Make snapshot projections self-diagnosing

cmd/agent/main.go

@@ -296,6 +296,9 @@ func (d *dispatcher) handle(ctx context.Context, env api.Envelope, tx wsclient.S
 		}
 		go d.handleTreeList(ctx, env.ID, p, tx)
 
+	case api.MsgSnapshotsRefresh:
+		go d.refreshSnapshots(ctx, tx)
+
 	case api.MsgScheduleSet:
 		var p api.ScheduleSetPayload
 		if err := env.UnmarshalPayload(&p); err != nil {
@@ -405,6 +408,22 @@ func (d *dispatcher) handle(ctx context.Context, env api.Envelope, tx wsclient.S
 	return nil
 }
 
+func (d *dispatcher) refreshSnapshots(ctx context.Context, tx wsclient.Sender) {
+	creds, err := d.secrets.Load()
+	if err != nil || creds.Empty() {
+		slog.Warn("ws agent: snapshots.refresh unavailable", "err", err)
+		return
+	}
+	r := runner.New(runner.Config{
+		ResticBin: d.resticBin, ResticVersion: d.resticVer,
+		RepoURL: creds.URL, RepoUsername: creds.Username, RepoPassword: creds.Password,
+		SupportsRestoreNoOwnership: d.resticSupportsNoOwnership,
+	}, tx, time.Second)
+	if err := r.RefreshSnapshots(ctx); err != nil {
+		slog.Warn("ws agent: snapshots.refresh failed", "err", err)
+	}
+}
+
 // handleTreeList runs `restic ls --json <snapshot> <path>` and ships
 // the matching tree.list.result envelope back, correlated by the
 // request envelope's ID. Errors (missing creds, restic failure)

internal/agent/runner/runner.go

@@ -77,6 +77,12 @@ func (r *Runner) resticEnv() restic.Env {
 	}
 }
 
+// RefreshSnapshots reconciles the server's cached projection without running
+// a mutating repository job.
+func (r *Runner) RefreshSnapshots(ctx context.Context) error {
+	return r.reportSnapshots(ctx, r.resticEnv())
+}
+
 // sendStarted ships a job.started envelope.
 func (r *Runner) sendStarted(jobID string, kind api.JobKind, startedAt time.Time) {
 	env, _ := api.Marshal(api.MsgJobStarted, jobID, api.JobStartedPayload{
@@ -226,14 +232,9 @@ func (r *Runner) RunBackup(ctx context.Context, jobID string, paths, excludes, t
 		}
 	}
 
-	r.sendFinished(ctx, jobID, finishedAt, err, statsBlob)
-
 	// On a successful backup, refresh the server's snapshot projection.
-	// We do this *after* job.finished so the UI sees the job land first;
-	// the snapshot list is a follow-up that the host detail page polls
-	// or the dashboard sees on its next refresh. A failure here is
-	// logged but doesn't fail the job — the next successful backup will
-	// catch the projection up.
+	// Do this before job.finished so a failure in terminal reporting cannot
+	// prevent the independently useful projection refresh.
 	if err == nil {
 		if rerr := r.reportSnapshots(ctx, env); rerr != nil {
 			slog.Warn("runner: snapshots.report failed", "job_id", jobID, "err", rerr)
@@ -242,6 +243,7 @@ func (r *Runner) RunBackup(ctx context.Context, jobID string, paths, excludes, t
 			slog.Warn("runner: stats.report after backup failed", "job_id", jobID, "err", rerr)
 		}
 	}
+	r.sendFinished(ctx, jobID, finishedAt, err, statsBlob)
 
 	if err != nil {
 		return fmt.Errorf("runner backup: %w", err)
@@ -282,8 +284,6 @@ func (r *Runner) RunForget(ctx context.Context, jobID string, groups []restic.Fo
 	var seq atomic.Int64
 	err := env.RunForget(ctx, groups, dryRun, r.streamHandler(jobID, &seq))
 	finishedAt := time.Now().UTC()
-	r.sendFinished(ctx, jobID, finishedAt, err, nil)
-
 	// Refresh the server's snapshot projection — forget rewrites the
 	// index so the host's snapshot list almost certainly shrunk.
 	if err == nil {
@@ -292,6 +292,7 @@ func (r *Runner) RunForget(ctx context.Context, jobID string, groups []restic.Fo
 				"job_id", jobID, "err", rerr)
 		}
 	}
+	r.sendFinished(ctx, jobID, finishedAt, err, nil)
 
 	if err != nil {
 		return fmt.Errorf("runner forget: %w", err)
@@ -318,6 +319,9 @@ func (r *Runner) RunPrune(ctx context.Context, jobID string) error {
 		if rerr := r.reportStats(ctx, env, api.RepoStatsPayload{LastPruneAt: &pruneAt}); rerr != nil {
 			slog.Warn("runner: stats.report after prune failed", "job_id", jobID, "err", rerr)
 		}
+		if rerr := r.reportSnapshots(ctx, env); rerr != nil {
+			slog.Warn("runner: snapshots.report after prune failed", "job_id", jobID, "err", rerr)
+		}
 	}
 
 	r.sendFinished(ctx, jobID, finishedAt, err, nil)

internal/agent/runner/runner_test.go

@@ -116,7 +116,8 @@ func envelopeOrder(envs []api.Envelope) []api.MessageType {
 // TestRunPruneShipsExpectedEnvelopes drives RunPrune with a fake
 // binary that prints "prune" on stdout (for the log.stream envelope)
 // and emits valid stats JSON so reportStats can populate size fields.
-// Expected sequence: job.started → log.stream → repo.stats → job.finished.
+// Expected sequence: job.started → log.stream → repo.stats → snapshots.report
+// → job.finished.
 func TestRunPruneShipsExpectedEnvelopes(t *testing.T) {
 	t.Parallel()
 
@@ -126,6 +127,7 @@ func TestRunPruneShipsExpectedEnvelopes(t *testing.T) {
 case "$1" in
   prune)  echo "prune" ;;
   stats)  echo '`+statsJSON+`' ;;
+	  snapshots) echo "[]" ;;
   *)      echo "unknown: $*" ;;
 esac
 `)
@@ -138,7 +140,7 @@ esac
 
 	order := envelopeOrder(tx.envs)
 	// Confirm landmark envelope types appear in the required order.
-	wantTypes := []api.MessageType{api.MsgJobStarted, api.MsgLogStream, api.MsgRepoStats, api.MsgJobFinished}
+	wantTypes := []api.MessageType{api.MsgJobStarted, api.MsgLogStream, api.MsgRepoStats, api.MsgSnapshotsRpt, api.MsgJobFinished}
 	positions := map[api.MessageType]int{}
 	for i, mt := range order {
 		if _, seen := positions[mt]; !seen {
@@ -379,6 +381,15 @@ func TestRunInitShipsStartedAndFinished(t *testing.T) {
 	_ = firstEnvOfType(t, tx.envs, api.MsgJobFinished)
 }
 
+func firstIndexOfType(envs []api.Envelope, typ api.MessageType) int {
+	for i, env := range envs {
+		if env.Type == typ {
+			return i
+		}
+	}
+	return -1
+}
+
 // TestRunForgetShipsStartedAndFinished confirms the refactored
 // RunForget still produces job.started and job.finished envelopes.
 func TestRunForgetShipsStartedAndFinished(t *testing.T) {
@@ -402,5 +413,9 @@ esac
 		t.Fatalf("RunForget: %v", err)
 	}
 	_ = firstEnvOfType(t, tx.envs, api.MsgJobStarted)
-	_ = firstEnvOfType(t, tx.envs, api.MsgJobFinished)
+	finished := firstIndexOfType(tx.envs, api.MsgJobFinished)
+	refreshed := firstIndexOfType(tx.envs, api.MsgSnapshotsRpt)
+	if refreshed < 0 || finished < 0 || refreshed >= finished {
+		t.Fatalf("snapshot refresh must precede terminal report: %v", envelopeOrder(tx.envs))
+	}
 }

internal/api/wire.go

@@ -36,12 +36,13 @@ const (
 
 // Server → agent message types.
 const (
-	MsgCommandRun    MessageType = "command.run"
-	MsgCommandCancel MessageType = "command.cancel"
-	MsgScheduleSet   MessageType = "schedule.set"
-	MsgConfigUpdate  MessageType = "config.update"
-	MsgCommandUpdate MessageType = "command.update"
-	MsgTreeList      MessageType = "tree.list" // sync RPC: list a snapshot's children
+	MsgCommandRun       MessageType = "command.run"
+	MsgCommandCancel    MessageType = "command.cancel"
+	MsgScheduleSet      MessageType = "schedule.set"
+	MsgConfigUpdate     MessageType = "config.update"
+	MsgCommandUpdate    MessageType = "command.update"
+	MsgTreeList         MessageType = "tree.list" // sync RPC: list a snapshot's children
+	MsgSnapshotsRefresh MessageType = "snapshots.refresh"
 )
 
 // Envelope is the framing for every WS message in either direction.

internal/server/http/server.go

@@ -262,6 +262,7 @@ func (s *Server) routes(r chi.Router) {
 		r.Post("/api/hosts/{id}/repo/unlock", s.handleRunRepoUnlock)
 		r.Post("/api/jobs/{id}/cancel", s.handleCancelJob)
 		r.Post("/api/hosts/{id}/snapshots/diff", s.handleSnapshotDiff)
+		r.Post("/api/hosts/{id}/snapshots/refresh", s.handleRefreshHostSnapshots)
 
 		// HTMX form variants outside /api.
 		r.Post("/hosts/{id}/snapshots/diff", s.handleSnapshotDiff)

internal/server/http/snapshots.go

@@ -5,6 +5,10 @@ import (
 	"time"
 
 	"github.com/go-chi/chi/v5"
+	"github.com/oklog/ulid/v2"
+
+	"gitea.dcglab.co.uk/steve/restic-manager/internal/api"
+	"gitea.dcglab.co.uk/steve/restic-manager/internal/store"
 )
 
 // snapshotView is the public JSON shape for a snapshot. Matches the
@@ -26,6 +30,7 @@ type listSnapshotsResponse struct {
 	HostID      string         `json:"host_id"`
 	Count       int            `json:"count"`
 	RefreshedAt *time.Time     `json:"refreshed_at,omitempty"`
+	Stale       bool           `json:"stale"`
 	Snapshots   []snapshotView `json:"snapshots"`
 }
 
@@ -45,7 +50,8 @@ func (s *Server) handleListHostSnapshots(w stdhttp.ResponseWriter, r *stdhttp.Re
 		return
 	}
 
-	if _, err := s.deps.Store.GetHost(r.Context(), hostID); err != nil {
+	host, err := s.deps.Store.GetHost(r.Context(), hostID)
+	if err != nil {
 		writeJSONError(w, stdhttp.StatusNotFound, "host_not_found", "")
 		return
 	}
@@ -61,10 +67,13 @@ func (s *Server) handleListHostSnapshots(w stdhttp.ResponseWriter, r *stdhttp.Re
 		Count:     len(snaps),
 		Snapshots: make([]snapshotView, len(snaps)),
 	}
-	if len(snaps) > 0 {
-		t := snaps[0].RefreshedAt
-		out.RefreshedAt = &t
+	out.RefreshedAt = host.SnapshotRefreshedAt
+	mutationAt, err := s.deps.Store.LatestSuccessfulRepoMutation(r.Context(), hostID)
+	if err != nil {
+		writeJSONError(w, stdhttp.StatusInternalServerError, "internal", "")
+		return
 	}
+	out.Stale = mutationAt != nil && (out.RefreshedAt == nil || out.RefreshedAt.Before(*mutationAt))
 	for i, sn := range snaps {
 		out.Snapshots[i] = snapshotView{
 			ID:        sn.ID,
@@ -80,3 +89,31 @@ func (s *Server) handleListHostSnapshots(w stdhttp.ResponseWriter, r *stdhttp.Re
 
 	writeJSON(w, stdhttp.StatusOK, out)
 }
+
+func (s *Server) handleRefreshHostSnapshots(w stdhttp.ResponseWriter, r *stdhttp.Request) {
+	user, ok := s.requireUser(r)
+	if !ok {
+		writeJSONError(w, stdhttp.StatusUnauthorized, "unauthorised", "")
+		return
+	}
+	hostID := chi.URLParam(r, "id")
+	if _, err := s.deps.Store.GetHost(r.Context(), hostID); err != nil {
+		writeJSONError(w, stdhttp.StatusNotFound, "host_not_found", "")
+		return
+	}
+	if s.deps.Hub == nil || !s.deps.Hub.Connected(hostID) {
+		writeJSONError(w, stdhttp.StatusConflict, "host_offline", "agent is not currently connected")
+		return
+	}
+	env, _ := api.Marshal(api.MsgSnapshotsRefresh, ulid.Make().String(), nil)
+	if err := s.deps.Hub.Send(r.Context(), hostID, env); err != nil {
+		writeJSONError(w, stdhttp.StatusConflict, "host_offline", err.Error())
+		return
+	}
+	now := time.Now().UTC()
+	_ = s.deps.Store.AppendAudit(r.Context(), store.AuditEntry{
+		ID: ulid.Make().String(), UserID: &user.ID, Actor: "user",
+		Action: "host.snapshots_refresh", TargetKind: ptr("host"), TargetID: &hostID, TS: now,
+	})
+	writeJSON(w, stdhttp.StatusAccepted, map[string]string{"status": "refresh_requested"})
+}

internal/server/http/snapshots_test.go

@@ -0,0 +1,83 @@
+package http
+
+import (
+	"context"
+	"encoding/json"
+	stdhttp "net/http"
+	"testing"
+	"time"
+
+	"gitea.dcglab.co.uk/steve/restic-manager/internal/api"
+	"gitea.dcglab.co.uk/steve/restic-manager/internal/store"
+)
+
+func TestSnapshotsFreshnessAndExplicitRefresh(t *testing.T) {
+	t.Parallel()
+	srv, ts, st := rawTestServerWithUI(t)
+	hostID, token := enrolHostForUI(t, srv, st, "snapshot-refresh-host")
+	c := agentDial(t, srv, ts, hostID, token)
+	sendHello(t, c, "snapshot-refresh-host")
+	_ = drainUntil(t, c, api.MsgScheduleSet)
+	cookie := loginAsAdmin(t, st)
+
+	mutationAt := time.Now().UTC().Add(-time.Minute).Truncate(time.Millisecond)
+	if err := st.CreateJob(context.Background(), store.Job{
+		ID: "mutation-job", HostID: hostID, Kind: "forget", ActorKind: "user", CreatedAt: mutationAt.Add(-time.Minute),
+	}); err != nil {
+		t.Fatalf("create mutation: %v", err)
+	}
+	if err := st.MarkJobFinished(context.Background(), "mutation-job", "succeeded", 0, nil, "", mutationAt); err != nil {
+		t.Fatalf("finish mutation: %v", err)
+	}
+
+	get := func() listSnapshotsResponse {
+		req, _ := stdhttp.NewRequest(stdhttp.MethodGet, ts.URL+"/api/hosts/"+hostID+"/snapshots", nil)
+		req.AddCookie(cookie)
+		res, err := stdhttp.DefaultClient.Do(req)
+		if err != nil {
+			t.Fatalf("get snapshots: %v", err)
+		}
+		defer res.Body.Close()
+		var body listSnapshotsResponse
+		if err := json.NewDecoder(res.Body).Decode(&body); err != nil {
+			t.Fatalf("decode snapshots: %v", err)
+		}
+		return body
+	}
+	if body := get(); !body.Stale || body.RefreshedAt != nil {
+		t.Fatalf("unrefreshed projection should be stale: %+v", body)
+	}
+
+	refreshedAt := mutationAt.Add(time.Second)
+	if err := st.ReplaceHostSnapshots(context.Background(), hostID, nil, refreshedAt); err != nil {
+		t.Fatalf("replace empty: %v", err)
+	}
+	if body := get(); body.Stale || body.RefreshedAt == nil || !body.RefreshedAt.Equal(refreshedAt) {
+		t.Fatalf("fresh empty projection reported incorrectly: %+v", body)
+	}
+
+	req, _ := stdhttp.NewRequest(stdhttp.MethodPost, ts.URL+"/api/hosts/"+hostID+"/snapshots/refresh", nil)
+	req.AddCookie(cookie)
+	res, err := stdhttp.DefaultClient.Do(req)
+	if err != nil {
+		t.Fatalf("request refresh: %v", err)
+	}
+	defer res.Body.Close()
+	if res.StatusCode != stdhttp.StatusAccepted {
+		t.Fatalf("refresh status = %d, want 202", res.StatusCode)
+	}
+
+	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Second)
+	defer cancel()
+	_, raw, err := c.Read(ctx)
+	if err != nil {
+		t.Fatalf("read refresh command: %v", err)
+	}
+	var env api.Envelope
+	if err := json.Unmarshal(raw, &env); err != nil {
+		t.Fatalf("decode envelope: %v", err)
+	}
+	if env.Type != api.MsgSnapshotsRefresh {
+		t.Fatalf("message type = %q, want %q", env.Type, api.MsgSnapshotsRefresh)
+	}
+}

internal/store/hosts.go

@@ -44,7 +44,7 @@ func (s *Store) LookupHostByAgentToken(ctx context.Context, tokenHash string) (*
 			repo_size_bytes, snapshot_count, open_alert_count,
 			applied_schedule_version, bandwidth_up_kbps, bandwidth_down_kbps,
 			pre_hook_default, post_hook_default,
-			repo_status, repo_status_error, always_on
+			repo_status, repo_status_error, always_on, snapshot_refreshed_at
 		 FROM hosts WHERE agent_token_hash = ?`,
 		tokenHash)
 	return scanHost(row)
@@ -59,7 +59,7 @@ func (s *Store) GetHost(ctx context.Context, id string) (*Host, error) {
 			repo_size_bytes, snapshot_count, open_alert_count,
 			applied_schedule_version, bandwidth_up_kbps, bandwidth_down_kbps,
 			pre_hook_default, post_hook_default,
-			repo_status, repo_status_error, always_on
+			repo_status, repo_status_error, always_on, snapshot_refreshed_at
 		 FROM hosts WHERE id = ?`, id)
 	return scanHost(row)
 }
@@ -227,7 +227,7 @@ func (s *Store) ListHosts(ctx context.Context) ([]Host, error) {
 			repo_size_bytes, snapshot_count, open_alert_count,
 			applied_schedule_version, bandwidth_up_kbps, bandwidth_down_kbps,
 			pre_hook_default, post_hook_default,
-			repo_status, repo_status_error, always_on
+			repo_status, repo_status_error, always_on, snapshot_refreshed_at
 		 FROM hosts ORDER BY name`)
 	if err != nil {
 		return nil, fmt.Errorf("store: list hosts: %w", err)
@@ -268,6 +268,7 @@ func scanHostRow(s hostScanner) (*Host, error) {
 		bwUp, bwDown                 sql.NullInt64
 		preHook, postHook            sql.NullString
 		alwaysOn                     int
+		snapshotRefreshedAt          sql.NullString
 	)
 	err := s.Scan(&h.ID, &h.Name, &h.OS, &h.Arch,
 		&h.AgentVersion, &h.ResticVersion, &h.ProtocolVersion,
@@ -276,7 +277,7 @@ func scanHostRow(s hostScanner) (*Host, error) {
 		&h.RepoSizeBytes, &h.SnapshotCount, &h.OpenAlertCount,
 		&h.AppliedScheduleVersion, &bwUp, &bwDown,
 		&preHook, &postHook,
-		&h.RepoStatus, &h.RepoStatusError, &alwaysOn)
+		&h.RepoStatus, &h.RepoStatusError, &alwaysOn, &snapshotRefreshedAt)
 	if err != nil {
 		if errors.Is(err, sql.ErrNoRows) {
 			return nil, ErrNotFound
@@ -332,6 +333,13 @@ func scanHostRow(s hostScanner) (*Host, error) {
 		h.PostHookDefault = postHook.String
 	}
 	h.AlwaysOn = alwaysOn != 0
+	if snapshotRefreshedAt.Valid {
+		t, err := time.Parse(time.RFC3339Nano, snapshotRefreshedAt.String)
+		if err != nil {
+			return nil, fmt.Errorf("store: parse snapshot_refreshed_at: %w", err)
+		}
+		h.SnapshotRefreshedAt = &t
+	}
 	return &h, nil
 }
 

internal/store/jobs.go

@@ -39,6 +39,27 @@ type RunningJobActivity struct {
 	LastActivity time.Time
 }
 
+// LatestSuccessfulRepoMutation returns the newest completion time for a job
+// that can change the repository's snapshot projection.
+func (s *Store) LatestSuccessfulRepoMutation(ctx context.Context, hostID string) (*time.Time, error) {
+	var raw sql.NullString
+	err := s.db.QueryRowContext(ctx, `
+		SELECT MAX(finished_at) FROM jobs
+		WHERE host_id = ? AND status = 'succeeded'
+		  AND kind IN ('backup', 'forget', 'prune')`, hostID).Scan(&raw)
+	if err != nil {
+		return nil, fmt.Errorf("store: latest successful repo mutation: %w", err)
+	}
+	if !raw.Valid {
+		return nil, nil
+	}
+	t, err := time.Parse(time.RFC3339Nano, raw.String)
+	if err != nil {
+		return nil, fmt.Errorf("store: parse latest repo mutation: %w", err)
+	}
+	return &t, nil
+}
+
 // ListRunningJobActivity returns every running job with its latest persisted
 // activity. Jobs without started_at are excluded because they have not actually
 // entered the running state coherently and cannot be aged safely here.

internal/store/migrations/0025_hosts_snapshot_refreshed_at.sql

@@ -0,0 +1 @@
+ALTER TABLE hosts ADD COLUMN snapshot_refreshed_at TEXT;

internal/store/snapshots.go

@@ -69,8 +69,8 @@ func (s *Store) ReplaceHostSnapshots(ctx context.Context, hostID string, snaps [
 	}
 
 	if _, err := tx.ExecContext(ctx,
-		`UPDATE hosts SET snapshot_count = ? WHERE id = ?`,
-		len(snaps), hostID); err != nil {
+		`UPDATE hosts SET snapshot_count = ?, snapshot_refreshed_at = ? WHERE id = ?`,
+		len(snaps), when.UTC().Format(time.RFC3339Nano), hostID); err != nil {
 		return fmt.Errorf("store: update host snapshot_count: %w", err)
 	}
 

internal/store/snapshots_test.go

@@ -138,7 +138,8 @@ func TestReplaceHostSnapshotsEmpty(t *testing.T) {
 		t.Fatalf("replace 1: %v", err)
 	}
 	// Then empty — host has been wiped.
-	if err := s.ReplaceHostSnapshots(ctx, hostID, nil, time.Now().UTC()); err != nil {
+	refreshedAt := time.Now().UTC().Truncate(time.Millisecond)
+	if err := s.ReplaceHostSnapshots(ctx, hostID, nil, refreshedAt); err != nil {
 		t.Fatalf("replace empty: %v", err)
 	}
 	out, err := s.ListSnapshotsByHost(ctx, hostID)
@@ -152,4 +153,7 @@ func TestReplaceHostSnapshotsEmpty(t *testing.T) {
 	if h.SnapshotCount != 0 {
 		t.Errorf("snapshot_count should reset to 0, got %d", h.SnapshotCount)
 	}
+	if h.SnapshotRefreshedAt == nil || !h.SnapshotRefreshedAt.Equal(refreshedAt) {
+		t.Errorf("empty projection refresh time = %v, want %s", h.SnapshotRefreshedAt, refreshedAt)
+	}
 }

internal/store/types.go

@@ -77,6 +77,7 @@ type Host struct {
 	LastBackupStatus       *string
 	RepoSizeBytes          int64
 	SnapshotCount          int
+	SnapshotRefreshedAt    *time.Time
 	OpenAlertCount         int
 	AppliedScheduleVersion int64
 	// Host-wide bandwidth caps applied to every restic invocation