Files
EvoBGP/internal/store/backend.go
T
DenozordecandCursor f8e0520817 feat(httpapi): harden control plane and speed up prefix refresh
Очередь job получила lease и fencing, телеметрия и прогресс deploy сохраняются, ревизия пишет manifest входов. Обновление префиксов не переписывает неизменённый снимок, большие наборы идут через COPY, а устаревшая ревизия не попадает в deploy. Бандл ноды кэшируется, prune и постановка job больше не держат горячий путь.

Co-authored-by: Cursor <[email protected]>
2026-11-02 17:23:59 +07:00

410 lines
19 KiB
Go

package store
import (
"context"
"strings"
"time"
)
// Backend is the persistence abstraction for the control plane (memory, PostgreSQL, SQLite).
type Backend interface {
JobQueue
Deployments
TelemetryStore
// ObserveStoreMetrics returns historical revision and peer inventory aggregates or an error.
ObserveStoreMetrics(ctx context.Context) (max, sum, peers int, states map[string]int, err error)
MaterializedPrefixStats() (max int, sum int)
PeerCount() int
PeerSessionCountsByState() map[string]int
// DemoIDs is non-empty only after SeedDemo (memory or seeded SQL).
DemoIDs() (tenant, moduleCDN, moduleIP, revision, speaker string)
// ListTenantIDs returns distinct tenant identifiers (for background workers).
ListTenantIDs() ([]string, error)
// GetRenderGeneration returns the tenant's monotonic render-input barrier.
GetRenderGeneration(ctx context.Context, tenantID string) (int64, error)
// ListModuleSnapshotHeaders returns snapshot versions for the tenant, ordered by module id.
// It does not load prefix rows.
ListModuleSnapshotHeaders(ctx context.Context, tenantID string) ([]ModuleSnapshotUse, error)
// LatestRenderManifest returns the manifest stored with the newest config revision.
// ok is false when that revision has no manifest yet.
LatestRenderManifest(ctx context.Context, tenantID string) (*RenderManifestRecord, bool, error)
// ListModules returns all modules for a tenant (control plane may paginate in httpapi).
ListModules(tenantID string) []*Module
// ListModulesPage returns one page of modules (limit capped by caller).
ListModulesPage(tenantID, cursor string, limit int) ([]*Module, string, bool)
GetModule(tenantID, moduleID string) (*Module, error)
CreateModule(tenantID string, in *Module) (*Module, error)
UpdateModule(tenantID, moduleID string, patch *ModulePatch) (*Module, error)
SoftDeleteModule(tenantID, moduleID string) error
ListCDNSources(tenantID, moduleID string) ([]*CDNSource, error)
CreateCDNSource(tenantID, moduleID string, in *CDNSource) (*CDNSource, error)
UpdateCDNSource(tenantID, moduleID, sourceID string, patch *CDNSourcePatch) (*CDNSource, error)
DeleteCDNSource(tenantID, moduleID, sourceID string) error
ListASEntries(tenantID, moduleID string) ([]*ASEntry, error)
CreateASEntry(tenantID, moduleID string, in *ASEntry) (*ASEntry, error)
UpdateASEntry(tenantID, moduleID, entryID string, patch *ASEntryPatch) (*ASEntry, error)
DeleteASEntry(tenantID, moduleID, entryID string) error
// UpdateASEntryResolveMeta записывает имя AS, число объявленных префиксов и время успешного резолва (pipeline).
UpdateASEntryResolveMeta(tenantID, moduleID, entryID string, asnName string, prefixCount int64, resolvedAt time.Time) error
UpdateASEntryResolveMetaBatch(tenantID, moduleID string, updates []ASEntryResolveMetaUpdate, resolvedAt time.Time) error
ListDomainEntries(tenantID, moduleID string) ([]*DomainEntry, error)
CreateDomainEntry(tenantID, moduleID string, in *DomainEntry) (*DomainEntry, error)
UpdateDomainEntry(tenantID, moduleID, entryID string, patch *DomainEntryPatch) (*DomainEntry, error)
DeleteDomainEntry(tenantID, moduleID, entryID string) error
ListIPRangeEntries(tenantID, moduleID string) ([]*IPRangeEntry, error)
CreateIPRangeEntry(tenantID, moduleID string, in *IPRangeEntry) (*IPRangeEntry, error)
UpdateIPRangeEntry(tenantID, moduleID, entryID string, patch *IPRangePatch) (*IPRangeEntry, error)
DeleteIPRangeEntry(tenantID, moduleID, entryID string) error
ListDohProfiles(tenantID string) ([]*DohProfile, error)
GetDohProfile(tenantID, id string) (*DohProfile, error)
CreateDohProfile(tenantID string, in *DohProfile) (*DohProfile, error)
UpdateDohProfile(tenantID, id string, patch *DohProfilePatch) (*DohProfile, error)
DeleteDohProfile(tenantID, id string) error
ListCommunities(tenantID string) ([]*Community, error)
GetCommunity(tenantID, id string) (*Community, error)
CreateCommunity(tenantID string, in *Community) (*Community, error)
UpdateCommunity(tenantID, id string, patch *CommunityPatch) (*Community, error)
DeleteCommunity(tenantID, id string) error
// ListCommunityPrefixes returns unique prefixes tagged with community from latest revision per module.
ListCommunityPrefixes(tenantID, communityID, cursor string, limit int) (prefixes []PrefixRow, nextCursor string, hasMore bool, err error)
// ListPeers returns all BGP peers for a tenant (control plane may paginate in httpapi).
ListPeers(tenantID string) []*BGPPeer
GetPeer(tenantID, id string) (*BGPPeer, error)
CreatePeer(tenantID string, in *BGPPeer) (*BGPPeer, error)
UpdatePeer(tenantID, id string, patch *PeerPatch) (*BGPPeer, error)
DeletePeer(tenantID, id string) error
// Peer discovery (dynamic BGP quarantine → approve/reject).
ListPeerDiscoveries(tenantID, status string) ([]*BGPPeerDiscovery, error)
GetPeerDiscovery(tenantID, id string) (*BGPPeerDiscovery, error)
UpsertPeerDiscovery(tenantID string, in *PeerDiscoveryUpsert) (*BGPPeerDiscovery, error)
ApprovePeerDiscovery(tenantID, id string, in *PeerDiscoveryApproveInput) (*BGPPeer, *BGPPeerDiscovery, error)
RejectPeerDiscovery(tenantID, id string) (*BGPPeerDiscovery, error)
ListSpeakersForTenant(tenantID string) []*Speaker
GetSpeaker(tenantID, speakerID string) (*Speaker, error)
GetSpeakerAnyTenant(speakerID string) (*Speaker, error)
CreateSpeaker(tenantID string, in *Speaker) (*Speaker, error)
UpdateSpeaker(tenantID, id string, patch *SpeakerPatch) (*Speaker, error)
DeleteSpeaker(tenantID, id string) error
GetRevision(tenantID, revisionID string) (*Revision, error)
// GetRevisionSummary returns revision metadata without preview_fragments payloads.
GetRevisionSummary(tenantID, revisionID string) (*Revision, error)
ListRevisions(tenantID, moduleID string, cursor string, limit int) (items []*Revision, nextCursor string, hasMore bool)
ListRevisionPrefixes(tenantID, revisionID string, cursor string, limit int) (prefixes []PrefixRow, next string, more bool)
CreateRollbackRevision(tenantID, sourceRevisionID string) (newID string, err error)
// CreateRenderRevision inserts a new config_revision (revID must be unique) with materialized prefixes and preview fragments.
// The same transaction records config_generation and snapshot versions in render_revision_manifest.
// ctx cancellation aborts the write. prefix_snapshot is keyed by a hash of prefixes only.
CreateRenderRevision(ctx context.Context, revisionID, tenantID, moduleID string, parentRevisionID *string, contentHash string, previewFragments map[string]string, prefixes []PrefixRow) error
RevisionDiff(tenantID, aID, bID string) (map[string]any, error)
EstimateRevisionPrune(tenantID string, cutoff time.Time, retentionMinutes int) (RevisionPruneEstimate, error)
PruneRevisionsWithStats(tenantID string, cutoff time.Time) (RevisionPruneResult, error)
// PruneRevisionsBefore deletes revisions older than cutoff, keeping the newest
// revision and any revision a speaker still references. ctx bounds the delete.
PruneRevisionsBefore(ctx context.Context, tenantID string, cutoff time.Time) (deleted int, err error)
SetLastAppliedRevision(tenantID, speakerID, revisionID string) error
PublishRevisionForSpeaker(speakerID, revisionID string) error
LatestPublishedRevision(speakerID string) (revisionID string, publishedAt time.Time, err error)
ListGlobalSettings(tenantID string) (map[string]any, error)
PatchGlobalSettings(tenantID string, patch map[string]any) error
ListAPIKeys(tenantID string) ([]*APIKey, error)
GetAPIKey(tenantID, id string) (*APIKey, error)
CreateAPIKey(tenantID string, in *APIKeyCreate) (*APIKeyWithSecret, error)
UpdateAPIKey(tenantID, id string, patch *APIKeyPatch) (*APIKey, error)
RevokeAPIKey(tenantID, id string) error
RotateAPIKey(tenantID, id string) (*APIKeyWithSecret, error)
ListActiveAPIKeyHashes() ([]APIKeyAuthRow, error)
TouchAPIKeyLastUsed(id string) error
// Module prefix snapshots cache last successful collect per module (pipeline ingest/render).
GetModulePrefixSnapshot(ctx context.Context, tenantID, moduleID string) (*ModulePrefixSnapshot, bool, error)
// ListModuleSnapshotSourceRows returns persisted rows for one source key.
// It does not load the rest of the module snapshot.
ListModuleSnapshotSourceRows(ctx context.Context, tenantID, moduleID, source string) ([]PrefixRow, error)
SetModulePrefixSnapshot(ctx context.Context, tenantID, moduleID, inputHash string, prefixes []PrefixRow) error
DeleteModulePrefixSnapshot(ctx context.Context, tenantID, moduleID string) error
// UpdateModulePrefixSnapshot runs a pure mutation against the latest snapshot atomically.
UpdateModulePrefixSnapshot(ctx context.Context, tenantID, moduleID string, update ModuleSnapshotUpdate) error
// SetModuleInputHash stores the ingest fingerprint on the module row (O(1) snapshot check).
SetModuleInputHash(ctx context.Context, tenantID, moduleID, hash string) error
// ASNPrefixCache stores RIPEstat announced-prefixes per ASN (global TTL cache).
GetASNPrefixCache(asn int64) (*ASNPrefixCacheEntry, bool, error)
SetASNPrefixCache(asn int64, holder string, prefixes []string) error
// DomainResolveCache stores DoH results per FQDN (global TTL cache).
GetDomainResolveCache(fqdn string) (*DomainResolveCacheEntry, bool, error)
SetDomainResolveCache(fqdn string, addrs []string) error
// Ping verifies backend connectivity (no-op for in-memory).
Ping(ctx context.Context) error
// RunPeriodicMaintenance prunes stale DB rows (no-op for in-memory).
RunPeriodicMaintenance(ctx context.Context)
// Maintenance policies (instance-scoped PostgreSQL maintenance configuration).
ListMaintenancePolicies(cursor string, limit int) ([]*MaintenancePolicy, string, bool, error)
GetMaintenancePolicy(id string) (*MaintenancePolicy, error)
CreateMaintenancePolicy(in *MaintenancePolicy) (*MaintenancePolicy, error)
UpdateMaintenancePolicy(id string, patch *MaintenancePolicyPatch) (*MaintenancePolicy, error)
DeleteMaintenancePolicy(id string) error
TouchMaintenancePolicyRun(id, status, errMsg string) error
AppendMaintenancePolicyConfigAudit(actor, policyID, action string, before, after map[string]any) error
ListMaintenancePolicyConfigAudit(cursor string, limit int) ([]*MaintenancePolicyConfigAudit, string, bool, error)
// Runtime log cleanup audit (filesystem ops logged per tenant).
AppendRuntimeLogCleanupAudit(tenantID, actor, filename, action string, sizeBefore int64, sizeAfter *int64, detail map[string]any) (string, error)
ListRuntimeLogCleanupAudit(tenantID, cursor string, limit int) ([]*RuntimeLogCleanupAudit, string, bool, error)
// CRUD audit log (tenant-scoped; optional portal ingest push from httpapi).
AppendAudit(in AuditAppendInput) (*AuditEntry, error)
ListAudit(tenantID, cursor string, limit int, filter AuditListFilter) ([]*AuditEntry, string, bool, error)
MarkAuditPortalPushed(id string) error
// Firewall blocklist clients and policy rules.
ListFirewallClients(tenantID string) ([]*FirewallClient, error)
GetFirewallClient(tenantID, id string) (*FirewallClient, error)
CreateFirewallClient(tenantID string, in *FirewallClientCreate) (*FirewallClient, error)
UpdateFirewallClient(tenantID, id string, patch *FirewallClientPatch) (*FirewallClient, error)
ApproveFirewallClient(tenantID, id, approverAPIKeyID string) (*FirewallClient, error)
RevokeFirewallClient(tenantID, id string) error
DeleteFirewallClient(tenantID, id string) error
LookupFirewallClientByTokenHash(hash []byte) (*FirewallClient, error)
TouchFirewallClientLastSeen(id, source, clientIP, clientVersion string) error
TouchFirewallClientLastApply(id, source, status, errMsg string, prefixCount, ipCount int, packetsDropped, packetsAccepted int64) error
ListActiveFirewallClientHashes() ([]FirewallClientAuthRow, error)
ListApprovedFirewallClientsForReplication(tenantID string) ([]FirewallClientReplicationRow, error)
ListFirewallRules(tenantID string, clientID *string) ([]*FirewallRule, error)
ListAllFirewallRulesForClient(tenantID, clientID string) ([]*FirewallRule, error)
ListAllFirewallRulesForReplication(tenantID string) ([]*FirewallRule, error)
CreateFirewallRule(tenantID string, clientID *string, in *FirewallRuleCreate) (*FirewallRule, error)
GetFirewallRule(tenantID, ruleID string) (*FirewallRule, error)
UpdateFirewallRule(tenantID, ruleID string, patch *FirewallRulePatch) (*FirewallRule, error)
DeleteFirewallRule(tenantID, ruleID string) error
ReorderFirewallRules(tenantID string, clientID *string, orderedIDs []string) error
}
// ASNPrefixCacheEntry is a cached RIPEstat response for one ASN.
type ASNPrefixCacheEntry struct {
ASN int64
Holder string
Prefixes []string
FetchedAt time.Time
}
// DomainResolveCacheEntry is a cached DoH A/AAAA result for one FQDN.
type DomainResolveCacheEntry struct {
FQDN string
Addrs []string
ResolvedAt time.Time
}
// ModulePrefixSnapshot is the cached materialization for one module between refreshes.
type ModulePrefixSnapshot struct {
InputHash string
CollectedAt time.Time
Version int64
Prefixes []PrefixRow
}
// ModulePatch is a partial update for module.
type ModulePatch struct {
Name *string `json:"name,omitempty"`
Enabled *bool `json:"enabled,omitempty"`
Priority *int `json:"priority,omitempty"`
RefreshIntervalSec *int `json:"refresh_interval_sec,omitempty"`
CronExpr *string `json:"cron_expr,omitempty"`
DefaultCommunityID *string `json:"default_community_id,omitempty"`
DohProfileID *string `json:"doh_profile_id,omitempty"`
DohProfileIDs *[]string `json:"doh_profile_ids,omitempty"`
DohResolverPolicy *string `json:"doh_resolver_policy,omitempty"`
LastRefreshedAt *time.Time `json:"-"`
}
// CDNSource is a row under a CDN module.
type CDNSource struct {
ID string `json:"id,omitempty"`
ModuleID string `json:"module_id,omitempty"`
SourceKind string `json:"source_kind"`
URL string `json:"url"`
PrefixPath string `json:"prefix_path,omitempty"`
Etag string `json:"etag"`
RefreshIntervalSec *int `json:"refresh_interval_sec"`
CommunityID *string `json:"community_id"`
LastRefreshedAt *time.Time `json:"-"`
}
type CDNSourcePatch struct {
SourceKind *string `json:"source_kind,omitempty"`
URL *string `json:"url,omitempty"`
PrefixPath *string `json:"prefix_path,omitempty"`
Etag *string `json:"etag,omitempty"`
RefreshIntervalSec *int `json:"refresh_interval_sec,omitempty"`
CommunityID *string `json:"community_id,omitempty"`
LastRefreshedAt *time.Time `json:"-"`
}
type ASEntry struct {
ID string `json:"id,omitempty"`
ModuleID string `json:"module_id,omitempty"`
ASN int64 `json:"asn"`
CommunityID *string `json:"community_id"`
ASNName string `json:"asn_name,omitempty"`
PrefixCount *int64 `json:"prefix_count,omitempty"`
ASNResolvedAt *time.Time `json:"asn_resolved_at,omitempty"`
}
type ASEntryPatch struct {
ASN *int64 `json:"asn,omitempty"`
CommunityID *string `json:"community_id,omitempty"`
}
// ValidASN reports whether n is a usable BGP ASN (1..4294967295).
func ValidASN(n int64) bool {
return n >= 1 && n <= 4294967295
}
type DomainEntry struct {
ID string `json:"id,omitempty"`
ModuleID string `json:"module_id,omitempty"`
FQDN string `json:"fqdn"`
CommunityID *string `json:"community_id"`
}
type DomainEntryPatch struct {
FQDN *string `json:"fqdn,omitempty"`
CommunityID *string `json:"community_id,omitempty"`
}
type IPRangeEntry struct {
ID string `json:"id,omitempty"`
ModuleID string `json:"module_id,omitempty"`
Prefix string `json:"prefix"`
CommunityID *string `json:"community_id"`
}
type IPRangePatch struct {
Prefix *string `json:"prefix,omitempty"`
CommunityID *string `json:"community_id,omitempty"`
}
type DohProfile struct {
ID string `json:"id,omitempty"`
TenantID string `json:"tenant_id,omitempty"`
Name string `json:"name"`
URL string `json:"url"`
TimeoutMs *int `json:"timeout_ms"`
SecretRef *string `json:"vault_secret_ref"`
}
type DohProfilePatch struct {
Name *string `json:"name,omitempty"`
URL *string `json:"url,omitempty"`
TimeoutMs *int `json:"timeout_ms,omitempty"`
SecretRef *string `json:"vault_secret_ref,omitempty"`
}
type Community struct {
ID string `json:"id,omitempty"`
TenantID string `json:"tenant_id,omitempty"`
Community string `json:"community"`
Title string `json:"title"`
ValueJSON string `json:"value_json"`
}
type CommunityPatch struct {
Community *string `json:"community,omitempty"`
Title *string `json:"title,omitempty"`
ValueJSON *string `json:"value_json,omitempty"`
}
// APIKey is tenant-scoped API key metadata (secret never stored in plaintext).
type APIKey struct {
ID string `json:"id"`
TenantID string `json:"tenant_id,omitempty"`
Name string `json:"name"`
Role string `json:"role"`
Prefix string `json:"prefix"`
CreatedAt time.Time `json:"created_at"`
UpdatedAt time.Time `json:"updated_at"`
ExpiresAt *time.Time `json:"expires_at,omitempty"`
RevokedAt *time.Time `json:"revoked_at,omitempty"`
LastUsedAt *time.Time `json:"last_used_at,omitempty"`
}
// APIKeyCreate is input for issuing a new key.
type APIKeyCreate struct {
Name string `json:"name"`
Role string `json:"role"`
ExpiresAt *time.Time `json:"expires_at,omitempty"`
}
// APIKeyPatch is a partial update (role change affects auth after resolver reload).
type APIKeyPatch struct {
Name *string `json:"name,omitempty"`
Role *string `json:"role,omitempty"`
ExpiresAt *time.Time `json:"expires_at,omitempty"`
ClearExpiresAt bool `json:"-"`
}
// APIKeyWithSecret is returned only on create/rotate.
type APIKeyWithSecret struct {
APIKey
Token string `json:"token"`
}
// APIKeyAuthRow is used to build the in-process auth index.
type APIKeyAuthRow struct {
ID string
TenantID string
Role string
TokenHash []byte
}
// ValidAPIKeyRole reports whether role is allowed for API keys.
func ValidAPIKeyRole(role string) bool {
switch strings.ToLower(strings.TrimSpace(role)) {
case "viewer", "editor", "operator", "node":
return true
default:
return false
}
}
type PeerPatch struct {
Neighbor *string `json:"neighbor,omitempty"`
RemoteASN *int64 `json:"remote_asn,omitempty"`
SpeakerID *string `json:"bgp_speaker_id,omitempty"`
Enabled *bool `json:"enabled,omitempty"`
Name *string `json:"name,omitempty"`
SessionState *string `json:"session_state,omitempty"`
PoliciesJSON *string `json:"policies_json,omitempty"`
}
type SpeakerPatch struct {
Role *string `json:"role,omitempty"`
Endpoint *string `json:"endpoint,omitempty"`
MetaJSON *string `json:"meta_json,omitempty"`
}
// PrefixRow is one materialized prefix for GET /revisions/.../prefixes.
type PrefixRow struct {
Prefix string `json:"prefix"`
CommunityID *string `json:"community_id,omitempty"`
Source string `json:"source,omitempty"`
}