5

Similar question: How to expose kube-dns service for queries outside cluster?

I have a PerconaDB instance in a VM in Google compute engine. Next to it is running Kubernetes cluster where services connect to the PerconaDB.

When I login with MySQL client and do show processlist;, I see following:

| 175 | user       | 10.12.142.24:46124 | user | Sleep   |   14 |                                       | NULL                                                                                                 |         0 |             0 |
| 176 | user       | 10.12.142.24:46126 | user | Sleep   |   14 |                                       | NULL                                                                                                 |         0 |             0 |
| 177 | user       | 10.12.122.42:60806 | user | Sleep   |    2 |                                       | NULL                                                                                                 |         0 |             0 |
| 178 | user       | 10.12.122.43:55164 | user | Sleep   |   14 |                                       | NULL                                                                                                 |         1 |             0 |
| 179 | user       | 10.12.122.43:55166 | user | Sleep   |    4 |                                       | NULL                                                                                                 |         1 |             0 |
| 180 | user       | 10.12.141.11:35944 | user | Sleep   |   14 |                                       | NULL                                                                                                 |         1 |             0 |

Notice the number of different IPs for which I have no idea what they belong to. These are the pods inside the Kubernetes cluster and I would like to know their names so instead of 10.12.142.24:46124 I could see myservice-0dkd0:46124.

I thought the solution would be to somehow link the kube-dns service to the PerconaDB VM, but I have no idea, how to do that correctly. Also this is now running in production, so I don't want to experiment too much.

Vojtěch
  • 11,312
  • 31
  • 103
  • 173

2 Answers2

3

At this moment, the reverse DNS lookup or PTR type lookup for a POD IP is possible only in case of pods that are part of a headless service (details: https://github.com/kubernetes/dns/pull/25) but even that has it's limitations. Furthermore, kubernetes has no default per POD dns name at all, even inside kubernetes cluster you are not able to say curl http://<pod_name>. You have services for that. What you are asking for is not really achievable with kubernetes and DNS as it is now. Please do remember that PTR record (IP->name) should go hand in hand with regular record for resolving name (name->IP) which also makes things complicated, and means you can not have just myservice-0dkd0 in there.

That said, you can achieve what you want in a non-dns way here. Assuming you run on linux, you can use /etc/hosts to maintain a name-to-ip and ip-to-name list that is exclusive to that particular system, and does not need to adhere to all the limitations of real DNS.

If, on your mysql host you run something like following say from cron every 1 min, you should get correctly mapped names in your /etc/hosts almost all the time :

NAMESPACE=default
sed -i "/^[0-9\.]*\t[a-zA-Z0-9-]*\.$NAMESPACE/d" /etc/hosts
kubectl get pod --namespace default --field-selector=status.phase==Running -o jsonpath='{range .items[*]}{.status.podIP}{"\t"}{.metadata.name}.{.metadata.namespace}{"\n"}{end}' >> /etc/hosts
Radek 'Goblin' Pieczonka
  • 21,554
  • 7
  • 52
  • 48
1

Looks like you need to make changes at your mysql server level. Checkout this documentation link for detailed info for skipping name resolution to IP addresses.

By default, MySQL makes a DNS lookup to resolve the client hostnames. You can disable this lookup with the option skip_name_resolve.

mysql> show variables like "skip_name_resolve"; // will show the current state of the variable.

Narendra
  • 382
  • 2
  • 8
  • This is not an answer to my question. – Vojtěch Aug 22 '18 at 04:38
  • This answer is not a solution, but more of an explanation of why it is the way it is, and the way that you can debug it when needed. There're a couple of options available, like, [Skydns](https://github.com/skynetservices/skydns), or[Coredns](https://github.com/coredns/coredns)You may also try using a StatefulSet as per this [post](https://stackoverflow.com/questions/48261375/make-pod-dns-name-resolvable-in-kubernetes-cluster) as StatefulSet maintains a sticky identity for each of their Pods. So since most of these options will require a custom tweak, it will be best to try with a test cluster – Sunny J Aug 23 '18 at 20:09
  • I am not sure I understand: 1. Disabling skip_name_resolve, won't get me the pods names. It is the other way round - I need the name resolving. 2. It doesn't explain how to map the DNS service to pods. – Vojtěch Aug 24 '18 at 06:11