Skip to content

CHANGELOG.md

Latest commit

 

History

History
200 lines (176 loc) · 14.5 KB

CHANGELOG.md

File metadata and controls

200 lines (176 loc) · 14.5 KB
Raw

Changelog

All notable changes to Checkgate are documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.1.20] - 2026-07-18

Added

  • SSR / bootstrap helpers (@checkgate/ssr) — server-render initial flag state and hydrate the client with zero flag flicker (Next.js, Remix, SvelteKit, any SSR framework). buildBootstrap() evaluates a page's flags for the current user on the server; bootstrapScriptTag() embeds the result as an XSS-safe <script> (escapes </script> and U+2028/U+2029, supports a CSP nonce); readBootstrap() + BootstrapValues render the first client paint from the server-resolved values (null-safe, version-checked). The Web SDK gained a bootstrap option that seeds its WASM core from the embedded snapshot so the live client is ready without waiting for its stream, and @checkgate/edge gained snapshotFlags() to expose the raw snapshot for embedding. Zero dependencies; unit-tested with node --test.

  • Infrastructure as code — manage flags and segments outside the dashboard, all on one shared Go API client (integrations/checkgate-go, dependency-free and unit-tested):

    • Terraform / OpenTofu provider (terraform-provider-checkgate) with checkgate_flag and checkgate_segment resources and a checkgate_flag data source. Polymorphic/nested fields (default_value, rules, variants, prerequisites) use semantic-JSON equality so re-serialisation never churns a plan; supports import and detects require_approval environments (surfaces the queued change request instead of hanging). Built on terraform-plugin-framework.
    • Kubernetes operator (checkgate-operator) reconciling a FeatureFlag CRD (flags.checkgate.io/v1alpha1) into Checkgate — create/update via the API, a finalizer that deletes the remote flag with the CR, periodic drift correction, and status conditions. The token is read from a referenced Secret. Reconcile logic is unit-tested with a fake client + fake server.

  • Edge Side Evaluation (@checkgate/edge) — a new, zero-dependency, runtime-agnostic edge evaluator that pulls a flag snapshot from /flags/snapshot, caches it with a TTL plus optional stale-while-revalidate, keeps the last-known-good snapshot through origin outages (fail-open), and delegates evaluation to the shared @checkgate/web WASM engine — so edge results are identical to every other SDK, with no re-implemented rule/rollout/segment logic. Ships with a Cloudflare Workers example (per-request edge evaluation in a warm isolate + a Cron Trigger that keeps the snapshot warm) and a Fly.io multi-region deploy recipe (fly.toml + guide). Lives in edge/ with a full node --test suite covering refresh de-duplication, TTL/SWR behavior, and fail-open resilience.

  • Type-Safe Schema CLI (@checkgate/cli) — a new checkgate typegen command generates type-safe flag accessors for TypeScript, Dart, and Rust from your flag definitions, read either from a running server (--url/--env/--token, or the CHECKGATE_* env vars) or a local JSON export (--input). TypeScript emits a FlagKey union, per-flag value types, defaults, and a typedFlags() wrapper so a wrong key or mis-typed value is a compile error; Dart emits a FlagKey enum + TypedFlags wrapper; Rust emits a FlagKey enum (as_str()/ALL/Display) and a dependency-free defaults module. Output is deterministic (sorted, archived flags excluded); non-identifier keys (hyphens, leading digits) are converted safely per language. Zero runtime dependencies; Node 18+. Lives in cli/ with a full node --test suite.

  • Exposure Dashboards — a new Exposure page visualizes which users are being exposed to which flag variant. Per-flag it shows the variant distribution (share of evaluations and unique users) and a 14-day stacked timeline of evaluations per variant, all derived from existing impression data. Backed by GET /api/environments/{env}/impressions/exposure?flag_key=….

  • A/B Testing (Beta) — an end-to-end conversion-measurement pipeline:

    • Goal events: SDKs gained a track(eventKey, userKey, { value, context }) method that batches and reports conversion events best-effort, mirroring impression reporting. Available in all four SDKs — Node, Web (WASM), React Native (JSI), and Flutter (FFI). Events land via a new POST /api/environments/{env}/events ingest endpoint (256 KB body limit, same as impressions). Both impression and event ingest now clamp the client-supplied timestamp to the server clock (LEAST(_, NOW())) so a future-fast client clock can't make a conversion sort ahead of the exposure that caused it; genuinely-old (offline-queued) timestamps are preserved.
    • Experiments: a new Experiments page (and full CRUD API) ties a flag (the variant source) to a goal event. Its results view computes each variant's conversion rate, relative uplift vs. the control, and a two-proportion z-test with p-value and a 95% significance verdict. Each user enters the experiment at their first exposure to the flag and is bucketed into the variant seen then; they count as converted only if they fired the goal at or after that exposure. The control is the configured control_variant or, if unset, the highest-exposure variant.
    • New tables: events and experiments. New sidebar entries: Exposure and Experiments.

[0.1.19] - 2026-07-05

Changed

  • Dashboard layout overhaul — the sidebar can now be collapsed to an icon-only rail (persisted across sessions via localStorage), reclaiming space on smaller screens. Every list/table page (Dashboard, Feature Flags, Environments, Compare, Change Requests, Projects, Users, Audit Log, Impressions, SDK Health, Scheduled, Webhooks) is now genuinely full width instead of capped at an arbitrary max-width with dead space on either side.
  • Flag Editor and Settings restructured into two columns — behavior/configuration (flag type, rollout, targeting rules; SDK snippets, personal access tokens) in a wider main column, with identity/metadata (key, tags, ownership, prerequisites; auth info, API endpoint, account) in a narrower sidebar. Previously everything was stacked in a single column, requiring far more scrolling and leaving most of the screen empty on anything wider than a laptop.
  • Setup and Login pages now fill the browser window edge to edge (previously framed/centered with growing dead space in the middle on wide monitors, then over-corrected to a narrow centered card — this settles on genuinely full-width panels with wider inner content blocks). The Setup welcome screen's headline now wraps to two lines instead of three, and gained an "Already set up? Sign in" link for anyone who lands there after setup is already complete.
  • ProjectSettings now supports deep-linking to a specific tab via ?tab=keys (etc.), so other pages can link straight to e.g. SDK Keys instead of dropping the user on the default tab.

Fixed

  • Setup/login redirect loop — /api/auth/me 401s for anyone without a valid session cookie, which the frontend was also using to determine "has setup been completed?" A fresh browser, a different device, or a session that expired after a server restart would incorrectly conclude setup had never been done and permanently redirect to /setup instead of /login, with no way back. Fixed by checking the public /api/auth/workspace endpoint (now also returns is_setup_complete) instead, and by fixing a related race where the redirect decision fired on a stale guess before that check even resolved.
  • Dead "SDK Keys" link on the Settings page — pointed to /settings (itself) instead of the actual key-management page. Now links to /projects/{id}?tab=keys and lands directly on the SDK Keys tab.
  • React Native SDK packaging — sdks/react-native/android/.gradle/ (a local Gradle build cache, accidentally committed) was being swept into the published npm tarball via the files field covering the whole android/ directory. Untracked and gitignored; harmless to existing consumers, just junk that shouldn't have shipped.

[0.1.18] - 2026-07-04

Added

  • Prerequisite (dependent) flags — a flag can require another flag to be enabled, or resolved to a specific value, before its own rules/rollout are considered. Evaluated recursively (a prerequisite can itself have prerequisites) with a depth guard that fails closed on cycles or misconfigured chains. Lives in the shared evaluation core, so every SDK supports it automatically.
  • Weighted multivariate rollouts — a variants field distributes traffic across multiple values by weight (e.g. a 60/30/10 A/B/C split), independent of the on/off rollout gate. The foundation for A/B testing.
  • getValue/getVariant in every SDK wrapper — the underlying bindings already supported multi-variant flags, but the public Node, Web, React Native, and Flutter wrappers only exposed isEnabled. All four now expose the full evaluation surface.
  • SDK impression reporting — all SDKs asynchronously batch and report evaluation events, feeding the dashboard's impression stats and evaluation stream. Off by default; user attributes are never sent unless sendEvaluationContext is enabled.
  • Numeric targeting operators — greater_than, greater_than_or_equal, less_than, less_than_or_equal, alongside the existing string operators.
  • SDK resilience — exponential backoff with jitter on SSE reconnect (replacing a fixed retry delay); flag-change listeners (onChange) with bootstrap/reconnect-resync suppression so only genuine live deltas fire; offline persistence via a pluggable storage adapter (hydrate on cold start, persist on bootstrap/delta); and an HTTP poll fallback (GET /flags/snapshot) for environments where SSE can't be established (e.g. a proxy blocking long-lived connections).
  • Unified connect()/ready semantics — a new server-emitted ready SSE event gives every SDK a consistent way to know when the initial flag set has loaded, instead of each platform inferring it differently.
  • Flag lifecycle hygiene — tags, an owner email, and archival (soft-delete, reversible, zero evaluation impact). Kept as dashboard-only metadata on a FlagWithMetadata wrapper so it never flows into the evaluation core or the SSE//flags/snapshot wire format.
  • Cross-environment diff — a "Compare environments" dashboard page showing flags that exist in only one environment, or differ in evaluation-relevant fields (rollout, rules, variants, prerequisites), with a one-click sync action per flag.
  • Scoped personal access tokens — user-owned, revocable API credentials for CI/CD, Terraform, and scripts, as an alternative to the always-admin-equivalent SDK key. A token acts as its owning user (same role, same project memberships) and can be capped to read_only. Tokens are SHA-256 hashed at rest, support optional expiry, and are strictly self-service (list/create/revoke your own only).
  • Change requests (approval workflow) — a per-environment require_approval toggle. When set, a flag PATCH is captured as a pending change request instead of applying immediately, and a different editor/admin must review it. Approve applies the original patch against the flag's current state; reject (with an optional reason) or withdraw never touch the flag.

Changed

  • evaluate/evaluate_variant in the core evaluator now take a &FlagStore parameter to support recursive prerequisite lookups. Internal API change — all four SDK bindings were updated accordingly; public SDK surfaces are unaffected.

Fixed

  • time::OffsetDateTime fields in impressions.rs, keys.rs, and users.rs were serializing in a proprietary, non-RFC-3339 format that JavaScript's Date cannot parse, despite doc comments claiming "ISO-8601." Corrected to genuine RFC 3339 output across all affected endpoints.
  • Node.js SDK: eventsource v4's named export and dropped headers option meant the Bearer token was silently omitted from SSE reconnect requests. Fixed via a custom fetch hook.
  • Dashboard Environments.tsx called bare /api/environments... routes that don't exist on the server (the real routes are nested under /api/projects/{project_id}/environments...), causing "Server returned 405" on environment creation and silent failures on delete/set-default.
  • Removed stale, git-tracked copies of sdks/react-native/rust-core/ — a generated build artifact that had drifted out of sync with core/. Now gitignored; regenerated fresh before every release.
  • Node.js SDK packaging — index.js (the hand-written wrapper) and napi build --platform's auto-generated multi-platform dispatch loader both wrote to the same filename, so whichever ran last won; the checked-in wrapper always won in practice, and it require()d a single-file index.node that never exists in the real per-platform release output (checkgate.darwin-x64.node, checkgate.linux-x64-gnu.node, etc.). Every install of @checkgate/node would have thrown Cannot find module './index.node'. Pre-existing since at least v0.1.17 and never caught because nothing in CI actually require()d the assembled package. Fixed by generating the dispatch loader to a separate native-binding.js/native-binding.d.ts (via napi build's --js/--dts flags) that index.js requires instead, and by adding a real require() smoke test to the release workflow so this class of bug fails loudly in CI instead of shipping silently.

Security

  • Personal access tokens are SHA-256 hashed at rest (SDK keys remain plaintext, by contrast, since they're a different, coarser-grained credential class), scoped to the owning user's actual role and project memberships rather than being blanket admin-equivalent, and a read_only token cannot mint a read_write replacement for itself.
  • Change-request self-approval is rejected — a reviewer other than the requester must approve.

[0.1.17] - 2026-06-14

  • Audit logs, user segmentation, webhooks with HMAC signing, time-based scheduled flag changes, and live SSE connection monitoring (SDK Health).

[0.1.0] - 0.1.16

See individual SDK changelogs (e.g. sdks/flutter/dart/CHANGELOG.md) and docs/roadmap.md for earlier history: initial local-evaluation core, multi-variant flags, editor RBAC, projects/multi- tenancy, impression tracking, and security hardening.