@pikku/queue-pg-boss 0.12.11 → 0.12.12

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pikku/queue-pg-boss",
-  "version": "0.12.11",
+  "version": "0.12.12",
   "repository": {
     "type": "git",
     "url": "git+https://github.com/pikkujs/pikku.git",
@@ -16,17 +16,23 @@
     "tsc": "tsc",
     "test": "bash run-tests.sh",
     "ncu": "npx npm-check-updates",
-    "build": "tsc -b",
+    "build": "bun run clean && tsc -b",
     "release": "npm run build && npm test",
-    "prepublishOnly": "bun run build"
+    "prepublishOnly": "bun run build",
+    "clean": "rm -rf dist tsconfig.tsbuildinfo"
   },
   "peerDependencies": {
-    "@pikku/core": "^0.12.108"
+    "@pikku/core": "^0.12.144"
   },
   "dependencies": {
-    "pg-boss": "^12.27.0"
+    "pg-boss": "^12.38.0"
   },
   "exports": {
-    ".": "./dist/index.js"
-  }
+    ".": "./dist/index.js",
+    "./package.json": "./package.json"
+  },
+  "files": [
+    "dist",
+    "!dist/**/*.tsbuildinfo"
+  ]
 }

package/CHANGELOG.md

@@ -1,437 +0,0 @@
-## 0.12.11
-
-### Patch Changes
-
-- 4c7a1b5: Run the monorepo's own scripts through bun instead of yarn. What moves is the
-  package manager each package's `prepublishOnly` and build scripts invoke, plus
-  the two manifest fixes bun needs to resolve the tree: `uWebSockets.js` is
-  declared with an explicit `github:` specifier, and `@pikku/uws-handler` marks
-  its `uWebSockets.js` peer optional so a bun install of a consumer that brings
-  its own uWS app does not try to fetch it from the registry.
-
-  Three published behaviours change, all of them cases where an isolated
-  `node_modules` or bun as the runtime had been papered over by yarn's hoisting:
-
-  - `@pikku/migrator-sql` turns foreign keys on when it opens a sqlite database
-    through bun. `node:sqlite` enforces them by default and `bun:sqlite` does not,
-    which silently turned every `ON DELETE CASCADE` into a no-op under
-    `bunx --bun pikku`.
-  - `@pikku/cli` resolves a deploy provider against the project being deployed
-    rather than against wherever the CLI itself is installed, which is what its
-    own "is not installed" error asks the user to arrange.
-  - `@pikku/cli` treats a specifier a runtime hands straight back — bun does this
-    for the modules it implements itself — as not resolved from the project, so
-    it falls back rather than loading the runtime's own copy.
-
-- Updated dependencies [4c7a1b5]
-  - @pikku/core@0.12.108
-
-## 0.12.10
-
-### Patch Changes
-
-- 266e3bc: One door per name: `@pikku/core/ecosystem/*` and the package root are gone
-
-  `@pikku/core` published every module twice. `ecosystem/http` re-exported
-  `./http`, `ecosystem/services` re-exported `./services`, and a name was
-  reachable through either — so every addition had to be made in two places, and a
-  consumer's import said nothing about what it actually used. The package root was
-  the same problem at a larger scale: a single barrel of 206 names that no bundler
-  could take apart, and the one specifier that revealed nothing at all.
-
-  Both are deleted. Every name now lives on the subpath that owns it, and every
-  import carries that subpath — `@pikku/core/http`, `@pikku/core/services`,
-  `@pikku/core/errors`, `@pikku/core/types`.
-
-  Deleting the facades meant the raw subpaths had to become a superset of them,
-  which they were not: the facade tree had accumulated 25 names with no raw home
-  and about 26 more filed under a different area than the module they came from.
-  Those names moved to the area that owns them, and three areas were published as
-  new entry points rather than left on a root that is going away — `./types` (the
-  shared type surface, the largest single destination), `./state` and
-  `./classification`.
-
-  `./classification` is one door onto one subject: what a value is and how it must
-  be handled. Its three halves would each have been an entry point — the brands
-  and manifest types, the stored-form helpers (`hashToken`, `unsafeAsSealed` and
-  friends), and `SecretValue` — split by whether a name happens to be a type or a
-  value, which is the same defect as the facades. The duration and versioned-id
-  helpers went to `./utils`, which already published, and `PikkuRequest` went to
-  `./function`: it is the transport-agnostic request base, not an HTTP one — HTTP
-  has `PikkuHTTPAbstractRequest`, and the only thing outside core that extends
-  `PikkuRequest` is Azure's timer request.
-
-  `./types` inherited the root barrel's habit before it inherited its contents, so
-  the names with an owner elsewhere were moved off it. The middleware types and the
-  five middleware factories — `pikkuMiddleware`, `pikkuMiddlewareFactory`,
-  `pikkuChannelMiddleware`, `pikkuChannelMiddlewareFactory` and
-  `pikkuAgentMiddleware`, runtime values on a types entry point — are now
-  `@pikku/core/middleware`; the function meta types are `@pikku/core/function`;
-  `SerializedError` is `@pikku/core/errors`; and the generic TypeScript helpers
-  (`MakeRequired`, `PickRequired`, `PickOptional`, `RequireAtLeastOne`,
-  `JSONPrimitive`, `JSONValue`) are `@pikku/core/utils`. What is left on `./types`
-  is the vocabulary the wirings share, which no single module owns.
-
-  `pikku` was itself a root barrel — `export * from '@pikku/core'` — and
-  now exports only the services it bundles.
-
-  One module survives at the old specifier, and only for the bootstrap:
-  `packages/cli` is generated by the _published_ CLI pinned in its `build.sh`, and
-  that CLI still writes a bare `@pikku/core` into the files it generates for the
-  CLI itself. `bootstrap-compat/root.ts` carries the eight types it names, a test
-  in core fails if that list grows, and it goes when the pin moves to a CLI
-  released from this branch. The adapter names the pinned CLI reaches for —
-  `pikkuState` and `CreateWireServices` — are rewritten to `@pikku/core/state` and
-  `@pikku/core/types` by the same `build.sh` patch pass, so no second shim is
-  needed for them.
-
-  A guard test keeps the root shut: it parses imports and rejects a bare
-  `@pikku/core` rather than grepping for it, because several tests hold a user's
-  file as a template literal, where `import … from '@pikku/core'` is fixture text
-  rather than an import this repo makes.
-
-  An agent scaffold a project generated under an older CLI is refreshed rather
-  than left to fail: `pikku all` already deleted one importing an entry point
-  `@pikku/core` no longer publishes, and the `#pikku` hub joins that list.
-  Without it a project that scaffolds the agent endpoint but
-  declares no agents keeps the old file forever — the generator that would rewrite
-  it only runs when agents exist, and the file being present is what stops it
-  being regenerated as missing.
-
-  `pikku new addon` also wrote a tsconfig `paths` map naming only the deleted hub.
-  An addon's `imports` map points into `dist`, so `paths` is what resolves
-  `#pikku/<leaf>` for the addon's own source build — it now names the two leaf
-  patterns, in both the addon and its test harness.
-
-- 6eef0a0: Bump every dependency to its latest compatible minor/patch across the monorepo.
-- Updated dependencies [7722ceb]
-- Updated dependencies [375c1ff]
-- Updated dependencies [02a70cd]
-- Updated dependencies [aeef159]
-- Updated dependencies [a281de6]
-- Updated dependencies [266e3bc]
-- Updated dependencies [02a70cd]
-- Updated dependencies [786dae5]
-- Updated dependencies [6eef0a0]
-- Updated dependencies [3561d67]
-- Updated dependencies [a91c433]
-- Updated dependencies [02a70cd]
-- Updated dependencies [9537f74]
-- Updated dependencies [2b57ca8]
-- Updated dependencies [266e3bc]
-- Updated dependencies [9fce0f1]
-- Updated dependencies [83683a0]
-- Updated dependencies [456c88b]
-- Updated dependencies [456c88b]
-- Updated dependencies [c127273]
-  - @pikku/core@0.12.85
-
-## 0.12.9
-
-### Patch Changes
-
-- 6794681: Publish the ecosystem surface as per-area sub-barrels under `@pikku/core/ecosystem/*`, and point generated code, the CLI, the inspector and the runtime adapters at them.
-
-  348 names that only generated code, the toolchain or a runtime adapter ever imports now have a second home on `@pikku/core/ecosystem/<area>` — one sub-barrel per area, matching how core already publishes its entrypoints, so no single barrel grows without bound and a consumer only pulls in the area it uses.
-
-  This step is additive: every name is still exported from the entrypoint it was published from before, so nothing downstream breaks. Removing them from the app-facing barrels is a later change, and needs a release carrying `./ecosystem/*` first.
-
-- Updated dependencies [7406bfe]
-- Updated dependencies [6794681]
-  - @pikku/core@0.12.84
-
-## 0.12.8
-
-### Patch Changes
-
-- 6847338: Stop one slow job from holding an entire queue (head-of-line blocking).
-
-  The worker mapped pikku's parallelism to pg-boss `batchSize`, whose handler
-  receives the whole fetched batch and only fetches again once that handler
-  resolves. Because the handler `Promise.all`s the batch, a single long-running
-  job blocked every sibling in its batch _and_ stalled the next fetch — so on a
-  shared queue one slow job could freeze the worker and starve everything behind
-  it (observed with workflow-orchestrator jobs, where a long step pinned all 10
-  slots for tens of minutes).
-
-  Parallelism now maps to pg-boss `localConcurrency` with `batchSize: 1`: N
-  independent workers each fetch and process a single job and refill the instant
-  they finish, so a slow job never holds up the others. Same effective
-  concurrency, no head-of-line blocking.
-
-- e3d4454: Add job groups, so one shared queue can stay fair without splitting into one
-  queue per producer.
-
-  A job may now carry `group: { id, tier }`, and a worker may cap how many jobs
-  of any one group run at once via `groupConcurrency`. On pg-boss this maps to
-  `localGroupConcurrency`, which excludes at-capacity groups from the fetch query
-  itself, so a capped group costs nothing rather than being fetched and restored.
-  BullMQ declares it unsupported (groups are a BullMQ Pro feature) — being
-  push-based, it can simply use a queue per group at no polling cost.
-
-  Workflow services accept a `queueStrategy`. The default `'per-workflow'` is
-  unchanged: every workflow gets its own `wf-orchestrator-*` / `wf-step-*` queue,
-  which is also what lets serverless providers deploy one unit per workflow. The
-  new `'shared-groups'` routes every workflow through the shared
-  orchestrator/step-worker queues and isolates them by group instead, so a
-  monolith runs one set of pollers rather than one per workflow — on a
-  pull-based backend with dozens of workflows that is the difference between
-  hundreds of poll loops and twenty. It is for single-process runtimes only; a
-  per-unit serverless deploy still needs the per-workflow queues to route to its
-  units.
-
-- Updated dependencies [24252b8]
-- Updated dependencies [e3d4454]
-  - @pikku/core@0.12.69
-
-## 0.12.7
-
-### Patch Changes
-
-- 6c30861: fix workflow step retry backoff firing immediately
-  - `@pikku/queue-pg-boss`: `backoff: 'exponential'` mapped to `retryBackoff: true` without a base `retryDelay`; pg-boss computes exponential backoff as `retry_delay * 2^n` with a queue default of 0, so every retry fired immediately. Exponential backoff now gets a 1s base delay, and sub-second fixed delays round up to 1s instead of flooring to 0 (= immediate).
-  - `@pikku/core`: a duration-string `retryDelay` (e.g. `'15s'`) on a workflow step was silently dropped (only numbers were honored) and fell back to exponential. It now resolves to a fixed backoff via `getDurationInMilliseconds`.
-
-- Updated dependencies [6c30861]
-  - @pikku/core@0.12.56
-
-## 0.12.6
-
-### Patch Changes
-
-- 41ce2cb: Upgrade to TypeScript 6 and raise the minimum Node.js version to 22.
-
-  All packages now build against `typescript@^6.0.3` and declare `engines.node >= 22`. Internal tooling (`ts-json-schema-generator`, `zod-to-ts`) was bumped to TypeScript 6-compatible releases.
-
-- Updated dependencies [41ce2cb]
-  - @pikku/core@0.12.44
-
-## 0.12.5
-
-### Patch Changes
-
-- 05aba57: fix(scheduler): prune orphaned recurring-task schedules on start
-
-  When a scheduled task is removed from code, its pg-boss schedule row survived and
-  kept producing jobs into a queue with no registered worker — the jobs piled up as
-  'created' forever and could flood the queue (thousands of stuck jobs). `start()`
-  now drops any `pikku-recurring-scheduled-task_*` schedule and queue that is no
-  longer registered, so task removals self-heal on the next boot.
-
-- Updated dependencies [a8022e5]
-  - @pikku/core@0.12.43
-
-## 0.12.4
-
-### Patch Changes
-
-- 92cd5b1: feat(workflow): workflow-owned step retries + stable invocationId
-
-  The workflow — not the queue — now owns step retry policy, and each step
-  invocation gets a stable idempotency key.
-  - **Default `retries: 5` with exponential backoff.** A step with no `retries`
-    previously inherited the queue's bare default (e.g. pg-boss `retry_limit 2`,
-    no backoff) so retries fired instantly and couldn't outlast a transient
-    outage. Retries now default to 5 with backoff, resolved at the workflow layer.
-  - **`retries: 0` is honored.** Dispatch previously passed `undefined` options
-    for `retries: 0`, letting the queue re-run a non-idempotent step up to its own
-    default. The resolved policy now always sets `attempts` (`retries: 0` →
-    `attempts: 1`), so the queue never second-guesses the workflow. The persisted
-    step retries and the dispatched `attempts` are resolved together so
-    "retries exhausted" and "no more redeliveries" are the same event.
-  - **`workflowStep.invocationId`** — a deterministic, dependency-free
-    `uuidv5(runId:stepName)` handed to every step. Unlike `stepId` (minted per
-    attempt), it is identical across retries, so a step can dedupe on it
-    (`ON CONFLICT (invocationId)`, Stripe idempotency keys, etc.).
-  - **queue-bullmq**: `mapPikkuJobToBull` now maps `backoff` (previously dropped,
-    so a step's backoff silently never applied on Redis), and `registerQueues`
-    throws a clear error when no logger is available (matching queue-pg-boss).
-  - **Dispatch failures are recoverable, not fatal.** A step is now marked
-    `scheduled` only _after_ it is successfully handed to its transport (queue or
-    scheduler) — a failed hand-off leaves it `pending` so a replay re-dispatches
-    it, instead of stranding it in `scheduled` (replay would pause forever on a
-    job that was never enqueued). A transport outage (e.g. pg-boss momentarily
-    down) is surfaced as a new `WorkflowDispatchException`, which the orchestrator
-    treats as transient: the run is left running and the orchestrator job is
-    rethrown for redelivery (it replays idempotently from the snapshot) rather
-    than the whole run being marked `failed`. The orchestrator job now also
-    carries its own retry policy, so this holds even when the orchestrator queue
-    is configured `retry_limit 0`. A genuine step error still fails the run.
-  - **Same step name can be invoked multiple times in one run.** Step rows are now
-    keyed per _invocation_: the Nth reach of a step name in a replay resolves to a
-    physical key (`name` for the first, `name#N` for repeats), so a literal
-    duplicate name no longer clobbers the earlier step's state. The first reach
-    keeps the bare name, so existing rows, graph-node matching and `invocationId`s
-    are unchanged. Ordinals are derived deterministically from DSL execution order
-    and reset each replay.
-  - **Step provenance (`fromStepName`) + graph cycles.** Every step now records
-    the predecessor it was scheduled from (`fromStepName`; entry steps have none),
-    persisted on the step row across all stores (in-memory, kysely, redis,
-    mongodb, cloudflare DO) and carried in the queued payload. The DSL wire
-    exposes the derived `fromInvocationId` (`uuidv5(runId:fromStepName)`) so
-    consumers get the stable predecessor key without a second persisted id —
-    `fromStepName` is the source of truth (it is replay-deterministic; `stepId`,
-    minted per row, is not). This makes the walked path reconstructable even when
-    a node is reached more than once: in `a → b → a → c` the second `a` is a
-    distinct ordinal instance (`a#1`) whose `fromStepName` is `b`.
-    The graph runner now supports **cycles**: a forward edge into an
-    already-started node still collapses to a single run (joins/diamonds are
-    unchanged), but a _back-edge_ — one whose target can reach its source — fires
-    a fresh ordinal instance, so a node can loop back to itself. Termination is
-    the graph's responsibility (branch routing must converge); the engine enforces
-    no visit cap.
-
-- Updated dependencies [92cd5b1]
-  - @pikku/core@0.12.38
-
-## 0.12.0
-
-## 0.12.3
-
-### Patch Changes
-
-- 8b9b2e9: Accept optional logger parameter in `registerQueues()` instead of reaching into pikku state directly. Falls back to `getSingletonServices()` for backwards compatibility.
-- 8b9b2e9: Fix child workflow completion in queued execution mode. When a sub-workflow completes, the parent step is now marked as succeeded and the parent orchestrator resumes automatically via `onChildWorkflowCompleted`. Adds `parentStepId` to `WorkflowRunWire` to track the parent step without querying. Retains advisory locks in PgKyselyWorkflowService for concurrency safety. Fixes pgboss `registerQueues` to accept an optional logger parameter.
-- Updated dependencies [e412b4d]
-- Updated dependencies [53dc8c8]
-- Updated dependencies [0a1cc51]
-- Updated dependencies [0a1cc51]
-- Updated dependencies [0a1cc51]
-- Updated dependencies [0a1cc51]
-- Updated dependencies [0a1cc51]
-- Updated dependencies [0a1cc51]
-- Updated dependencies [0a1cc51]
-- Updated dependencies [0a1cc51]
-- Updated dependencies [0a1cc51]
-- Updated dependencies [8b9b2e9]
-- Updated dependencies [8b9b2e9]
-- Updated dependencies [b973d44]
-- Updated dependencies [8b9b2e9]
-- Updated dependencies [8b9b2e9]
-  - @pikku/core@0.12.9
-
-## 0.12.2
-
-### Patch Changes
-
-- 3e04565: chore: update dependencies to latest minor/patch versions
-- Updated dependencies [cc4c9e9]
-- Updated dependencies [3e04565]
-  - @pikku/core@0.12.2
-
-## 0.12.1
-
-### Patch Changes
-
-- 62a8725: Fix pg-boss queue initialisation: create the queue before registering a worker to avoid a race condition on first startup. Also sanitise scheduler names to meet pg-boss naming constraints.
-- Updated dependencies [62a8725]
-- Updated dependencies [a3bdb0d]
-- Updated dependencies [e0349ff]
-- Updated dependencies [62a8725]
-- Updated dependencies [e04531f]
-- Updated dependencies [62a8725]
-- Updated dependencies [a83efb8]
-- Updated dependencies [8eed717]
-- Updated dependencies [62a8725]
-- Updated dependencies [62a8725]
-- Updated dependencies [62a8725]
-  - @pikku/core@0.12.1
-
-### New Features
-
-- `setServices()`, `start()`, `stop()` lifecycle methods on scheduler and queue workers
-
-## 0.11.0
-
-## 0.11.2
-
-### Patch Changes
-
-- db9c7bf: Simplify CreateWireServices type signature for custom Config compatibility
-- Updated dependencies [db9c7bf]
-  - @pikku/core@0.11.2
-
-### Fixes
-
-- ddd87eaf: Simplify CreateWireServices type signature for custom Config compatibility
-
-## 0.11.1
-
-### Patch Changes
-
-- 4b811db: chore: updating all dependencies
-- 06e1a31: breaking: change session services to wire services
-- Updated dependencies [4b811db]
-- Updated dependencies [e12a00c]
-- Updated dependencies [4579434]
-- Updated dependencies [28aeb7f]
-- Updated dependencies [ce902b1]
-- Updated dependencies [06e1a31]
-  - @pikku/core@0.11.1
-
-### Minor Changes
-
-- Add scheduler service implementation
-- Add service factory for queue and scheduler creation
-
-# @pikku/queue-pg-boss
-
-## 0.10.0
-
-This release includes significant improvements across the framework including tree-shaking support, middleware/permission factories, enhanced CLI functionality, improved TypeScript type safety, and comprehensive test strategies.
-
-For complete details, see https://pikku.dev/changelogs/0_10_0.md
-
-## 0.9.5-next.0
-
-### Patch Changes
-
-- Updated dependencies
-  - @pikku/core@0.9.12-next.0
-
-## 0.9.4
-
-### Patch Changes
-
-- 3bb5824: fix: forcing release to see if dist is generated
-- Updated dependencies [99c2b3a]
-  - @pikku/core@0.9.9
-
-## 0.9.3
-
-### Patch Changes
-
-- c18800d: feat: adding queue and scheduledTask to interactions
-- Updated dependencies [c18800d]
-  - @pikku/core@0.9.4
-
-## 0.9.2
-
-### Patch Changes
-
-- a5905a9: chore: updating all dependencies
-- Updated dependencies [1256238]
-- Updated dependencies [6cf8efd]
-- Updated dependencies [d3a9a09]
-- Updated dependencies [840e078]
-- Updated dependencies [667d23c]
-- Updated dependencies [a5905a9]
-  - @pikku/core@0.9.2
-
-## 0.9.1
-
-### Patch Changes
-
-- fdb1593: core: bumping everything with a patch to sync up the major release inconsistencies in dependencies
-- Updated dependencies [fdb1593]
-  - @pikku/core@0.9.1
-
-## 0.9.0
-
-### Breaking Changes
-
-- Normalized all transports to use "wirings" instead of events/routes/transports for consistency across the framework
-
-## 0.8.0
-
-- **PG Boss integration**: Postgres queue service implementation

package/run-tests.sh

@@ -1,68 +0,0 @@
-#!/bin/bash
-
-# Enable nullglob to handle cases where no files match the pattern
-shopt -s nullglob
-
-# Initialize variables for options
-watch_mode=false
-coverage_mode=false
-
-# Parse command-line options
-while [[ $# -gt 0 ]]; do
-  case $1 in
-    --watch)
-      watch_mode=true
-      shift
-      ;;
-    --coverage)
-      coverage_mode=true
-      shift
-      ;;
-    *)
-      echo "Unknown option: $1"
-      exit 1
-      ;;
-  esac
-done
-
-# Define the pattern to match your test files
-pattern="src/*.test.ts"
-
-# Expand the pattern into an array of files
-files=($(find src -type f -name "*.test.ts"))
-
-# Check if any files matched the pattern
-if [ ${#files[@]} -eq 0 ]; then
-  echo "No test files found matching pattern: $pattern"
-  if [ "${STRICT_TEST_DISCOVERY}" = "1" ]; then
-    exit 1
-  fi
-  exit 0
-fi
-
-# Construct the bun command
-#
-# --parallel gives every test file its own worker process, which is what node's
-# `--test` did. Several suites keep module-level caches (a TypeScript program, a
-# temp fixture directory) that a shared process would leak between files.
-#
-# --timeout because bun caps a test at 5s by default where node had no cap, and
-# the suites that build real TypeScript programs run well past that.
-bun_cmd=(bun test --parallel --timeout 120000)
-
-# Append options based on flags
-if [ "$watch_mode" = true ]; then
-  # --watch reruns in one process; parallel workers have nothing to watch.
-  bun_cmd=(bun test --watch --timeout 120000)
-fi
-
-if [ "$coverage_mode" = true ]; then
-  # --coverage-dir=. keeps lcov.info at the package root: CI merges every
-  # package's report by prefixing its SF paths with the file's own directory,
-  # so a nested coverage/ directory would re-root them one level too deep.
-  bun_cmd+=(--coverage --coverage-reporter=lcov --coverage-reporter=text --coverage-dir=.)
-  export PIKKU_TEST_COVERAGE=1
-fi
-
-# Execute the bun command with the expanded list of files
-"${bun_cmd[@]}" "${files[@]}"