package frostfs import ( "encoding/hex" "github.com/TrueCloudLab/frostfs-node/pkg/morph/event" frostfsEvent "github.com/TrueCloudLab/frostfs-node/pkg/morph/event/neofs" "github.com/nspcc-dev/neo-go/pkg/util/slice" "go.uber.org/zap" ) func (np *Processor) handleDeposit(ev event.Event) { deposit := ev.(frostfsEvent.Deposit) np.log.Info("notification", zap.String("type", "deposit"), zap.String("id", hex.EncodeToString(slice.CopyReverse(deposit.ID())))) // send event to the worker pool err := np.pool.Submit(func() { np.processDeposit(&deposit) }) if err != nil { // there system can be moved into controlled degradation stage np.log.Warn("frostfs processor worker pool drained", zap.Int("capacity", np.pool.Cap())) } } func (np *Processor) handleWithdraw(ev event.Event) { withdraw := ev.(frostfsEvent.Withdraw) np.log.Info("notification", zap.String("type", "withdraw"), zap.String("id", hex.EncodeToString(slice.CopyReverse(withdraw.ID())))) // send event to the worker pool err := np.pool.Submit(func() { np.processWithdraw(&withdraw) }) if err != nil { // there system can be moved into controlled degradation stage np.log.Warn("frostfs processor worker pool drained", zap.Int("capacity", np.pool.Cap())) } } func (np *Processor) handleCheque(ev event.Event) { cheque := ev.(frostfsEvent.Cheque) np.log.Info("notification", zap.String("type", "cheque"), zap.String("id", hex.EncodeToString(cheque.ID()))) // send event to the worker pool err := np.pool.Submit(func() { np.processCheque(&cheque) }) if err != nil { // there system can be moved into controlled degradation stage np.log.Warn("frostfs processor worker pool drained", zap.Int("capacity", np.pool.Cap())) } } func (np *Processor) handleConfig(ev event.Event) { cfg := ev.(frostfsEvent.Config) np.log.Info("notification", zap.String("type", "set config"), zap.String("key", hex.EncodeToString(cfg.Key())), zap.String("value", hex.EncodeToString(cfg.Value()))) // send event to the worker pool err := np.pool.Submit(func() { np.processConfig(&cfg) }) if err != nil { // there system can be moved into controlled degradation stage np.log.Warn("frostfs processor worker pool drained", zap.Int("capacity", np.pool.Cap())) } } func (np *Processor) handleBind(ev event.Event) { e := ev.(frostfsEvent.Bind) np.log.Info("notification", zap.String("type", "bind"), ) // send event to the worker pool err := np.pool.Submit(func() { np.processBind(e) }) if err != nil { // there system can be moved into controlled degradation stage np.log.Warn("frostfs processor worker pool drained", zap.Int("capacity", np.pool.Cap())) } } func (np *Processor) handleUnbind(ev event.Event) { e := ev.(frostfsEvent.Unbind) np.log.Info("notification", zap.String("type", "unbind"), ) // send event to the worker pool err := np.pool.Submit(func() { np.processBind(e) }) if err != nil { // there system can be moved into controlled degradation stage np.log.Warn("frostfs processor worker pool drained", zap.Int("capacity", np.pool.Cap())) } }