search

tsuna-server / build-server-ansible

1 stars 0 forks source link

Instances on specific node can NOT pass packet that is over about 1500 byte. #80

Closed TsutomuNakamura closed 2 years ago

TsutomuNakamura commented 2 years ago

When I built OpenStack environment with the component below.

Name of node type
dev-controller01 Controller node
dev-compute01 Compute node
dev-compute02 Compute node

Instances only on dev-compute01 can NOT pass packet that is over about 1500 byte. ICMP packets can be passed but HTTP packets can NOT be passed.

TsutomuNakamura commented 2 years ago
Name of node type
dev-controller01 Controller node
dev-compute01 Compute node
dev-compute02 Compute node
dev-compute03 Compute node

When building OpenStack like above with MTU 1442, instances on only dev-compute01 can pass packet that is over about 1500 byte.

If I set MTU 1300, all instances on all of them can it. It it related https://github.com/tsuna-server/build-server-ansible/issues/76 ?

TsutomuNakamura commented 2 years ago

Testing MTU

Size of MTU Result
1442 ❌
1410 ❌
1395 ❌
1393 ❌
1392 ❌
1391 ✅
1388 ✅
1380 ✅
1300 ✅

Even if number of compute nodes over 6, any HTTP request was succeeded.

TsutomuNakamura commented 2 years ago

Setting MTU 1450 on each worker might make any problem?

# ip a
......
2: enp1s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc fq_codel master ovs-system state UP group default qlen 1000
    link/ether 52:54:00:00:00:21 brd ff:ff:ff:ff:ff:ff
    inet6 fe80::5054:ff:fe00:21/64 scope link
       valid_lft forever preferred_lft forever
3: enp2s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1450 qdisc fq_codel state UP group default qlen 1000
    link/ether 52:54:00:00:00:22 brd ff:ff:ff:ff:ff:ff
    inet 172.22.1.12/16 brd 172.22.255.255 scope global enp2s0
       valid_lft forever preferred_lft forever
    inet6 fe80::5054:ff:fe00:22/64 scope link
       valid_lft forever preferred_lft forever
4: ovs-system: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN group default qlen 1000
    link/ether 5a:29:e0:4f:a0:64 brd ff:ff:ff:ff:ff:ff
......

There are few probabilities because any request on each instances through enp1s0 that is belonging in provider network.

TsutomuNakamura commented 2 years ago

It can be solved by aligning MTU as 1500 each node and routers in the cluster. The fix is already merged in https://github.com/tsuna-server/build-dev-server-ansible .