merge master (count-caps #263) into scale-to-zero

# Conflicts:
#	internal/config/config.go
#	internal/config/config_test.go

Author: mastermanas805

internal/config/config.go

@@ -206,6 +206,17 @@ type Config struct {
 	// Enabling it is an operator action (see infra runbook) after a canary.
 	DeployScaleToZeroEnabled bool
 
+	// ResourceCountCapsEnabled gates per-service resource-count enforcement
+	// (Task #55). Default FALSE: when off, the count-check block in every
+	// provision handler (db/vector/cache/nosql/storage) is skipped entirely —
+	// zero behavior change, so shipping the caps cannot surprise-break an
+	// existing heavy tenant with a 402. When on, each handler counts the team's
+	// active resources of that type and rejects over-cap provisions with 402 +
+	// agent_action, mirroring the always-on queue_count cap. Enabling it is an
+	// operator action (kubectl set env RESOURCE_COUNT_CAPS_ENABLED=true) after a
+	// usage audit so no current tenant is over the new per-tier caps.
+	ResourceCountCapsEnabled bool
+
 	// GitHub App (P4) — install-once push-to-deploy + short-lived installation
 	// tokens for private-repo clones. Distinct from the GitHub OAuth *login* app
 	// above (GitHubClientID/Secret). GitHubAppEnabled gates the whole feature:
@@ -520,6 +531,17 @@ func Load() *Config {
 		cfg.DeployScaleToZeroEnabled = false
 	}
 
+	// RESOURCE_COUNT_CAPS_ENABLED: default FALSE (Task #55). Off → the per-service
+	// count-check block in every provision handler is skipped (zero behavior
+	// change). On → over-cap provisions get 402. Operator action after a usage
+	// audit so no current tenant is retroactively over a new per-tier cap.
+	switch strings.ToLower(strings.TrimSpace(os.Getenv("RESOURCE_COUNT_CAPS_ENABLED"))) {
+	case "true", "1", "yes":
+		cfg.ResourceCountCapsEnabled = true
+	default:
+		cfg.ResourceCountCapsEnabled = false
+	}
+
 	// GITHUB_APP_ENABLED: default FALSE (off until the operator registers the
 	// App and provisions GITHUB_APP_* secrets — see infra/GITHUB-APP-RUNBOOK.md).
 	switch strings.ToLower(strings.TrimSpace(os.Getenv("GITHUB_APP_ENABLED"))) {

internal/config/config_test.go

@@ -64,6 +64,7 @@ func allKeys() []string {
 		"DELETION_CONFIRMATION_TTL_MINUTES", "FAMILY_BINDINGS_ENABLED",
 		"DEPLOY_SOURCE_IMAGE_ENABLED", "DEPLOY_SOURCE_GIT_ENABLED",
 		"DEPLOY_SCALE_TO_ZERO_ENABLED",
+		"RESOURCE_COUNT_CAPS_ENABLED",
 		"GITHUB_APP_ENABLED", "GITHUB_APP_ID", "GITHUB_APP_SLUG", "GITHUB_APP_PRIVATE_KEY",
 		"GITHUB_APP_WEBHOOK_SECRET", "GITHUB_APP_CLIENT_ID", "GITHUB_APP_CLIENT_SECRET",
 		"BREVO_WEBHOOK_SECRET", "SES_SNS_SUBSCRIPTION_ARN",
@@ -403,6 +404,21 @@ func TestLoad_DeployScaleToZeroEnabled(t *testing.T) {
 	}
 }
 
+func TestLoad_ResourceCountCapsEnabled(t *testing.T) {
+	for _, val := range []string{"true", "1", "yes", "TRUE", "  Yes  "} {
+		applyBaselineEnv(t, map[string]string{"RESOURCE_COUNT_CAPS_ENABLED": val})
+		if !Load().ResourceCountCapsEnabled {
+			t.Errorf("RESOURCE_COUNT_CAPS_ENABLED=%q should enable", val)
+		}
+	}
+	for _, val := range []string{"false", "0", "no", "maybe", ""} {
+		applyBaselineEnv(t, map[string]string{"RESOURCE_COUNT_CAPS_ENABLED": val})
+		if Load().ResourceCountCapsEnabled {
+			t.Errorf("RESOURCE_COUNT_CAPS_ENABLED=%q should stay disabled (default OFF)", val)
+		}
+	}
+}
+
 func TestLoad_GitHubAppEnabled(t *testing.T) {
 	// When enabling the App, Load() fails closed unless the webhook secret +
 	// private key + app id are present (review HIGH-1), so set them here.

internal/handlers/billing_usage.go

@@ -68,6 +68,13 @@ type usageMetric struct {
 	LimitBytes int64 `json:"limit_bytes,omitempty"`
 	Count      int   `json:"count,omitempty"`
 	Limit      int   `json:"limit,omitempty"`
+	// CountLimit is the per-tier resource-COUNT cap (Task #55) for the
+	// byte-metered storage services (postgres/redis/mongodb), where `Limit`
+	// already carries no value (those use LimitBytes). It lets the dashboard
+	// show "3 / 5 databases" alongside "120 MB / 1024 MB". For the
+	// count-metered services (deployments/webhooks/vault/members) the existing
+	// Count/Limit pair is unchanged. -1 means unlimited.
+	CountLimit int `json:"count_limit,omitempty"`
 }
 
 // GetUsage handles GET /api/v1/billing/usage.
@@ -152,9 +159,15 @@ func (h *BillingUsageHandler) computeUsage(ctx context.Context, teamID uuid.UUID
 			return usageSummary{}, sumErr
 		}
 		limitMB := h.plans.StorageLimitMB(tier, svc)
+		// Task #55: also surface the active-resource COUNT + per-tier count cap
+		// so the dashboard can render "3 / 5 databases" next to the byte gauge.
+		// Best-effort: a count error must not fail the byte rows.
+		count, _ := models.CountActiveResourcesByTeamAndType(ctx, h.db, teamID, svc)
 		usage[svc] = usageMetric{
 			Bytes:      bytes,
 			LimitBytes: mbToBytes(limitMB),
+			Count:      count,
+			CountLimit: h.plans.ResourceCountLimit(tier, svc),
 		}
 	}
 

internal/handlers/cache.go

@@ -333,6 +333,12 @@ func (h *CacheHandler) newCacheAuthenticated(
 		tier = "growth"
 	}
 
+	// Task #55: per-tier redis count cap (flag-gated, default OFF). Redis is the
+	// binding COGS constraint ($6.50/GB) so this is the most-conservative cap.
+	if handled, capErr := h.enforceResourceCountCap(c, teamUUID, team.PlanTier, models.ResourceTypeRedis, requestID); handled {
+		return capErr
+	}
+
 	parentRootID, perr := resolveFamilyParent(c, h.db, parentResourceID, teamUUID, models.ResourceTypeRedis, env)
 	if perr != nil {
 		return perr

internal/handlers/capabilities.go

@@ -32,36 +32,43 @@ func NewCapabilitiesHandler(p *plans.Registry) *CapabilitiesHandler {
 }
 
 type tierCapabilities struct {
-	Tier                  string         `json:"tier"`
-	DisplayName           string         `json:"display_name"`
-	PriceUSDMonthly       int            `json:"price_usd_monthly"`
-	PaidFromDayOne        bool           `json:"paid_from_day_one"`
-	StorageLimitMB        map[string]int `json:"storage_limit_mb"`
-	ConnectionsLimit      map[string]int `json:"connections_limit"`
-	Deployments           int            `json:"deployments_apps"`
-	BackupRetentionDays   int            `json:"backup_retention_days"`
-	BackupRestoreEnabled  bool           `json:"backup_restore_enabled"`
-	ManualBackupsPerDay   int            `json:"manual_backups_per_day"`
+	Tier             string         `json:"tier"`
+	DisplayName      string         `json:"display_name"`
+	PriceUSDMonthly  int            `json:"price_usd_monthly"`
+	PaidFromDayOne   bool           `json:"paid_from_day_one"`
+	StorageLimitMB   map[string]int `json:"storage_limit_mb"`
+	ConnectionsLimit map[string]int `json:"connections_limit"`
+	// ResourceCountLimit is the per-service max number of active resources a
+	// team may hold (Task #55). Keyed by the same service strings as
+	// StorageLimitMB. -1 means unlimited; a positive value is the hard cap.
+	// Enforcement is flag-gated (RESOURCE_COUNT_CAPS_ENABLED) — this surface
+	// always advertises the cap so an agent can plan around it even while the
+	// operator hasn't yet flipped enforcement on.
+	ResourceCountLimit   map[string]int `json:"resource_count_limit"`
+	Deployments          int            `json:"deployments_apps"`
+	BackupRetentionDays  int            `json:"backup_retention_days"`
+	BackupRestoreEnabled bool           `json:"backup_restore_enabled"`
+	ManualBackupsPerDay  int            `json:"manual_backups_per_day"`
 	// RPOMinutes / RTOMinutes — FIX-H #Q50 (B36). 0 means
 	// "not promised" (no scheduled backups / no self-serve restore on
 	// the tier). Lets an agent reason about durability requirements
 	// per-tier without a second round-trip.
-	RPOMinutes            int            `json:"rpo_minutes"`
-	RTOMinutes            int            `json:"rto_minutes"`
-	AnnualDiscountPercent int            `json:"annual_discount_percent"`
+	RPOMinutes            int `json:"rpo_minutes"`
+	RTOMinutes            int `json:"rto_minutes"`
+	AnnualDiscountPercent int `json:"annual_discount_percent"`
 	// UpgradeURL — pointer so the terminal tier (Team — there is nothing
 	// to upgrade to) emits an explicit JSON `null` instead of the pricing
 	// URL. DOG-26 (QA 2026-05-29): every tier including Team used to
 	// return the pricing page, which would have SDKs / dashboards
 	// rendering an "Upgrade" CTA on the Team plan with no destination.
 	// `null` is the contract-stable terminal-tier marker; a non-null
 	// string is the "click here to upgrade" signal.
-	UpgradeURL            *string        `json:"upgrade_url"`
+	UpgradeURL *string `json:"upgrade_url"`
 	// IsTerminalTier — explicit boolean so clients don't have to encode
 	// the "is upgrade_url null" check at every render site. True for the
 	// top tier (Team today), false for everything below. Pairs with
 	// UpgradeURL — when IsTerminalTier=true, UpgradeURL is null.
-	IsTerminalTier        bool           `json:"is_terminal_tier"`
+	IsTerminalTier bool `json:"is_terminal_tier"`
 }
 
 // capabilityResourceTypes is the list of service types the /capabilities
@@ -71,6 +78,13 @@ var capabilityResourceTypes = []string{
 	"postgres", "redis", "mongodb", "queue", "storage", "webhook", "vector",
 }
 
+// countCapResourceTypes is the set of services that carry a per-tier
+// resource-COUNT cap (Task #55). Webhook is omitted — it is byte/request-capped
+// via webhook_requests_stored, not count-capped. Order is contract-stable.
+var countCapResourceTypes = []string{
+	"postgres", "vector", "redis", "mongodb", "storage", "queue",
+}
+
 // upgradeURL is the marketing pricing page that every tier row in the
 // /capabilities response points back to. Hoisted to a package const so
 // the URL fragment isn't scattered as a string literal across the handler.
@@ -158,10 +172,17 @@ func (h *CapabilitiesHandler) Get(c *fiber.Ctx) error {
 	for _, e := range entries {
 		storage := map[string]int{}
 		conns := map[string]int{}
+		counts := map[string]int{}
 		for _, rt := range capabilityResourceTypes {
 			storage[rt] = h.plans.StorageLimitMB(e.name, rt)
 			conns[rt] = h.plans.ConnectionsLimit(e.name, rt)
 		}
+		// Task #55: per-service resource-count caps. Only the count-capped
+		// services appear (webhook is byte-capped via webhook_requests_stored,
+		// not count-capped). ResourceCountLimit returns -1 for unlimited.
+		for _, rt := range countCapResourceTypes {
+			counts[rt] = h.plans.ResourceCountLimit(e.name, rt)
+		}
 		priceUSD := e.plan.PriceMonthly / 100 // cents → dollars
 		// DOG-26: terminal tier marker — top of the rank ladder has
 		// nothing to upgrade to. upgrade_url is null + is_terminal_tier
@@ -179,6 +200,7 @@ func (h *CapabilitiesHandler) Get(c *fiber.Ctx) error {
 			PaidFromDayOne:        priceUSD > 0,
 			StorageLimitMB:        storage,
 			ConnectionsLimit:      conns,
+			ResourceCountLimit:    counts,
 			Deployments:           h.plans.DeploymentsAppsLimit(e.name),
 			BackupRetentionDays:   h.plans.BackupRetentionDays(e.name),
 			BackupRestoreEnabled:  h.plans.BackupRestoreEnabled(e.name),

internal/handlers/capabilities_test.go

@@ -34,6 +34,7 @@ type capabilityTier struct {
 	PaidFromDayOne        bool           `json:"paid_from_day_one"`
 	StorageLimitMB        map[string]int `json:"storage_limit_mb"`
 	ConnectionsLimit      map[string]int `json:"connections_limit"`
+	ResourceCountLimit    map[string]int `json:"resource_count_limit"`
 	Deployments           int            `json:"deployments_apps"`
 	BackupRetentionDays   int            `json:"backup_retention_days"`
 	BackupRestoreEnabled  bool           `json:"backup_restore_enabled"`
@@ -214,6 +215,44 @@ func TestCapabilities_LimitsResolveFromRegistry(t *testing.T) {
 	assert.Equal(t, 5, hp.ManualBackupsPerDay, "hobby_plus manual backups/day")
 }
 
+// TestCapabilities_SurfacesResourceCountLimit is the Task #55 rule-18 surface
+// guard: GET /api/v1/capabilities must expose resource_count_limit for EVERY
+// count-capped service on every paid tier, with the value matching the live
+// registry. Iterates the registry rather than hand-typing tiers so a new tier or
+// service can't silently ship without the cap appearing on the public matrix.
+func TestCapabilities_SurfacesResourceCountLimit(t *testing.T) {
+	reg := plans.Default()
+	app := newCapabilitiesApp(t, reg)
+	_, body := callCapabilities(t, app)
+	require.NotEmpty(t, body.Tiers)
+
+	countServices := []string{"postgres", "vector", "redis", "mongodb", "storage", "queue"}
+	for _, tier := range body.Tiers {
+		require.NotNil(t, tier.ResourceCountLimit,
+			"tier %q must carry resource_count_limit", tier.Tier)
+		for _, svc := range countServices {
+			got, ok := tier.ResourceCountLimit[svc]
+			require.True(t, ok, "tier %q resource_count_limit must include %q", tier.Tier, svc)
+			assert.Equal(t, reg.ResourceCountLimit(tier.Tier, svc), got,
+				"tier %q %s count limit must match the registry", tier.Tier, svc)
+		}
+		// Webhook is request-capped, not count-capped — must NOT appear.
+		_, hasWebhook := tier.ResourceCountLimit["webhook"]
+		assert.False(t, hasWebhook, "webhook must not appear in resource_count_limit (it is request-capped)")
+	}
+
+	// Spot-pin a couple of binding values so a loosened cap is a visible diff.
+	for _, tier := range body.Tiers {
+		switch tier.Tier {
+		case "pro":
+			assert.Equal(t, 3, tier.ResourceCountLimit["redis"], "pro redis_count")
+			assert.Equal(t, 5, tier.ResourceCountLimit["postgres"], "pro postgres_count")
+		case "team":
+			assert.Equal(t, 4, tier.ResourceCountLimit["redis"], "team redis_count")
+		}
+	}
+}
+
 // TestCapabilities_PlansUnavailable — when the registry pointer is nil
 // (boot-time failure in dev with no fallback), the handler must return
 // 503 instead of panicking. Lifted contract from the original handler.

internal/handlers/db.go

@@ -381,6 +381,12 @@ func (h *DBHandler) newDBAuthenticated(
 		tier = "growth"
 	}
 
+	// Task #55: per-tier postgres count cap (flag-gated, default OFF — inert
+	// unless RESOURCE_COUNT_CAPS_ENABLED). Mirrors queue.go's A6 block.
+	if handled, capErr := h.enforceResourceCountCap(c, teamUUID, team.PlanTier, models.ResourceTypePostgres, requestID); handled {
+		return capErr
+	}
+
 	// Family-link validation runs BEFORE provisioning so a cross-team /
 	// cross-type / duplicate-twin parent_resource_id never causes us to
 	// create-then-fail (which would leak a database we can't link).

internal/handlers/nosql.go

@@ -326,6 +326,11 @@ func (h *NoSQLHandler) newNoSQLAuthenticated(
 		tier = "growth"
 	}
 
+	// Task #55: per-tier mongodb count cap (flag-gated, default OFF). Mirrors queue.go.
+	if handled, capErr := h.enforceResourceCountCap(c, teamUUID, team.PlanTier, models.ResourceTypeMongoDB, requestID); handled {
+		return capErr
+	}
+
 	parentRootID, perr := resolveFamilyParent(c, h.db, parentResourceID, teamUUID, models.ResourceTypeMongoDB, env)
 	if perr != nil {
 		return perr

internal/handlers/openapi.go

@@ -3292,8 +3292,9 @@ const openAPISpec = `{
         "properties": {
           "bytes":       { "type": "integer", "format": "int64", "description": "Current storage usage in bytes. Present on postgres/redis/mongodb." },
           "limit_bytes": { "type": "integer", "format": "int64", "description": "Storage cap in bytes (plans.yaml storage_mb × 1024 × 1024). -1 = unlimited." },
-          "count":       { "type": "integer", "description": "Current count. Present on deployments/webhooks/vault/members." },
-          "limit":       { "type": "integer", "description": "Count cap from plans.yaml. -1 = unlimited." }
+          "count":       { "type": "integer", "description": "Current count. Present on deployments/webhooks/vault/members, and (Task #55) on postgres/redis/mongodb as the active-resource count alongside bytes." },
+          "limit":       { "type": "integer", "description": "Count cap from plans.yaml. -1 = unlimited." },
+          "count_limit": { "type": "integer", "description": "Task #55: per-tier resource-COUNT cap for the byte-metered storage services (postgres/redis/mongodb), where the limit field is unused. -1 = unlimited. Enforcement is flag-gated (RESOURCE_COUNT_CAPS_ENABLED) but the cap is always advertised." }
         }
       },
       "TeamSummaryResponse": {
@@ -3383,6 +3384,7 @@ const openAPISpec = `{
           "paid_from_day_one":       { "type": "boolean", "description": "True iff price_usd_monthly > 0. Mirrors project policy: no trial — paid tiers are paid from signup." },
           "storage_limit_mb":        { "type": "object", "additionalProperties": { "type": "integer" }, "description": "Per-service storage cap in MB. Keys: postgres, redis, mongodb, queue, storage, webhook, vector. -1 sentinel means 'unlimited'." },
           "connections_limit":       { "type": "object", "additionalProperties": { "type": "integer" }, "description": "Per-service concurrent-connection cap. Keys mirror storage_limit_mb. -1 = unlimited." },
+          "resource_count_limit":    { "type": "object", "additionalProperties": { "type": "integer" }, "description": "Task #55: per-service max number of active resources a team may hold. Keys: postgres, vector, redis, mongodb, storage, queue (webhook is request-capped, not count-capped). -1 = unlimited. Enforcement is flag-gated (RESOURCE_COUNT_CAPS_ENABLED) but the cap is always advertised so an agent can plan around it." },
           "deployments_apps":        { "type": "integer", "description": "Max number of /deploy/new apps allowed. -1 = unlimited." },
           "backup_retention_days":   { "type": "integer" },
           "backup_restore_enabled":  { "type": "boolean" },