon
> > - Web based gui
> > - On the fly configuration
> > - Cluster Statistics reporting
> >
> > The LVSM homepage is at http://lvsm.theory.org
>
> Cool, yet another cluster manager software :)
>
> > Please send any questions / feedback / complaints to the
> > lvsm mailing list at lvsm@xxxxxxxxxxx
>
> Ok, I just downloaded the code, extracted it and had a 15 minute look
> at it (random stuff I would complain about):
>
> o It's very difficult to understand the code:
> - indentation is only 2 chars
> - c++ , just kidding :) [very advanced c++ for me!]
> - some prototype definitions are missing or should go into *.h files
> e.g, Select.cc is ugly as it gets :))
Well, I suppose readability is subjective, but I'm sorry you find my
code hard to read.
I'm one of those people who thinks 2 space indentation is a good thing :)
Also I like c++. I know most free projects disavow c++ but I think it
has gotten a bad rap. I feel I have used basic c++ features such as
templates and inheritance appropriately.
And yes, Select.cc is an ugly hack. You seem to be good at finding them :)
> o Why do you use the ``ipchains -L -vnx --line-numbers'' approach to get the
> statistics? [ipchains_stats.cc]:
>
Well, let's just say it's a first cut.
> if(use_chains)
> ipch = popen("/sbin/ipchains -L -vnx --line-numbers", "r");
> else
> ipch = popen("/sbin/iptables -L -vnx --line-numbers", "r");
>
> fgets(buf, 4096, ipch);
> keep_going = true;
> char * ptr = NULL;
> int c = 0;
> while(keep_going) {
> if(strstr(buf, "Chain ")==buf && strstr(buf, chain) - buf == 6) {
> fgets(buf, 4096, ipch);
> fgets(buf, 4096, ipch);
> while(strstr(buf, "Chain")!=buf) {
> if(num_rules == c) {
> if(use_chains) {
> ptr = strchr(buf, '-');
> if(!ptr) {
> l->log(LL_WARN, "couldn't find str '-' in buf '%s'\n", buf);
> }
> } else {
> ptr = buf; // aye aye aye.. /\d+[^\d+]\d+[^\d+]\d+\s+/
> while(isdigit(*ptr)) ptr++;
> while(!isdigit(*ptr)) ptr++;
> while(isdigit(*ptr)) ptr++;
> while(!isdigit(*ptr)) ptr++;
> while(isdigit(*ptr)) ptr++;
> while(isspace(*ptr)) ptr++;
> }
> break;
> }
> c++;
>
> o very interesting code, but very CPU intensive! I have 15000 iptables rules,
> and all together 5000 concurrent connections -> iptables -L -vnx is a pain
> in the ass, that's why we have /proc/net/ip_masq/vs_stats.
Ok, yes I see your complaint with ipchains now. That is a _lot_ of rules.
The stats are modular so I will write one for /proc/net/ip_masq/vs_stats
(already on my TODO)
> o you plvsd daemon is not really portable to other architectures or at least
> it takes quite some time: You seem to use the pthreads in a way Plan9 or
> HP/UX would not support them. You assume, everybody is running Linux as
> real servers.
Well, portability was not an initial design goal. Also, I work at a linux
shop and have only linux at home so I develop for linux.
My use of pthreads is sparing, and only once do I call pthread_create().
I'm not really sure why other OS's that claim to have pthread support
would not work.
> o not everybody has the pidof(8) installed on the system, but everybody for
> sure has ps. Generally, make sure you use basic Unix tools and not for
> example distro-specific tools.
>
Well, at the end of the INSTALL file is a known problems section. This is
listed :)
The init scripts are for redhat now. Sucks but writing them portably is
really hard. Yes everyone has ps but the switches are different in many
places.
Did I mention I hate shell scripting :)
Anyways, thanks for your input.
Brian
|
