package saas import ( "context" "crypto/ed25519" "crypto/sha256" "encoding/base64" "encoding/hex" "encoding/json" "errors" "fmt" "sort" "strings" "time" "data-terminal-cloud/internal/audit" "data-terminal-cloud/internal/database" "data-terminal-cloud/internal/identity" "github.com/jackc/pgx/v5" ) type opsFirmwareAccess struct { tenantID, tenantPublicID, userID, userPublicID, role string } type FirmwareDownload struct { ReleaseID string `json:"release_id"` Version string `json:"version"` ExpectedSize int64 `json:"size"` SHA256 string `json:"sha256"` Signature string `json:"signature"` SigningKeyID string `json:"signing_key_id"` DownloadURL string `json:"download_url"` DownloadUntil time.Time `json:"expires_at"` } func resolveOpsFirmwareAccess(ctx context.Context, q queryRow, principal identity.Principal, tenantPublicID string, mutate bool) (opsFirmwareAccess, error) { var access opsFirmwareAccess err := q.QueryRow(ctx, `SELECT t.id::text,t.public_id::text,ua.id::text,ua.public_id::text,ua.platform_role FROM tenant t JOIN user_account ua ON ua.tenant_id=t.id AND ua.status='active' AND ua.account_kind='operator' WHERE t.public_id=$1::uuid AND t.status='active' AND ua.oidc_issuer=$2 AND ua.oidc_subject=$3`, tenantPublicID, principal.Issuer, principal.Subject). Scan(&access.tenantID, &access.tenantPublicID, &access.userID, &access.userPublicID, &access.role) if errors.Is(err, pgx.ErrNoRows) { return opsFirmwareAccess{}, ErrAccessDenied } if err != nil { return opsFirmwareAccess{}, fmt.Errorf("resolve operations firmware access: %w", err) } if mutate && access.role != "ops_operator" && access.role != "ops_admin" { return opsFirmwareAccess{}, ErrAccessDenied } return access, nil } func (s *Service) ListOpsFirmwareSigningKeys(ctx context.Context, principal identity.Principal, tenantPublicID string) ([]FirmwareSigningKey, error) { access, err := resolveOpsFirmwareAccess(ctx, s.pool, principal, tenantPublicID, false) if err != nil { return nil, err } return s.listFirmwareSigningKeys(ctx, access.tenantID) } func (s *Service) RegisterOpsFirmwareSigningKey(ctx context.Context, principal identity.Principal, tenantPublicID, requestID, keyID, publicKeyBase64 string) error { decoded, err := base64.StdEncoding.DecodeString(publicKeyBase64) if err != nil || len(decoded) != ed25519.PublicKeySize || base64.StdEncoding.EncodeToString(decoded) != publicKeyBase64 || keyID == "" || len([]rune(keyID)) > 120 { return errors.New("invalid Ed25519 public key") } return database.Transact(ctx, s.pool, func(tx pgx.Tx) error { access, err := resolveOpsFirmwareAccess(ctx, tx, principal, tenantPublicID, true) if err != nil { return err } var existingPublicKey, status string err = tx.QueryRow(ctx, `SELECT public_key_base64,status FROM firmware_signing_key WHERE tenant_id=$1::uuid AND key_id=$2 FOR UPDATE`, access.tenantID, keyID).Scan(&existingPublicKey, &status) if errors.Is(err, pgx.ErrNoRows) { if _, err := tx.Exec(ctx, `INSERT INTO firmware_signing_key(tenant_id,key_id,public_key_base64,created_by) VALUES($1::uuid,$2,$3,$4::uuid)`, access.tenantID, keyID, publicKeyBase64, access.userID); err != nil { return err } } else if err != nil { return err } else if existingPublicKey != publicKeyBase64 || status != "ACTIVE" { return ErrFirmwareSigningKeyConflict } fingerprint := sha256.Sum256(decoded) return s.audit.Record(ctx, tx, audit.Event{TenantID: &access.tenantID, ActorType: "user", ActorPublicID: &access.userPublicID, Action: "ops.firmware.signing_key.register", ResourceType: "firmware_signing_key", RequestID: requestID, Result: "success", Detail: map[string]any{"key_id": keyID, "fingerprint_sha256": hex.EncodeToString(fingerprint[:])}}) }) } func (s *Service) RevokeOpsFirmwareSigningKey(ctx context.Context, principal identity.Principal, tenantPublicID, requestID, keyID string) (FirmwareSigningKey, error) { var internalTenantID string err := database.Transact(ctx, s.pool, func(tx pgx.Tx) error { access, err := resolveOpsFirmwareAccess(ctx, tx, principal, tenantPublicID, true) if err != nil { return err } internalTenantID = access.tenantID var status string if err := tx.QueryRow(ctx, `SELECT status FROM firmware_signing_key WHERE tenant_id=$1::uuid AND key_id=$2 FOR UPDATE`, access.tenantID, keyID).Scan(&status); errors.Is(err, pgx.ErrNoRows) { return ErrFirmwareSigningKeyConflict } else if err != nil { return err } if status == "REVOKED" { return nil } var blocked bool if err := tx.QueryRow(ctx, `SELECT EXISTS( SELECT 1 FROM firmware_release fr WHERE fr.tenant_id=$1::uuid AND fr.signing_key_id=$2 AND fr.status='UPLOADING' UNION ALL SELECT 1 FROM firmware_rollout ro JOIN firmware_release fr ON fr.id=ro.release_id AND fr.tenant_id=ro.tenant_id WHERE ro.tenant_id=$1::uuid AND fr.signing_key_id=$2 AND ro.status='RUNNING' )`, access.tenantID, keyID).Scan(&blocked); err != nil { return err } if blocked { return ErrFirmwareSigningKeyInUse } if _, err := tx.Exec(ctx, `UPDATE firmware_signing_key SET status='REVOKED',revoked_at=now() WHERE tenant_id=$1::uuid AND key_id=$2 AND status='ACTIVE'`, access.tenantID, keyID); err != nil { return err } return s.audit.Record(ctx, tx, audit.Event{TenantID: &access.tenantID, ActorType: "user", ActorPublicID: &access.userPublicID, Action: "ops.firmware.signing_key.revoke", ResourceType: "firmware_signing_key", RequestID: requestID, Result: "success", Detail: map[string]any{"key_id": keyID}}) }) if err != nil { return FirmwareSigningKey{}, err } items, err := s.listFirmwareSigningKeys(ctx, internalTenantID) if err != nil { return FirmwareSigningKey{}, err } for _, item := range items { if item.KeyID == keyID { return item, nil } } return FirmwareSigningKey{}, ErrFirmwareSigningKeyConflict } func (s *Service) ListOpsFirmwareReleases(ctx context.Context, principal identity.Principal, tenantPublicID, model, status string) ([]FirmwareRelease, error) { access, err := resolveOpsFirmwareAccess(ctx, s.pool, principal, tenantPublicID, false) if err != nil { return nil, err } rows, err := s.pool.Query(ctx, `SELECT public_id::text,COALESCE(household_id::text,''),model,COALESCE(hardware_revision,''),version,expected_size,expected_digest,signature_algorithm,signature,signing_key_id,COALESCE(minimum_security_version,''),COALESCE(release_notes,''),release_channel,mandatory,published_at,created_source,status,verified_at,created_at,updated_at,object_ref FROM firmware_release WHERE tenant_id=$1::uuid AND (NULLIF($2,'') IS NULL OR model=$2) AND (NULLIF($3,'') IS NULL OR status=$3) ORDER BY created_at DESC LIMIT 200`, access.tenantID, strings.TrimSpace(model), strings.ToUpper(strings.TrimSpace(status))) if err != nil { return nil, fmt.Errorf("list operations firmware releases: %w", err) } defer rows.Close() items := make([]FirmwareRelease, 0) for rows.Next() { var item FirmwareRelease if err := rows.Scan(firmwareReleaseTargets(&item)...); err != nil { return nil, fmt.Errorf("scan operations firmware release: %w", err) } items = append(items, item) } return items, rows.Err() } func (s *Service) PrepareOpsFirmwareRelease(ctx context.Context, principal identity.Principal, tenantPublicID, requestID string, input PrepareFirmwareInput) (FirmwareRelease, error) { if s.firmwareStore == nil { return FirmwareRelease{}, ErrFirmwareStateConflict } var item FirmwareRelease err := database.Transact(ctx, s.pool, func(tx pgx.Tx) error { access, err := resolveOpsFirmwareAccess(ctx, tx, principal, tenantPublicID, true) if err != nil { return err } var signingKeyActive bool if err := tx.QueryRow(ctx, `SELECT EXISTS(SELECT 1 FROM firmware_signing_key WHERE tenant_id=$1::uuid AND key_id=$2 AND status='ACTIVE')`, access.tenantID, input.SigningKeyID).Scan(&signingKeyActive); err != nil { return err } if !signingKeyActive { return ErrFirmwareSigningKeyInactive } tenantHash := sha256.Sum256([]byte(access.tenantID)) objectRef := fmt.Sprintf("tenants/%s/firmware/%s/%s/%s.bin", hex.EncodeToString(tenantHash[:16]), safePath(input.Model), safePath(input.HardwareRevision), safePath(input.Version)) err = tx.QueryRow(ctx, `INSERT INTO firmware_release(tenant_id,household_id,model,hardware_revision,version,object_ref,expected_size,expected_digest,signature,signing_key_id,minimum_security_version,release_notes,release_channel,mandatory,created_source,created_by) VALUES($1::uuid,NULL,$2,NULLIF($3,''),$4,$5,$6,$7,$8,$9,NULLIF($10,''),NULLIF($11,''),$12,$13,'OPS',$14::uuid) RETURNING public_id::text,COALESCE(household_id::text,''),model,COALESCE(hardware_revision,''),version,expected_size,expected_digest,signature_algorithm,signature,signing_key_id,COALESCE(minimum_security_version,''),COALESCE(release_notes,''),release_channel,mandatory,published_at,created_source,status,verified_at,created_at,updated_at,object_ref`, access.tenantID, input.Model, input.HardwareRevision, input.Version, objectRef, input.ExpectedSize, input.ExpectedDigest, input.Signature, input.SigningKeyID, input.MinimumSecurityVersion, input.ReleaseNotes, normalizeReleaseChannel(input.ReleaseChannel), input.Mandatory, access.userID).Scan(firmwareReleaseTargets(&item)...) if err != nil { return err } return s.audit.Record(ctx, tx, audit.Event{TenantID: &access.tenantID, ActorType: "user", ActorPublicID: &access.userPublicID, Action: "ops.firmware.release.prepare", ResourceType: "firmware_release", ResourceID: &item.ID, RequestID: requestID, Result: "success", Detail: map[string]any{"model": input.Model, "version": input.Version, "digest": input.ExpectedDigest, "release_channel": item.ReleaseChannel, "mandatory": item.Mandatory}}) }) if err != nil { return FirmwareRelease{}, err } url, err := s.firmwareStore.PresignUpload(ctx, item.objectRef, 15*time.Minute) if err != nil { return FirmwareRelease{}, err } until := time.Now().Add(15 * time.Minute) item.UploadURL, item.UploadUntil = url, &until return item, nil } func (s *Service) resolveOpsFirmwareRelease(ctx context.Context, principal identity.Principal, tenantPublicID, releasePublicID string, mutate bool) (FirmwareRelease, opsFirmwareAccess, error) { access, err := resolveOpsFirmwareAccess(ctx, s.pool, principal, tenantPublicID, mutate) if err != nil { return FirmwareRelease{}, opsFirmwareAccess{}, err } var item FirmwareRelease err = s.pool.QueryRow(ctx, `SELECT public_id::text,COALESCE(household_id::text,''),model,COALESCE(hardware_revision,''),version,expected_size,expected_digest,signature_algorithm,signature,signing_key_id,COALESCE(minimum_security_version,''),COALESCE(release_notes,''),release_channel,mandatory,published_at,created_source,status,verified_at,created_at,updated_at,object_ref FROM firmware_release WHERE tenant_id=$1::uuid AND public_id=$2::uuid`, access.tenantID, releasePublicID).Scan(firmwareReleaseTargets(&item)...) if errors.Is(err, pgx.ErrNoRows) { return FirmwareRelease{}, opsFirmwareAccess{}, ErrNotFound } if err != nil { return FirmwareRelease{}, opsFirmwareAccess{}, err } return item, access, nil } func (s *Service) FinalizeOpsFirmwareRelease(ctx context.Context, principal identity.Principal, tenantPublicID, releasePublicID, requestID string) (FirmwareRelease, error) { item, access, err := s.resolveOpsFirmwareRelease(ctx, principal, tenantPublicID, releasePublicID, true) if err != nil { return FirmwareRelease{}, err } if item.Status == "READY" { return item, nil } if item.Status != "UPLOADING" || s.firmwareStore == nil { return FirmwareRelease{}, ErrFirmwareStateConflict } if err := s.firmwareStore.VerifyDigest(ctx, item.objectRef, item.ExpectedDigest, item.ExpectedSize); err != nil { return FirmwareRelease{}, err } var publicKeyBase64 string if err := s.pool.QueryRow(ctx, `SELECT public_key_base64 FROM firmware_signing_key WHERE tenant_id=$1::uuid AND key_id=$2 AND status='ACTIVE'`, access.tenantID, item.SigningKeyID).Scan(&publicKeyBase64); err != nil { return FirmwareRelease{}, ErrFirmwareSigningKeyInactive } publicKey, keyErr := base64.StdEncoding.DecodeString(publicKeyBase64) signature, signatureErr := base64.StdEncoding.DecodeString(item.Signature) if keyErr != nil || signatureErr != nil || len(publicKey) != ed25519.PublicKeySize || !ed25519.Verify(ed25519.PublicKey(publicKey), []byte(firmwareManifest(item.Model, item.HardwareRevision, item.Version, item.ExpectedDigest, item.ExpectedSize)), signature) { return FirmwareRelease{}, ErrFirmwareStateConflict } err = database.Transact(ctx, s.pool, func(tx pgx.Tx) error { command, err := tx.Exec(ctx, `UPDATE firmware_release SET status='READY',verified_at=now(),published_at=COALESCE(published_at,now()),updated_at=now() WHERE tenant_id=$1::uuid AND public_id=$2::uuid AND status='UPLOADING'`, access.tenantID, releasePublicID) if err != nil { return err } if command.RowsAffected() != 1 { return ErrFirmwareStateConflict } return s.audit.Record(ctx, tx, audit.Event{TenantID: &access.tenantID, ActorType: "user", ActorPublicID: &access.userPublicID, Action: "ops.firmware.release.ready", ResourceType: "firmware_release", ResourceID: &releasePublicID, RequestID: requestID, Result: "success", Detail: map[string]any{"digest": item.ExpectedDigest, "size": item.ExpectedSize}}) }) if err != nil { return FirmwareRelease{}, err } return s.currentOpsFirmwareRelease(ctx, principal, tenantPublicID, releasePublicID) } func (s *Service) RenewOpsFirmwareUpload(ctx context.Context, principal identity.Principal, tenantPublicID, releasePublicID, requestID string) (FirmwareRelease, error) { item, access, err := s.resolveOpsFirmwareRelease(ctx, principal, tenantPublicID, releasePublicID, true) if err != nil { return FirmwareRelease{}, err } if item.Status != "UPLOADING" || s.firmwareStore == nil { return FirmwareRelease{}, ErrFirmwareStateConflict } url, err := s.firmwareStore.PresignUpload(ctx, item.objectRef, 15*time.Minute) if err != nil { return FirmwareRelease{}, err } until := time.Now().Add(15 * time.Minute) err = database.Transact(ctx, s.pool, func(tx pgx.Tx) error { var currentStatus string if err := tx.QueryRow(ctx, `SELECT status FROM firmware_release WHERE tenant_id=$1::uuid AND public_id=$2::uuid FOR SHARE`, access.tenantID, releasePublicID).Scan(¤tStatus); err != nil { return err } if currentStatus != "UPLOADING" { return ErrFirmwareStateConflict } return s.audit.Record(ctx, tx, audit.Event{TenantID: &access.tenantID, ActorType: "user", ActorPublicID: &access.userPublicID, Action: "ops.firmware.release.upload_renew", ResourceType: "firmware_release", ResourceID: &releasePublicID, RequestID: requestID, Result: "success", Detail: map[string]any{"upload_until": until}}) }) if err != nil { return FirmwareRelease{}, err } item.UploadURL, item.UploadUntil = url, &until return item, nil } func (s *Service) PauseOpsFirmwareRelease(ctx context.Context, principal identity.Principal, tenantPublicID, releasePublicID, requestID string) (FirmwareRelease, error) { item, access, err := s.resolveOpsFirmwareRelease(ctx, principal, tenantPublicID, releasePublicID, true) if err != nil { return FirmwareRelease{}, err } if item.Status == "PAUSED" { return item, nil } if item.Status != "READY" { return FirmwareRelease{}, ErrFirmwareStateConflict } err = database.Transact(ctx, s.pool, func(tx pgx.Tx) error { var releaseID string if err := tx.QueryRow(ctx, `UPDATE firmware_release SET status='PAUSED',updated_at=now() WHERE tenant_id=$1::uuid AND public_id=$2::uuid AND status='READY' RETURNING id::text`, access.tenantID, releasePublicID).Scan(&releaseID); err != nil { return ErrFirmwareStateConflict } paused, err := tx.Exec(ctx, `UPDATE firmware_rollout SET status='PAUSED',updated_at=now() WHERE release_id=$1::uuid AND status='RUNNING'`, releaseID) if err != nil { return err } cancelled, err := tx.Exec(ctx, `UPDATE device_command dc SET status='CANCELLED',completed_at=now(),version=version+1,updated_at=now() FROM firmware_assignment fa JOIN firmware_rollout ro ON ro.id=fa.rollout_id WHERE fa.command_id=dc.id AND ro.release_id=$1::uuid AND ro.status='PAUSED' AND dc.status='QUEUED'`, releaseID) if err != nil { return err } return s.audit.Record(ctx, tx, audit.Event{TenantID: &access.tenantID, ActorType: "user", ActorPublicID: &access.userPublicID, Action: "ops.firmware.release.pause", ResourceType: "firmware_release", ResourceID: &releasePublicID, RequestID: requestID, Result: "success", Detail: map[string]any{"paused_rollouts": paused.RowsAffected(), "cancelled_commands": cancelled.RowsAffected()}}) }) if err != nil { return FirmwareRelease{}, err } return s.currentOpsFirmwareRelease(ctx, principal, tenantPublicID, releasePublicID) } func (s *Service) ResumeOpsFirmwareRelease(ctx context.Context, principal identity.Principal, tenantPublicID, releasePublicID, requestID string) (FirmwareRelease, error) { item, access, err := s.resolveOpsFirmwareRelease(ctx, principal, tenantPublicID, releasePublicID, true) if err != nil { return FirmwareRelease{}, err } if item.Status == "READY" { return item, nil } if item.Status != "PAUSED" { return FirmwareRelease{}, ErrFirmwareStateConflict } err = database.Transact(ctx, s.pool, func(tx pgx.Tx) error { command, err := tx.Exec(ctx, `UPDATE firmware_release fr SET status='READY',updated_at=now() FROM firmware_signing_key fsk WHERE fr.tenant_id=$1::uuid AND fr.public_id=$2::uuid AND fr.status='PAUSED' AND fsk.tenant_id=fr.tenant_id AND fsk.key_id=fr.signing_key_id AND fsk.status='ACTIVE'`, access.tenantID, releasePublicID) if err != nil { return err } if command.RowsAffected() != 1 { return ErrFirmwareSigningKeyInactive } return s.audit.Record(ctx, tx, audit.Event{TenantID: &access.tenantID, ActorType: "user", ActorPublicID: &access.userPublicID, Action: "ops.firmware.release.resume", ResourceType: "firmware_release", ResourceID: &releasePublicID, RequestID: requestID, Result: "success"}) }) if err != nil { return FirmwareRelease{}, err } return s.currentOpsFirmwareRelease(ctx, principal, tenantPublicID, releasePublicID) } func (s *Service) RetireOpsFirmwareRelease(ctx context.Context, principal identity.Principal, tenantPublicID, releasePublicID, requestID string) (FirmwareRelease, error) { item, access, err := s.resolveOpsFirmwareRelease(ctx, principal, tenantPublicID, releasePublicID, true) if err != nil { return FirmwareRelease{}, err } if item.Status != "UPLOADING" && item.Status != "READY" && item.Status != "PAUSED" && item.Status != "RETIRED" { return FirmwareRelease{}, ErrFirmwareStateConflict } removeObject := item.Status == "UPLOADING" || item.Status == "RETIRED" && item.VerifiedAt == nil if item.Status != "RETIRED" { err = database.Transact(ctx, s.pool, func(tx pgx.Tx) error { var releaseID, status string var verifiedAt *time.Time if err := tx.QueryRow(ctx, `SELECT id::text,status,verified_at FROM firmware_release WHERE tenant_id=$1::uuid AND public_id=$2::uuid FOR UPDATE`, access.tenantID, releasePublicID).Scan(&releaseID, &status, &verifiedAt); err != nil { return err } if status == "READY" || status == "PAUSED" { var running bool if err := tx.QueryRow(ctx, `SELECT EXISTS(SELECT 1 FROM firmware_rollout WHERE release_id=$1::uuid AND status='RUNNING')`, releaseID).Scan(&running); err != nil { return err } if running { return ErrFirmwareReleaseInUse } removeObject = false } if _, err := tx.Exec(ctx, `UPDATE firmware_release SET status='RETIRED',updated_at=now() WHERE id=$1::uuid AND status=$2`, releaseID, status); err != nil { return err } return s.audit.Record(ctx, tx, audit.Event{TenantID: &access.tenantID, ActorType: "user", ActorPublicID: &access.userPublicID, Action: "ops.firmware.release.retire", ResourceType: "firmware_release", ResourceID: &releasePublicID, RequestID: requestID, Result: "success", Detail: map[string]any{"from_status": status, "object_preserved": verifiedAt != nil}}) }) if err != nil { return FirmwareRelease{}, err } item.Status, item.UpdatedAt = "RETIRED", time.Now() } if removeObject { if s.firmwareStore == nil { return FirmwareRelease{}, ErrFirmwareStateConflict } if err := s.firmwareStore.Remove(ctx, item.objectRef); err != nil { return FirmwareRelease{}, fmt.Errorf("remove retired operations firmware object: %w", err) } } return item, nil } func (s *Service) currentOpsFirmwareRelease(ctx context.Context, principal identity.Principal, tenantPublicID, releasePublicID string) (FirmwareRelease, error) { item, _, err := s.resolveOpsFirmwareRelease(ctx, principal, tenantPublicID, releasePublicID, false) return item, err } func (s *Service) OpsFirmwareDownload(ctx context.Context, principal identity.Principal, tenantPublicID, releasePublicID, requestID string) (FirmwareDownload, error) { item, access, err := s.resolveOpsFirmwareRelease(ctx, principal, tenantPublicID, releasePublicID, false) if err != nil { return FirmwareDownload{}, err } if item.VerifiedAt == nil || (item.Status != "READY" && item.Status != "PAUSED" && item.Status != "RETIRED") || s.firmwareStore == nil { return FirmwareDownload{}, ErrFirmwareStateConflict } ttl := s.firmwareURLTTL if ttl <= 0 { ttl = 5 * time.Minute } url, err := s.firmwareStore.PresignDownload(ctx, item.objectRef, ttl, "firmware-"+safePath(item.Model)+"-"+safePath(item.Version)+".bin") if err != nil { return FirmwareDownload{}, err } until := time.Now().Add(ttl) err = database.Transact(ctx, s.pool, func(tx pgx.Tx) error { return s.audit.Record(ctx, tx, audit.Event{TenantID: &access.tenantID, ActorType: "user", ActorPublicID: &access.userPublicID, Action: "ops.firmware.release.download", ResourceType: "firmware_release", ResourceID: &releasePublicID, RequestID: requestID, Result: "success", Detail: map[string]any{"expires_at": until}}) }) if err != nil { return FirmwareDownload{}, err } return FirmwareDownload{ReleaseID: item.ID, Version: item.Version, ExpectedSize: item.ExpectedSize, SHA256: item.ExpectedDigest, Signature: item.Signature, SigningKeyID: item.SigningKeyID, DownloadURL: url, DownloadUntil: until}, nil } func (s *Service) StartOpsFirmwareRollout(ctx context.Context, principal identity.Principal, tenantPublicID, releasePublicID, requestID, name string, percentage int, deviceIDs []string) (FirmwareRollout, error) { if strings.TrimSpace(name) == "" || percentage < 1 || percentage > 100 || len(deviceIDs) == 0 { return FirmwareRollout{}, ErrFirmwareStateConflict } deviceIDs = append([]string(nil), deviceIDs...) sort.Strings(deviceIDs) uniqueDeviceIDs := deviceIDs[:0] for _, deviceID := range deviceIDs { if len(uniqueDeviceIDs) == 0 || uniqueDeviceIDs[len(uniqueDeviceIDs)-1] != deviceID { uniqueDeviceIDs = append(uniqueDeviceIDs, deviceID) } } deviceIDs = uniqueDeviceIDs var result FirmwareRollout err := database.Transact(ctx, s.pool, func(tx pgx.Tx) error { access, err := resolveOpsFirmwareAccess(ctx, tx, principal, tenantPublicID, true) if err != nil { return err } var releaseID, model, hardware, version, digest, signature, keyID string err = tx.QueryRow(ctx, `SELECT fr.id::text,fr.model,COALESCE(fr.hardware_revision,''),fr.version,fr.expected_digest,fr.signature,fr.signing_key_id FROM firmware_release fr JOIN firmware_signing_key fsk ON fsk.tenant_id=fr.tenant_id AND fsk.key_id=fr.signing_key_id AND fsk.status='ACTIVE' WHERE fr.tenant_id=$1::uuid AND fr.public_id=$2::uuid AND fr.status='READY' FOR SHARE OF fr,fsk`, access.tenantID, releasePublicID). Scan(&releaseID, &model, &hardware, &version, &digest, &signature, &keyID) if errors.Is(err, pgx.ErrNoRows) { return ErrFirmwareStateConflict } if err != nil { return err } var rolloutID string err = tx.QueryRow(ctx, `INSERT INTO firmware_rollout(tenant_id,household_id,release_id,name,status,target_percentage,created_by,started_at) VALUES($1::uuid,NULL,$2::uuid,$3,'RUNNING',$4,$5::uuid,now()) RETURNING id::text,public_id::text,created_at,updated_at`, access.tenantID, releaseID, strings.TrimSpace(name), percentage, access.userID).Scan(&rolloutID, &result.ID, &result.CreatedAt, &result.UpdatedAt) if err != nil { return err } result.ReleaseID, result.Name, result.Status, result.TargetPercentage = releasePublicID, strings.TrimSpace(name), "RUNNING", percentage selected := 0 for _, devicePublicID := range deviceIDs { hash := sha256.Sum256([]byte(devicePublicID)) if int(hash[0])%100 >= percentage && selected > 0 { continue } var deviceID, previous string err := tx.QueryRow(ctx, `SELECT d.id::text,COALESCE(dt.firmware_version,'') FROM device d JOIN device_twin dt ON dt.device_id=d.id AND dt.tenant_id=d.tenant_id WHERE d.tenant_id=$1::uuid AND d.public_id=$2::uuid AND d.status='registered' AND d.model=$3 AND ($4='' OR d.hardware_revision=$4)`, access.tenantID, devicePublicID, model, hardware).Scan(&deviceID, &previous) if err != nil { return fmt.Errorf("%w: rollout target %s is incompatible or inaccessible", ErrFirmwareStateConflict, devicePublicID) } payload, _ := json.Marshal(map[string]any{"release_id": releasePublicID, "version": version, "sha256": digest, "signature": signature, "signing_key_id": keyID, "download_path": "/device/v2/firmware-releases/" + releasePublicID + "/download"}) requestDigest := sha256.Sum256(payload) var commandID string err = tx.QueryRow(ctx, `INSERT INTO device_command(tenant_id,device_id,command_type,payload,requested_by,idempotency_key,request_digest,expires_at) VALUES($1::uuid,$2::uuid,'firmware_update',$3::jsonb,$4::uuid,$5,$6,now()+interval '24 hours') RETURNING id::text`, access.tenantID, deviceID, payload, access.userID, "ops-firmware-rollout:"+rolloutID+":"+deviceID, hex.EncodeToString(requestDigest[:])).Scan(&commandID) if err != nil { return err } if _, err := tx.Exec(ctx, `INSERT INTO firmware_assignment(rollout_id,tenant_id,device_id,command_id,previous_version,status) VALUES($1::uuid,$2::uuid,$3::uuid,$4::uuid,NULLIF($5,''),'ASSIGNED')`, rolloutID, access.tenantID, deviceID, commandID, previous); err != nil { return err } selected++ } if selected == 0 { return errors.New("rollout selected no devices") } result.Targeted = selected return s.audit.Record(ctx, tx, audit.Event{TenantID: &access.tenantID, ActorType: "user", ActorPublicID: &access.userPublicID, Action: "ops.firmware.rollout.start", ResourceType: "firmware_rollout", ResourceID: &result.ID, RequestID: requestID, Result: "success", Detail: map[string]any{"release_id": releasePublicID, "targeted": selected, "percentage": percentage}}) }) return result, err } func (s *Service) ListOpsFirmwareRollouts(ctx context.Context, principal identity.Principal, tenantPublicID string) ([]FirmwareRollout, error) { access, err := resolveOpsFirmwareAccess(ctx, s.pool, principal, tenantPublicID, false) if err != nil { return nil, err } rows, err := s.pool.Query(ctx, `SELECT ro.public_id::text,fr.public_id::text,COALESCE(h.public_id::text,''),ro.name,ro.status,ro.target_percentage, count(fa.id),count(fa.id) FILTER(WHERE fa.status='CONFIRMED'),count(fa.id) FILTER(WHERE fa.status='FAILED'),ro.created_at,ro.updated_at FROM firmware_rollout ro JOIN firmware_release fr ON fr.id=ro.release_id LEFT JOIN household h ON h.id=ro.household_id LEFT JOIN firmware_assignment fa ON fa.rollout_id=ro.id WHERE ro.tenant_id=$1::uuid GROUP BY ro.id,fr.public_id,h.public_id ORDER BY ro.created_at DESC LIMIT 200`, access.tenantID) if err != nil { return nil, err } defer rows.Close() items := make([]FirmwareRollout, 0) for rows.Next() { var item FirmwareRollout var targeted, succeeded, failed int64 if err := rows.Scan(&item.ID, &item.ReleaseID, &item.HouseholdID, &item.Name, &item.Status, &item.TargetPercentage, &targeted, &succeeded, &failed, &item.CreatedAt, &item.UpdatedAt); err != nil { return nil, err } item.Targeted, item.Succeeded, item.Failed = int(targeted), int(succeeded), int(failed) items = append(items, item) } return items, rows.Err() } func (s *Service) GetOpsFirmwareRollout(ctx context.Context, principal identity.Principal, tenantPublicID, rolloutPublicID string) (FirmwareRollout, error) { access, err := resolveOpsFirmwareAccess(ctx, s.pool, principal, tenantPublicID, false) if err != nil { return FirmwareRollout{}, err } var item FirmwareRollout var targeted, succeeded, failed int64 err = s.pool.QueryRow(ctx, `SELECT ro.public_id::text,fr.public_id::text,COALESCE(h.public_id::text,''),ro.name,ro.status,ro.target_percentage, count(fa.id),count(fa.id) FILTER(WHERE fa.status='CONFIRMED'),count(fa.id) FILTER(WHERE fa.status='FAILED'),ro.created_at,ro.updated_at FROM firmware_rollout ro JOIN firmware_release fr ON fr.id=ro.release_id LEFT JOIN household h ON h.id=ro.household_id LEFT JOIN firmware_assignment fa ON fa.rollout_id=ro.id WHERE ro.tenant_id=$1::uuid AND ro.public_id=$2::uuid GROUP BY ro.id,fr.public_id,h.public_id`, access.tenantID, rolloutPublicID). Scan(&item.ID, &item.ReleaseID, &item.HouseholdID, &item.Name, &item.Status, &item.TargetPercentage, &targeted, &succeeded, &failed, &item.CreatedAt, &item.UpdatedAt) if errors.Is(err, pgx.ErrNoRows) { return FirmwareRollout{}, ErrNotFound } if err != nil { return FirmwareRollout{}, err } item.Targeted, item.Succeeded, item.Failed = int(targeted), int(succeeded), int(failed) item.Assignments, err = s.listFirmwareAssignments(ctx, rolloutPublicID) return item, err } func (s *Service) PauseOpsFirmwareRollout(ctx context.Context, principal identity.Principal, tenantPublicID, rolloutPublicID, requestID string) (FirmwareRollout, error) { access, err := resolveOpsFirmwareAccess(ctx, s.pool, principal, tenantPublicID, true) if err != nil { return FirmwareRollout{}, err } err = database.Transact(ctx, s.pool, func(tx pgx.Tx) error { command, err := tx.Exec(ctx, `UPDATE firmware_rollout SET status='PAUSED',updated_at=now() WHERE tenant_id=$1::uuid AND public_id=$2::uuid AND status='RUNNING'`, access.tenantID, rolloutPublicID) if err != nil { return err } if command.RowsAffected() != 1 { return ErrFirmwareStateConflict } if _, err := tx.Exec(ctx, `UPDATE device_command dc SET status='CANCELLED',completed_at=now(),version=version+1,updated_at=now() FROM firmware_assignment fa JOIN firmware_rollout ro ON ro.id=fa.rollout_id WHERE fa.command_id=dc.id AND ro.public_id=$1::uuid AND dc.status='QUEUED'`, rolloutPublicID); err != nil { return err } return s.audit.Record(ctx, tx, audit.Event{TenantID: &access.tenantID, ActorType: "user", ActorPublicID: &access.userPublicID, Action: "ops.firmware.rollout.pause", ResourceType: "firmware_rollout", ResourceID: &rolloutPublicID, RequestID: requestID, Result: "success"}) }) if err != nil { return FirmwareRollout{}, err } return s.GetOpsFirmwareRollout(ctx, principal, tenantPublicID, rolloutPublicID) }