Local Redis Enterprise TLS environment validation

Date: 2026-11-01 (America/Los_Angeles). Local execution, not hosted CI.

Provisioned with python3 scripts/enterprise-env.py up, using the separately qualified public/private source pair under ~/sandbox/work. The builder's full module fingerprint check remained enabled. This does not qualify this worktree's newer public Go changes. The Linux ARM64 enterprise executable runs in Docker.

Environment: Redis Software image redislabs/redis:8.2.0-78.18, database Redis 8.6.2 with Search, one cluster node, three active primary shards, no replicas, RAM storage, 1 GiB configured database memory, active 4x query performance factor. This is functional coverage, not failover, multi-node or performance evidence. All published ports are bound to loopback. Existing containers were left intact.

Passed

  • Provisioning and repeated up against the existing environment.
  • verify: trusted CA and hostname validation, AUTH with the dedicated flight_test identity, ACL WHOAMI, PING, Search index creation, hash write, indexed readback and cleanup.
  • Rejected unauthenticated TLS, wrong password, untrusted CA and plaintext Redis.
  • Enterprise Flight backend health: SERVING.
  • Focused qualified enterprise race suite: TestBuildIntegrationRoundTrip, TestIntegrationSQL, TestIntegrationWritesRunOnce, TestIntegrationIngest.
  • smoke: SQL CREATE, INSERT, UPDATE and exact readback through the running Flight endpoint, followed by fixture cleanup.
  • Python syntax compilation, Go vet for the smoke client, Go formatting and git diff --check.

Initial full-suite failures (resolved by the follow-up below)

python3 scripts/enterprise-env.py test ran all packages with -race and failed. No test assertions were removed or converted to skips:

  • TestIntegrationCancellationIdentity, TestIntegrationCancellationReadOnlyFallback, TestIntegrationStaleCredentials, TestIntegrationRedisACL and TestIntegrationSessionOwner attempt ACL SETUSER. Redis Software rejects that command. Their fixture setup needs the Software management API with equivalent role/ACL assertions. Redis Software also does not implement ACL DELUSER.
  • TestIntegrationSIGTERM and subtests of TestIntegrationStreamResults failed cursor/streaming expectations. Examples: no cursor after the first batch and 285 cursors observed where a single-shard expectation was 96. These failures require investigation of the streaming path and shard-aware assertions; this record does not establish their root cause or treat them as passing.

The environment is usable for reproducing these failures. A working environment, healthy Flight endpoint and focused successes do not constitute a green full integration suite or release qualification.

Reproduce

python3 scripts/enterprise-env.py verify
python3 scripts/enterprise-env.py smoke
python3 scripts/enterprise-env.py test

See setup and lifecycle. Passwords, private keys and authenticated URIs are confined to ignored local state and are omitted from this evidence. .local/enterprise/verification.json records the latest nonsecret environment checks; build.json records the composition manifest hash.

Full-suite repair and release gate

The follow-up private composition was built in an isolated worktree, preserving unrelated local private changes. It qualifies the public PR source, rather than using the older source pair described above. The complete invocation python3 scripts/enterprise-env.py test passed with go test -count=1 -p 1 -race ./...: command, integration helper and server packages passed; the smoke client package compiled with no unit tests.

ACL fixtures now use the hostname-verified Software management API with a separate admin identity. They create database-only roles, wait for observable credential/permission propagation, and clean up users, roles and ACLs. The stale token test retains Unauthenticated/PermissionDenied/Unauthenticated expectations; it denies an actual SQL metadata read because Software keeps PING available.

Streaming tests assert active Flight stream gauges and keep exact row, pool, cancellation, fallback-budget and cleanup checks. The current driver's bounded row-ID scans can exhaust their Search cursor within a range while continuing as a live stream, so persistent shard-cursor presence is not a valid streaming oracle. Wide results prevent the complete result fitting in gRPC buffers. External cursor-pressure tests still verify real Search cursor counts, and a new reservation regression checks per-shard cursor costs. Software command permission errors map to PermissionDenied, with a public regression check.

Public releases now require a fresh successful private qualification run and receipt matching the exact public commit, private commit, run ID and manifest hash. Binary and image publication depend on that gate and use its resolved immutable source SHA. Seven release-gate tests and actionlint passed.

ENTERPRISE_CI_TOKEN was created with user approval and saved in the public repository's encrypted Actions secrets. It is restricted to the private Enterprise repository with Actions read/write and Contents read, and expires 2026-12-01. GitHub reports organization approval pending. No credential value is included here. Private workflow execution from the release gate remains blocked until that approval and the coordinated workflow changes are merged; local full-suite success is not a claimed hosted CI success.