agentleFS
kagentikagenti/kagenti/.claude/skills/istio:mesh-selfheal/SKILL.md

istio:mesh-selfheal

Skill304 starsChanged 4 months agoApache-2.0View on GitHub

Detect and recover from Istio Ambient mesh outages where ztunnel/waypoint proxies serve expired mTLS certs and every gateway route returns HTTP 503 (long-running or suspended dev/Kind clusters)

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Scope: dev / Kind: Do not auto-run recovery against a production mesh.

    line 13

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Istio Ambient Mesh Self-Heal

Detect and recover the classic "everything returns 503" outage on a long-running or suspended dev/Kind cluster (rossoctl/rossoctl#1899): after a host suspend the Istio Ambient data plane (ztunnel + waypoints) keeps serving expired istiod-issued mTLS certs and never re-fetches, so all *.localtest.me routes 503 — while every pod still looks Running.

Scope: dev / Kind. Do not auto-run recovery against a production mesh.

When to use

Reach for this when:

  • http://*.localtest.me:8080/ (UI, API, MLflow, Kiali…) all return HTTP 503, and
  • kubectl get pods -A looks healthy, Gateway/HTTPRoute status is Accepted/ResolvedRefs, and
  • the 503 body is upstream connect error … connection termination with header server: istio-envoy.

That pattern = the data plane is reachable and routing, but the upstream mTLS hop fails on an expired cert.

Two failure domains (don't conflate)

  1. Mesh 503 (this skill). ztunnel/waypoint certs are istiod-issued (SDS), not SPIRE. The durable fix is upstream (istio/ztunnel#1679); the dev-cluster mitigation is to restart the data plane so it re-fetches fresh certs.
  2. SPIRE agent reattest deadlock (a separate, coincident suspend casualty) breaks AuthBridge / token-exchange identity, not mesh routing. Check it with --include-spire.

Detect + recover (one command)

# Detect only (read-only) — reports status and prints the recommended recovery commands
scripts/k8s/mesh-recover.sh

# Apply the recovery (rollout restart ztunnel + waypoints + gateway)
scripts/k8s/mesh-recover.sh --fix

# Also check/restart the SPIRE agent (separate AuthBridge domain)
scripts/k8s/mesh-recover.sh --fix --include-spire

Exit codes: 0 healthy · 2 degraded (or fix attempted) · 3 inconclusive (unreachable / not Ambient) · 1 usage/precondition error. Useful flags: --probe-host <host>, --gateway-url <url> (default http://127.0.0.1:8080), --json (machine-readable). The script discovers the ztunnel namespace (Kind installs put it in istio-system; the chart uses istio-ztunnel) rather than hardcoding it.

Manual fallback (what --fix runs)

# 1. ztunnel — the node proxy all mesh traffic flows through (discover its namespace first)
ZT_NS=$(kubectl get ds -A -l app=ztunnel -o jsonpath='{.items[0].metadata.namespace}')
kubectl rollout restart daemonset/ztunnel -n "$ZT_NS"

# 2. waypoint proxies (L7), if any
kubectl rollout restart deploy -A -l gateway.networking.k8s.io/managed-by=istio.io-mesh-controller

# 3. the ingress gateway
kubectl rollout restart deploy/http-istio -n rossoctl-system

# verify
curl -s -o /dev/null -w "HTTP %{http_code}\n" -H "Host: rossoctl-ui.localtest.me" http://127.0.0.1:8080/

Diagnostic one-liners

# Is it a mesh cert problem? (CertificateExpired / expired peer certificate)
kubectl logs -n "$ZT_NS" -l app=ztunnel --tail=100 | grep -iE "certificate expired|CertificateExpired"

# ztunnel cert status (if istioctl is installed)
istioctl ztunnel-config certificates "$(kubectl get pod -n "$ZT_NS" -l app=ztunnel -o jsonpath='{.items[0].metadata.name}')" -n "$ZT_NS"

# SPIRE agent failing to re-attest? (separate domain)
kubectl logs -n spire-system -l app=spire-agent --previous --tail=40 | grep -iE "reattest|token has expired|Agent crashed"

Testing (dev/Kind)

A real suspend/clock-expiry can't be forced deterministically, so a helper induces the same symptom (CA down + ztunnel restarted → data plane has no valid cert → mesh-wide 503) and runs the full detect → --fix → re-probe loop. Destructive to the mesh; Kind-guarded.

scripts/k8s/induce-mesh-outage.sh --self-test   # break → assert detect(exit 2) → restore CA → --fix → assert green
scripts/k8s/induce-mesh-outage.sh --break       # induce and leave broken (to test manually)
scripts/k8s/induce-mesh-outage.sh --restore     # undo

References

  • rossoctl/rossoctl#1899 — the mesh-wide 503 outage this addresses.
  • istio/ztunnel#1679 — the upstream bug for the durable data-plane self-heal fix.
  • Related skills: k8s:health, istio:ambient-waypoint.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.