On 01/12/16 14:35, Saverio Proto wrote:
Your policy routing looks good.
The problem must be somewhere else, where you do the nat maybe ?
Go in the network namespace where there is the neutron router with
address 10.0.16.1
If you tcpdump there what do you see ?
to be 100% sure about the policy routing just go in the network node
where you do the nat.
ip netns exec qrouter-<uuid> wget -O /dev/nullhttp://10.0.16.11/
uuid is the uuid of the neutron router where you are natting
I guess this will work.
Yes, this does seem to work as expected, in both namespaces;
# Determine the controller hosting the router for 10.0.0.11
[stack@osp-director-prod ~]$ neutron l3-agent-list-hosting-router
5f9d983c-3b51-4411-b921-1a523652d55f
+--------------------------------------+--------------------------+----------------+-------+----------+
| id | host |
admin_state_up | alive | ha_state |
+--------------------------------------+--------------------------+----------------+-------+----------+
| 37051003-636f-4fb7-b6d9-1ff9d9182e9d | clc-sby4f-n3.mgt.cluster |
True | :-) | standby |
| b14fcc29-67c0-4420-8abf-433dafde980d | clc-rb15-n1.mgt.cluster |
True | :-) | standby |
| 37518a78-4b39-463a-8387-2866989bba06 | clc-ra15-n2.mgt.cluster |
True | :-) | active |
+--------------------------------------+--------------------------+----------------+-------+----------+
# Change into the namespace and test the 10.0.0.11 web-server
[root@clc-ra15-n2 ~]# ip netns exec
qrouter-5f9d983c-3b51-4411-b921-1a523652d55f wget -O /tmp/test
http://10.0.0.11/; head -n 10 /tmp/test
--2016-12-01 14:54:09-- http://10.0.0.11/
Connecting to 10.0.0.11:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 10701 (10K) [text/html]
Saving to: ‘/tmp/test’
2016-12-01 14:54:09 (318 MB/s) - ‘/tmp/test’ saved [10701/10701]
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN"
"http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8" />
<title>Apache2 Debian Default Page: It works</title>
<style type="text/css" media="screen">
* {
margin: 0px 0px 0px 0px;
padding: 0px 0px 0px 0px;
# Determine the controller hosting the router for 10.0.16.11
[stack@osp-director-prod ~]$ neutron l3-agent-list-hosting-router
4fcdbc75-f53b-4833-902f-689258cd82ea
+--------------------------------------+--------------------------+----------------+-------+----------+
| id | host |
admin_state_up | alive | ha_state |
+--------------------------------------+--------------------------+----------------+-------+----------+
| 37051003-636f-4fb7-b6d9-1ff9d9182e9d | clc-sby4f-n3.mgt.cluster |
True | :-) | standby |
| b14fcc29-67c0-4420-8abf-433dafde980d | clc-rb15-n1.mgt.cluster |
True | :-) | standby |
| 37518a78-4b39-463a-8387-2866989bba06 | clc-ra15-n2.mgt.cluster |
True | :-) | active |
+--------------------------------------+--------------------------+----------------+-------+----------
# Change into the namespace and test the 10.0.16.11 web-server
[root@clc-ra15-n2 ~]# ip netns exec
qrouter-4fcdbc75-f53b-4833-902f-689258cd82ea wget -O /tmp/test
http://10.0.16.11/; head -n 10 /tmp/test
--2016-12-01 14:51:41-- http://10.0.16.11/
Connecting to 10.0.16.11:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 10701 (10K) [text/html]
Saving to: ‘/tmp/test’
2016-12-01 14:51:41 (317 MB/s) - ‘/tmp/test’ saved [10701/10701]
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN"
"http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8" />
<title>Apache2 Debian Default Page: It works</title>
<style type="text/css" media="screen">
* {
margin: 0px 0px 0px 0px;
padding: 0px 0px 0px 0px;
Oh, did you double check the security groups ?
Oh yes, I've fallen for that one too many times! Both Neutron ports
share the same security groups, to allow SSH (TCP 22), ICMP and HTTP
(TCP 80)
I think I may have seen now where I was going wrong though. It's a
peculiarity of our Red Hat OSP8 (so, Liberty) setup that we run VXLANs
with a network_device_mtu of 1400, as the physical network is set 1500.
We should probably have set physical network MTU up to 1600 before
deployment, allowing naive selection of 1500MTU by VXLAN connected
vNICs, and may in future.
In this case;
* The first vNIC was getting a dnsmasq provided IP from DHCP agent, and
so getting the 1400 MTU
* The second vNIC was instead taking a statically configured IP and so
using the naive MTU of 1500, too large for the physical network when
VXLAN overhead is added on top.
Once I revised the second vNIC to also take a DHCP IP, but keep the
post-up routing rules on it, the web-server responds to remote clients
on both interfaces as expected.
Many thanks for your help in chasing this down, it's given me some extra
tools in the Neutron troubleshooting tool-box!
Kind regards,
Paul Browne
Saverio
2016-12-01 15:18 GMT+01:00 Paul Browne<pf...@cam.ac.uk>:
Hello Saverio,
Many thanks for the reply, I'll answer your queries below;
On 01/12/16 12:49, Saverio Proto wrote:
Hello,
while the problem is in place, you should share the output of
ip rule show
ip route show table 1
It could be just a problem in your ruleset
Of course, these are those outputs ;
root@test1:~# ip rule show
0: from all lookup local
32764: from all to 10.0.16.11 lookup rt2
32765: from 10.0.16.11 lookup rt2
32766: from all lookup main
32767: from all lookup default
root@test1:~# ip route show table 1
default via 10.0.16.1 dev eth1
10.0.16.0/24 dev eth1 scope link src 10.0.16.11
and, which one is your webserver ? can you tcpdump to make sure reply
packets get out on the NIC with src address 10.0.16.11 ?
Saverio
The instance has its two vNICs with source addresses 10.0.0.11 & 10.0.16.11,
and the web-server is listening on both.
The HTTP packets do seem to be getting out from 10.0.16.11 as source, but
are stopped elsewhere upstream.
I've attached two pcaps showing HTTP reply packets, one from 10.0.0.11
(first vNIC; HTTP request and reply works to a remote client) and one from
10.0.16.11 (second vNIC; HTTP request is sent, reply not received by remote
client). In the latter case, the server starts to make retransmissions to
the remote client.
Kind regards,
Paul Browne
2016-12-01 13:08 GMT+01:00 Paul Browne<pf...@cam.ac.uk>:
Hello Operators,
For reasons not yet amenable to persuasion otherwise, a customer of our
ML2+OVS classic implemented OpenStack would like to map two floating IPs
pulled from two separate external network floating IP pools, to two
different vNICs on his instances.
The floating IP pools correspond to one pool routable from the external
Internet and another, RFC1918 pool routable from internal University
networks.
The tenant private networks are arranged as two RFC1918 VXLANs, each with
a
router to one of the two external networks.
10.0.0.0/24 -> route to -> 128.232.226.0/23
10.0.16.0/24 -> route to -> 172.24.46.0/23
Mapping two floating IPs to instances isn't possible in Horizon, but is
possible from command-line. This doesn't immediately work, however, as
the
return traffic from the instance needs to be sent back through the
correct
router gateway interface and not the instance default gateway.
I'd initially thought this would be possible by placing a second routing
table on the instances to handle the return traffic;
debian@test1:/etc/iproute2$ less rt_tables
#
# reserved values
#
255 local
254 main
253 default
0 unspec
#
# local
#
#1 inr.ruhep
1 rt2
debian@test1:/etc/network$ less interfaces
# The loopback network interface
auto lo
iface lo inet loopback
# The first vNIC, eth0
auto eth0
iface eth0 inet dhcp
# The second vNIC, eth1
auto eth1
iface eth1 inet static
address 10.0.16.11
netmask 255.255.255.0
post-up ip route add 10.0.16.0/24 dev eth1 src 10.0.16.11 table
rt2
post-up ip route add default via 10.0.16.1 dev eth1 table rt2
post-up ip rule add from 10.0.16.11/32 table rt2
post-up ip rule add to 10.0.16.11/32 table rt2
And this works well for SSH and ICMP, but curiously not for HTTP traffic.
Requests to a web-server listening on all vNICs are sent but replies not
received when the requests are sent to the second mapped floating IP
(HTTP
requests and replies work as expected when sent to the first mapped
floating
IP). The requests are logged in both cases however, so traffic is making
it
to the instance in both cases.
I'd say this is clearly an unusual (and possibly un-natural) arrangement,
but I was wondering whether anyone else on Operators had come across a
similar situation in trying to map floating IPs from two different
external
networks to an instance?
Kind regards,
Paul Browne
--
*******************
Paul Browne
Research Computing Platforms
University Information Services
Roger Needham Building
JJ Thompson Avenue
University of Cambridge
Cambridge
United Kingdom
E-Mail:pf...@cam.ac.uk
Tel: 0044-1223-46548
*******************
_______________________________________________
OpenStack-operators mailing list
OpenStack-operators@lists.openstack.org
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
--
*******************
Paul Browne
Research Computing Platforms
University Information Services
Roger Needham Building
JJ Thompson Avenue
University of Cambridge
Cambridge
United Kingdom
E-Mail:pf...@cam.ac.uk
Tel: 0044-1223-46548
*******************
--
*******************
Paul Browne
Research Computing Platforms
University Information Services
Roger Needham Building
JJ Thompson Avenue
University of Cambridge
Cambridge
United Kingdom
E-Mail:pf...@cam.ac.uk
Tel: 0044-1223-46548
*******************
_______________________________________________
OpenStack-operators mailing list
OpenStack-operators@lists.openstack.org
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators