feat: add configuration, logging, metrics, scheduler and storage foundation

- config: defaults -> JSON file -> NOEMA_* env with validation and redaction - observability: slog with secret-redacting handler; Prometheus registry with the metric set from the plan (events, workspace, memory, LLM, tools, queues) - scheduler: supervised Group and bounded Pool so no goroutine is unowned - storage: pgx pool with statement timeout, embedded migration runner (advisory lock, checksums, up/down/status), per-test schema helper - migrations/0001: append-only guard trigger, users, sessions, api tokens, recovery codes, audit log, settings - pkg/ids: UUIDv7 with monotonic counter
2026-06-14 23:08:37 +01:00
parent 3de6e53ab0
commit 6374e174c6
19 changed files with 1422 additions and 0 deletions
@@ -0,0 +1,166 @@
+// Package scheduler owns every background goroutine in the daemon. Nothing
+// outside this package should use the go statement for long-lived work: that
+// rule is what makes "no unbounded goroutines" auditable.
+package scheduler
+
+import (
+	"context"
+	"errors"
+	"log/slog"
+	"runtime/debug"
+	"sync"
+	"sync/atomic"
+	"time"
+)
+
+// Group runs named goroutines bound to one context and waits for all of them.
+// Panics inside a task are recovered, logged and reported as errors so a
+// malformed input in one worker cannot take the daemon down.
+type Group struct {
+	ctx    context.Context
+	cancel context.CancelFunc
+	log    *slog.Logger
+	wg     sync.WaitGroup
+	mu     sync.Mutex
+	errs   []error
+	active atomic.Int64
+}
+
+// NewGroup creates a group whose tasks are cancelled when ctx ends or Stop is called.
+func NewGroup(ctx context.Context, log *slog.Logger) *Group {
+	ctx, cancel := context.WithCancel(ctx)
+	return &Group{ctx: ctx, cancel: cancel, log: log}
+}
+
+// Context returns the group's context.
+func (g *Group) Context() context.Context { return g.ctx }
+
+// Go starts fn as a supervised goroutine.
+func (g *Group) Go(name string, fn func(ctx context.Context) error) {
+	g.wg.Add(1)
+	g.active.Add(1)
+	go func() {
+		defer g.wg.Done()
+		defer g.active.Add(-1)
+		defer func() {
+			if r := recover(); r != nil {
+				g.log.Error("task panicked", "task", name, "panic", r, "stack", string(debug.Stack()))
+				g.record(errors.New("task " + name + " panicked"))
+			}
+		}()
+		if err := fn(g.ctx); err != nil && !errors.Is(err, context.Canceled) {
+			g.log.Error("task failed", "task", name, "err", err)
+			g.record(err)
+		}
+	}()
+}
+
+// Every runs fn on a fixed interval until the context ends. Each run is
+// recovered individually so a single failure does not stop the ticker.
+func (g *Group) Every(name string, interval time.Duration, fn func(ctx context.Context) error) {
+	g.Go(name, func(ctx context.Context) error {
+		t := time.NewTicker(interval)
+		defer t.Stop()
+		for {
+			select {
+			case <-ctx.Done():
+				return nil
+			case <-t.C:
+				g.safeRun(name, ctx, fn)
+			}
+		}
+	})
+}
+
+func (g *Group) safeRun(name string, ctx context.Context, fn func(context.Context) error) {
+	defer func() {
+		if r := recover(); r != nil {
+			g.log.Error("periodic task panicked", "task", name, "panic", r, "stack", string(debug.Stack()))
+		}
+	}()
+	if err := fn(ctx); err != nil && !errors.Is(err, context.Canceled) {
+		g.log.Warn("periodic task failed", "task", name, "err", err)
+	}
+}
+
+func (g *Group) record(err error) {
+	g.mu.Lock()
+	g.errs = append(g.errs, err)
+	g.mu.Unlock()
+}
+
+// Active returns the number of running tasks.
+func (g *Group) Active() int64 { return g.active.Load() }
+
+// Stop cancels the group and waits for all tasks, returning any errors.
+func (g *Group) Stop() error {
+	g.cancel()
+	g.wg.Wait()
+	g.mu.Lock()
+	defer g.mu.Unlock()
+	return errors.Join(g.errs...)
+}
+
+// Wait blocks until all tasks have returned without cancelling them.
+func (g *Group) Wait() error {
+	g.wg.Wait()
+	g.mu.Lock()
+	defer g.mu.Unlock()
+	return errors.Join(g.errs...)
+}
+
+// Pool bounds concurrent execution of short tasks. Submit blocks when the pool
+// is full (backpressure) rather than queueing without limit.
+type Pool struct {
+	sem  chan struct{}
+	busy atomic.Int64
+}
+
+// NewPool creates a pool allowing size concurrent tasks.
+func NewPool(size int) *Pool {
+	if size < 1 {
+		size = 1
+	}
+	return &Pool{sem: make(chan struct{}, size)}
+}
+
+// Run executes fn once a slot is free, or returns ctx.Err() while waiting.
+// It runs fn on the caller's goroutine; combine with Group.Go for async work.
+func (p *Pool) Run(ctx context.Context, fn func(ctx context.Context) error) error {
+	select {
+	case p.sem <- struct{}{}:
+	case <-ctx.Done():
+		return ctx.Err()
+	}
+	defer func() { <-p.sem }()
+	p.busy.Add(1)
+	defer p.busy.Add(-1)
+	return fn(ctx)
+}
+
+// Busy returns the number of occupied slots.
+func (p *Pool) Busy() int64 { return p.busy.Load() }
+
+// Size returns the pool capacity.
+func (p *Pool) Size() int { return cap(p.sem) }
+
+// Parallel runs each task through the pool concurrently and waits for all of
+// them, collecting errors. It never starts more than the pool size at once.
+func Parallel(ctx context.Context, p *Pool, tasks []func(ctx context.Context) error) error {
+	var wg sync.WaitGroup
+	errs := make([]error, len(tasks))
+	for i, task := range tasks {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			defer func() {
+				if r := recover(); r != nil {
+					errs[i] = errors.New("parallel task panicked")
+				}
+			}()
+			errs[i] = p.Run(ctx, task)
+		}()
+	}
+	wg.Wait()
+	return errors.Join(errs...)
+}
@@ -0,0 +1,63 @@
+package scheduler
+
+import (
+	"context"
+	"log/slog"
+	"sync/atomic"
+	"testing"
+	"time"
+)
+
+func TestPoolBoundsConcurrency(t *testing.T) {
+	p := NewPool(3)
+	var cur, peak atomic.Int64
+	tasks := make([]func(context.Context) error, 20)
+	for i := range tasks {
+		tasks[i] = func(ctx context.Context) error {
+			n := cur.Add(1)
+			for {
+				old := peak.Load()
+				if n <= old || peak.CompareAndSwap(old, n) {
+					break
+				}
+			}
+			time.Sleep(5 * time.Millisecond)
+			cur.Add(-1)
+			return nil
+		}
+	}
+	if err := Parallel(context.Background(), p, tasks); err != nil {
+		t.Fatal(err)
+	}
+	if peak.Load() > 3 {
+		t.Fatalf("peak concurrency %d exceeded pool size", peak.Load())
+	}
+}
+
+func TestGroupRecoversPanicAndCancels(t *testing.T) {
+	g := NewGroup(context.Background(), slog.New(slog.DiscardHandler))
+	g.Go("boom", func(ctx context.Context) error { panic("x") })
+	blocked := make(chan struct{})
+	g.Go("waiter", func(ctx context.Context) error { close(blocked); <-ctx.Done(); return ctx.Err() })
+	<-blocked
+	err := g.Stop()
+	if err == nil {
+		t.Fatal("expected panic error")
+	}
+	if g.Active() != 0 {
+		t.Fatal("tasks still active")
+	}
+}
+
+func TestPoolRunHonoursCancellation(t *testing.T) {
+	p := NewPool(1)
+	ctx, cancel := context.WithCancel(context.Background())
+	release := make(chan struct{})
+	go p.Run(context.Background(), func(context.Context) error { <-release; return nil })
+	time.Sleep(time.Millisecond)
+	cancel()
+	if err := p.Run(ctx, func(context.Context) error { return nil }); err != context.Canceled {
+		t.Fatalf("expected cancellation, got %v", err)
+	}
+	close(release)
+}