[
https://issues.apache.org/jira/browse/FLINK-6096?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15946708#comment-15946708
]
ASF GitHub Bot commented on FLINK-6096:
---------------------------------------
Github user shixiaogang commented on a diff in the pull request:
https://github.com/apache/flink/pull/3558#discussion_r108615377
--- Diff:
flink-runtime/src/main/java/org/apache/flink/migration/v0/SavepointV0.java ---
@@ -0,0 +1,207 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.flink.migration.v0;
+
+import org.apache.flink.migration.v0.runtime.TaskStateV0;
+import org.apache.flink.runtime.checkpoint.TaskState;
+import org.apache.flink.runtime.checkpoint.savepoint.Savepoint;
+import org.apache.flink.util.Preconditions;
+
+import java.util.Collection;
+import java.util.HashMap;
+import java.util.Map;
+
+/**
+ * Savepoint version 0.
+ *
+ * <p>This format was introduced with Flink 1.1.0.
+ * <pre>
+ * checkpointId: long
+ * numTaskStates: int
+ * |----jobVertexID: long[2]
+ * |----parallelism: int
+ * |----numSubtaskStates: int
+ * | |----subtaskIndex: int
+ * | |----serializedValueLength: int
+ * | |----serializedValue: byte[] (null if serializedValueLength is
-1)
+ * | |----subtaskStateSize: long
+ * | |----subtaskStateDuration: long
+ * |----numKeyGroupStates: int
+ * | |----subtaskIndex: int
+ * | |----serializedValueLength: int
+ * | |----serializedValue: byte[] (null if serializedValueLength is
-1)
+ * | |----keyGroupStateSize: long
+ * | |----keyGroupStateDuration: long
+ * </pre>
+ */
+@Deprecated
+@SuppressWarnings("deprecation")
+public class SavepointV0 implements Savepoint {
+
+ /** The classes that are migrated in SavepointV0 */
+ public static final Map<String, String> MigrationMapping = new
HashMap<String, String>() {{
+
+ /* migrated state descriptors */
+ put("org.apache.flink.api.common.state.StateDescriptor",
--- End diff --
@StefanRRichter +1 for option b)
I planned to use a tool to automatically create the mapping, but i found it
very difficult to obtain the changes in the savepoints. It seems we must find
all migrated classes manually.
I agree with you that the maintenance of the mapping is error-prone. The
usage of a naming scheme convention will help us to avoid typo in new class
names. But we still have to carefully find all classes that are changed in the
savepoints. I think it's necessary for us. We must keep in mind what are
changed in the savepoint format.
In my opinion, the number of migrated classes will be very small in the
future. Most changes should happen in the serialization formats of actual
states (e.g., the snapshots of keyed states and timer services) instead of the
state handles. Therefore, there will be few efforts needed to maintain the
mapping.
> Refactor the migration of old versioned savepoints
> --------------------------------------------------
>
> Key: FLINK-6096
> URL: https://issues.apache.org/jira/browse/FLINK-6096
> Project: Flink
> Issue Type: Improvement
> Components: State Backends, Checkpointing
> Reporter: Xiaogang Shi
> Assignee: Xiaogang Shi
>
> Existing code for the migration of old-versioned savepoints does not allow to
> correctly deserialize those classes changed in different versions. I think
> we should create a migration package for each old-versioned savepoint and put
> all migrated classes in the savepoint there. A mapping can be deployed to
> record those migrated classes in the savepoint so that we can correctly
> deserialize them.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
