Network SLOs

Please consider opening a PR to improve this runbook if you gain new information about causes of the alert, or how to debug or resolve the alert. Click "Edit this Page" in the top right corner to create a PR directly on GitHub.

Cluster Network Packet Loss


The SLI measures the percentage of dropped ICMP ping messages between canary pods on every worker node. Every second on each worker node a canary pod pings canary pods on every other worker node and reports the success as a Prometheus metric.

If this SLI results in an alert, it means that some network packets between pods are lost.

There are two types of alerts that fire if we expect to miss the configured objective.

  • A ticket alert means that the error rate is slightly higher than the objective. If we don’t intervene at some point after receiving this alert, we can expect to miss the objective. However no immediate, urgent action is necessary. A ticket alert should have a label severity: warning.

  • A page alert means that the error rate is significantly higher than the objective. Immediate action is necessary to not miss the objective. A page alert should have a label severity: critical and should page on-call.

Steps for debugging

Packet loss can have multiple root causes and the impact of this alert might vary. Given that the cluster was able to send this alert, a complete outage of the network is unlikely.

In any case, first try to connect to the affected cluster and to open the Prometheus UI. If that works successfully you can run the following query to see between which pods and nodes network packets are lost.

sum(rate(ping_rtt_seconds_count{reason="lost"}[5m])) by (target, source)
* on (target) group_left(target_node)
     max(kube_pod_info{pod_ip!~""}) by (pod_ip, node),
    "target", "$1", "pod_ip", "(.+)"),
   "target_node", "$1", "node", "(.+)")
* on (source) group_left(source_node)
     max(kube_pod_info{pod_ip!~""}) by (pod_ip, node),
    "source", "$1", "pod_ip", "(.+)"),
   "source_node", "$1", "node", "(.+)")
> 0

This query should return one or more time series, indicating packet loss between canary pods and which nodes they run on. From that you should be able to see which nodes are effected. If the target or source node is always the same, this is most likely an issue with that specific node, maybe it just crashed and this isn’t a network issue after all.

Replicate Issue

If the issue isn’t obvious, it’s a good idea to first replicate the issue.

First look at the logs of one of the source canary pods reporting packet loss. Find the relevant pod from the reported source IP.

kubectl -n appuio-network-canary get pod -o wide
kubectl -n appuio-network-canary logs ${CANARY_POD}

In the logs you should be able to see if the issue is actually DNS related or if there are any other unexpected issues. If DNS isn’t working properly, service discovery of the network canary won’t work, which can lead to misreported packet loss.

If the canary doesn’t log anything unusual, the next step is to reproduce the packet loss from within the pod.

kubectl --as cluster-admin -n appuio-network-canary debug -it -c debugger --image=ubuntu ${CANARY_POD}

> apt update
> apt install iputils-ping
> ping ${TARGET_IP}

If you can’t reproduce the packet loss, this might be a false alarm. Check if any other system is reporting network issues. If not this might just be a bug in the network canary. Please open an issue on Github.

Debug Node Issue

Especially if a single node is affected, but also in other cases, it makes sense to connect to an affected node and try to reproduce the packet loss on a node to node level.

Connect to one of the affected source node and ping a target node.

oc --as=cluster-admin -n syn-debug-nodes debug "node/${nodename}"

> chroot /host
> ping ${TARGET_NODE_IP}

If you’re able to reproduce the packet loss between the nodes themselves, the issue is most likely with with the underlying host network or the host itself.

If you can’t reproduce it, there is probably an issue with the overlay network. If you’re using Cilium, the troubleshooting guide might be a good next step. For ovn-kubernetes, check the logs of the ovnkube-master pods or follow the debugging steps provided by ovn-kubernetes.

Cluster inaccessible

If you’re unable to connect to the cluster first check if the cloud provider reports any issues.

Next try to connect to the cluster using the admin credentials in the password manager. Direct access to the cluster API might still work, even if the console isn’t accessible.

If this doesn’t work, try to connect to the loadbalancer VMs using SSH and check if they’re able to reach the cluster nodes. In that case you should be able to connect to the cluster nodes using SSH via the loadbalancers.

# For example:
# IMPORTANT: do NOT add a trailing `/`. Commands below will fail.
export COMMODORE_API_URL=<lieutenant-api-endpoint>

# Set Project Syn cluster and tenant ID
export CLUSTER_ID=<lieutenant-cluster-id> # Looks like: c-<something>
export TENANT_ID=$(curl -sH "Authorization: Bearer $(commodore fetch-token)" ${COMMODORE_API_URL}/clusters/${CLUSTER_ID} | jq -r .tenant)

# Login to vault
export VAULT_ADDR=
vault login -method=oidc

# Fetch SSH key
vault kv get -format=json clusters/kv/${TENANT_ID}/${CLUSTER_ID}/cloudscale/ssh \
  | jq -r '' | base64 --decode > ssh_key
chmod 400 ssh_key

# Connect to a cluster node
LB_HOST=$(grep -E "^Host.*${CLUSTER_ID}" ~/.ssh/sshop_config | head -1 | awk '{print $2}')
ssh -J "${LB_HOST}" -i ssh_key "core@${NODE}"
We don’t have a lot of experience with this alert yet. If you had to debug this alert, please consider adding any insight, tips, or code snippets you gained to this runbook.


If this alert isn’t actionable, noisy, or was raised too late you may want to tune the SLO.

You have the option tune the SLO through the component parameters. You can modify the objective, disable the page or ticket alert, or completely disable the SLO.

The example below will set the SLO set the objective to 99.25% and disable the page alert. This means this SLO won’t alert on-call anymore.

      objective: 99.25
          enabled: false
Disabling the SLO or changing the objective will also impact the SLO dashboard and SLA reporting. Only disable SLOs if they’re not relevant, not if the alerts are noisy.