Splunk Enterprise Kubernetes Setup

Prerequisites

Tool or access	Purpose	Verify
Bash and Python 3	Run bundled setup and validation helpers	bash --version && python3 --version
Required product/platform access	Inspect or configure the selected target	Complete the documented preflight
Credential files for live modes	Keep secrets out of chat	Verify paths only

Workflow Overview

+-- Preflight --+ -> +-- Render/review --+ -> +-- Apply/handoff --+ -> +-- Validate evidence --+

When to Activate

• Planning, installing, upgrading, or validating either runtime.
• Preview and review the splunk enterprise kubernetes setup workflow before any live apply phase.
• Diagnose failed prerequisites, generated assets, configuration, or validation evidence.

Scope

Follow the documented read-only or render-first path whenever it is available. This skill does not imply permission to mutate live systems. Require explicit apply flags, protected credentials, and operator review for state changes.

Examples

Inspect the supported setup modes before selecting one:

bash skills/splunk-enterprise-kubernetes-setup/scripts/setup.sh --help

Expected output: usage, supported modes, and required arguments are displayed without changing the target environment.

Inspect validation modes before running completion checks:

bash skills/splunk-enterprise-kubernetes-setup/scripts/validate.sh --help

Expected output: offline, live, and completion options are displayed when the skill supports them; help exits without mutation.

Troubleshooting

Issue	Cause	Resolution
Preflight fails	A required tool or access path is missing	Resolve it before rendering or applying
Rendered assets are incomplete	Required non-secret inputs are absent	Complete intake and render again
Apply is blocked	Review, credentials, or explicit acceptance is missing	Use the documented handoff
Validation is incomplete	Live evidence is unavailable	Record the gap and keep completion open

Use this skill for customer-managed Splunk Enterprise on Kubernetes. Resolve the target before rendering:

1. sok installs the Splunk Operator and Splunk Enterprise Helm charts on an existing Kubernetes cluster.
2. pod drives the coupled Splunk POD installer on pre-built Cisco UCS infrastructure.

Do not map POD sizes to SVA names. POD requires an explicit --pod-profile. Read coverage.md before promising a feature: it distinguishes first-class flags from reviewed overlays, external handoffs, and unsupported operations.

Supported Baselines

• SOK: Operator and charts 3.1.0, Splunk Enterprise 10.4.1 by default. The renderer enforces the current 3.1.0 release matrix. Kubernetes 1.25-1.34 is supported subject to the Splunk version rules in the official release.
• POD: coupled bundle 10.4.0_1.6.0. Do not independently select its SOK, Kubernetes, or Splunk Enterprise versions. The implementation hard-rejects bundles older than 10.2.1_1.5.0 because its native preflight-only command is mandatory; name-based routing and License Manager apps require 10.4.0_1.6.0 or later.
• Python 3.9+ with PyYAML 6.x, Bash, kubectl, and Helm are required for SOK live phases. POD live phases require the executable installer on the bastion.

Version truth comes from the official release pages, not a generic latest Splunk default:

• https://github.com/splunk/splunk-operator/releases/tag/3.1.0
• https://help.splunk.com/en/splunk-enterprise/splunk-pod-guide/10.4/splunk-pod-release-notes

Secret and Safety Rules

Never request or place passwords, HEC tokens, private keys, object-store keys, or license contents in chat, templates, overlays, or command arguments.

• Pass license and private-key paths only. POD private keys must not be readable or writable by group or others; use mode 0600 or stricter.
• Reference existing Kubernetes Secrets and, where the verified path permits it, service accounts by name. The SOK 3.1 separated-ingestion path in this skill is a deliberate exception: it requires Queue Secret authentication and rejects workload-identity service accounts because that upstream path has not been verified.
• Keep overlays non-secret. Use a secret manager, an explicitly supported workload-identity path, or a separately created Kubernetes Secret for credentials.
• Treat get-creds.sh, POD logs, and diagnostic bundles as sensitive output.
• Confirm the target kubecontext, namespace, release names, storage class, object-store path, and upgrade intent before a live phase.
• Never automate POD destruction or SOK downgrade. Those are explicit external handoffs.

The default output directory, ./splunk-enterprise-k8s-rendered/, is ignored by Git. An external output path is the operator's responsibility.

Phase Contract

The default phase is render and does not contact or mutate a cluster.

• render writes a private reviewable bundle and bundle-manifest.json. The manifest records each tracked file's SHA-256 and mode; live helpers require the bundle root and tracked files to remain owned by the current user at the reviewed modes. This detects accidental drift and unsafe replacement, but is not a cryptographic signature or external provenance attestation.
• preflight, apply, and status operate on an existing bundle and reject file drift. They do not rerender it.
• apply runs preflight first, but post-apply readiness is a separate status phase.
• all renders, preflights, applies, and waits for status in one command. Use it only when every input has already been reviewed.
• --apply on a render run applies immediately but does not run the final status phase. Prefer the explicit phases for production.
• SOK overlays are the supported customization mechanism. Editing rendered files invalidates bundle integrity.
• The POD installer may add termsConditionsAccepted: true to cluster-config.yaml; deploy.sh refreshes the bundle hash afterward. Remove that field only from a separate copy made for sharing.

Dry-run plans do not render or execute; use --dry-run --json for a machine-readable preview.

SOK Workflow

Collect the SVA, Kubernetes server version, namespaces/RBAC scope, storage, license ownership, SmartStore authentication, topology zones, sizing, and upgrade intent.

For EKS work in this repository, authenticate to the approved AWS account and role with duo-sso before preflight. Never capture the resulting credentials in the rendered bundle or logs.

Namespace-scoped mode is the default and requires the operator and Enterprise namespace to be the same. For separate namespaces, use a cluster-scoped operator and include the Enterprise namespace in --watch-namespaces.

A fresh SOK bundle creates an absent namespace explicitly. A healthy Active, non-terminating namespace may already exist so reviewed SmartStore/Queue/App Framework Secrets, IRSA ServiceAccounts, image-pull references, or an existing LicenseManager can be staged; the bundle leaves that namespace's metadata unchanged. Exact Helm, Operator, and global CR collision checks prevent this from becoming an existing-deployment adoption path. Use --allow-upgrade only for the exact already-managed Helm releases; its preflight proves release, Operator, CR, and namespace ownership before mutation. Fresh preflight also inventories enterprise.splunk.com CRDs cluster-wide. No CRDs is the normal creation path. If any exist, the complete live CRD set and normalized specs must exactly equal the SHA-verified reviewed 3.1.0 release manifest, every CRD must be established on reviewed stored versions, and no SOK CR may exist anywhere in the cluster, except an explicitly reviewed existing LicenseManager identity. Partial, drifted, extra, unreadable, terminating, or otherwise populated inventories fail before namespace or CRD mutation. validate.sh --live retains its separate existing-deployment readback mode and is not an apply bypass. The LicenseManager exception requires an exact v4 API identity, a separate currently deployed Helm owner release, complete API-server identity, no deletion/pause/admin-managed-PV state, current generation, and a clean Ready status. For the remote development path, apply.sh downloads the pinned CRD manifest once into a private staging directory, verifies its SHA-256, and uses that same staged file for preflight comparison, server-side apply, and the Established wait. Production uses the hash-verified local CRD snapshot in the reviewed bundle. Generated Helm inventory helpers detect Helm 3 versus Helm 4 and fail closed if compatibility discovery or release listing fails.

The chart exposes first-class SVA presets only for S1, C3, and M4. The upstream CRDs can compose C1/C11, C13, M2/M12, M3/M13, and M14, but this skill does not claim those direct-CR topologies as rendered presets. D1/D11 is not recommended by the upstream Applied SVA guidance. Use the explicit handoff in coverage.md instead of relabeling one of the three presets. The verified chart 3.1.0 M4 preset is exactly two sites because it hardcodes multisite RF/SF totals of two. Its SHC and deployer are site-affined and pinned to one selected Kubernetes zone, so this is not a stretched multi-zone search tier and does not establish search continuity through that zone's failure. M4 with 3-63 sites or custom factors, and SHC site0/stretched scheduling, are direct-CR or specifically reviewed-overlay handoffs, not preset overrides.

Render a development C3 bundle:

bash skills/splunk-enterprise-kubernetes-setup/scripts/setup.sh \
  --target sok \
  --architecture c3 \
  --kubernetes-version 1.33 \
  --accept-splunk-general-terms

For separate namespaces, render cluster-scoped RBAC with --operator-scope cluster, distinct operator/Enterprise namespaces, and the Enterprise namespace in --watch-namespaces.

--deployment-profile production is a guardrail, not sizing approval. It requires reviewed local 3.1 charts/CRDs, digest-pinned images, exact cluster identity, Guaranteed-QoS role resources, storage, licensing, complete SmartStore inventory/migration and exclusive-path attestations, and Secret or reviewed AWS IRSA authentication. IRSA requires the service account, exact role ARN, token TTL, AWS region, matching annotations, and regional STS; it is scoped only to S1 Standalone or C3/M4 indexers and is server-dry-run/live validated. Production M4 also locks its two site/zone placements. Follow the complete intake and command in reference.md and template.example; capacity, path ownership across clusters, and zone-failure continuity remain external evidence.

Advanced SOK Options

C3 indexing/ingestion separation and constrained, non-secret overlays are advanced fail-closed paths. Read the complete separated-ingestion contract, overlay rules, and coverage boundaries before selecting either path.

Review, Validate, and Apply SOK

bash skills/splunk-enterprise-kubernetes-setup/scripts/validate.sh \
  --target sok

bash skills/splunk-enterprise-kubernetes-setup/scripts/setup.sh \
  --target sok \
  --phase preflight

bash skills/splunk-enterprise-kubernetes-setup/scripts/setup.sh \
  --target sok \
  --phase apply

bash skills/splunk-enterprise-kubernetes-setup/scripts/setup.sh \
  --target sok \
  --phase status

If the reviewed bundle contains create-license-configmap.sh, the apply phase detects and executes it from the bundle; do not repeat render inputs on an existing-bundle phase. Existing Helm releases fail closed unless the bundle was rendered with --allow-upgrade. Upgrade preflight rejects chart downgrades and immutable Queue/ObjectStorage/reference changes. An upgrade still requires an external backup, release-note review, and recovery plan; SOK does not support downgrade. When the target Splunk release is 10.4 or later, --allow-upgrade also requires --confirm-splunk-10-4-upgrade-readiness. That attests a restorable platform and KV Store backup, a healthy KV Store server at version 7.0 or later, compatible premium apps/add-ons, TLS 1.2 or later, and review of the target release's upgrade notes. The confirmation flag is invalid without --allow-upgrade.

server-dry-run.sh accepts operator, enterprise, or all. Upgrade apply uses the least-mutating valid order: Operator server dry-run before CRD changes, then reviewed CRD apply, then Enterprise server dry-run against the resulting schemas, before either Helm release is mutated. A new install creates the reviewed namespaces/CRDs first because the Enterprise resources need their APIs, then runs the combined server dry-run before Helm install.

Splunk POD Workflow

POD assumes Cisco CVD infrastructure already exists. The installer preflight does not replace Cisco Intersight, network, RAID, RHEL, DNS, certificate, or capacity ownership. Review coverage.md and the current CVD before live work.

Accepted selectors are:

• pod-small, pod-medium, pod-large, pod-xlarge
• the same names with -es or -itsi

The suffix is a skill selector. cluster-config.yaml always contains the official base profile. Base worker counts are 8, 11, 15, and 30; an ES or ITSI secondary search tier makes them 9, 14, 18, and 33. Every profile has exactly three controllers.

Do not infer capacity or high availability from those counts alone. The POD SOK workloads allocate/limit each indexer pod at 36 CPU/96 GB and each search-head pod at 24 CPU/96 GB; the infrastructure must provide those allocatable resources plus system and failure headroom. The design also requires profile-specific disk/RAID layouts, three SeaweedFS managers, at least three filers, one volume process per volume worker, and three-way volume replication (two-volume-node failure tolerance). Small, Medium, and Large use a 90-day local cache with one-year SmartStore retention; X-Large uses a 60-day cache with 180-day SmartStore retention. X-Large spans two racks/four switches and requires complete cross-rack reachability. The renderer checks node counts and addresses, not rack placement, CPU/RAM, RAID, SeaweedFS process placement, retention behavior, or failure tolerance; retain Cisco/Splunk infrastructure evidence for each. There is a product-documentation conflict for custom indexes: the Manage guide says POD injects frozenTimePeriodInSecs=31536000 (one year), while the architecture gives X-Large 180-day retention. Do not assume which wins for an X-Large custom index; require post-deploy readback and Splunk guidance.

The POD schema supports up to two standalone search heads on Small and two SHCs on Medium, Large, and X-Large. This skill renders a second tier only for the first-class -es and -itsi selectors. A second generic search tier, or generic apps assigned specifically to that secondary tier, requires a reviewed manual configuration and Cisco/Splunk vendor handoff; do not relabel it as ES or ITSI. The selectors always retain a primary core search tier. A sole ES/ITSI premium search tier without that primary tier is not represented and is a manual/vendor handoff.

A bundle intended for live use requires exact, unique controller/worker IPs, the executable installer, license and SSH-key files, and an explicit immutable primary search-tier name. ES and ITSI profiles also require an immutable secondary name. Existing-bundle phases reuse these reviewed values and reject bundle or external-file drift rather than accepting render inputs again. An automated first deployment additionally requires --confirm-new-pod-install, a one-time reviewed attestation that the exact configuration and node set do not contain an existing POD deployment.

bash skills/splunk-enterprise-kubernetes-setup/scripts/setup.sh \
  --target pod \
  --pod-profile pod-medium \
  --phase all \
  --pod-version 10.4.0_1.6.0 \
  --confirm-new-pod-install \
  --installer-path /opt/splunk-pod/kubernetes-installer-standalone \
  --installer-sha256 '<independently-reviewed-64-hex-sha256>' \
  --controller-ips "${CONTROLLER_IPS}" \
  --worker-ips "${WORKER_IPS}" \
  --primary-search-name core-shc \
  --license-file /secure/path/splunk.lic \
  --ssh-private-key-file /secure/path/pod-ssh-key

POD App Scopes

• --indexer-apps: clustermanager.apps.cluster
• --cluster-manager-apps: clustermanager.apps.local
• --search-apps: primary SHC apps.cluster
• --search-deployer-apps: primary SHC deployer apps.local
• --standalone-apps: primary Small standalone apps.local
• --premium-apps: secondary ES tier apps.premium; ES only
• --itsi-apps: secondary ITSI standalone local scope or SHC cluster scope
• --license-manager-apps: licensemanager.apps.local

Strict validation inspects every supplied app archive by internal top-level app directory, not by filename. Unsafe tar paths and member types, multiple roots, duplicate internal names, and mismatched optional [package] id values fail. The skill does not provide a separate generic-secondary-tier app list.

Enterprise Security on POD

ES requires SplunkEnterpriseSecuritySuite in --premium-apps and the matching Splunk_TA_ForIndexers in --indexer-apps. Their internal versions must match and must be one of the current Splunk Enterprise 10.4-compatible releases: 8.3.0, 8.4.1, or 8.5.1. The 8.1.1 package shown in an older POD YAML example is not in the current compatibility matrix and is intentionally rejected. A live ES profile also requires two physically distinct .lic files for Enterprise and ES; duplicate paths or hard links to the same inode fail.

Source: https://help.splunk.com/en/splunk-enterprise/release-notes-and-updates/compatibility-matrix/splunk-products-version-compatibility/splunk-products-version-compatibility-matrix

ITSI on POD

POD 10.4 requires ITSI 4.21.2, distinct Enterprise/ITSI licenses, the original bundle plus reviewed SHA-256, the reviewed OpenJDK 17 x86-64 app, and exact search/indexer/License Manager package placement. Validation checks internal app roots and canonical content against the source bundle, not filenames; SA-ITSI-Licensechecker belongs on the License Manager and is forbidden on the search tier. The target-specific inventory deliberately resolves omissions in the generic package page. Use the complete list and input contract in reference.md.

POD Name-Based Routing and Certificates

IP-based routing is the default and needs no domain or custom ingress material. Name-based routes can use POD's product certificate without a customer certificate; pass --ingress-domain to record the reviewed DNS suffix and expect browser trust warnings unless that product certificate is trusted. POD's default Splunk UI ingress certificates are Cert Manager certificates that automatically renew every 90 days.

A customer-provided wildcard certificate is optional and is the path for trusted name-based TLS. When used, all four inputs are required together: --ingress-domain, --ingress-certificate-file, --ingress-private-key-file, and --ingress-ca-file. The customer owns DNS, certificate issuance, trust distribution, renewal, and the official certificate-redeployment handoff. Strict validation requires:

• a PEM chain and a non-interactively readable matching PEM private key
• explicit CA-bundle path validation with server purpose and hostname checks
• a wildcard DNS SAN, CA:FALSE, serverAuth, and TLS-compatible key usage
• RSA 2048-bit or stronger, or prime256v1, secp384r1, or secp521r1 EC
• no MD5/SHA-1 certificate signatures and at least 30 days validity remaining
• private-key permissions of 0600 or stricter

Configure forward and reverse DNS for POD servers and a wildcard A record containing every worker IP. The installer flags expired custom certificates and certificates within 30 days of expiry. Name-based routing and this certificate stanza require POD 10.4.0_1.6.0 or later.

Sources:

• https://help.splunk.com/en/splunk-enterprise/splunk-pod-guide/10.4/network-routing-and-ingress-for-splunk-pod
• https://help.splunk.com/en/splunk-enterprise/splunk-pod-guide/10.4/splunk-pod-requirements
• https://help.splunk.com/en/splunk-enterprise/splunk-pod-guide/10.4/splunk-pod-release-notes

POD First Deploy and Day-2 Handoffs

The official docs define installer -status as listing pods and -status.workers as listing workers. They do not define exit codes or output for a cluster that has never been deployed. The generated fresh-deploy helper therefore requires the reviewed --confirm-new-pod-install attestation, blocks when -status succeeds, blocks every unrecognized status failure, and proceeds only when the attestation is present with a narrow recognized no-cluster message. The attestation is mutually exclusive with --allow-upgrade; it does not authorize reuse against a partial or existing deployment. Do not broaden the text match to treat arbitrary SSH, configuration, or installer errors as proof that no cluster exists.

Automated POD upgrade and day-2 app reconciliation are intentionally disabled. Rendering with --allow-upgrade records reviewed intent and permits preflight evidence, but deploy.sh fails closed instead of invoking an upgrade. Follow the official lockstep upgrade or certificate-redeployment runbook with backup, release-note review, the exact coupled installer, and Cisco/Splunk vendor review. App removal, downgrade, mutation of an existing cluster, and POD destruction are manual/vendor handoffs and are not exposed by this skill.

Universal Forwarder rollout is also a separate handoff. For UF 9.4.1 or later, the POD multi-worker receiver path requires forcedTimeBasedAutoLB=true; configure and validate it through splunk-universal-forwarder-setup or splunk-agent-management-setup rather than treating POD installation as forwarder onboarding.

Sources:

• https://help.splunk.com/en/splunk-enterprise/splunk-pod-guide/10.4/deploy-splunk-pod
• https://help.splunk.com/en/splunk-enterprise/splunk-pod-guide/10.4/troubleshoot-splunk-pod

Validation

Static validation checks bundle integrity and shell helpers. SOK additionally pulls, lints, templates, and semantically checks both pinned charts when Helm is available; --strict requires Helm. POD checks profile, node counts, addresses, immutable names, and local references. Strict POD validation also runs the archive, ES/ITSI provenance, and ingress cryptographic gates described above.

bash skills/splunk-enterprise-kubernetes-setup/scripts/validate.sh \
  --target pod \
  --strict

bash skills/splunk-enterprise-kubernetes-setup/scripts/validate.sh \
  --target sok \
  --live \
  --json

Live SOK validation runs compatibility/RBAC/reference preflight and waits on Splunk custom-resource phases. It also exactly compares the Operator's Helm-owned Deployment, RBAC, ServiceAccount, staging PVC, and Services with a fresh server dry-run; validates the expected Splunk child Service inventory and ready EndpointSlices; and checks StatefulSet/pod placement. Production requires replicas of each replicated StatefulSet on distinct nodes, while M4 additionally checks every reviewed pod against its node's expected zone. The preset expects the SHC/deployer in one zone, so this does not prove search continuity after a zone failure. Live POD validation reruns native preflight and then uses the generated bounded readiness loop over both worker and pod status. Validation does not prove application-level search, ingest, premium-app setup, backup quality, DNS propagation, client trust, or external cloud/Cisco configuration; record those as separate evidence.