WireGuard key rotation
Rotate the WireGuard private key of one site. Rotate one site at a time.
What happens
| When | What happens |
|---|---|
| The site's Secret holds the new key | Within 2s, the active gateway agent puts the new key on every WireGuard interface, and publishes the new public key in Site.status.wireGuardPublicKey. |
Until a peer's SitePeer holds the new public key | The WireGuard links between the site and that peer fail, in both directions: each end rejects the other's key. |
A peer's SitePeer holds the new public key | That peer's agent applies it at once. The link passes its health checks again after the next handshake. |
While the WireGuard links to a peer fail, traffic and the observation exchange take another healthy link to that peer, such as a direct link. With no other healthy link, each of the two sites counts the other's members Unknown: a routed member is held for policy.unknownHold, then the name routes nowhere. So the time from step 2 to step 4 must stay under the smallest unknownHold of the names with a member at either site, unless another link carries the traffic.
Steps
The site is a, its peer b, the Secret strait-wireguard in strait-system. The private key goes from wg genkey into the Secret, and is never printed or written to a file. Public keys are not secrets.
| Step | What to do | Check |
|---|---|---|
| 1. Check the links | strait --context site-a links, and keep the old public key to compare in step 3. | Whether a link other than WireGuard is HEALTHY for each peer. If not, plan steps 2 to 4 within the budget above. |
| 2. Replace the key | Run the dry run, then the same command without --dry-run=server. | The dry run prints secret/strait-wireguard replaced (server dry run). The Secret keeps its name, which the chart's RBAC names. |
| 3. Read the new public key | Read Site.status.wireGuardPublicKey until it differs from the old one. | |
| 4. Update each peer | At each peer, set the SitePeer of site a to the new public key. | |
| 5. Check | strait links and strait peers at both sites. | Each link HEALTHY, and each peer REACHABLE. |
# 1. The links, and the old public key.
strait --context site-a links
OLD_KEY=$(kubectl --context site-a get site a -o jsonpath='{.status.wireGuardPublicKey}')
# 2. Replace the private key.
wg genkey | kubectl --context site-a -n strait-system create secret generic strait-wireguard \
--from-file=privateKey=/dev/stdin --dry-run=client -o yaml \
| kubectl --context site-a -n strait-system replace --dry-run=server -f -
wg genkey | kubectl --context site-a -n strait-system create secret generic strait-wireguard \
--from-file=privateKey=/dev/stdin --dry-run=client -o yaml \
| kubectl --context site-a -n strait-system replace -f -
# 3. The new public key.
A_KEY=$OLD_KEY
while [ "$A_KEY" = "$OLD_KEY" ]; do
sleep 1
A_KEY=$(kubectl --context site-a get site a -o jsonpath='{.status.wireGuardPublicKey}')
done
# 4. At each peer.
kubectl --context site-b patch sitepeer a --type merge \
-p "{\"spec\":{\"wireguard\":{\"publicKey\":\"${A_KEY}\"}}}" --dry-run=server
kubectl --context site-b patch sitepeer a --type merge \
-p "{\"spec\":{\"wireguard\":{\"publicKey\":\"${A_KEY}\"}}}"
# 5. Both ends.
strait --context site-a links
strait --context site-b linkskubectl replace writes the Secret as given. kubectl apply would also copy the key into the annotation kubectl.kubernetes.io/last-applied-configuration.
If a link stays unhealthy: Troubleshooting, the Healthy condition. Why each peer must be updated: RFC0007, Limitations.