From sanzem at yahoo.com Thu Jul 19 17:03:19 2007 From: sanzem at yahoo.com (Sanjay Mahapatra) Date: Thu, 19 Jul 2007 10:03:19 -0700 (PDT) Subject: Piranha load balancing, all requests from client end up with one real server, not load balancing Message-ID: <681687.54767.qm@web55813.mail.re3.yahoo.com> Hello I have a pirahna load balancer set up with a virtual server /IP that routes UDP protocol messages to TWO real server host-ports. One host runs pirahna and two other hosts act as the real servers in the is scenario, all three hosts run on Redhat EL 4 cluster. The network type is : Direct Routing, and there is no firewall involved between these hosts. The LVS routing table shows both the routes while both real servers are healthy. Problem : Even though I have persistence turned off ( Persistence set to zero seconds ), it appears that all requests from the same client IP get routed to ONE of the real servers. I am unable to achieve load distribution between the two available real servers ( running an automated load test that originates from a single client IP ) I used round robin at first and later used least connection based scheduling algorithm, but in each case only one of the real servers appears to receive all messages from the same client. Is there something I am missing ? Thanks in advance. Sanjay Mahapatra -------------- next part -------------- An HTML attachment was scrubbed... URL: From herta.vandeneynde at gmail.com Thu Jul 19 19:14:11 2007 From: herta.vandeneynde at gmail.com (Herta Van den Eynde) Date: Thu, 19 Jul 2007 21:14:11 +0200 Subject: Piranha load balancing, all requests from client end up with one real server, not load balancing In-Reply-To: <681687.54767.qm@web55813.mail.re3.yahoo.com> References: <681687.54767.qm@web55813.mail.re3.yahoo.com> Message-ID: On 19/07/07, Sanjay Mahapatra wrote: > Hello > > > I have a pirahna load balancer set up with a virtual server /IP that routes > UDP protocol messages to TWO real server host-ports. > > One host runs pirahna and two other hosts act as the real servers in the is > scenario, all three hosts run on Redhat EL 4 cluster. > > > The network type is : Direct Routing, and there is no firewall involved > between these hosts. > > > The LVS routing table shows both the routes while both real servers are > healthy. > > Problem : > > Even though I have persistence turned off ( Persistence set to zero seconds > ), it appears that all requests from the same client IP get routed to > ONE of the real servers. I am unable to achieve load distribution between > the two available real servers ( running an automated load test that > originates from a single client IP ) > > I used round robin at first and later used least connection based scheduling > algorithm, but in each case only one of the real servers appears to receive > all messages from the same client. > > > Is there something I am missing ? > > Thanks in advance. > > Sanjay Mahapatra Hi Sanjay, Could you post your config file? This sounds like a test environment. Is the client on the same LAN as the real servers? In that case, did you verify that the packets pass through the load balancer? Kind regards, Herta From sanzem at yahoo.com Fri Jul 20 16:28:49 2007 From: sanzem at yahoo.com (Sanjay Mahapatra) Date: Fri, 20 Jul 2007 09:28:49 -0700 (PDT) Subject: Piranha load balancing, all requests from client end up with one real server, not load balancing Message-ID: <726061.67331.qm@web55812.mail.re3.yahoo.com> Thanks Herta, Yes the client in on the same LAN. I am working on tallying the throughput via the loadbalancer, with what reaches the real servers. On the surface it appears that the traffic is going through the The content of lvs.cf is as under: virtual SanjaysTestCluster_SIP { active = 1 address = 10.20.1.99 eth0:2 vip_nmask = 255.255.240.0 port = 5060 expect = "OK" use_regex = 0 send_program = "/usr/local/monitor_scripts/check_sip_server.sh %h" load_monitor = none scheduler = rr protocol = udp timeout = 6 reentry = 15 quiesce_server = 0 server Engine_One { address = bluehost active = 1 weight = 1 } server Engine_Two { address = redhost active = 1 weight = 1 } Thanks Sanjay >>Hi Sanjay, >> >>Could you post your config file? >> >>This sounds like a test environment. Is the client on the same LAN as >>the real servers? In that case, did you verify that the packets pass >>through the load balancer? >>Kind regards, >>Herta -------------- next part -------------- An HTML attachment was scrubbed... URL: From herta.vandeneynde at gmail.com Fri Jul 20 18:35:46 2007 From: herta.vandeneynde at gmail.com (Herta Van den Eynde) Date: Fri, 20 Jul 2007 20:35:46 +0200 Subject: Piranha load balancing, all requests from client end up with one real server, not load balancing In-Reply-To: <726061.67331.qm@web55812.mail.re3.yahoo.com> References: <726061.67331.qm@web55812.mail.re3.yahoo.com> Message-ID: On 20/07/07, Sanjay Mahapatra wrote: > Thanks Herta, Yes the client in on the same LAN. > > I am working on tallying the throughput via the loadbalancer, > with what reaches the real servers. On the surface it appears that the > traffic is going through the > > The content of lvs.cf is as under: > virtual SanjaysTestCluster_SIP { > active = 1 > address = 10.20.1.99 eth0:2 > vip_nmask = 255.255.240.0 > port = 5060 > expect = "OK" > use_regex = 0 > send_program = > "/usr/local/monitor_scripts/check_sip_server.sh %h" > load_monitor = none > scheduler = rr > protocol = udp > timeout = 6 > reentry = 15 > quiesce_server = 0 > server Engine_One { > address = bluehost > active = 1 > weight = 1 > } > server Engine_Two { > address = redhost > active = 1 > weight = 1 > } > > Thanks > Sanjay > > >>Hi > Sanjay, > >> > >>Could you post your config file? > >> > >>This sounds like a test environment. Is the client on the same LAN as > >>the real servers? In that case, did you verify that the packets pass > >>through the load balancer? > > >>Kind regards, > > >>Herta Part of your second sentence got lost somewhere. You can easily check whether the packets pass through the load balancer by executing "tcpdump host [client-hostname]" as root on the load balancer. That will print all packets to and from the client. Kind regards, Herta From sanzem at yahoo.com Fri Jul 20 19:46:45 2007 From: sanzem at yahoo.com (Sanjay Mahapatra) Date: Fri, 20 Jul 2007 12:46:45 -0700 (PDT) Subject: Piranha load balancing, all requests from client end up with one real server, not load balancing In-Reply-To: Message-ID: <974736.21340.qm@web55807.mail.re3.yahoo.com> Thanks Herta, I am able to verify that every message is passing through the loadbalancer host using tshark. That was the part of my sentence that got truncated. I have now modified my experiment a little bit : Its turns out that all requests from a particular client keep getting routed to a particular RealServer ( #1 or #2 depending on the run) persistently although I have persistence turned off. I have two clients sending requests to the load balancer /cluster. I tested the scenario when requests from Test Client #1 ( TC#1) got sent via the load balancer to RealServer#1 while all requests from TC#2 went to Real Server #2. ( As soon as I shutdown one of the RealServer service ports say #1, the piranha routing table, promptly excluded the dead real server/ Realserver#1 [ both in the piranha UI as well in the output of /sbin/ipvsadm ] TC#1 to got failures for almost 5 minutes while TC#2 which was being serviced by RealServer#2 continue to go throuh and did not have any failures. Requests from TC#1 took about 5 minutes to start getting routed to RealServer#2. In a subsequent run of the same scenario as above, I restarted pulse to see if that would cause immediate re routing of requests from TC#1 to RealServer#2. But it made no difference, furthermore, on yet another run, I restarted both /etc/network and pulse and that did not make any difference either. Thanks and Regards Sanjay Herta Van den Eynde wrote: On 20/07/07, Sanjay Mahapatra wrote: > Thanks Herta, Yes the client in on the same LAN. > > I am working on tallying the throughput via the loadbalancer, > with what reaches the real servers. On the surface it appears that the > traffic is going through the > > The content of lvs.cf is as under: > virtual SanjaysTestCluster_SIP { > active = 1 > address = 10.20.1.99 eth0:2 > vip_nmask = 255.255.240.0 > port = 5060 > expect = "OK" > use_regex = 0 > send_program = > "/usr/local/monitor_scripts/check_sip_server.sh %h" > load_monitor = none > scheduler = rr > protocol = udp > timeout = 6 > reentry = 15 > quiesce_server = 0 > server Engine_One { > address = bluehost > active = 1 > weight = 1 > } > server Engine_Two { > address = redhost > active = 1 > weight = 1 > } > > Thanks > Sanjay > > >>Hi > Sanjay, > >> > >>Could you post your config file? > >> > >>This sounds like a test environment. Is the client on the same LAN as > >>the real servers? In that case, did you verify that the packets pass > >>through the load balancer? > > >>Kind regards, > > >>Herta Part of your second sentence got lost somewhere. You can easily check whether the packets pass through the load balancer by executing "tcpdump host [client-hostname]" as root on the load balancer. That will print all packets to and from the client. Kind regards, Herta _______________________________________________ Piranha-list mailing list Piranha-list at redhat.com https://www.redhat.com/mailman/listinfo/piranha-list -------------- next part -------------- An HTML attachment was scrubbed... URL: From sebastien.bonnet at experian.fr Wed Jul 25 08:27:44 2007 From: sebastien.bonnet at experian.fr (=?ISO-8859-1?Q?S=E9bastien_BONNET?=) Date: Wed, 25 Jul 2007 10:27:44 +0200 Subject: Piranha load balancing, all requests from client end up with one real server, not load balancing In-Reply-To: <974736.21340.qm@web55807.mail.re3.yahoo.com> References: <974736.21340.qm@web55807.mail.re3.yahoo.com> Message-ID: <46A70980.1000008@experian.fr> Sanjay Mahapatra a ?crit : > In a subsequent run of the same scenario as above, I restarted pulse > to see if that would cause immediate re routing of requests from TC#1 > to RealServer#2. But it made no difference, furthermore, on yet > another run, I restarted both /etc/network and pulse and that did not > make any difference either. I'm guessing you have NOT solved the ARP problem. On your clients, check (arp -an) that the VIP is correctly associated to the MAC address of the load balancer. Regards, -- *S?bastien BONNET* Ing?nieur syst?me Tel: 04.42.25.15.40 GSM: 06.64.44.58.98 CONFIDENTIALIT? Les informations contenues dans ce document sont de nature confidentielle et ? l'usage exclusif des destinataires pr?vus. Si vous l'avez re?u par erreur, son utilisation sous quelque forme qu'elle soit est strictement interdite. Nous vous remercions dans ce cas de d?truire le message et de prendre contact avec son exp?diteur. L'int?grit? des messages n'?tant pas assur?e sur Internet, Experian ne saurait ?tre tenu responsable si ce message s'av?rait modifi? ou falsifi?. CONFIDENTIAL The information contained in this message and any attachments hereto are confidential and intended solely for the addressee(s) intended. If you are not the intended addressee of this message, you are hereby notified that you have received this document in error. Any review, dissemination, distribution, or copying of this message is prohibited. Please delete it and inform the sender immediately. Thank you for your help. Messages are susceptible to alteration. Experian shall not be liable for the message if altered, changed or falsified. -------------- next part -------------- An HTML attachment was scrubbed... URL: From sanzem at yahoo.com Wed Jul 25 12:56:19 2007 From: sanzem at yahoo.com (Sanjay Mahapatra) Date: Wed, 25 Jul 2007 05:56:19 -0700 (PDT) Subject: Piranha load balancing, all requests from client end up with one real server, not load balancing In-Reply-To: <46A70980.1000008@experian.fr> Message-ID: <642090.79666.qm@web55805.mail.re3.yahoo.com> Thank you Sebastien, Thats right, the problem remains unsolved, I have opened a case with Redhat. And I have checked the network packets on the Test Client hosts, to verify that the Mac address of the Load Balancer/ VIP shows up rather than one of the Real Servers. Thanks for you suggestions Sanjay S?bastien BONNET wrote: Sanjay Mahapatra a ?crit : In a subsequent run of the same scenario as above, I restarted pulse to see if that would cause immediate re routing of requests from TC#1 to RealServer#2. But it made no difference, furthermore, on yet another run, I restarted both /etc/network and pulse and that did not make any difference either. I'm guessing you have NOT solved the ARP problem. On your clients, check (arp -an) that the VIP is correctly associated to the MAC address of the load balancer. Regards, -- div.signature span { width: 150px; display: block; float: left; color: #000099; } div.signature hr { color: #000099; } S?bastien BONNET Ing?nieur syst?me Tel: 04.42.25.15.40 GSM: 06.64.44.58.98 --> CONFIDENTIALIT? Les informations contenues dans ce document sont de nature confidentielle et ? l'usage exclusif des destinataires pr?vus. Si vous l'avez re?u par erreur, son utilisation sous quelque forme qu'elle soit est strictement interdite. Nous vous remercions dans ce cas de d?truire le message et de prendre contact avec son exp?diteur. L'int?grit? des messages n'?tant pas assur?e sur Internet, Experian ne saurait ?tre tenu responsable si ce message s'av?rait modifi? ou falsifi?. CONFIDENTIAL The information contained in this message and any attachments hereto are confidential and intended solely for the addressee(s) intended. If you are not the intended addressee of this message, you are hereby notified that you have received this document in error. Any review, dissemination, distribution, or copying of this message is prohibited. Please delete it and inform the sender immediately. Thank you for your help. Messages are susceptible to alteration. Experian shall not be liable for the message if altered, changed or falsified. _______________________________________________ Piranha-list mailing list Piranha-list at redhat.com https://www.redhat.com/mailman/listinfo/piranha-list -------------- next part -------------- An HTML attachment was scrubbed... URL: From christopherx at xandros.com Mon Jul 30 17:24:55 2007 From: christopherx at xandros.com (Christopher Xu) Date: Mon, 30 Jul 2007 13:24:55 -0400 Subject: Howtos Message-ID: <46AE1EE7.5080904@xandros.com> Hello: I just sub. to this list, and I got a couple question for you: 1) how to search the archive 2) on Piranha, how do I add, and remove the real server when the cluster is running. When I delete a real server, and related nanny process is still running. And nanny for newly add the real server will not start. I did this in the order of remove then add. Is this allowed from daemon point view. (from the GUI, it's fine) Thanks, Chris Xu