[mod_cluster-issues] [JBoss JIRA] (MODCLUSTER-487) Advertise: UDP Multicast on Linux systems with more NICs: datagrams could be delayed or lost

Michal Karm Babacek (JIRA) issues at jboss.org
Mon Mar 21 15:32:00 EDT 2016 

Michal Karm Babacek created MODCLUSTER-487:
----------------------------------------------

             Summary: Advertise: UDP Multicast on Linux systems with more NICs: datagrams could be delayed or lost
                 Key: MODCLUSTER-487
                 URL: https://issues.jboss.org/browse/MODCLUSTER-487
             Project: mod_cluster
          Issue Type: Bug
          Components: Native (httpd modules)
    Affects Versions: 1.3.1.Final, 1.2.11.Final
         Environment: Linux, multiple NICs environment 
            Reporter: Michal Karm Babacek
            Assignee: Michal Karm Babacek
            Priority: Critical


Credit where it's due: the issue was first spotted by [~rhatlapa].

h3. Problem
It appears that trying to send to all interfaces with {{NULL}} or {{"0.0.0.0"}} in the following statement actually picks the first non-loopback interface and sends to it.
The result is that there is no datagram on other interfaces. Surprisingly, this is not deterministic though: After dozens or hundreds of messages, eventually one datagram reaches another interface.
{code}
if ((rv = apr_sockaddr_info_get(&ma_niface_sa, NULL,
                                ma_mgroup_sa->family, 0,
                                APR_UNSPEC, pool)) != APR_SUCCESS) {
    ap_log_error(APLOG_MARK, APLOG_ERR, rv, s,
                 "mod_advertise: ma_group_join apr_sockaddr_info_get(0.0.0.0:0) failed");
    return rv;
}
{code}

h3. Impact
Picture this simple scenario: There are two interfaces, e.g. 
{noformat}
enp1s0 10.16.88.187
enp2s0 172.18.0.1
{noformat}
listed in this exact order with {{ip addr show}}.

One has an EAP 7 (Wildfly 10) instance with mod_cluster bound to {{172.18.0.1}} IP address, which implies {{enp2s0}} interface.

Furthermore, one has an Apache HTTP Server instance with mod_cluster bound to {{172.18.0.1}} IP address, i.e. MCMP VirtualHost and main VirtualHost all Listen on this IP address.

Result: Without advertising, using an explicit {{proxy-list}}, all is well. MCMP works, requests work, balancing works.
On the other hand, relying on advertisement, it could take EAP 7 (Wildfly 10) *minutes* to register with the balancer.
The reason is that a vast majority of UDP Multicast datagrams arrives at enp1s0 and EAP 7 (Wildfly 10) doesn't see them.

h3. Reproducer
Lemme demonstrate with a recently refactored [advertise.c|https://github.com/Karm/mod_cluster/tree/advertise-native-test/test/native/advertise] utility for sending datagrams and the well known [Advertize.java|https://raw.githubusercontent.com/modcluster/mod_cluster/master/test/java/Advertize.java] utility for receiving them.
Your your convenience, here are binaries built from the aforementioned sources:

* Advertize java utility: [^Advertize.class]
* advertise native utility (Linux3 x86_64): [^advertise]
* advertise native utility (WIndows x86): [^advertise.exe]

h3. Demonstration on Linux

h4. System
{noformat}
[mbabacek at perf09 ~]$ ip addr show
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN 
    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
    inet 127.0.0.1/8 scope host lo
       valid_lft forever preferred_lft forever
2: enp1s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP qlen 1000
    link/ether 00:18:8b:7a:46:04 brd ff:ff:ff:ff:ff:ff
    inet 10.16.88.187/21 brd 10.16.95.255 scope global enp1s0
       valid_lft forever preferred_lft forever
    inet 10.16.93.253/21 brd 10.16.95.255 scope global secondary enp1s0
       valid_lft forever preferred_lft forever
3: enp2s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP qlen 1000
    link/ether 00:18:8b:7a:46:05 brd ff:ff:ff:ff:ff:ff
    inet 172.17.72.254/19 brd 172.17.95.255 scope global enp2s0
       valid_lft forever preferred_lft forever
4: docker0: <NO-CARRIER,BROADCAST,MULTICAST,UP> mtu 1500 qdisc noqueue state DOWN 
    link/ether 02:42:07:ab:74:f9 brd ff:ff:ff:ff:ff:ff
    inet 172.18.0.1/16 scope global docker0
       valid_lft forever preferred_lft forever
{noformat}

h4. Java
{noformat}
[mbabacek at perf09 ~]$ java -version
openjdk version "1.8.0_71"
OpenJDK Runtime Environment (build 1.8.0_71-b15)
OpenJDK 64-Bit Server VM (build 25.71-b15, mixed mode)
{noformat}

h4.Advertise SENT
{noformat}
[mbabacek at perf09 ~]$ date;./advertise -a 224.0.1.102 -p 33364
Mon Mar 21 12:39:51 EDT 2016
UDP Multicast address to send datagrams to. Value: 224.0.1.102
UDP Multicast port. Value: 33364
IP address of the NIC to bound to. Value: NULL
apr_socket_bind on 0.0.0.0:0
apr_mcast_join on 0.0.0.0:0
apr_socket_sendto to 224.0.1.102:33364
{noformat}

h4. Advertize RECEIVED

YES (/)
{noformat}
[mbabacek at perf09 ~]$ java Advertize 224.0.1.102 33364
Linux like OS
ready waiting...
received: Advertize !!! Mon, 21 Mar 2016 16:39:51 GMT
received from /10.16.88.187:38907
{noformat}

YES (/)
{noformat}
[mbabacek at perf09 ~]$ java Advertize 224.0.1.102 33364 10.16.88.187
Linux like OS
ready waiting...
received: Advertize !!! Mon, 21 Mar 2016 16:39:51 GMT
received from /10.16.88.187:38907
{noformat}

NO (x)
{noformat}
[mbabacek at perf09 ~]$ java Advertize 224.0.1.102 33364 172.17.72.254
Linux like OS
ready waiting...
{noformat}

YES (/)
{noformat}
[mbabacek at perf09 ~]$ java Advertize 224.0.1.102 33364 0.0.0.0
Linux like OS
ready waiting...
received: Advertize !!! Mon, 21 Mar 2016 16:39:51 GMT
received from /10.16.88.187:38907
{noformat}

And now let's take a look at {{172.17.72.254}}, i.e. {{enp2s0}}

h4. Advertise SENT
{noformat}
[mbabacek at perf09 ~]$ date;./advertise -a 224.0.1.102 -p 33364 -n 172.17.72.254
Mon Mar 21 12:42:57 EDT 2016
UDP Multicast address to send datagrams to. Value: 224.0.1.102
UDP Multicast port. Value: 33364
IP address of the NIC to bound to. Value: 172.17.72.254
apr_socket_bind on 172.17.72.254:0
apr_mcast_join on 172.17.72.254:0
apr_socket_sendto to 224.0.1.102:33364
{noformat}

h4. Advertize RECEIVED
NO (x)
{noformat}
[mbabacek at perf09 ~]$ java Advertize 224.0.1.102 33364
Linux like OS
ready waiting...
{noformat}

NO (x)
{noformat}
[mbabacek at perf09 ~]$ java Advertize 224.0.1.102 33364 10.16.88.187
Linux like OS
ready waiting...
{noformat}

YES (/)
{noformat}
[mbabacek at perf09 ~]$ java Advertize 224.0.1.102 33364 172.17.72.254
Linux like OS
ready waiting...
received: Advertize !!! Mon, 21 Mar 2016 16:42:57 GMT
received from /172.17.72.254:35452
{noformat}

NO (x)
{noformat}
[mbabacek at perf09 ~]$ java Advertize 224.0.1.102 33364 0.0.0.0
Linux like OS
ready waiting...
{noformat}

h3. Demonstration on Windows
One could note that the problem doesn't exist on Windows. All interfaces receive advertising.

h4. Advertise SENT
{noformat}
C:\Users\karm\advertise-build
λ advertise.exe -a 224.0.1.102 -p 33364
UDP Multicast address to send datagrams to. Value: 224.0.1.102
UDP Multicast port. Value: 33364
IP address of the NIC to bound to. Value: NULL
apr_socket_bind on 0.0.0.0:0
apr_mcast_join on 0.0.0.0:0
apr_socket_sendto to 224.0.1.102:33364
{noformat}

h4. Advertize RECEIVED
YES (/)
{noformat}
C:\Users\karm\WORKSPACE
λ "C:\Program Files\Java\jdk1.8.0_74\bin\java" Advertize 224.0.1.102 33364
ready waiting...
received: Advertize !!! Mon, 21 Mar 2016 18:07:50 GMT
received from /192.168.122.52:61805
{noformat}

YES (/)
{noformat}
C:\Users\karm\WORKSPACE
λ "C:\Program Files\Java\jdk1.8.0_74\bin\java" Advertize 224.0.1.102 33364 192.168.122.52
ready waiting...
received: Advertize !!! Mon, 21 Mar 2016 18:07:50 GMT
received from /192.168.122.52:61805
{noformat}

YES (/)
{noformat}
C:\Users\karm\WORKSPACE
λ "C:\Program Files\Java\jdk1.8.0_74\bin\java" Advertize 224.0.1.102 33364 192.168.122.199
ready waiting...
received: Advertize !!! Mon, 21 Mar 2016 18:07:50 GMT
received from /192.168.122.52:61805
{noformat}

h4. Advertise SENT
{noformat}
C:\Users\karm\advertise-build
λ advertise.exe -a 224.0.1.102 -p 33364 -n 192.168.122.199
UDP Multicast address to send datagrams to. Value: 224.0.1.102
UDP Multicast port. Value: 33364
IP address of the NIC to bound to. Value: 192.168.122.199
apr_socket_bind on 192.168.122.199:0
apr_mcast_join on 192.168.122.199:0
apr_socket_sendto to 224.0.1.102:33364
{noformat}

h4. Advertize RECEIVED
YES (/)
{noformat}
C:\Users\karm\WORKSPACE
λ "C:\Program Files\Java\jdk1.8.0_74\bin\java" Advertize 224.0.1.102 33364
ready waiting...
received: Advertize !!! Mon, 21 Mar 2016 18:09:55 GMT
received from /192.168.122.199:52781
{noformat}

YES (/)
{noformat}
C:\Users\karm\WORKSPACE
λ "C:\Program Files\Java\jdk1.8.0_74\bin\java" Advertize 224.0.1.102 33364 192.168.122.52
ready waiting...
received: Advertize !!! Mon, 21 Mar 2016 18:09:55 GMT
received from /192.168.122.199:52781
{noformat}

YES (/)
{noformat}
C:\Users\karm\WORKSPACE
λ "C:\Program Files\Java\jdk1.8.0_74\bin\java" Advertize 224.0.1.102 33364 192.168.122.199
ready waiting...
received: Advertize !!! Mon, 21 Mar 2016 18:09:55 GMT
received from /192.168.122.199:52781
{noformat}

Ideas? :) [~jfclere], [~rhusar] 
I suggest adding {{bindaddr}} instead of NULL or perhaps adding a configuration directive that lets one to pick either MCMP VirtualHost ip address or default -- NULL.  





--
This message was sent by Atlassian JIRA
(v6.4.11#64026)

More information about the mod_cluster-issues mailing list