resolve issue #77

* handle vshard error the same way as lua vshard router.

Author: nurzhan-saktaganov

CHANGELOG.md

@@ -6,6 +6,7 @@ BUG FIXES:
 CHANGES:
 * Add comment why and how we handle "NON_MASTER" vshard error.
 * Don't support 'type Error struct' anymore.
+* Handle vshard error the same way as lua vshard router (resolve issue #77).
 
 FEATURES:
 

api.go

@@ -52,6 +52,7 @@ type StorageCallVShardError struct {
 	MasterUUID     string `msgpack:"master" mapstructure:"master"`
 	ReplicasetUUID string `msgpack:"replicaset" mapstructure:"replicaset"`
 	ReplicaUUID    string `msgpack:"replica" mapstructure:"replica"`
+	Destination    string `msgpack:"destination" mapstructure:"destination"`
 }
 
 func (s StorageCallVShardError) Error() string {
@@ -171,17 +172,42 @@ func (r *Router) RouterCallImpl(ctx context.Context,
 			case VShardErrNameWrongBucket, VShardErrNameBucketIsLocked:
 				r.BucketReset(bucketID)
 
-				// TODO we should inspect here err.destination like lua vshard router does,
-				// but we don't support vshard error fully yet:
-				// https://github.com/KaymeKaydex/go-vshard-router/issues/94
-				// So we just retry here as a temporary solution.
-				r.metrics().RetryOnCall("bucket_migrate")
-
-				r.log().Debugf(ctx, "retrying fnc '%s' cause got vshard error: %v", fnc, &vshardError)
-
-				// this vshardError will be returned to a caller in case of timeout
-				err = &vshardError
-				continue
+				if vshardError.Destination == "" {
+					break // leads to retry
+				}
+
+				destinationUUID, err := uuid.Parse(vshardError.Destination)
+				if err != nil {
+					return nil, nil, fmt.Errorf("protocol violation %s: malformed destination %w: %w",
+						vshardStorageClientCall, vshardError, err)
+				}
+
+				var loggedOnce bool
+				for {
+					idToReplicasetRef := r.getIDToReplicaset()
+					if _, ok := idToReplicasetRef[destinationUUID]; ok {
+						_, err := r.BucketSet(bucketID, destinationUUID)
+						if err == nil {
+							break // loop
+						}
+						r.log().Warnf(ctx, "Failed set bucket %d to %v (possible race): %v", bucketID, destinationUUID, err)
+					}
+
+					if !loggedOnce {
+						r.log().Warnf(ctx, "Replicaset '%v' was not found, but received from storage as destination - please "+
+							"update configuration", destinationUUID)
+						loggedOnce = true
+					}
+
+					const defaultPoolingPause = 50 * time.Millisecond
+					time.Sleep(defaultPoolingPause)
+
+					if time.Since(timeStart) > timeout {
+						return nil, nil, &vshardError
+					}
+				}
+
+				// leads to retry
 			case VShardErrNameTransferIsInProgress:
 				// Since lua vshard router doesn't retry here, we don't retry too.
 				// There is a comment why lua vshard router doesn't retry:
@@ -197,6 +223,15 @@ func (r *Router) RouterCallImpl(ctx context.Context,
 			default:
 				return nil, nil, &vshardError
 			}
+
+			// retry for VShardErrNameWrongBucket, VShardErrNameBucketIsLocked
+
+			r.metrics().RetryOnCall("bucket_migrate")
+
+			r.log().Debugf(ctx, "retrying fnc '%s' cause got vshard error: %v", fnc, &vshardError)
+
+			// this vshardError will be returned to a caller in case of timeout
+			err = &vshardError
 		}
 
 		var isVShardRespOk bool