Skip to content

Purpose

This page contains various troubleshooting tasks.

Find node for an instance

You might want to check that the node has the required pods running, or do other troubleshooting tasks.

This will tell you what node an instance runs on.

openstack server show <INSTANCE_UUID> -c hypervisor_hostname

Kube-OVN kubectl plugin

  • Kube-OVN has a kubectl plugin.
  • You can see documentation at the Kube-OVN documentation
  • You should install it from wherever you would like to use kubectl from, as described here
    • However, you will also find it already on installed on Genestack Kubernetes nodes, so you can use it with kubectl there, if desired.
  • You can use this to run many OVS commands for a particular node
    • if, for instance, you've found a node for a particular instances as discussed above

You can get help like:

kubectl ko help

kubectl ko {subcommand} [option...]
Available Subcommands:
  [nb|sb] [status|kick|backup|dbstatus|restore]     ovn-db operations show cluster status, kick stale server, backup database, get db consistency status or restore ovn nb db when met 'inconsistent data' error
  nbctl [ovn-nbctl options ...]    invoke ovn-nbctl
  sbctl [ovn-sbctl options ...]    invoke ovn-sbctl
  vsctl {nodeName} [ovs-vsctl options ...]   invoke ovs-vsctl on the specified node
  ofctl {nodeName} [ovs-ofctl options ...]   invoke ovs-ofctl on the specified node
  dpctl {nodeName} [ovs-dpctl options ...]   invoke ovs-dpctl on the specified node
  appctl {nodeName} [ovs-appctl options ...]   invoke ovs-appctl on the specified node
  tcpdump {namespace/podname} [tcpdump options ...]     capture pod traffic
  trace ...    trace ovn microflow of specific packet
    trace {namespace/podname} {target ip address} [target mac address] {icmp|tcp|udp} [target tcp/udp port]    trace ICMP/TCP/UDP
    trace {namespace/podname} {target ip address} [target mac address] arp {request|reply}                     trace ARP request/reply
  diagnose {all|node} [nodename]    diagnose connectivity of all nodes or a specific node
  env-check    check the environment configuration
  tuning {install-fastpath|local-install-fastpath|remove-fastpath|install-stt|local-install-stt|remove-stt} {centos7|centos8}} [kernel-devel-version]    deploy kernel optimisation components to the system
  reload    restart all kube-ovn components

For instance,

kubectl ko vsctl <K8S_NODE_NAME> show

works as if had ran ovs-vsctl show when logged into the ovs-ovn or kube-ovn-cni pod for the specified node.

Usefully, you can check the status of the NB and SB:

kubectl ko nb status
kubectl ko sb status

and check dbstatus:

kubectl ko nb dbstatus

See the full documentation for more details, for Kube-OVN, OVN, and OVS, as applicable.

List all pods on a node

  • You may want to list all pods on a node. You may have found an instance above and now want to check the pods on the host.
    • you can use kubectl ko to run OVS commands for particular nodes as described above without identifying the node for a pod manually as described here.
  • This will show all pods running on a given Kubernetes node.
  • You can use this to see if a node has all of the pods it should running, or to see their status, or find them to check their logs.
  • As described in the "Introduction and Overview" page, you find several pods on a node relevant to operation of OVN, OVS, and Neutron, like:
    • kube-ovn-cni
    • kube-ovn-pinger
    • ovs-ovn
    • neutron-ovn-metadata-agent-default
      • this one only runs on compute nodes
    • other pods that run on all nodes or all computes not directly to OVN.
      • (e.g., kube-proxy, and several others.)

You can use a command like:

kubectl get pods --all-namespaces --field-selector spec.nodeName=$node

where you can use a value found from following steps like those outlined in "Find node for an instance" to see what node you wanted to list the pods for, and use that instead of $node, if, for instance, you wanted to check on the node that has a particular instance.

Finding OVN and OVS pods

  • In addition to the list of pods on each node for OVS and OVN operation, Kube-OVN has several centralized pods, which all run in the kube-system namespace, that you should ensure as running normally when troubleshooting:
    • ovn-central pods
    • kube-ovn-controller pods
    • kube-ovn-monitor pods
      • although this only collects metrics, so may not adversely affect normal operation
  • You may want to find the OVS and OVN pods to use commands on or in them.
    • but, as noted above, you can often use kubectl ko to execute relevant commands for a particular node without manually identifying it as show here, and it seems less confusing to stick with running OVS and OVN commands that way.
      • additionally, note that OVS appears the same from the Kubernetes node itself, ovs-ovn pods, and kube-ovn-cni pods, but the Kubernetes nodes themselves don't have the OVS and OVN commands, while the pods do.
        • (e.g., same process IDs, same UUID identifiers. OVS runs on the Kubernetes nodes without a pod, but appears visible in certain pods.)
  • Nodes have OVS pods as indicated above in the "List all pods on a node" section
    • only compute nodes have neutron-ovn-metadata-agent pods, but you should find kube-ovn-cni, kube-ovn-pinger, and ovs-ovn on all nodes.
  • Kube-OVN has some central pods that don't run per node:
    • You can find these all in the kube-system namespace
    • ovn-central
      • these have the NB and SB DB and ovn-northd as described in the "Introduction and Overview" page.
    • kube-ovn-controller
      • acts as control plane for Kube-OVN
    • kube-ovn-monitor
      • collects OVN status information and the monitoring metrics
  • You can list pods for a node as shown above in "List all pods on a node".

You can list OVN-central pods like:

kubectl -n kube-system get pod -l app=ovn-central

Getting a shell or running commands on the OVS/OVN pods

  • As mentioned, you will probably find it easier to use kubectl ko to run OVS and OVN commands instead of finding pods.
    • For some rare cases, you may wish to do something like run ovsdb-client to dump the OVSDB (although or do some other thing not supported directly from kubectl ko.
  • You may want to check the status of OVS or OVN pods for a node found by following steps like "Find node for an instance" and "List all pods on a node" above, if you have networking trouble with an instance
    • to see if it has all necessary pods running, or check logs for a pod
  • Within the ovs-ovn and kube-ovn-cni pods for a node, you can use OVS commands if needed.

You can get a shell in the ovs-ovn pod like:

kubectl -n kube-system exec -it ovs-ovn-XXXXX -- /bin/bash

You will probably have 5 lower case letters in place of XXXXX because each node has a copy of this pod. You may also have used an OVN central pod instead of an ovs-ovn-XXXXX pod.

Additionally, while mostly not shown here, many OVS commands can and do simply return results, so you might not want or need to spawn an interactive shell as above. As an example:

kubectl -n kube-system exec -it ovs-ovn-XXXX -- ovs-vsctl list manager

gives you the output just like you would get if you executed it from an interactive shell.

You can find all OVS and OVN commands from bin directories in the pod like this:

dpkg -l | perl -lane '$package=$F[1];
  next unless /ovn/ or /openv/;
  chomp(@FILES = `dpkg -L $package`);
  for (@FILES) {
     next unless /bin/;
    -f and print
  }'

These rarely change, so the list produced will look similar to this:

/usr/bin/ovs-appctl
/usr/bin/ovs-docker
/usr/bin/ovs-ofctl
/usr/bin/ovs-parse-backtrace
/usr/bin/ovs-pki
/usr/bin/ovsdb-client
/usr/sbin/ovs-bugtool
/usr/bin/ovs-dpctl
/usr/bin/ovs-dpctl-top
/usr/bin/ovs-pcap
/usr/bin/ovs-tcpdump
/usr/bin/ovs-tcpundump
/usr/bin/ovs-vlan-test
/usr/bin/ovs-vsctl
/usr/bin/ovsdb-tool
/usr/sbin/ovs-vswitchd
/usr/sbin/ovsdb-server
/usr/bin/ovn-ic
/usr/bin/ovn-northd
/usr/bin/ovn-appctl
/usr/bin/ovn-ic-nbctl
/usr/bin/ovn-ic-sbctl
/usr/bin/ovn-nbctl
/usr/bin/ovn-sbctl
/usr/bin/ovn-trace
/usr/bin/ovn_detrace.py
/usr/bin/ovn-controller
  • Different commands work in different pods
    • You can expect ovs-{app,of,vs}ctl to work in ovs-ovn pods, but not ovn-* commands, mostly. Similarly, in ovn-central pods, some ovn-* commands will work, but OVS commands probably won't.

full usage of these goes beyond the scope of this page. However, you can find more information:

Run ovs-vsctl list manager

For an OVS pod, you can check that it has a manager connection. Nodes should have an OVS manager connection for normal operation.

kubectl ko vsctl <node> list manager

_uuid               : 43c682c2-a6c3-493f-9f6c-079ca55a5aa8
connection_mode     : []
external_ids        : {}
inactivity_probe    : []
is_connected        : true
max_backoff         : []
other_config        : {}
status              : {bound_port="6640", n_connections="2", sec_since_connect="0", sec_since_disconnect="0"}
target              : "ptcp:6640:127.0.0.1"

Run ovs-vsctl show

This shows various useful output, such as ports on the bridges, including:

  • br-int, which has the tap devices (instance network interfaces)
  • br-ext, usually for the public Internet
kubectl ko vsctl <node> show

As an aside, you can just list the bridges without the more verbose output of ovs-vsctl show:

kubectl ko vsctl <node> list-br

br-ex
br-int

You will probably have a br-int for things like instance tap devices, and br-ex for a public Internet connection, but the names could vary depending on how you installed Genestack.

Find the tap devices for an instance

KVM creates tap devices for the instance NICs. You might want to identify the tap device on the correct bridge from the output of ovs-vsctl show described above. In that case, you need to find the name of the tap device and which Kubernetes node you can find it on. You can also tcpdump the interface from the Kubernetes node when you find it this way.

This shows you the instance name as used by KVM, which does not match the nova UUID, and the Kubernetes node as the hypervisor hostname:

openstack server show $UUID  -c OS-EXT-SRV-ATTR:instance_name -c hypervisor_hostname -f json

Thereafter, you can get the tap devices from virsh in the libvirt-libvirt-default pod for the Kubernetes node, using the instance_name from the previous command by first getting the domain ID:

kubectl -n openstack exec libvirt-libvirt-default-25vcr -- virsh domid instance-000014a6

and then the tap devices for the domain ID:

kubectl -n openstack exec libvirt-libvirt-default-25vcr -- virsh domiflist 1025

virsh domiflist also provides the MAC address.

Then, you can see that the integration bridge has ports:

kubectl ko <node> ofctl show br-int | grep -iE 'tap28144317-cd|tap3e6fb108-a4'

where the tap devices from grep com from the virsh domiflist. You should also see the MAC addresses match. If grep finds this, you have identified the network interface(s) for the instance on the integration bridge on the correct Kubernetes node.

This information will tell you what to look for regarding the instance in OVS, and you can see these in the output of ip a sh on the compute node itself:

ip a sh

and you can use tcpdump on it there, e.g., tcpdump -i tap28144317-cd.

Use the the MySQL command line client for the DB

Neutron has a pretty comprehensive API, but you might want or need to see the database sometimes. In very bad cases, you may need to adjust the schema.

You need to use a node with access to the service network (with the Kubernetes cluster IPs), e.g., use one of your Kubernetes nodes

If you don't have it, you will need to install a mysql command line client (on a Kubernetes node or a node on the Kubernetes service network):

# On Ubuntu
sudo apt install mariadb-client-core-10.6

Then you can connect to the database:

mysql -u root \
-p$(kubectl --namespace openstack get secret mariadb -o jsonpath='{.data.root-password}' | base64 -d) \
-h mariadb-galera-primary.openstack.svc.cluster.local

Make sure to change svc.cluster.local if you have set the name of your cluster away from this default value.

Maria has databases for Neutron, etc, so you may want use neutron; after starting the client, or add neutron to the MySQL command.

use neutron;

From here, you can use MySQL client commands on the database.