chore(deps): update registry.gitlab.com/gitlab-com/gl-infra/common-ci-tasks-images/golang-fips docker tag to v1.26.2

fix(deps): update module gitlab.com/gitlab-com/gl-infra/platform/runway/fairway to v1.15.2

See merge request !475

Merged-by: Matias Alvarez <malvarez@gitlab.com>
Approved-by: Matias Alvarez <malvarez@gitlab.com>
Co-authored-by: GitLab Renovate Bot <ops-contact+gl-common-ci-tasks-self-renovate@gitlab.com>

Restructures the Health endpoints section into Liveness, Readiness,
and a new Writing safe readiness checks block.

The new section explains the readiness semantic (can-I-serve-traffic,
not are-my-deps-healthy), what to include and exclude, the cascading
failure pattern, and Kubernetes probe configuration guidance.

Adds a ### Startup section explaining when to use a startup probe,
why /-/readiness is the right endpoint to reuse, how to size
failureThreshold for worst-case startup time, and when a custom
startup endpoint makes sense.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Health endpoints (/-/liveness, /-/readiness) and /-/metrics are now
served exclusively on Config.ProbeAddr (default :9090). The main port
serves application traffic only.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Go 1.22+ loop variables are scoped per iteration; the name := name copy
is unnecessary and flagged by golangci-lint.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Tests that call Start without ProbeAddr defaulted to :9090, causing
port conflicts when tests ran concurrently.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Reflow long single-line paragraphs to one sentence per line, apply
M_Alvarez's wording suggestions for the startup probe section, improve
active voice on the readiness what-to-check bullet, align YAML comment
spacing, and remove the unrelated docs/ gitignore entry.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

TestLiveness_IncludesBuildInfo, TestLiveness_OmitsBuildInfoWhenEmpty,
and TestReadiness_IncludesBuildInfo were written against srv.ServeHTTP
which no longer routes /-/liveness or /-/readiness now that those
endpoints moved to the dedicated probe server. Rewrite them to use
startProbeServer and hit the probe address via HTTP, consistent with
every other health endpoint test in the file.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

We're currently in a position where we incorrectly have `v2/X.X.X` tags.
Our major version subdirectory set up requries us to use tags in the form
`v2.X.X`. These tags previously existed and were deleted.

While recreating these old tags, deleting incorrect formats, and fixing
the release script; we should temporarily disable the release process by
calling the script in dry-run mode.

This release script was built out with the incorrect assumption that v2
module tags should be in the form `v2/v2.X.X`.

This is incorrect and we should instead be using the format `v2.X.X`

ci: set release jobs to dry run while fixing release tags

See merge request !477

Merged-by: Matias Alvarez <malvarez@gitlab.com>
Approved-by: Matias Alvarez <malvarez@gitlab.com>
Co-authored-by: Luke Hollinda <lhollinda@gitlab.com>

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Co-authored-by: Matias Alvarez <malvarez@gitlab.com>

ci: release script to use correct v2 tag format

See merge request !478

Merged-by: Luke Hollinda <lhollinda@gitlab.com>
Approved-by: Matias Alvarez <malvarez@gitlab.com>

Adjusts release script comments to account for recent changes.
- V2 module tags in format v2.X.X
- Patch releases are only created for `fix` commits

Updates docs to align with our removal of Semantic Release

Add a new `infrastructure` package that centralises loading of the
infrastructure config and instantiation of the secrets provider, replacing the
per-package loading logic that previously lived in `postgres`.

`infrastructure.DefaultConfig()` lazily loads `/etc/labkit/infrastructure.yaml`
and a file-backed secret provider rooted at `/secrets/infrastructure`, caching
the result for subsequent callers.

Refactor the PostgreSQL client to consume `*infrastructure.Config` via a new
`WithInfrastructure` option, which is primarily useful for testing. When no
config is provided, `postgres.New(ctx)` falls back to `DefaultConfig()`, so the
developer-friendly zero-argument API is preserved and LabKit will do the right
thing.

As part of the refactor:

- Replace `ErrUndefinedDependency` with the more specific `ErrMissingConfig`
  and `ErrMissingSecrets`.
- Move secret lookups from a postgres-specific mount (`/secrets/infrastructure/postgresql`)
  to keyed lookups (`postgresql/username`, `postgresql/password`) against the
  shared provider rooted at `/secrets/infrastructure`.
- Rewrite `postgres` tests to construct `*infrastructure.Config` directly with
  a `secret.MemoryProvider`, avoiding filesystem fixtures and adding coverage
  for missing-secret error paths.

Issue: gitlab-com/gl-infra/platform/runway/team#884

Infrastructure clients should return this error if the client was not
configured, giving callers a consistent way to check for this condition.

chore: correct release script misnomers

See merge request !482

Merged-by: Matias Alvarez <malvarez@gitlab.com>
Approved-by: Matias Alvarez <malvarez@gitlab.com>
Co-authored-by: Luke Hollinda <lhollinda@gitlab.com>

ci: remove dry-run, enabling release CI jobs

See merge request !483

Merged-by: Matias Alvarez <malvarez@gitlab.com>
Approved-by: Matias Alvarez <malvarez@gitlab.com>
Co-authored-by: Luke Hollinda <lhollinda@gitlab.com>

chore(deps): update pre-commit hook gitlab-com/gl-infra/common-ci-tasks to v4.1

See merge request !481

Merged-by: Bob Van Landuyt <bob@gitlab.com>
Approved-by: Bob Van Landuyt <bob@gitlab.com>
Co-authored-by: GitLab Renovate Bot <ops-contact+gl-common-ci-tasks-self-renovate@gitlab.com>

fix(deps): update module gitlab.com/gitlab-com/gl-infra/platform/runway/fairway to v1.15.3

See merge request !480

Merged-by: Elliot Forbes <eforbes@gitlab.com>
Approved-by: Elliot Forbes <eforbes@gitlab.com>
Co-authored-by: GitLab Renovate Bot <ops-contact+gl-common-ci-tasks-self-renovate@gitlab.com>

Merge branch 'renovate/registry.gitlab.com-gitlab-com-gl-infra-common-ci-tasks-images-golang-fips-1.x' into 'master'

chore(deps): update registry.gitlab.com/gitlab-com/gl-infra/common-ci-tasks-images/golang-fips docker tag to v1.26.2

See merge request !451

Merged-by: Elliot Forbes <eforbes@gitlab.com>
Approved-by: Elliot Forbes <eforbes@gitlab.com>
Co-authored-by: GitLab Renovate Bot <ops-contact+gl-common-ci-tasks-self-renovate@gitlab.com>

docs(v2/httpserver): add readiness check best practices

Closes gitlab-org/quality/quality-engineering/team-tasks#4381

See merge request !402

Merged-by: Elliot Forbes <eforbes@gitlab.com>
Approved-by: Matias Alvarez <malvarez@gitlab.com>
Reviewed-by: GitLab Duo <gitlab-duo@gitlab.com>
Reviewed-by: Matias Alvarez <malvarez@gitlab.com>

Add httptest slow-server timeout test to featureflag client

See merge request !485

Merged-by: Luke Hollinda <lhollinda@gitlab.com>
Approved-by: Luke Hollinda <lhollinda@gitlab.com>
Reviewed-by: GitLab Duo <gitlab-duo@gitlab.com>
Co-authored-by: Elliot Forbes <eforbes@gitlab.com>

chore(deps): update registry.gitlab.com/gitlab-com/gl-infra/common-ci-tasks-images/golang-fips docker tag to v1.26.3

This commit adds the string equivilant of
TestBooleanValueDetails_DefaultTimeoutReturnsDefault, introduced in:
!485

Refactors the slowServerForTimeout function to accept an optional
timeout duration. When no timeout is configured, this timer channel
blocks indefinitely.

This allows us to use the helper function for all tests requiring a
"slow" server.

Merge branch 'renovate/registry.gitlab.com-gitlab-com-gl-infra-common-ci-tasks-images-golang-fips-1.x' into 'master'

chore(deps): update registry.gitlab.com/gitlab-com/gl-infra/common-ci-tasks-images/golang-fips docker tag to v1.26.3

See merge request !490

Merged-by: Bob Van Landuyt <bob@gitlab.com>
Approved-by: Bob Van Landuyt <bob@gitlab.com>
Co-authored-by: GitLab Renovate Bot <ops-contact+gl-common-ci-tasks-self-renovate@gitlab.com>

test(v2/featureflag): test improvements following !485

See merge request !493

Merged-by: Elliot Forbes <eforbes@gitlab.com>
Approved-by: Elliot Forbes <eforbes@gitlab.com>
Co-authored-by: Luke Hollinda <lhollinda@gitlab.com>

- Use log.Error(err) instead of slog.Any("error_message", err) to
  align with the standardised fields.ErrorMessage field from LabKit
- Replace bytes.Buffer/slog.NewTextHandler with logtest.NewWithRecorder
  in logging tests for structured record assertions

Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>

chore(deps): update pre-commit hook gitlab-com/gl-infra/common-ci-tasks to v4.2

See merge request !491

Merged-by: Steve Xuereb <sxuereb@gitlab.com>
Approved-by: Steve Xuereb <sxuereb@gitlab.com>
Co-authored-by: GitLab Renovate Bot <ops-contact+gl-common-ci-tasks-self-renovate@gitlab.com>

fix(deps): update module gitlab.com/gitlab-com/gl-infra/platform/runway/fairway to v1.16.0

See merge request !492

Merged-by: Elliot Forbes <eforbes@gitlab.com>
Approved-by: Elliot Forbes <eforbes@gitlab.com>
Co-authored-by: GitLab Renovate Bot <ops-contact+gl-common-ci-tasks-self-renovate@gitlab.com>

chore(deps): update dependency https://gitlab.com/gitlab-com/gl-infra/common-template-copier.git to v1.49.1

Co-authored-by: Duo Fix CI/CD Pipeline <service_account_group_9970_fe9437f62858fed38f53a853f2d1d219@noreply.gitlab.com>

chore(deps): update dependency https://gitlab.com/gitlab-com/gl-infra/common-template-copier.git to v1.49.1

See merge request !288

Merged-by: Bob Van Landuyt <bob@gitlab.com>
Approved-by: Bob Van Landuyt <bob@gitlab.com>
Co-authored-by: GitLab Renovate Bot <ops-contact+gl-common-ci-tasks-self-renovate@gitlab.com>

Add slog.Logger support for feature flag eval failure observability

See merge request !486

Merged-by: Matias Alvarez <malvarez@gitlab.com>
Approved-by: Matias Alvarez <malvarez@gitlab.com>
Co-authored-by: Elliot Forbes <eforbes@gitlab.com>

Document error propagation behaviour and update Config godoc

See merge request !487

Merged-by: Matias Alvarez <malvarez@gitlab.com>
Approved-by: Luke Hollinda <lhollinda@gitlab.com>
Approved-by: Matias Alvarez <malvarez@gitlab.com>
Co-authored-by: Elliot Forbes <eforbes@gitlab.com>

Adds two integration tests using a hung httptest.Server to verify that
evaluation completes within the configured timeout, returns the default
flag value, propagates an observable error, and records the failure on
the OTel span. Closes criterion 5 of team-tasks#4382.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

test(v2/featureflag): add integration tests for timeout fallback behaviour

See merge request !495

Merged-by: Matias Alvarez <malvarez@gitlab.com>
Approved-by: Matias Alvarez <malvarez@gitlab.com>
Reviewed-by: GitLab Duo <gitlab-duo@gitlab.com>
Co-authored-by: e_forbes <eforbes@gitlab.com>

Heavily diven by Claude code. To be cut down and refactored in following
commits.

Aligns with the postgres package on how we handle non-valid server
address.

Removes the New constructor; which we can reintroduce if and when we
need to set up automatic client construction from Fairway / known
config locations

Initial registration and collection of pool metrics. Will require
refactoring to more cleanly register these metrics along with attaching the
hook required for the command metrics.

All metrics now use prometheus.BuildFQName with configurable Namespace
(default "gitlab") and Subsystem (default "redis"), producing names like
gitlab_redis_command_duration_seconds. A "name" constant label is added to
all metrics to identify the client instance.

Replaces registerOrReuse with registerCollector, which errors on duplicate
descriptors rather than silently reusing an existing collector. The previous
approach dropped pool stats and merged command metrics across distinct clients.

Merges newMetricsHook and registerPoolMetrics into a single newMetrics entry
point.

Unexports StatGetter and NewPoolCollector as internal implementation details.

- Updates metric name strings (redis_command_* → gitlab_redis_command_*)
- Splits TestMetrics_MultipleClientsSharedRegisterer into two tests:
  - TestMetrics_DistinctNamesSharedRegisterer — distinct names share a registerer without error
  - TestMetrics_DuplicateNameSharedRegistererErrors — same name returns an error with the expected message

Adds 14 fields covering the connection, timeout, pool, and retry parameters
that teams commonly need to tune. ClientName is auto-set to cfg.Name so every
client is identifiable in CLIENT LIST output without additional configuration.

feat(infrastructure): Introduce shared infrastructure config package.

See merge request !467

Merged-by: Elliot Forbes <eforbes@gitlab.com>
Approved-by: Elliot Forbes <eforbes@gitlab.com>
Reviewed-by: GitLab Duo <gitlab-duo@gitlab.com>
Co-authored-by: Florian Forster <fforster@gitlab.com>

Rework the DSN construction in `clientOptions.newFromProto` to support
secrets produced by a variety of PostgreSQL operators. Presets are
shipped for CNPG, Crunchy Data, and Zalando; users can also supply a
custom field projection via `PostgreSQLSecret.Projection`.

Each DSN field is resolved either from inline config (`Host`, `Port`,
`DatabaseName`, `Username`) or from a secret field looked up via the
projection. `Password` is always resolved from a secret. Errors across
all fields are joined so callers see every misconfiguration at once,
not just the first.

As defense in depth, `resolveSecret` rejects secret references
containing a slash to prevent path traversal into other secret
namespaces. The schema already constrains secret names to valid
Kubernetes resource names.

Depends on the matching Fairway schema change in
gitlab-com/gl-infra/platform/runway/fairway!63, which introduces
`PostgreSQLSecret`, `PostgreSQLFormat`, and `PostgreSQLProjection` and
turns `PostgreSQL.Host` into a `*string`. The `go.mod` entry points at
a pre-release commit until that MR is merged and tagged.

Issue: gitlab-com/gl-infra/platform/runway/team#873

Adds a Logger *slog.Logger field to Config. When set, a loggingHook is
registered that emits a structured log entry for every command and
pipeline: redis_command, redis_key, redis_duration_ms, and error_message on
failures.

redis.Nil is not treated as an error. Pipeline executions are logged as a single
entry with redis_command="pipeline" and redis_pipeline_size.

Errors trigger Warn level; normal completions use Debug. Start and
Shutdown log at Info.

The client name is baked into a child logger via .With() at construction
time so all entries carry it automatically without repeating it at each
call site.

When Logger is nil, the logging hook is not registered (no per-command
overhead). A discard logger is stored on the client so Start and
Shutdown can log unconditionally without nil guards.

This avoids logging `redis_key=""` for pipelines and other commands which
do not have a redis_key.

db.system    -> db.system.name
db.statement -> db.operation.name

From: https://opentelemetry.io/docs/specs/semconv/db/redis/

From https://opentelemetry.io/docs/specs/semconv/db/redis/:

> For transactions and pipelined calls, if the individual operations are known
  to have the same command then that command SHOULD be used prepended by MULTI
  or PIPELINE . Otherwise db.operation.name SHOULD be MULTI or PIPELINE.

From https://opentelemetry.io/docs/specs/semconv/db/redis/:

> db.operation.batch.size: Operations are only considered batches when they
  contain two or more operations, and so db.operation.batch.size SHOULD never
  be 1.

This commit splits the logic for setting up and registering the pool
metrics into their own file. This better shows that the pool metrics are
pull metrics which do not depend on the metrics_hook, which is entirely
responsible for the per-command metrics.

Small corrections to documentation and comments

Adds three integration jobs covering supported Redis topologies:
- integration:redis (single-node)
- integration:redis:cluster
- integration:redis:sentinel

Topology setup is handled by small bash scripts under
v2/redis/testdata/ci/.

I would have prefered to re-use out docker-compose files, which can be
used to run these tests locally. However, setting the Cluster and
Sentinal instance via scripts allows us to avoid messing with
`network_mode: host` / DIND networking complexities.

Use `set -euo pipefail` per Duo review feedback

Uses -skip 'Cluster|Sentinel' to exclude topology-specific tests from the
single-node job, preventing future tests starting with C or S from being
silently excluded.

feat(postgres): Support multiple PostgreSQL operator secret formats.

See merge request !484

Merged-by: Elliot Forbes <eforbes@gitlab.com>
Approved-by: Elliot Forbes <eforbes@gitlab.com>
Co-authored-by: Florian Forster <fforster@gitlab.com>

When PostgreSQLFormat is set to CUSTOM but no projection is supplied,
projection() returned (nil, nil). The subsequent field accesses in
dsn() (proj.Host, proj.Port, ...) then panic with a nil pointer
dereference instead of surfacing the misconfiguration.

Return ErrSecretProjectionMissing in that case, and switch the CUSTOM
branch to use the nil-safe getter chain for consistency with the
UNSPECIFIED handling above it.

AI-assisted: Claude Code (claude-opus-4-7)

Errors returned by dsn() (and the projection() helper it calls) had
no package prefix, so they reached the caller as bare strings like
"unknown secret format: -1" or a multi-line errors.Join output. The
sibling error paths in newFromProto wrap with "postgres:", so the
dsn() path was the only one without it.

Wrap the dsn() result once at the call site to keep newFromProto's
error surface uniform. errors.Is on the underlying sentinels keeps
working since fmt.Errorf with %w preserves the chain.

AI-assisted: Claude Code (claude-opus-4-7)

Three small typos in client_test.go:
- "requies" -> "requires"
- "withou" -> "without"
- "clientOptions.dns()" -> "clientOptions.dsn()" in the t.Fatalf
  format string

AI-assisted: Claude Code (claude-opus-4-7)

Duo Developer authored May 22, 2026 and Alessio Caiazza committed May 22, 2026

Resolved conflict in v2/go.sum:
- Kept fairway v1.18.0 hashes (from master, matching go.mod)
- Kept gopher-lua v1.1.1 hashes (indirect dep still in go.mod)
- Dropped zeebo/xxh3 v1.1.0 (no longer referenced in go.mod)

Run go mod tidy

fix(v2/postgres): reject CUSTOM format without projection

See merge request !503

Merged-by: Elliot Forbes <eforbes@gitlab.com>
Approved-by: Elliot Forbes <eforbes@gitlab.com>
Co-authored-by: Alessio Caiazza <code.git@caiazza.info>

feat(v2/redis): add instrumented Redis client

See merge request !497

Merged-by: Elliot Forbes <eforbes@gitlab.com>
Approved-by: Alessio Caiazza <code.git@caiazza.info>
Approved-by: Elliot Forbes <eforbes@gitlab.com>
Co-authored-by: Duo Developer <service_account_group_9970_1976e9ecde3c53c783b64edb9ae993eb@noreply.gitlab.com>
Co-authored-by: Luke Hollinda <lhollinda@gitlab.com>

ci: wire v2/redis integration tests into CI pipeline

See merge request !502

Merged-by: Matias Alvarez <malvarez@gitlab.com>
Approved-by: Matias Alvarez <malvarez@gitlab.com>
Reviewed-by: GitLab Duo <gitlab-duo@gitlab.com>
Reviewed-by: Matias Alvarez <malvarez@gitlab.com>
Co-authored-by: Luke Hollinda <lhollinda@gitlab.com>