You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@rocketmq.apache.org by GitBox <gi...@apache.org> on 2022/11/24 07:35:56 UTC
[GitHub] [rocketmq] hzh0425 opened a new issue, #5585: Support DLedger Controller Snapshot
hzh0425 opened a new issue, #5585:
URL: https://github.com/apache/rocketmq/issues/5585
# Background
The DLedger Controller mode was launched after RocketMQ 5.0, which can provide the function of automatic master-slave switching.
The principle is as follows:
DLedger Controller is a strongly consistent metadata center built on DLedger (Raft), which stores metadata related to master selection.
However, because DLedger did not provide the ability of Snapshot before, when the Controller restarts, it needs to replay all the logs in DLedger to restore the state before Shutdown. Restarting all logs can cause the restart to take a lot of time, especially when there are a lot of logs.
Recently, we added the Snapshot capability to DLedger, thanks to @ for his contribution.
The basic principle is as follows: For example, for the log before LogIndex = 6, DLedger will create a snapshot for it. This Snapshot contains the state machine state before LogIndex = 6. In this way, when restarting, we only need to load this snapshot, and only replay logs 6 and 7, instead of replaying logs 1 ~ 7, which can save a lot of time.
# Implementation
The Snapshot function has been implemented in DLdeger, and we currently require the following improvements in the Controller module:
- Design a ControllerSnapshotFile and its file format.
- Construct SnapshotGenerator for creating and reading SnapshotFile.
- Implement the onSnapshotLoad and onSnapshotSave APIs in ControllerStateMachine.
## MetadataManager
MetadataManager represents a metadata manager in DLedgerController, and can also be considered as the memory state machine of DLedgerController. DLedgerController will apply Event to MetadataManager.
Currently, there is only one MetadataManager in DLedgerController, that is, ReplicasInfoManager. Other MetadataManagers, such as TopicManager, may be expanded later, so our SnapshotFile needs to be able to carry the data of multiple MetadataManagers.
For each specific MetadataManager, it needs to implement the SnapshotAbleMetadataManager interface and define its own Snapshot format.
```
public interface SnapshotAbleMetadataManager {
/**
* Encode the metadata contained in this MetadataManager
* @return encoded metadata
*/
byte[] encodeMetadata();
/**
*
* According to the param data, load metadata into the MetadataManager
* @param data encoded metadata
*/
void loadMetadata(byte[] data);
/**
* Get the type of this MetadataManager
* @return MetadataManagerType
*/
MetadataManagerType getMetadataManagerType();
}
```
## SnapshotFile Format
SnapshotFile consists of the following two parts
- SnapshotFilHeader: record version, CRC and other informations
- Sections array: composed of multiple Sections, each Section records the metadata of a MetadataManager.
The specific Format is as follows:
```
Snapshot File Header:
- Version: Char
- TotalSections: Int (number of sections)
- Magic: Int
- CRC: Int (CRC check digit)
- Reversed: Long (reserved, for subsequent expansion)
Sections:
- SectionHeader:
- SectionType: Char (refers to the category of MetadataManager)
- Length: Int (total length of SectionBody)
- SectionBody: bytes
```
### ReplicasInfoManager Snapshot Format
The metadata that needs to be stored in ReplicasInfoManager are:
```
private final Map<String/* brokerName */, BrokerInfo> replicaInfoTable;
private final Map<String/* brokerName */, SyncStateInfo> syncStateSetInfoTable;
```
Its Snapshot Format is as follows:
```
- ReplicaInfoTableLength: Int
- ReplicaInfoTableBytes
- syncStateSetInfoTableLength: Int
- syncStateSetInfoTableBytes
```
## Whole snapshot process
The Snapshot Save process is as follows:
- When DLedger triggers Snapshot, it will call the onSnapshotSave interface of DLedgerControllerStateMachine.
- onSnapshotSave builds the SnapshotGenerator.
- SnapshotGenerator creates a corresponding SectionBuilder based on each MetadataManager, which is responsible for building the metadata snapshot of the MetadataManager.
- SnapshotGenerator creates SnapshotFile, populates Header and Sections.
- SnapshotGenerator passes SnapshotFile to DLedger.
- Snapshot ends.
The Snapshot Load process is similar to the Save process.
## Steps
- [ ] Update the version of DLedger in RocketMQ.
- [ ] Build the SnapshotAbleMetadataManager interface.
- [ ] Let ReplicasInfoManager implement the SnapshotAbleMetadataManager interface to implement metadata snapshots.
- [ ] Build SnapshotGenerator and SectionBuilder.
- [ ] Implement the onSnapshotLoad and onSnapshotSave APIs in ControllerStateMachine.
- [ ] Add more integration tests to prove the correctness of Controller Snapshot.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@rocketmq.apache.org.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [I] Support DLedger Controller Snapshot [rocketmq]
Posted by "github-actions[bot] (via GitHub)" <gi...@apache.org>.
github-actions[bot] commented on issue #5585:
URL: https://github.com/apache/rocketmq/issues/5585#issuecomment-1858625937
This issue is stale because it has been open for 365 days with no activity. It will be closed in 3 days if no further activity occurs.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@rocketmq.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [I] Support DLedger Controller Snapshot [rocketmq]
Posted by "github-actions[bot] (via GitHub)" <gi...@apache.org>.
github-actions[bot] closed issue #5585: Support DLedger Controller Snapshot
URL: https://github.com/apache/rocketmq/issues/5585
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@rocketmq.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [I] Support DLedger Controller Snapshot [rocketmq]
Posted by "github-actions[bot] (via GitHub)" <gi...@apache.org>.
github-actions[bot] commented on issue #5585:
URL: https://github.com/apache/rocketmq/issues/5585#issuecomment-1861889271
This issue was closed because it has been inactive for 3 days since being marked as stale.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@rocketmq.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org