If your systems have more than one interface then you probably need to
tell gmond which interface to bind to by setting the mcast_if option to
eth1 in the gmond.conf file.  The exact details will vary depending on
which version of ganglia you are using, 2.5.x or 3.0.x.

~Jason


On Tue, 2005-09-20 at 14:41 +0100, Alex Davies wrote:
> Dear All,
> 
> Thank you for your continued support. The firewall is reporting errors
> like these:
> 
> iptables: dropped input: IN=eth0 OUT=
> MAC=ff:ff:ff:ff:ff:ff:00:e0:81:32:00:1a:08:00 SRC= 0.0.0.0
> DST=255.255.255.255 LEN=42 TOS=0x00 PREC=0x00 TTL=1 ID=62540 PROTO=UDP
> SPT=3711 DPT=3711 LEN=22
> 
> This on all the servers.
> 
> My current firewall, in english, is 
>       * All outgoing traffic allowed
>       * All incomming traffic on eth1 allowed without going through
>         firewall
>       * All incomming traffic on eth0 allowed if it comes from another
>         cluster node or on a standard port ( e.g. 80; I did try the
>         gmond ports in here to no avail).
> Is there a way of *forcing* gmond to use eth1 for these bizarre
> packets which are going out to 255.255.255.255 (which I take it is
> part of multicasting since this address does not exist)? 
> 
> Failing that, is there a route command I can use to force these
> packets out of eth1?
> 
> Falingthat, is there an IP tables command I can use to allow these
> packets? (I tried allowing all ports from 0.0.0.0 to no avail).
> 
> Many thanks,
> 
> Alex
> 
> 
> On 20/09/05, Jason A. Smith <[EMAIL PROTECTED]> wrote:
> > If you think it is the iptables firewall that is causing you your 
> > problems then try turning on logging of dropped/rejected packets to
> help
> > debug your problem:
> > 
> > iptables -A INPUT -m limit --limit 3/m -j LOG --log-level info \
> >   --log-prefix "iptables: dropped input: " 
> > 
> > this rule needs to go before the DROP/REJECT rule that is at the end
> of
> > your INPUT chain, or just at the bottom if you have a default policy
> set
> > to DROP/REJECT.  Then watch /var/log/messages for iptables logs to
> see 
> > what ganglia related traffic is getting blocked.
> > 
> > Also, I assume that you have a rule allowing your gmetad server to
> > connect to your cluster, something like:
> > 
> > iptables -A INPUT -p tcp -m state --state NEW -m tcp -s
> gmetad.host.ip \
> >   --dport 8649 -j ACCEPT
> > 
> > 
> > ~Jason
> > 
> > 
> > On Tue, 2005-09-20 at 00:12, Alex Davies wrote:
> > > I am afraid that I still experience the complete loss of
> monitoring as 
> > > soon as I start my firewall even with those rules added...
> > >
> > > I cant seem to find any clear instructions on this, but is there
> any
> > > way to get each gmond daemon just to collect statistics from its
> local 
> > > host and have one server collect all the xml files every 20
> seconds or
> > > so?
> > >
> > > Many thanks,
> > >
> > > Alex
> > >
> > >
> > > On 20/09/05, Jason A. Smith < [EMAIL PROTECTED]> wrote:
> > > > If your network switches are configured to do igmp, then you
> will
> > > > probably want to add an iptables rule like this:
> > > > 
> > > > iptables -A INPUT -p igmp -j ACCEPT
> > > >
> > > > We have iptables configured on all of our systems running
> ganglia
> > > > without any problems and only have 2 related rules, the igmp one
> above 
> > > > and a multicast rule like this:
> > > >
> > > > iptables -A INPUT -p udp -m udp -d 239.2.11.71 --dport 8649 -j
> ACCEPT
> > > >
> > > > ~Jason 
> > > >
> > > >
> > > > On Mon, 2005-09-19 at 22:47, Alex Davies wrote:
> > > > > Dear Mike,
> > > > >
> > > > > Many thanks for your very fast reply :) 
> > > > >
> > > > > All my nodes are on the same switch, but we are using software
> > > > > firewalls. The cluster is trying to use the second ethernet
> port which
> > > > > is plugged into a dedicated switch, which is "trusted" and
> not 
> > > > > supposed to be firewalled but I have a hunch that the
> multicasting
> > > > > thing is not going over that port, and I am not sure how to
> force it
> > > > > to (nor am I sure if it matters). 
> > > > >
> > > > > However, could you confirm how you forced ganglia to use your
> tunnel -
> > > > > particularly for the multicast packets?
> > > > >
> > > > > Many thanks, 
> > > > >
> > > > > Alex
> > > > >
> > > > > On 20/09/05, michael chang <[EMAIL PROTECTED]> wrote:
> > > > > > On 9/19/05, Alex Davies < [EMAIL PROTECTED]> wrote:
> > > > > > > I have been  trying to install ganglia on my 13-node
> cluster and had
> > > > > > > it all working wonderfully and was amazed how easilly
> until I 
> > > > > > > restarted my firewall :)
> > > > > >
> > > > > > This is probably impratical, and the worst advice I can give
> you, so
> > > > > > I'm going to BEG one of the others on the list to come up
> with a 
> > > > > > better answer, but I can tell you that I've got Ganglia
> working just
> > > > > > fine over a OpenVPN tunnel (provided that peer-to-peer
> client
> > > > > > communication is online, and the PC hosting the OpenVPN
> tunnel is 
> > > > > > online).  Maybe that's something you want...?  I doubt it,
> but I
> > > > > > figured I'd mention it anyways, just in case.
> > > > > >
> > > > > > I'm wondering, when the firewall restarts, what happens to
> the 
> > > > > > interfaces that Ganglia is multicasting over?  Maybe you
> need
> > > > > > something that automatically restarts ganglia when the
> firewall
> > > > > > restarts ...?  Like I said, someone else probably has a
> better answer. 
> > > > > >
> > > > > > My apologies for any unhelpfulnesses.
> > > > > >
> > > > > > --
> > > > > > ~Mike
> > > > > >  - Just my two cents 
> > > > > >  - No man is an island, and no man is unable.
> > > > > >
> > > > >
> > > >
> > >
> > 
> 
> 
> -- 
> Alex Davies // http://www.davz.net
> 
> This email and any files transmitted with it are confidential and
> intended solely for the use of the individual or entity to whom they
> are addressed. If you have received this email in error please notify 
> the sender immediately by e-mail and delete this e-mail permanently.
> 
> Contact me - MSN: [EMAIL PROTECTED] SKYPE: alex.davies
-- 
/------------------------------------------------------------------\
|  Jason A. Smith                          Email:  [EMAIL PROTECTED] |
|  Atlas Computing Facility, Bldg. 510M    Phone:  (631)344-4226   |
|  Brookhaven National Lab, P.O. Box 5000  Fax:    (631)344-7616   |
|  Upton, NY 11973-5000                                            |
\------------------------------------------------------------------/



Reply via email to