feat: add Connect/Workbench PPM authenticated repos via Identity Federation

[ian-flores]

Description

Enable Connect and Workbench to authenticate against PPM using Kubernetes Identity Federation (RFC 8693 token exchange). When authenticatedRepos: true is set on a product spec, the operator automatically injects init container + sidecar pods that exchange K8s projected service account tokens for PPM API tokens, writing netrc/curlrc files for Python and R package installation.

Issue

• PTDC-223: Configure Connect to automatically use PPM authenticated repos
• PTDC-222: Configure Workbench to automatically use PPM authenticated repos

Dependencies

• Depends on feat: add PPM OIDC/SSO web UI configuration #112 (PPM OIDC/SSO web UI configuration) merging first. Once merged, this PR should be rebased onto main to show only the auth repos diff.

Code Flow

Authentication Flow

K8s Projected SA Token → Init Container → PPM /__api__/token (RFC 8693) → PPM API Token
                                                                                │
                                                                        Written to netrc + curlrc
                                                                                │
                         Connect/Workbench ← reads netrc/curlrc ←───────────────┘
                         (R: CURL_HOME, Python: NETRC env var)

Sidecar refreshes the token on a timer (default 50min for 60min token lifetime).

Key Components

1. PPM Identity Federation Config (package_manager_config.go): PackageManagerIdentityFederationConfig struct with full field coverage. Extended GenerateGcfg() to emit named [IdentityFederation "connect"] sections.

2. Token Exchange Helpers (ppm_auth.go): Shared functions for init container, sidecar container, volumes, volume mounts, and env vars. Shell script (mounted from ConfigMap) uses wget + sed (BusyBox built-ins in alpine:3) for zero-dependency operation.

3. Product Integration (connect.go, workbench.go): When AuthenticatedRepos=true, adds projected SA token volume, shared emptyDir, script ConfigMap volume, init container, sidecar container, NETRC + CURL_HOME env vars.

4. Site Controller (site_controller.go): Creates {site}-ppm-auth-script ConfigMap when any product has authenticated repos enabled. Auto-configures Identity Federation entries on PPM based on product flags.

Testing

Tested end-to-end on ganso01-staging with adhoc image adhoc-ppm-auth-repos-v1.16.1-15-gedc6aaa.

Test	Status
Connect pods 3/3 Running (connect + chronicle + ppm-auth-sidecar)	✅
Workbench pods 3/3 Running (workbench + chronicle + ppm-auth-sidecar)	✅
Connect init container exchanges SA token for PPM API token	✅
Workbench init container exchanges SA token for PPM API token	✅
/mnt/ppm-auth/netrc contains valid JWT with correct subject claim	✅
/mnt/ppm-auth/.curlrc contains --netrc-file directive	✅
NETRC and CURL_HOME env vars set on main product containers	✅
PPM auth_available: true and sso_enabled: true after enabling repo auth	✅

Category of change

• New feature (non-breaking change which adds functionality)

[claude]

Claude finished @ian-flores's task —— View job

---

Review: PPM Authenticated Repos via Identity Federation

Well-structured feature with good separation of concerns. The auth flow (projected SA token → init container → PPM token exchange → netrc/curlrc) is sound. Here are findings organized by severity.

---

Security (Elevated Scrutiny)

1. SA token visible in process args — ppm_auth.go:53
The shell script passes the SA token via wget --post-data="...subject_token=${SA_TOKEN}...", which means the full token is visible in /proc/<pid>/cmdline. Consider writing the post data to a temp file and using --post-file instead.

2. Error logging leaks response data — ppm_auth.go:60
On failure, the script logs "Response was: $RESPONSE" to stderr. If PPM returns error details containing internal information, this ends up in container logs. Consider truncating or omitting the raw response.

3. extract_json_field sed regex is fragile — ppm_auth.go:45
The sed-based JSON parser won't handle escaped quotes in values. While PPM access tokens (JWTs) won't contain literal quotes, this is a latent risk if the pattern is reused for other fields.

---

Validation Gaps

4. Missing validation: AuthenticatedRepos=true without PPMUrl — connect.go:595, workbench.go:754
When AuthenticatedRepos is true but PPMUrl is empty, SanitizePPMUrl("") returns "", and the init container will wget to /__api__/token — failing cryptically. Add a validation guard here or at the API level.

5. OIDCIssuerURL empty with AuthenticatedRepos=true — site_controller_package_manager.go:139
If AuthenticatedRepos is enabled on Connect/Workbench but OIDCIssuerURL is empty, no Identity Federation entries are created on PPM, yet product pods still have the auth sidecar trying to exchange tokens. This will fail at runtime with no operator-level warning. Consider logging a warning or adding validation.

6. EKS-specific default audience — site_controller_package_manager.go:142
Defaulting audience to "sts.amazonaws.com" is EKS-specific. AKS users would get silent misconfiguration. Consider making both OIDCIssuerURL and OIDCAudience required when authenticated repos are enabled, or at minimum logging when the default is used.

---

Code Quality

7. Unused parameter ppmURL — ppm_auth.go:175
PPMAuthVolumes(siteName, ppmURL, audience string) accepts ppmURL but never uses it in the function body. Remove to avoid confusion.

8. Manual field serialization for IdentityFederation — package_manager_config.go:118-170
The field-by-field serialization is brittle — if a new field is added to PackageManagerIdentityFederationConfig, this code must be updated in lockstep. The main loop uses reflection for other sections. Could the same approach work here, with just the section header [IdentityFederation "name"] being special-cased?

9. Inconsistent JSON tag casing — package_manager_config.go:20
IdentityFederation uses lowercase json:"identityFederation" while all other fields in PackageManagerConfig use capitalized tags matching gcfg section names ("Server", "Http", etc.). Was this intentional?

10. Duplicated PPM auth setup logic — connect.go:589-606 and workbench.go:748-765
The PPM auth setup blocks in Connect and Workbench reconcilers are nearly identical. Consider extracting to a shared helper that returns all five slices, reducing duplication.

---

API Design

11. PPMUrl and PPMAuthAudience on product specs — connect_types.go:158, workbench_types.go:120
These fields exist on both ConnectSpec and WorkbenchSpec, but when used via the Site controller, they're derived from site-level config. These are only meaningful for standalone CRs. Consider documenting this distinction or adding validation.

---

Minor

12. Cleanup doesn't check ownership — site_controller.go:522
cleanupPPMAuthConfigMap deletes the ConfigMap without checking the ManagedByLabelKey label. If a user created a ConfigMap with the same name, it would be deleted.

---

What looks good

• Non-root security context on init/sidecar containers (RunAsUser: 65534, AllowPrivilegeEscalation: false)
• Atomic netrc writes via temp file + rename
• Resource limits on sidecar container
• Sidecar resilience (token refresh failures don't crash the loop)
• Good test coverage for the helper functions
• IdentityFederation name validation against injection chars (", ], newlines)
• CRDs and Helm chart are in sync

[claude]

Review of PPM authenticated repos feature — see the main Claude comment for detailed findings.