Agent-Skills.md

Agent Skills: Istio Knowledge Patch

Istio

UncategorizedID: nevaberry/nevaberry-plugins/istio-knowledge-patch

Repository

nevaberry/nevaberry-plugins
NevaberryLicense: MIT
23

Install this agent skill to your local

pnpm dlx add-skill https://github.com/Nevaberry/nevaberry-plugins/tree/HEAD/plugins/knowledge-patch/patches-codex/istio-knowledge-patch

Skill Files

Browse the full folder contents for istio-knowledge-patch.

Download Skill

Loading file tree…

plugins/knowledge-patch/patches-codex/istio-knowledge-patch/SKILL.md

Skill Metadata

Name
istio-knowledge-patch
Description
Istio

Istio Knowledge Patch

Use this skill when designing, upgrading, configuring, or troubleshooting Istio. Resolve the deployed Istio, Kubernetes, Gateway API, CNI, and Helm versions from manifests and live configuration before applying version-sensitive advice. Trust the project's manifests, rendered resources, tests, and observed behavior when they differ from guidance here.

How to use this skill

  1. Identify whether the task concerns an upgrade, installation, ambient mode, traffic management, security, telemetry, or command-line operations.
  2. Read the matching reference file from the index. Read more than one when the task crosses control-plane, data-plane, and installation boundaries.
  3. Check feature gates, environment variables, chart values, API versions, and defaults against the deployed release.
  4. Treat entries marked experimental or alpha as opt-in and verify their Kubernetes API prerequisites.
  5. During mixed-version upgrades, preserve compatibility until every affected proxy, ztunnel, gateway, and control-plane component has moved forward.

Reference index

| Reference | Topics | | --- | --- | | installation-and-upgrades.md | Upgrade gates, removals, Helm behavior, platform requirements, chart resources, and installation security contexts | | ambient-cni-and-multicluster.md | Ambient enrollment, DNS capture, CNI reconciliation, nftables, waypoints, ztunnel, and cross-network routing | | traffic-routing-and-gateways.md | ServiceEntry, DestinationRule, retries, traffic distribution, Gateway API, DNS routing, and EnvoyFilter behavior | | security-identity-and-policy.md | Authorization, JWT, mTLS, certificates, trust bundles, JWKS, Wasm safety, and TLS controls | | telemetry-and-observability.md | Metrics, tracing, logs, dashboards, baggage, workload identity tags, and OpenTelemetry | | operations-and-istioctl.md | istioctl changes, debug access, Pilot controls, resource watching, proxy administration, and diagnostics |

Breaking upgrade checks

Prepare Kubernetes and Gateway API first

  • Istio 1.30 requires Kubernetes 1.32 or newer.
  • Before a 1.30 upgrade, install Gateway API v1.5.x CRDs. Istiod reads TLSRoute and ReferenceGrant from gateway.networking.k8s.io/v1; older CRDs can silently leave TLS passthrough routes unattached.
  • Migrate BackendTLSPolicy from v1alpha3 to v1 before moving to 1.28.
  • Migrate InferencePool objects to v1; use spec.endpointPickerRef.port.number and set the port explicitly.

Account for data-plane transitions

  • Native sidecars are enabled by default for eligible pods from 1.27. Test mutating webhooks and controllers that expect istio-proxy to be a regular container.
  • During a mixed-version 1.28 ambient rollout, older ztunnels NACK the new PASSTHROUGH ServiceEntry type. Upgrade ztunnel before relying on newly added resolution: NONE entries or waypoint enforcement for them.
  • Ambient startup reconciliation changed from opt-in to default-on by 1.29. Use cni.ambient.reconcileIptablesOnStartup=false only when deliberately disabling repair of enrolled-pod iptables or nftables state.
  • Sidecar listeners that explicitly bind ports accept one connection per socket event by default from 1.28. Set MAX_CONNECTIONS_PER_SOCKET_EVENT_LOOP=0 to restore the prior behavior.

Update renamed or removed settings

  • Replace OpenCensus and Lightstep tracing providers with OpenTelemetry.
  • Remove reliance on ENABLE_AUTO_SNI; the flag and its code paths are gone.
  • Replace traffic.sidecar.istio.io/kubevirtInterfaces with istio.io/reroute-virtual-interfaces.
  • Replace ISTIO_META_DNS_AUTO_ALLOCATE with status-based ServiceEntry address allocation.
  • Replace METRIC_ROTATION_INTERVAL and METRIC_GRACEFUL_DELETION_INTERVAL with sidecar.istio.io/statsEvictionInterval.
  • Replace sidecar.istio.io/statsCompression with ProxyConfig.statsCompression or a per-pod proxy.istio.io/config override.
  • Remove --recursive from istioctl analyze; analysis is always recursive. Remove --xds-via-agents from istioctl proxy-status.

Protect chart-driven upgrades

  • A nondefault ztunnel release name can rename resources in 1.25. Pin resourceName=ztunnel; 1.26 restores ztunnel as the default resource name.
  • The 1.29 chart split moves duplicated RBAC and service-account objects from base to istiod and changes several names. Update ownership and automation before applying the charts.
  • Istio-written CNI configuration defaults to mode 0600 from 1.30. Set values.cni.env.CNI_CONF_GROUP_READ=true when non-root group readers require 0640.
  • Istio 1.30 images default to registry.istio.io; allow or mirror that registry before rollout.
  • With Helm 4 server-side apply, set base.validationFailurePolicy: Fail when rendering charts and applying them separately.

High-value configuration changes

Installation and CNI

  • Set global.platform=gke for Helm or values.global.platform=gke for istioctl. GKE installations then create required ResourceQuotas and set cniBinDir; istioctl can detect GKE.
  • Use global.resourceScope: namespace or cluster for persona-based installs. The ztunnel chart uses top-level resourceScope.
  • Enable chart NetworkPolicies with global.networkPolicy.enabled=true; newer defaults include istiod, gateways, and istio-cni.
  • For an Istio-owned ambient CNI conflist, enable cni.istioOwnedCNIConfig=true with both ambient and chained CNI. Ensure a custom filename sorts before the primary CNI file.
  • Choose iptables explicitly with FORCE_IPTABLES_BINARY, or enable native nftables with values.global.nativeNftables=true.

Service discovery and routing

  • ServiceEntries without spec.address receive addresses in status.addresses by default; proxies consume them only with DNS proxying.
  • Use PILOT_IP_AUTOALLOCATE_IPV4_PREFIX and PILOT_IP_AUTOALLOCATE_IPV6_PREFIX to choose controller VIP ranges.
  • Service.spec.trafficDistribution and networking.istio.io/traffic-distribution apply across data planes. A namespace annotation supplies a default unless the Service overrides it.
  • Sidecars and ambient egress waypoints can route wildcard DYNAMIC_DNS ServiceEntries under their documented constraints. Treat SNI and Host values as trusted-client inputs.
  • Use istio.io/connect-strategy: RACE_FIRST_TCP_CONNECT to race DNS A-record endpoints; configure proxy DNS upstream timeout with DNS_FORWARD_TIMEOUT.
  • DestinationRule retry budgets default to 20%; preserve both top-level and subset policies when subsets define their own traffic policy.

Gateway API

  • BackendTLSPolicy can target a named Service port with sectionName and can target a ServiceEntry.
  • Gateway API TLSRoute supports termination, mixed mode, and GA TLS passthrough listeners without the alpha Gateway API flag.
  • Generated Gateway resources inherit labels and annotations unless EnableGatewayAPICopyLabelsAnnotations is disabled.
  • Use ENABLE_GATEWAY_API_MANUAL_DEPLOYMENT=false to stop automatic attachment to existing gateway deployments.
  • Enable experimental Inference Extension with SUPPORT_GATEWAY_API_INFERENCE_EXTENSION; enable agentgateway with PILOT_ENABLE_AGENTGATEWAY.
  • Use PILOT_GATEWAY_TRANSPORT_SOCKET_CONNECT_TIMEOUT; 15s is the default and 0s disables the gateway transport-socket timeout.

Security and policy

  • A failed Wasm fetch with FAIL_CLOSE installs deny-all RBAC. Do not assume fetch failure permits traffic.
  • Ambient port-level STRICT PeerAuthentication rules enforce strict mTLS.
  • JWT rules may omit issuer only when jwksUri is explicit.
  • Enable ambient dry-run authorization only after all connected ztunnels are at least 1.29; older ztunnels can enforce a policy intended for dry-run.
  • Keep debug endpoint authorization enabled where possible. Use DEBUG_ENDPOINT_AUTH_ALLOWED_NAMESPACES for explicit namespace access.
  • For post-quantum policy, set COMPLIANCE_POLICY=PQC in both Pilot and ztunnel in ambient mode.
  • Set ISTIO_WASM_MAX_BINARY_SIZE_BYTES to bound fetched and decompressed Wasm modules.

Telemetry and operations

  • Gateway upstream spans are enabled by default from 1.28. Configure the earlier opt-in with PILOT_SPAWN_UPSTREAM_SPAN_FOR_GATEWAY.
  • Use service.istio.io/workload-name to override telemetry workload names.
  • Set per-pod flush and eviction with sidecar.istio.io/statsFlushInterval and sidecar.istio.io/statsEvictionInterval.
  • Use serviceAttributeEnrichment: OTEL_SEMANTIC_CONVENTIONS for Kubernetes service attributes in OpenTelemetry traces.
  • Expect headers from istioctl ztunnel-config all and istioctl proxy-config all when parsing tabular output.
  • Use root --kubeclient-timeout for Kubernetes-client calls and --proxy-admin-port when proxy administration is not on the default port.