Add per-agent log retention floor, owner-only to set or override

api/agents.ConfigOverride gains LogRetentionDays: a per-agent setting
edited on the same remote-config page as extra_file_paths, but unlike
every other field there it's central-policy metadata api/logretention
reads, never something the agent process itself sees. Any change to
it -- setting, raising, lowering, or clearing -- requires RoleOwner,
not just RoleAdmin: the whole point of the field is a floor an admin
can't move, so an admin able to freely edit it would defeat that.

api/logretention now checks the largest LogRetentionDays configured
across any agent (AgentRetentionStore, new) before every preview/delete:
a non-owner's request is rejected with a clear 403 if it would reach
into that protected window. An owner always bypasses it, matching "make
the log retention override any attempts to delete logs by anyone other
than owner role."

Verified live end-to-end: owner sets a 90-day floor on an agent, admin
is blocked deleting anything newer than that (both preview and delete),
allowed beyond it, and owner bypasses it entirely -- confirmed against
real ClickHouse data, not just the fake-backed unit tests. Also caught
and fixed a real pre-existing latent bug while verifying in-browser: a
type="number" Input's bind:value becomes an actual JS number once a
user types into it (only the initial value is a string), which broke a
bare .trim() call on the new field.

api/logretention/agent_floor.go

@@ -0,0 +1,46 @@
+package logretention
+
+import (
+	"context"
+
+	"github.com/jackc/pgx/v5/pgxpool"
+)
+
+// AgentRetentionStore reads the protective retention floor set on
+// agents.ConfigOverride.LogRetentionDays -- a separate, Postgres-backed
+// concern from Store's ClickHouse access above, so it lives in its own
+// file. Deliberately its own narrow query against the same `agents`
+// table api/agents.Store manages, rather than importing api/agents for
+// a shared type, matching this codebase's "each package owns direct
+// SQL access to what it needs" convention (e.g. alerting and api both
+// read dashboards-adjacent tables independently rather than sharing a
+// store type across a package boundary).
+type AgentRetentionStore struct {
+	pool *pgxpool.Pool
+}
+
+func NewAgentRetentionStore(pool *pgxpool.Pool) *AgentRetentionStore {
+	return &AgentRetentionStore{pool: pool}
+}
+
+// MaxRetentionDays reports the largest log_retention_days configured
+// across any agent's desired_override, if any are set at all -- this is
+// the floor a non-owner's deletion request must not reach into (see
+// Handler.checkRetentionFloor). The second return value is false when
+// no agent has this field configured, distinct from a configured floor
+// of 0 (which validateOverride never allows to be stored in the first
+// place).
+func (s *AgentRetentionStore) MaxRetentionDays(ctx context.Context) (int, bool, error) {
+	var days *int
+	err := s.pool.QueryRow(ctx, `
+		SELECT max((desired_override->>'log_retention_days')::int)
+		FROM agents
+		WHERE desired_override->>'log_retention_days' IS NOT NULL`).Scan(&days)
+	if err != nil {
+		return 0, false, err
+	}
+	if days == nil {
+		return 0, false, nil
+	}
+	return *days, true, nil
+}

api/logretention/handler.go

@@ -3,6 +3,7 @@ package logretention
 import (
 	"context"
 	"encoding/json"
+	"fmt"
 	"log/slog"
 	"net/http"
 	"strconv"
@@ -19,6 +20,13 @@ type store interface {
 	DeleteOlderThan(ctx context.Context, cutoff time.Time) error
 }
 
+// retentionFloor is the narrow interface backing the owner-only
+// override check -- *AgentRetentionStore (agent_floor.go) is the
+// production implementation.
+type retentionFloor interface {
+	MaxRetentionDays(ctx context.Context) (int, bool, error)
+}
+
 // maxOlderThanHours bounds the age a caller can specify -- 10 years is
 // far beyond any real retention window this feature exists for, and
 // exists only to reject an obviously-wrong input (e.g. a stray extra
@@ -28,11 +36,12 @@ const maxOlderThanHours = 10 * 365 * 24
 type Handler struct {
 	logger     *slog.Logger
 	store      store
+	floor      retentionFloor
 	authorizer authz.Authorizer
 }
 
-func NewHandler(logger *slog.Logger, store store, authorizer authz.Authorizer) *Handler {
-	return &Handler{logger: logger, store: store, authorizer: authorizer}
+func NewHandler(logger *slog.Logger, store store, floor retentionFloor, authorizer authz.Authorizer) *Handler {
+	return &Handler{logger: logger, store: store, floor: floor, authorizer: authorizer}
 }
 
 // RegisterRoutes: both routes are RoleAdmin -- RoleOwner satisfies it
@@ -58,6 +67,37 @@ func parseOlderThanHours(r *http.Request) (int, bool) {
 	return hours, true
 }
 
+// checkRetentionFloor enforces api/agents.ConfigOverride.LogRetentionDays
+// as a hard floor against anyone but an owner: if any agent has a
+// configured retention, the largest one across all agents is the
+// earliest boundary a non-owner may delete up to. An owner always
+// bypasses this (identity.Role == RoleOwner short-circuits before ever
+// querying the floor) -- "owner and admin" gates the routes themselves
+// (RegisterRoutes), but this narrows what admin specifically can do
+// once inside them, the same shape handleSetConfig's own
+// log_retention_days gate uses on the agents side. A nil identity (no
+// authorizer configured at all, Phase 0-3 default-open) skips this
+// too, consistent with every other RBAC check in this codebase being a
+// no-op when there's no RBAC to begin with.
+func (h *Handler) checkRetentionFloor(ctx context.Context, cutoff time.Time) (string, error) {
+	identity, ok := authz.IdentityFromContext(ctx)
+	if !ok || identity.Role == authz.RoleOwner {
+		return "", nil
+	}
+	maxDays, hasFloor, err := h.floor.MaxRetentionDays(ctx)
+	if err != nil {
+		return "", err
+	}
+	if !hasFloor {
+		return "", nil
+	}
+	protectedBoundary := time.Now().UTC().Add(-time.Duration(maxDays) * 24 * time.Hour)
+	if cutoff.After(protectedBoundary) {
+		return fmt.Sprintf("a configured log retention policy protects logs newer than %d days; only an owner can override this", maxDays), nil
+	}
+	return "", nil
+}
+
 type previewResponse struct {
 	Count  uint64    `json:"count"`
 	Cutoff time.Time `json:"cutoff"`
@@ -71,6 +111,15 @@ func (h *Handler) handlePreview(w http.ResponseWriter, r *http.Request) {
 	}
 	cutoff := time.Now().UTC().Add(-time.Duration(hours) * time.Hour)
 
+	if msg, err := h.checkRetentionFloor(r.Context(), cutoff); err != nil {
+		h.logger.Error("checking log retention floor", "error", err)
+		writeError(w, http.StatusInternalServerError, "checking retention policy failed")
+		return
+	} else if msg != "" {
+		writeError(w, http.StatusForbidden, msg)
+		return
+	}
+
 	count, err := h.store.CountOlderThan(r.Context(), cutoff)
 	if err != nil {
 		h.logger.Error("counting logs for retention preview", "error", err)
@@ -101,6 +150,15 @@ func (h *Handler) handleDelete(w http.ResponseWriter, r *http.Request) {
 	}
 	cutoff := time.Now().UTC().Add(-time.Duration(hours) * time.Hour)
 
+	if msg, err := h.checkRetentionFloor(r.Context(), cutoff); err != nil {
+		h.logger.Error("checking log retention floor", "error", err)
+		writeError(w, http.StatusInternalServerError, "checking retention policy failed")
+		return
+	} else if msg != "" {
+		writeError(w, http.StatusForbidden, msg)
+		return
+	}
+
 	count, err := h.store.CountOlderThan(r.Context(), cutoff)
 	if err != nil {
 		h.logger.Error("counting logs before retention delete", "error", err)

api/logretention/handler_test.go

@@ -8,6 +8,7 @@ import (
 	"log/slog"
 	"net/http"
 	"net/http/httptest"
+	"strconv"
 	"testing"
 	"time"
 
@@ -50,8 +51,21 @@ func (f fakeAuthorizer) Authorize(*http.Request) (authz.Identity, error) {
 	return authz.Identity{TenantID: "default", UserID: "u1", Role: f.role}, nil
 }
 
+// fakeFloor stands in for AgentRetentionStore -- hasFloor false (the
+// zero value) means no agent has log_retention_days configured, same
+// as every existing test in this file assumed before the floor existed.
+type fakeFloor struct {
+	days     int
+	hasFloor bool
+	err      error
+}
+
+func (f fakeFloor) MaxRetentionDays(context.Context) (int, bool, error) {
+	return f.days, f.hasFloor, f.err
+}
+
 func newTestHandler(s *fakeStore, role authz.Role) *Handler {
-	return NewHandler(discardLogger(), s, fakeAuthorizer{role: role})
+	return NewHandler(discardLogger(), s, fakeFloor{}, fakeAuthorizer{role: role})
 }
 
 func doRequest(t *testing.T, h *Handler, method, path string) *httptest.ResponseRecorder {
@@ -193,7 +207,7 @@ func TestViewerAndEditorAreForbiddenFromRetentionRoutes(t *testing.T) {
 
 func TestRetentionRoutesRequireAuth(t *testing.T) {
 	s := &fakeStore{}
-	h := NewHandler(discardLogger(), s, nil)
+	h := NewHandler(discardLogger(), s, fakeFloor{}, nil)
 
 	// A nil authorizer is Phase 0-3's default-open behavior (see
 	// authz.RequireRole's doc comment) -- confirm that posture applies
@@ -205,3 +219,81 @@ func TestRetentionRoutesRequireAuth(t *testing.T) {
 		t.Fatalf("status with nil authorizer = %d, want 200 (default-open, matches RequireRole elsewhere)", rec.Code)
 	}
 }
+
+// TestAdminBlockedByRetentionFloor is the core regression test for the
+// owner-only override: an agent configured with a 90-day retention
+// floor must block an admin's attempt to delete anything newer than
+// that, on both preview and delete.
+func TestAdminBlockedByRetentionFloor(t *testing.T) {
+	s := &fakeStore{count: 100}
+	h := NewHandler(discardLogger(), s, fakeFloor{days: 90, hasFloor: true}, fakeAuthorizer{role: authz.RoleAdmin})
+
+	// 30 days is newer than the 90-day floor -- must be blocked.
+	preview := doRequest(t, h, "GET", "/logs/retention/preview?older_than_hours="+hoursForDays(30))
+	if preview.Code != http.StatusForbidden {
+		t.Fatalf("preview at 30d against a 90d floor: status = %d, want 403, body=%s", preview.Code, preview.Body.String())
+	}
+	del := doRequest(t, h, "DELETE", "/logs/retention?older_than_hours="+hoursForDays(30))
+	if del.Code != http.StatusForbidden {
+		t.Fatalf("delete at 30d against a 90d floor: status = %d, want 403, body=%s", del.Code, del.Body.String())
+	}
+	if len(s.countedWith) != 0 || len(s.deletedWith) != 0 {
+		t.Error("a blocked request must never reach the store at all")
+	}
+}
+
+// TestAdminAllowedBeyondRetentionFloor confirms the floor only blocks
+// requests that would actually reach into the protected window -- a
+// request older than the floor itself is unaffected by it.
+func TestAdminAllowedBeyondRetentionFloor(t *testing.T) {
+	s := &fakeStore{count: 5}
+	h := NewHandler(discardLogger(), s, fakeFloor{days: 90, hasFloor: true}, fakeAuthorizer{role: authz.RoleAdmin})
+
+	// 120 days is older than the 90-day floor -- must be allowed.
+	del := doRequest(t, h, "DELETE", "/logs/retention?older_than_hours="+hoursForDays(120))
+	if del.Code != http.StatusOK {
+		t.Fatalf("delete at 120d against a 90d floor: status = %d, want 200, body=%s", del.Code, del.Body.String())
+	}
+}
+
+// TestOwnerBypassesRetentionFloor confirms the whole point of the
+// feature: an owner can still delete within a configured retention
+// window that blocks everyone else.
+func TestOwnerBypassesRetentionFloor(t *testing.T) {
+	s := &fakeStore{count: 100}
+	h := NewHandler(discardLogger(), s, fakeFloor{days: 90, hasFloor: true}, fakeAuthorizer{role: authz.RoleOwner})
+
+	del := doRequest(t, h, "DELETE", "/logs/retention?older_than_hours="+hoursForDays(1))
+	if del.Code != http.StatusOK {
+		t.Fatalf("owner deleting within the floor: status = %d, want 200, body=%s", del.Code, del.Body.String())
+	}
+}
+
+// TestNoConfiguredFloorNeverBlocksAdmin confirms the default, common
+// case (no agent has log_retention_days set) behaves exactly as before
+// this feature existed -- fakeFloor{} (hasFloor: false) is what every
+// other test in this file already relies on, this just makes the
+// no-floor-configured case explicit.
+func TestNoConfiguredFloorNeverBlocksAdmin(t *testing.T) {
+	s := &fakeStore{count: 9}
+	h := NewHandler(discardLogger(), s, fakeFloor{hasFloor: false}, fakeAuthorizer{role: authz.RoleAdmin})
+
+	del := doRequest(t, h, "DELETE", "/logs/retention?older_than_hours=1")
+	if del.Code != http.StatusOK {
+		t.Fatalf("status = %d, want 200 with no configured floor", del.Code)
+	}
+}
+
+func TestRetentionFloorCheckPropagatesStoreErrors(t *testing.T) {
+	s := &fakeStore{}
+	h := NewHandler(discardLogger(), s, fakeFloor{err: errors.New("postgres unreachable")}, fakeAuthorizer{role: authz.RoleAdmin})
+
+	rec := doRequest(t, h, "GET", "/logs/retention/preview?older_than_hours=24")
+	if rec.Code != http.StatusInternalServerError {
+		t.Fatalf("status = %d, want 500", rec.Code)
+	}
+}
+
+func hoursForDays(days int) string {
+	return strconv.Itoa(days * 24)
+}