kube-ovn ovn 偶现ttl跳变

我测试的时候出现一个很奇怪的情况，我在underlay pod中ping 这个ecmp下一跳网关的时候会偶现254的ttl，正常应该是64的ttl。感觉应该不影响功能，但还是很好奇

image.png

我统计了下

image.png

可以确定的是，该情况并不影响网络功能，也不影响性能。

To understand what is going on here we should start with the question of where and how the TTL is set. When you send a ping to some device you set the TTL on the ping request. The device sends a response and the responding device sets the TTL in the response. The TTL value that you see in your results is the TTL of the response coming to you. Various OS will set the TTL differently. Some, like the router will start with 255. Some like the server may start at 128. And some, like your default gateway may start at 64.

So the differences that you are seeing in TTL are no more than the differences in how various OS will set the TTL. It should have no impact on network performance and is not anything to be concerned about.

参考: https://community.cisco.com/t5/routing/strange-ttl-problem/td-p/1884916

但是另一个点，neutron 计算节点也存在ns，基于kolla-ansible 维护的ovn neturon metadata 服务也在docker里面运行，也是host network模式。 所以可能是容器相关的模式的问题，可以再对照代码看一下

[root@compute028 ~]# ip netns
ovnmeta-8569c79c-4184-406a-ad55-19bf6d47d549 (id: 0)
ovnmeta-1e8d676f-5d0a-4371-aa3d-a0029598e27a (id: 2)
ovnmeta-fcfd0f86-989f-47c3-9830-cd40e75ba873 (id: 1)
ovnmeta-b052976f-29e0-4a3c-ad66-fa23b058427b (id: 10)

# docker ps -a | grep neutron
f8d24cbde993        registry.yealinkops.com/third_party/kolla/centos-source-neutron-metadata-agent:victoria       "dumb-init --single-…"   11 months ago       Up 8 weeks (healthy)                        neutron_ovn_metadata_agent