[management] Rework Agent Network endpoint identity and settings bootstrap (#7085)

Store the per-account gateway endpoint as {domain, proxy_address} with a
global unique index on the full hostname; dedicated = (domain ==
proxy_address). Bootstrap becomes an explicit POST carrying exactly one
of proxy_address (server allocates an adjective-noun label beneath it)
or endpoint (claimed verbatim, address-first); provider create loses its
bootstrap side effect. PUT is a full replace with every field required —
the immutable identity fields must be echoed unchanged and a mismatch is
rejected with 422. A guarded DELETE releases the endpoint: refused with
412 while providers exist or a proxy is actively serving the endpoint
hostname (matched case-insensitively); re-creating bootstraps fresh. A
self-addressed pin excludes its address from the account's cluster allow
list, and the live mapping update path now addresses the serving proxy
from the synthesized service. Existing rows are migrated on all three
store engines.

management/server/activity/codes.go

@@ -281,6 +281,9 @@ const (
 	// AccountMetricsPushDisabled indicates that a user disabled metrics push for the account
 	AccountMetricsPushDisabled Activity = 141
 
+	// AgentNetworkSettingsDeleted indicates that a user deleted the Agent Network account settings, releasing the endpoint
+	AgentNetworkSettingsDeleted Activity = 142
+
 	AccountDeleted Activity = 99999
 )
 
@@ -453,6 +456,7 @@ var activityMap = map[Activity]Code{
 	AgentNetworkBudgetRuleDeleted: {"Agent Network budget rule deleted", "agent_network.budget_rule.delete"},
 
 	AgentNetworkSettingsUpdated: {"Agent Network settings updated", "agent_network.settings.update"},
+	AgentNetworkSettingsDeleted: {"Agent Network settings deleted", "agent_network.settings.delete"},
 
 	AccountMetricsPushEnabled:  {"Account metrics push enabled", "account.setting.metrics.push.enable"},
 	AccountMetricsPushDisabled: {"Account metrics push disabled", "account.setting.metrics.push.disable"},

management/server/agentnetwork_budgetrule_realstack_test.go

@@ -68,7 +68,10 @@ func TestAgentNetwork_BudgetRuleCRUD_RealManager(t *testing.T) {
 
 // TestAgentNetwork_UpdateSettings_PreservesImmutableAndTogglesCollection is the
 // GC-1 guard for UpdateSettings: it must apply the collection toggles while
-// preserving the immutable Cluster/Subdomain pinned at bootstrap.
+// preserving the immutable Domain/ProxyAddress assigned at bootstrap. The
+// request echoes the identity fields back — the PUT convention every other
+// endpoint follows — and a request echoing anything else is rejected outright
+// rather than quietly ignored.
 func TestAgentNetwork_UpdateSettings_PreservesImmutableAndTogglesCollection(t *testing.T) {
 	am, _, err := createManager(t)
 	require.NoError(t, err, "createManager must succeed")
@@ -84,7 +87,14 @@ func TestAgentNetwork_UpdateSettings_PreservesImmutableAndTogglesCollection(t *t
 
 	mgr := agentnetwork.NewManager(am.Store, permissions.NewManager(am.Store), am, nil)
 
-	// Creating a provider bootstraps the settings row (cluster + subdomain).
+	// Bootstrap is an explicit settings create; providers have no settings
+	// side effects anymore.
+	before, err := mgr.CreateSettings(ctx, adminUserID, agenttypes.DefaultSettings(accountID), clusterAddr, "")
+	require.NoError(t, err, "CreateSettings must bootstrap the row")
+	require.Equal(t, clusterAddr, before.ProxyAddress, "proxy address pinned at bootstrap")
+	require.NotEmpty(t, before.Domain, "endpoint allocated at bootstrap")
+	assert.False(t, before.EnablePromptCollection, "prompt collection defaults off")
+
 	_, err = mgr.CreateProvider(ctx, adminUserID, &agenttypes.Provider{
 		AccountID:   accountID,
 		ProviderID:  "openai_api",
@@ -93,43 +103,64 @@ func TestAgentNetwork_UpdateSettings_PreservesImmutableAndTogglesCollection(t *t
 		APIKey:      "sk-test",
 		Enabled:     true,
 		Models:      []agenttypes.ProviderModel{{ID: "gpt-5.4"}},
-	}, clusterAddr)
-	require.NoError(t, err, "CreateProvider must bootstrap settings")
-
-	before, err := mgr.GetSettings(ctx, accountID, adminUserID)
-	require.NoError(t, err, "GetSettings must succeed after bootstrap")
-	require.Equal(t, clusterAddr, before.Cluster, "cluster pinned at bootstrap")
-	require.NotEmpty(t, before.Subdomain, "subdomain pinned at bootstrap")
-	assert.False(t, before.EnablePromptCollection, "prompt collection defaults off")
-
-	// A cluster different from the one pinned at bootstrap must be rejected
-	// outright — never silently swapped or ignored.
-	_, err = mgr.UpdateSettings(ctx, adminUserID, &agenttypes.Settings{
-		AccountID:           accountID,
-		Cluster:             "attacker.cluster",
-		EnableLogCollection: true,
 	})
-	require.Error(t, err, "UpdateSettings with a mismatched cluster must fail")
+	require.NoError(t, err, "CreateProvider must succeed")
 
-	// Flipping the toggles works with the pinned cluster echoed back (and
-	// with it omitted); the subdomain is never taken from the request.
+	// Flipping the toggles works when the request echoes the assigned
+	// identity. Retention is echoed too: UpdateSettings takes it verbatim, so
+	// omitting it would zero the account's retention.
 	updated, err := mgr.UpdateSettings(ctx, adminUserID, &agenttypes.Settings{
 		AccountID:              accountID,
-		Cluster:                clusterAddr,
-		Subdomain:              "evil",
+		Domain:                 before.Domain,
+		ProxyAddress:           before.ProxyAddress,
 		EnableLogCollection:    true,
 		EnablePromptCollection: true,
 		RedactPii:              true,
+		AccessLogRetentionDays: before.AccessLogRetentionDays,
 	})
 	require.NoError(t, err, "UpdateSettings must succeed")
-	assert.Equal(t, before.Cluster, updated.Cluster, "cluster is immutable and must be preserved")
-	assert.Equal(t, before.Subdomain, updated.Subdomain, "subdomain is immutable and must be preserved")
+	assert.Equal(t, before.Domain, updated.Domain, "domain is immutable and must be preserved")
+	assert.Equal(t, before.ProxyAddress, updated.ProxyAddress, "proxy address is immutable and must be preserved")
 	assert.True(t, updated.EnableLogCollection, "log collection toggle must apply")
 	assert.True(t, updated.EnablePromptCollection, "prompt collection toggle must apply")
 	assert.True(t, updated.RedactPii, "redact toggle must apply")
+	assert.Equal(t, before.AccessLogRetentionDays, updated.AccessLogRetentionDays, "echoed retention must survive")
+
+	// Neither identity field can be smuggled into the row: a hand-rolled
+	// Settings value carrying a different endpoint or proxy address is
+	// rejected, not silently ignored.
+	for _, tc := range []struct {
+		name         string
+		domain       string
+		proxyAddress string
+	}{
+		{name: "foreign endpoint", domain: "evil.example.com", proxyAddress: before.ProxyAddress},
+		{name: "foreign proxy address", domain: before.Domain, proxyAddress: "attacker.cluster"},
+		{name: "empty identity echo", domain: "", proxyAddress: ""},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			_, err := mgr.UpdateSettings(ctx, adminUserID, &agenttypes.Settings{
+				AccountID:              accountID,
+				Domain:                 tc.domain,
+				ProxyAddress:           tc.proxyAddress,
+				EnableLogCollection:    false,
+				EnablePromptCollection: false,
+				RedactPii:              false,
+				AccessLogRetentionDays: before.AccessLogRetentionDays,
+			})
+			assert.Error(t, err, "a mismatched identity echo must be rejected")
+			assert.ErrorContains(t, err, "immutable", "the rejection must name the immutability rule")
+		})
+	}
+
+	// The rejected updates left the row exactly as the accepted one wrote it.
+	afterRejects, err := mgr.GetSettings(ctx, accountID, adminUserID)
+	require.NoError(t, err, "GetSettings must succeed")
+	assert.True(t, afterRejects.EnablePromptCollection, "a rejected update must not roll back the accepted toggles")
 
 	reloaded, err := am.Store.GetAgentNetworkSettings(ctx, store.LockingStrengthNone, accountID)
 	require.NoError(t, err)
-	assert.Equal(t, before.Cluster, reloaded.Cluster, "persisted cluster unchanged")
+	assert.Equal(t, before.Domain, reloaded.Domain, "persisted domain unchanged")
+	assert.Equal(t, before.ProxyAddress, reloaded.ProxyAddress, "persisted proxy address unchanged")
 	assert.True(t, reloaded.EnablePromptCollection, "persisted prompt collection toggled on")
 }

management/server/agentnetwork_realstack_test.go

@@ -92,6 +92,14 @@ func TestAgentNetwork_ProviderCRUD_FansOutToProxyAndClientPeers(t *testing.T) {
 	// UpdateAccountPeers, which is the path under test.
 	agentMgr := agentnetwork.NewManager(am.Store, permissions.NewManager(am.Store), am, nil)
 
+	_, err = agentMgr.CreateSettings(ctx, adminUserID, agenttypes.DefaultSettings(accountID), clusterAddr, "")
+	require.NoError(t, err, "CreateSettings must bootstrap the endpoint")
+	// The bootstrap itself reconciles and queues updates on both channels;
+	// drain them so the fan-out assertions below can only be satisfied by the
+	// operation under test, not by this leftover.
+	drain(clientCh)
+	drain(proxyCh)
+
 	provider, err := agentMgr.CreateProvider(ctx, adminUserID, &agenttypes.Provider{
 		AccountID:   accountID,
 		ProviderID:  "openai_api",
@@ -100,7 +108,7 @@ func TestAgentNetwork_ProviderCRUD_FansOutToProxyAndClientPeers(t *testing.T) {
 		APIKey:      "sk-test-key",
 		Enabled:     true,
 		Models:      []agenttypes.ProviderModel{{ID: "gpt-5.4"}},
-	}, clusterAddr)
+	})
 	require.NoError(t, err, "CreateProvider must succeed")
 
 	policy, err := agentMgr.CreatePolicy(ctx, adminUserID, &agenttypes.Policy{

management/server/migration/migration_agentnetwork.go

@@ -0,0 +1,112 @@
+package migration
+
+import (
+	"context"
+	"fmt"
+
+	log "github.com/sirupsen/logrus"
+	"gorm.io/gorm"
+)
+
+// agentNetworkSettingsMigration is a local view of the agent_network_settings
+// table spanning both the legacy identity columns (cluster, subdomain) and
+// their replacement (domain, proxy_address), so the migrator can address all
+// four during the reshape without importing the current model.
+type agentNetworkSettingsMigration struct {
+	AccountID    string `gorm:"primaryKey"`
+	Cluster      string
+	Subdomain    string
+	Domain       string `gorm:"type:varchar(255)"`
+	ProxyAddress string `gorm:"type:varchar(255)"`
+}
+
+func (agentNetworkSettingsMigration) TableName() string { return "agent_network_settings" }
+
+// MigrateAgentNetworkSettingsToDomain reshapes agent_network_settings from the
+// legacy (cluster, subdomain) identity columns to (domain, proxy_address):
+// domain becomes `<subdomain>.<cluster>` — the endpoint hostname the old
+// columns derived — and proxy_address becomes the cluster address, preserving
+// which proxy serves the account. Runs before AutoMigrate, which then creates
+// the unique index on the freshly backfilled domain column.
+//
+// A legacy row missing either half cannot be given an endpoint; the old
+// bootstrap always wrote both, so such a row indicates corruption and the
+// migration fails loudly rather than leaving an empty domain to collide with
+// the unique index confusingly.
+//
+// The transaction is real only on sqlite and postgres, where DDL is
+// transactional. MySQL implicitly commits around every ALTER TABLE, so there
+// each step stands alone; what makes an interrupted run resumable on MySQL is
+// that every step is guarded by the schema state it changes — the entry check
+// fires while either legacy column remains, the adds skip existing columns,
+// the backfill and its loud-failure check run only while the legacy cluster
+// column exists (they provably completed before any drop), and each drop
+// skips what is already gone.
+func MigrateAgentNetworkSettingsToDomain(ctx context.Context, db *gorm.DB) error {
+	model := &agentNetworkSettingsMigration{}
+	migrator := db.Migrator()
+
+	if !migrator.HasTable(model) {
+		return nil
+	}
+	hasCluster := migrator.HasColumn(model, "cluster")
+	if !hasCluster && !migrator.HasColumn(model, "subdomain") {
+		// Fresh schema or already migrated — nothing to reshape.
+		return nil
+	}
+
+	return db.Transaction(func(tx *gorm.DB) error {
+		txMigrator := tx.Migrator()
+		for _, field := range []string{"Domain", "ProxyAddress"} {
+			if !txMigrator.HasColumn(model, field) {
+				if err := txMigrator.AddColumn(model, field); err != nil {
+					return fmt.Errorf("add %s column to agent_network_settings: %w", field, err)
+				}
+			}
+		}
+
+		if hasCluster {
+			concat := "subdomain || '.' || cluster"
+			if tx.Name() == "mysql" {
+				concat = "CONCAT(subdomain, '.', cluster)"
+			}
+			res := tx.Exec(fmt.Sprintf(
+				"UPDATE agent_network_settings SET domain = %s, proxy_address = cluster WHERE (domain IS NULL OR domain = '') AND cluster <> '' AND subdomain <> ''",
+				concat,
+			))
+			if res.Error != nil {
+				return fmt.Errorf("backfill agent_network_settings domain: %w", res.Error)
+			}
+
+			var unmigratable int64
+			if err := tx.Model(model).Where("domain IS NULL OR domain = ''").Count(&unmigratable).Error; err != nil {
+				return fmt.Errorf("count unmigratable agent_network_settings rows: %w", err)
+			}
+			if unmigratable > 0 {
+				return fmt.Errorf(
+					"%d agent_network_settings row(s) have no cluster/subdomain to derive an endpoint from; resolve them manually before upgrading",
+					unmigratable,
+				)
+			}
+
+			if res.RowsAffected > 0 {
+				log.WithContext(ctx).Infof("migrated %d agent_network_settings row(s) to domain/proxy_address", res.RowsAffected)
+			}
+		}
+
+		if txMigrator.HasIndex(model, "idx_agent_network_settings_cluster_subdomain") {
+			if err := txMigrator.DropIndex(model, "idx_agent_network_settings_cluster_subdomain"); err != nil {
+				return fmt.Errorf("drop legacy agent_network_settings index: %w", err)
+			}
+		}
+		for _, field := range []string{"Cluster", "Subdomain"} {
+			if txMigrator.HasColumn(model, field) {
+				if err := txMigrator.DropColumn(model, field); err != nil {
+					return fmt.Errorf("drop legacy agent_network_settings column %s: %w", field, err)
+				}
+			}
+		}
+
+		return nil
+	})
+}

management/server/migration/migration_test.go

@@ -736,3 +736,125 @@ func TestFoldCostAggregatesIntoBuckets_SkipsAlreadyMigrated(t *testing.T) {
 	assert.InDelta(t, 0.004, row.OutputCostUSD, 1e-9)
 	assert.InDelta(t, 0.01, row.TotalCostUSD(), 1e-9, "derived total sums the four buckets")
 }
+
+// legacyAgentNetworkSettings is the pre-reshape schema: identity carried as
+// (cluster, subdomain) instead of (domain, proxy_address).
+type legacyAgentNetworkSettings struct {
+	AccountID           string `gorm:"primaryKey"`
+	Cluster             string
+	Subdomain           string
+	EnableLogCollection bool
+}
+
+func (legacyAgentNetworkSettings) TableName() string { return "agent_network_settings" }
+
+// TestMigrateAgentNetworkSettingsToDomain_BackfillsAndDropsLegacyColumns pins
+// the reshape: domain becomes `<subdomain>.<cluster>`, proxy_address becomes
+// the cluster, the legacy columns are dropped, and non-identity fields ride
+// through untouched.
+func TestMigrateAgentNetworkSettingsToDomain_BackfillsAndDropsLegacyColumns(t *testing.T) {
+	ctx := context.Background()
+	db := setupDatabase(t)
+	require.NoError(t, db.Migrator().DropTable(&legacyAgentNetworkSettings{}))
+	require.NoError(t, db.AutoMigrate(&legacyAgentNetworkSettings{}))
+	require.NoError(t, db.Create(&legacyAgentNetworkSettings{
+		AccountID: "acct-1", Cluster: "eu.proxy.netbird.io", Subdomain: "violet", EnableLogCollection: true,
+	}).Error)
+	require.NoError(t, db.Create(&legacyAgentNetworkSettings{
+		AccountID: "acct-2", Cluster: "us.proxy.netbird.io", Subdomain: "violet",
+	}).Error)
+
+	require.NoError(t, migration.MigrateAgentNetworkSettingsToDomain(ctx, db))
+	require.NoError(t, db.AutoMigrate(&agentNetworkTypes.Settings{}),
+		"AutoMigrate must create the domain unique index over the backfilled values")
+
+	var one, two agentNetworkTypes.Settings
+	require.NoError(t, db.First(&one, "account_id = ?", "acct-1").Error)
+	assert.Equal(t, "violet.eu.proxy.netbird.io", one.Domain, "domain must combine subdomain and cluster")
+	assert.Equal(t, "eu.proxy.netbird.io", one.ProxyAddress, "proxy address must carry the cluster")
+	assert.True(t, one.EnableLogCollection, "non-identity fields must ride through")
+	require.NoError(t, db.First(&two, "account_id = ?", "acct-2").Error)
+	assert.Equal(t, "violet.us.proxy.netbird.io", two.Domain,
+		"duplicate labels on different clusters are distinct hostnames and must both survive")
+
+	migrator := db.Migrator()
+	assert.False(t, migrator.HasColumn(&legacyAgentNetworkSettings{}, "cluster"), "legacy cluster column must be dropped")
+	assert.False(t, migrator.HasColumn(&legacyAgentNetworkSettings{}, "subdomain"), "legacy subdomain column must be dropped")
+}
+
+// TestMigrateAgentNetworkSettingsToDomain_SkipsAlreadyMigrated proves the
+// migration is safe to re-run: with no legacy column present it is a no-op.
+func TestMigrateAgentNetworkSettingsToDomain_SkipsAlreadyMigrated(t *testing.T) {
+	ctx := context.Background()
+	db := setupDatabase(t)
+	require.NoError(t, db.Migrator().DropTable(&agentNetworkTypes.Settings{}))
+	require.NoError(t, db.AutoMigrate(&agentNetworkTypes.Settings{}))
+	require.NoError(t, db.Create(&agentNetworkTypes.Settings{
+		AccountID: "acct-1", Domain: "gw.example.com", ProxyAddress: "gw.example.com",
+	}).Error)
+
+	require.NoError(t, migration.MigrateAgentNetworkSettingsToDomain(ctx, db),
+		"running against an already-migrated table must be a no-op, not an error")
+
+	var row agentNetworkTypes.Settings
+	require.NoError(t, db.First(&row, "account_id = ?", "acct-1").Error)
+	assert.Equal(t, "gw.example.com", row.Domain, "migrated rows must be untouched")
+}
+
+// TestMigrateAgentNetworkSettingsToDomain_FailsOnUnmigratableRow pins the
+// loud-failure contract: a legacy row missing its identity halves cannot be
+// given an endpoint, and silently leaving an empty domain would collide with
+// the unique index confusingly later.
+func TestMigrateAgentNetworkSettingsToDomain_FailsOnUnmigratableRow(t *testing.T) {
+	ctx := context.Background()
+	db := setupDatabase(t)
+	require.NoError(t, db.Migrator().DropTable(&legacyAgentNetworkSettings{}))
+	require.NoError(t, db.AutoMigrate(&legacyAgentNetworkSettings{}))
+	require.NoError(t, db.Create(&legacyAgentNetworkSettings{
+		AccountID: "acct-broken", Cluster: "", Subdomain: "",
+	}).Error)
+
+	err := migration.MigrateAgentNetworkSettingsToDomain(ctx, db)
+	require.Error(t, err, "a row with no identity to derive an endpoint from must fail the migration")
+	assert.Contains(t, err.Error(), "resolve them manually", "the error must tell the operator what to do")
+}
+
+// partialAgentNetworkSettings models the one non-atomic state a MySQL run can
+// be interrupted in: DDL auto-commits there, so a crash between the two legacy
+// column drops leaves subdomain behind while cluster (and the completed
+// backfill) are already committed.
+type partialAgentNetworkSettings struct {
+	AccountID    string `gorm:"primaryKey"`
+	Subdomain    string
+	Domain       string `gorm:"type:varchar(255)"`
+	ProxyAddress string `gorm:"type:varchar(255)"`
+}
+
+func (partialAgentNetworkSettings) TableName() string { return "agent_network_settings" }
+
+// TestMigrateAgentNetworkSettingsToDomain_ResumesAfterPartialDrop pins MySQL
+// resumability: a rerun over the interrupted state must remove the leftover
+// subdomain column without re-running the backfill (the cluster column that
+// feeds it is gone) and without touching the migrated values.
+func TestMigrateAgentNetworkSettingsToDomain_ResumesAfterPartialDrop(t *testing.T) {
+	ctx := context.Background()
+	db := setupDatabase(t)
+	require.NoError(t, db.Migrator().DropTable(&partialAgentNetworkSettings{}))
+	require.NoError(t, db.AutoMigrate(&partialAgentNetworkSettings{}))
+	require.NoError(t, db.Create(&partialAgentNetworkSettings{
+		AccountID: "acct-1", Subdomain: "violet",
+		Domain: "violet.eu.proxy.netbird.io", ProxyAddress: "eu.proxy.netbird.io",
+	}).Error)
+
+	require.NoError(t, migration.MigrateAgentNetworkSettingsToDomain(ctx, db),
+		"a rerun over a partially-dropped schema must resume, not error")
+
+	migrator := db.Migrator()
+	assert.False(t, migrator.HasColumn(&partialAgentNetworkSettings{}, "subdomain"),
+		"the leftover legacy column must be dropped on resume")
+
+	var row agentNetworkTypes.Settings
+	require.NoError(t, db.First(&row, "account_id = ?", "acct-1").Error)
+	assert.Equal(t, "violet.eu.proxy.netbird.io", row.Domain, "migrated values must be untouched")
+	assert.Equal(t, "eu.proxy.netbird.io", row.ProxyAddress, "migrated values must be untouched")
+}

management/server/store/sql_store.go

@@ -6340,6 +6340,30 @@ func (s *SqlStore) CountProxiesByAccountID(ctx context.Context, accountID string
 	return count, nil
 }
 
+// HasActiveProxyAtClusterAddress reports whether any proxy — shared or
+// account-scoped — is currently active at the given cluster address, using
+// the same connected-within-threshold window as the other active-proxy
+// queries. Backs the agent-network settings delete guard: settings cannot be
+// deleted while a proxy declares the endpoint hostname as its address.
+//
+// The comparison folds case on both sides: the caller passes a normalized
+// (lowercase) hostname, but proxies declare their cluster address verbatim
+// and Connect stores it unchanged, so on case-sensitive collations a proxy
+// declaring "GW.Example.com" would otherwise slip past the guard. Hostnames
+// are case-insensitive per RFC 4343; the guard must be too.
+func (s *SqlStore) HasActiveProxyAtClusterAddress(ctx context.Context, clusterAddress string) (bool, error) {
+	var count int64
+	result := s.db.
+		Model(&proxy.Proxy{}).
+		Where("LOWER(cluster_address) = LOWER(?) AND status = ? AND last_seen > ?", clusterAddress, proxy.StatusConnected, time.Now().Add(-proxyActiveThreshold)).
+		Count(&count)
+	if result.Error != nil {
+		log.WithContext(ctx).Errorf("failed to count active proxies at cluster address: %v", result.Error)
+		return false, status.Errorf(status.Internal, "failed to count active proxies at cluster address")
+	}
+	return count > 0, nil
+}
+
 func (s *SqlStore) IsClusterAddressConflicting(ctx context.Context, clusterAddress, accountID string) (bool, error) {
 	var count int64
 	result := s.db.

management/server/store/sql_store_agentnetwork.go

@@ -315,25 +315,65 @@ func (s *SqlStore) GetAllAgentNetworkSettings(ctx context.Context, lockStrength
 	return settings, nil
 }
 
-// GetAgentNetworkSettingsByCluster returns every Settings row pinned to
-// the given proxy cluster. Used by the bootstrap label generator to
-// build the set of subdomains already taken on a cluster.
-func (s *SqlStore) GetAgentNetworkSettingsByCluster(ctx context.Context, lockStrength LockingStrength, cluster string) ([]*agentNetworkTypes.Settings, error) {
+// GetAgentNetworkSettingsByProxyAddress returns every Settings row whose
+// gateway is served by the proxy declaring the given cluster address. Used by
+// cluster-scoped synthesis to find the accounts a shared proxy serves.
+func (s *SqlStore) GetAgentNetworkSettingsByProxyAddress(ctx context.Context, lockStrength LockingStrength, proxyAddress string) ([]*agentNetworkTypes.Settings, error) {
 	tx := s.db
 	if lockStrength != LockingStrengthNone {
 		tx = tx.Clauses(clause.Locking{Strength: string(lockStrength)})
 	}
 
 	var settings []*agentNetworkTypes.Settings
-	result := tx.Find(&settings, "cluster = ?", cluster)
+	result := tx.Find(&settings, "proxy_address = ?", proxyAddress)
 	if result.Error != nil {
-		log.WithContext(ctx).Errorf("failed to get agent network settings by cluster from store: %v", result.Error)
-		return nil, status.Errorf(status.Internal, "failed to get agent network settings by cluster from store")
+		log.WithContext(ctx).Errorf("failed to get agent network settings by proxy address from store: %v", result.Error)
+		return nil, status.Errorf(status.Internal, "failed to get agent network settings by proxy address from store")
 	}
 
 	return settings, nil
 }
 
+// GetAgentNetworkSettingsByDomain resolves the single Settings row holding the
+// given endpoint hostname — a point query on the domain unique index. Returns
+// status.NotFound when no account owns the domain.
+func (s *SqlStore) GetAgentNetworkSettingsByDomain(ctx context.Context, lockStrength LockingStrength, domain string) (*agentNetworkTypes.Settings, error) {
+	tx := s.db
+	if lockStrength != LockingStrengthNone {
+		tx = tx.Clauses(clause.Locking{Strength: string(lockStrength)})
+	}
+
+	var settings agentNetworkTypes.Settings
+	result := tx.Take(&settings, "domain = ?", domain)
+	if result.Error != nil {
+		if errors.Is(result.Error, gorm.ErrRecordNotFound) {
+			return nil, status.Errorf(status.NotFound, "agent network settings for domain %s not found", domain)
+		}
+
+		log.WithContext(ctx).Errorf("failed to get agent network settings by domain from store: %v", result.Error)
+		return nil, status.Errorf(status.Internal, "failed to get agent network settings by domain from store")
+	}
+
+	return &settings, nil
+}
+
+// CreateAgentNetworkSettings inserts a new settings row.
+//
+// Unlike SaveAgentNetworkSettings (an upsert) this is a plain INSERT, and it
+// returns the driver error unwrapped. Both properties are required by the
+// bootstrap allocator: an upsert would overwrite whichever row it collided
+// with, and the allocator classifies the rejection by matching the driver's
+// message — a unique violation on the account primary key means a concurrent
+// bootstrap for the same account won, and one on the domain index means the
+// hostname is taken.
+func (s *SqlStore) CreateAgentNetworkSettings(ctx context.Context, settings *agentNetworkTypes.Settings) error {
+	if err := s.db.Create(settings).Error; err != nil {
+		log.WithContext(ctx).Debugf("failed to create agent network settings: %v", err)
+		return err
+	}
+	return nil
+}
+
 // SaveAgentNetworkSettings upserts the per-account Agent Network
 // settings row.
 func (s *SqlStore) SaveAgentNetworkSettings(ctx context.Context, settings *agentNetworkTypes.Settings) error {
@@ -346,6 +386,25 @@ func (s *SqlStore) SaveAgentNetworkSettings(ctx context.Context, settings *agent
 	return nil
 }
 
+// DeleteAgentNetworkSettings removes the per-account Agent Network settings
+// row, releasing the account's endpoint. Returns status.NotFound when no row
+// exists. The guards on the delete (no providers, no proxy actively serving
+// the endpoint) live in the manager, which runs this inside a transaction
+// after re-checking them under a row lock.
+func (s *SqlStore) DeleteAgentNetworkSettings(ctx context.Context, accountID string) error {
+	result := s.db.Delete(&agentNetworkTypes.Settings{}, "account_id = ?", accountID)
+	if result.Error != nil {
+		log.WithContext(ctx).Errorf("failed to delete agent network settings from store: %v", result.Error)
+		return status.Errorf(status.Internal, "failed to delete agent network settings from store")
+	}
+
+	if result.RowsAffected == 0 {
+		return status.Errorf(status.NotFound, "agent network settings for account %s not found", accountID)
+	}
+
+	return nil
+}
+
 // IncrementAgentNetworkConsumption atomically upserts the consumption
 // row keyed on (account, dim_kind, dim_id, window_seconds, window_start)
 // and adds the supplied deltas. Concurrent calls from multiple proxy

management/server/store/sql_store_agentnetwork_budgetrule_test.go

@@ -88,9 +88,9 @@ func TestAgentNetworkSettings_RealStore_CollectionTogglesRoundTrip(t *testing.T)
 
 	const accountID = "acc-settings-toggles"
 	require.NoError(t, s.SaveAgentNetworkSettings(ctx, &agentNetworkTypes.Settings{
-		AccountID: accountID,
-		Cluster:   "eu.proxy.netbird.io",
-		Subdomain: "violet",
+		AccountID:    accountID,
+		Domain:       "violet.eu.proxy.netbird.io",
+		ProxyAddress: "eu.proxy.netbird.io",
 	}))
 
 	got, err := s.GetAgentNetworkSettings(ctx, LockingStrengthNone, accountID)

management/server/store/store.go

@@ -328,6 +328,7 @@ type Store interface {
 	GetProxyByAccountID(ctx context.Context, accountID string) (*proxy.Proxy, error)
 	CountProxiesByAccountID(ctx context.Context, accountID string) (int64, error)
 	IsClusterAddressConflicting(ctx context.Context, clusterAddress, accountID string) (bool, error)
+	HasActiveProxyAtClusterAddress(ctx context.Context, clusterAddress string) (bool, error)
 	DeleteAccountCluster(ctx context.Context, clusterAddress, accountID string) error
 
 	GetCustomDomainsCounts(ctx context.Context) (total int64, validated int64, err error)
@@ -360,8 +361,11 @@ type Store interface {
 	DeleteAgentNetworkGuardrail(ctx context.Context, accountID, guardrailID string) error
 	GetAgentNetworkSettings(ctx context.Context, lockStrength LockingStrength, accountID string) (*agentNetworkTypes.Settings, error)
 	GetAllAgentNetworkSettings(ctx context.Context, lockStrength LockingStrength) ([]*agentNetworkTypes.Settings, error)
-	GetAgentNetworkSettingsByCluster(ctx context.Context, lockStrength LockingStrength, cluster string) ([]*agentNetworkTypes.Settings, error)
+	GetAgentNetworkSettingsByProxyAddress(ctx context.Context, lockStrength LockingStrength, proxyAddress string) ([]*agentNetworkTypes.Settings, error)
+	GetAgentNetworkSettingsByDomain(ctx context.Context, lockStrength LockingStrength, domain string) (*agentNetworkTypes.Settings, error)
+	CreateAgentNetworkSettings(ctx context.Context, settings *agentNetworkTypes.Settings) error
 	SaveAgentNetworkSettings(ctx context.Context, settings *agentNetworkTypes.Settings) error
+	DeleteAgentNetworkSettings(ctx context.Context, accountID string) error
 	IncrementAgentNetworkConsumption(ctx context.Context, accountID string, kind agentNetworkTypes.ConsumptionDimension, dimID string, windowSeconds int64, windowStart time.Time, tokensIn, tokensOut int64, costUSD float64) error
 	IncrementAgentNetworkConsumptionBatch(ctx context.Context, accountID string, keys []agentNetworkTypes.ConsumptionKey, tokensIn, tokensOut int64, costUSD float64) error
 	GetAgentNetworkConsumption(ctx context.Context, lockStrength LockingStrength, accountID string, kind agentNetworkTypes.ConsumptionDimension, dimID string, windowSeconds int64, windowStart time.Time) (*agentNetworkTypes.Consumption, error)
@@ -608,6 +612,9 @@ func getMigrationsPreAuto(ctx context.Context) []migrationFunc {
 		func(db *gorm.DB) error {
 			return migration.BackfillPublicIDs[posture.Checks](ctx, db)
 		},
+		func(db *gorm.DB) error {
+			return migration.MigrateAgentNetworkSettingsToDomain(ctx, db)
+		},
 	}
 }
 

management/server/store/store_mock.go

@@ -268,6 +268,20 @@ func (mr *MockStoreMockRecorder) CreateAgentNetworkAccessLog(ctx, entry, groups
 	return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "CreateAgentNetworkAccessLog", reflect.TypeOf((*MockStore)(nil).CreateAgentNetworkAccessLog), ctx, entry, groups)
 }
 
+// CreateAgentNetworkSettings mocks base method.
+func (m *MockStore) CreateAgentNetworkSettings(ctx context.Context, settings *types.Settings) error {
+	m.ctrl.T.Helper()
+	ret := m.ctrl.Call(m, "CreateAgentNetworkSettings", ctx, settings)
+	ret0, _ := ret[0].(error)
+	return ret0
+}
+
+// CreateAgentNetworkSettings indicates an expected call of CreateAgentNetworkSettings.
+func (mr *MockStoreMockRecorder) CreateAgentNetworkSettings(ctx, settings interface{}) *gomock.Call {
+	mr.mock.ctrl.T.Helper()
+	return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "CreateAgentNetworkSettings", reflect.TypeOf((*MockStore)(nil).CreateAgentNetworkSettings), ctx, settings)
+}
+
 // CreateAgentNetworkUsage mocks base method.
 func (m *MockStore) CreateAgentNetworkUsage(ctx context.Context, usage *types.AgentNetworkUsage, groups []types.AgentNetworkUsageGroup) error {
 	m.ctrl.T.Helper()
@@ -493,6 +507,20 @@ func (mr *MockStoreMockRecorder) DeleteAgentNetworkProvider(ctx, accountID, prov
 	return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "DeleteAgentNetworkProvider", reflect.TypeOf((*MockStore)(nil).DeleteAgentNetworkProvider), ctx, accountID, providerID)
 }
 
+// DeleteAgentNetworkSettings mocks base method.
+func (m *MockStore) DeleteAgentNetworkSettings(ctx context.Context, accountID string) error {
+	m.ctrl.T.Helper()
+	ret := m.ctrl.Call(m, "DeleteAgentNetworkSettings", ctx, accountID)
+	ret0, _ := ret[0].(error)
+	return ret0
+}
+
+// DeleteAgentNetworkSettings indicates an expected call of DeleteAgentNetworkSettings.
+func (mr *MockStoreMockRecorder) DeleteAgentNetworkSettings(ctx, accountID interface{}) *gomock.Call {
+	mr.mock.ctrl.T.Helper()
+	return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "DeleteAgentNetworkSettings", reflect.TypeOf((*MockStore)(nil).DeleteAgentNetworkSettings), ctx, accountID)
+}
+
 // DeleteCustomDomain mocks base method.
 func (m *MockStore) DeleteCustomDomain(ctx context.Context, accountID, domainID string) error {
 	m.ctrl.T.Helper()
@@ -1687,19 +1715,34 @@ func (mr *MockStoreMockRecorder) GetAgentNetworkSettings(ctx, lockStrength, acco
 	return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "GetAgentNetworkSettings", reflect.TypeOf((*MockStore)(nil).GetAgentNetworkSettings), ctx, lockStrength, accountID)
 }
 
-// GetAgentNetworkSettingsByCluster mocks base method.
-func (m *MockStore) GetAgentNetworkSettingsByCluster(ctx context.Context, lockStrength LockingStrength, cluster string) ([]*types.Settings, error) {
+// GetAgentNetworkSettingsByDomain mocks base method.
+func (m *MockStore) GetAgentNetworkSettingsByDomain(ctx context.Context, lockStrength LockingStrength, domain string) (*types.Settings, error) {
 	m.ctrl.T.Helper()
-	ret := m.ctrl.Call(m, "GetAgentNetworkSettingsByCluster", ctx, lockStrength, cluster)
+	ret := m.ctrl.Call(m, "GetAgentNetworkSettingsByDomain", ctx, lockStrength, domain)
+	ret0, _ := ret[0].(*types.Settings)
+	ret1, _ := ret[1].(error)
+	return ret0, ret1
+}
+
+// GetAgentNetworkSettingsByDomain indicates an expected call of GetAgentNetworkSettingsByDomain.
+func (mr *MockStoreMockRecorder) GetAgentNetworkSettingsByDomain(ctx, lockStrength, domain interface{}) *gomock.Call {
+	mr.mock.ctrl.T.Helper()
+	return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "GetAgentNetworkSettingsByDomain", reflect.TypeOf((*MockStore)(nil).GetAgentNetworkSettingsByDomain), ctx, lockStrength, domain)
+}
+
+// GetAgentNetworkSettingsByProxyAddress mocks base method.
+func (m *MockStore) GetAgentNetworkSettingsByProxyAddress(ctx context.Context, lockStrength LockingStrength, proxyAddress string) ([]*types.Settings, error) {
+	m.ctrl.T.Helper()
+	ret := m.ctrl.Call(m, "GetAgentNetworkSettingsByProxyAddress", ctx, lockStrength, proxyAddress)
 	ret0, _ := ret[0].([]*types.Settings)
 	ret1, _ := ret[1].(error)
 	return ret0, ret1
 }
 
-// GetAgentNetworkSettingsByCluster indicates an expected call of GetAgentNetworkSettingsByCluster.
-func (mr *MockStoreMockRecorder) GetAgentNetworkSettingsByCluster(ctx, lockStrength, cluster interface{}) *gomock.Call {
+// GetAgentNetworkSettingsByProxyAddress indicates an expected call of GetAgentNetworkSettingsByProxyAddress.
+func (mr *MockStoreMockRecorder) GetAgentNetworkSettingsByProxyAddress(ctx, lockStrength, proxyAddress interface{}) *gomock.Call {
 	mr.mock.ctrl.T.Helper()
-	return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "GetAgentNetworkSettingsByCluster", reflect.TypeOf((*MockStore)(nil).GetAgentNetworkSettingsByCluster), ctx, lockStrength, cluster)
+	return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "GetAgentNetworkSettingsByProxyAddress", reflect.TypeOf((*MockStore)(nil).GetAgentNetworkSettingsByProxyAddress), ctx, lockStrength, proxyAddress)
 }
 
 // GetAgentNetworkUsageRows mocks base method.
@@ -2956,6 +2999,21 @@ func (mr *MockStoreMockRecorder) GetZoneDNSRecordsByName(ctx, lockStrength, acco
 	return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "GetZoneDNSRecordsByName", reflect.TypeOf((*MockStore)(nil).GetZoneDNSRecordsByName), ctx, lockStrength, accountID, zoneID, name)
 }
 
+// HasActiveProxyAtClusterAddress mocks base method.
+func (m *MockStore) HasActiveProxyAtClusterAddress(ctx context.Context, clusterAddress string) (bool, error) {
+	m.ctrl.T.Helper()
+	ret := m.ctrl.Call(m, "HasActiveProxyAtClusterAddress", ctx, clusterAddress)
+	ret0, _ := ret[0].(bool)
+	ret1, _ := ret[1].(error)
+	return ret0, ret1
+}
+
+// HasActiveProxyAtClusterAddress indicates an expected call of HasActiveProxyAtClusterAddress.
+func (mr *MockStoreMockRecorder) HasActiveProxyAtClusterAddress(ctx, clusterAddress interface{}) *gomock.Call {
+	mr.mock.ctrl.T.Helper()
+	return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "HasActiveProxyAtClusterAddress", reflect.TypeOf((*MockStore)(nil).HasActiveProxyAtClusterAddress), ctx, clusterAddress)
+}
+
 // IncrementAgentNetworkConsumption mocks base method.
 func (m *MockStore) IncrementAgentNetworkConsumption(ctx context.Context, accountID string, kind types.ConsumptionDimension, dimID string, windowSeconds int64, windowStart time.Time, tokensIn, tokensOut int64, costUSD float64) error {
 	m.ctrl.T.Helper()