demo #1732
demo #1732
Conversation
ti-chi-bot
bot
added
do-not-merge/needs-linked-issue
release-note
|
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: The full list of commands accepted by this bot can be found here.
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing |
ti-chi-bot
bot
added
the
size/XXL
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Summary of Changes
Hello @hongyunyan, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!
This pull request introduces a significant architectural enhancement to the TiCDC sink component by implementing a new TxnSink for MySQL and TiDB. The core purpose of this change is to shift from row-level DML application to transaction-level processing, which offers substantial benefits in terms of performance and data consistency. The new sink collects DML events, groups them into complete transactions, and then generates optimized SQL statements that are executed in batches. A sophisticated conflict detection mechanism ensures that transactions can be processed concurrently where possible, without violating causal dependencies. This change lays the groundwork for more robust and efficient data replication to transactional databases.
Highlights
- New Transaction-Level Sink: This PR introduces a new
TxnSink(Transaction Sink) specifically designed for MySQL and TiDB targets. Unlike row-by-row processing, this sink processes and applies DML events at the transaction level, aiming to improve performance and ensure transactional atomicity at the destination. - Optimized SQL Execution with Batching: A new
DBExecutorcomponent is added to handle the actual database execution. It intelligently batches SQL statements, combining multiple DMLs from a single transaction into a singleBEGIN;...;COMMIT;block. This significantly reduces network round-trips and improves write efficiency. - Smart SQL Generation: The
SQLGeneratoris responsible for converting raw DML events into optimized SQL. For inserts and updates, it leveragesINSERT ON DUPLICATE KEY UPDATEstatements, which can handle both new rows and existing rows efficiently within a single SQL command. - Concurrency with Conflict Detection: A
ConflictDetectoris implemented to manage dependencies between transactions. This allows theTxnSinkto process non-conflicting transactions concurrently across multiple workers, maximizing throughput while preserving the correct causal order of operations. - Transaction Event Buffering: The
TxnStoreacts as an in-memory buffer, collecting DML events and grouping them by their commit and start timestamps. This enables the system to form complete transactions before processing, which is crucial for the transaction-level sink's operation.
Using Gemini Code Assist
The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.
Invoking Gemini
You can request assistance from Gemini at any point in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.
| Feature | Command | Description |
|---|---|---|
| Code Review | /gemini review |
Performs a code review for the current pull request in its current state. |
| Pull Request Summary | /gemini summary |
Provides a summary of the current pull request in its current state. |
| Comment | @gemini-code-assist | Responds in comments when explicitly tagged, both in issue comments and review comments. |
| Help | /gemini help |
Displays a list of available commands. |
Customization
To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.
Limitations & Feedback
Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.
You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.
Footnotes
-
Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩
There was a problem hiding this comment.
Code Review
This pull request introduces a new txnsink for transaction-level SQL output, which is a significant feature. The implementation uses a pipelined architecture with channels and goroutines for processing events, which is a solid approach. However, there are several critical issues in the database execution logic, including incorrect transaction batching and a buggy rollback mechanism, which could lead to data inconsistency or errors. Additionally, there are multiple places with hardcoded configuration values (such as timeouts, buffer sizes, and batch sizes) that should be made configurable for production use. There's also some duplicated code for conflict detection that could be refactored for better maintainability.
| if len(batch) == 0 { | ||
| return nil | ||
| } | ||
|
|
||
| log.Debug("txnSink: executing SQL batch", | ||
| zap.Int("batchSize", len(batch))) | ||
|
|
||
| // If batch size is 1, execute directly (SQL already contains BEGIN/COMMIT) | ||
| if len(batch) == 1 { | ||
| txnSQL := batch[0] | ||
| for _, sql := range txnSQL.SQLs { | ||
| ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second) | ||
| _, execErr := e.db.ExecContext(ctx, sql) | ||
| cancel() | ||
|
|
||
| if execErr != nil { | ||
| log.Error("txnSink: failed to execute single SQL", | ||
| zap.String("sql", sql), | ||
| zap.Uint64("commitTs", txnSQL.TxnGroup.CommitTs), | ||
| zap.Uint64("startTs", txnSQL.TxnGroup.StartTs), | ||
| zap.Error(execErr)) | ||
| return errors.Trace(execErr) | ||
| } | ||
| } | ||
|
|
||
| log.Debug("txnSink: successfully executed single transaction", | ||
| zap.Uint64("commitTs", txnSQL.TxnGroup.CommitTs), | ||
| zap.Uint64("startTs", txnSQL.TxnGroup.StartTs)) | ||
| return nil | ||
| } | ||
|
|
||
| // For multiple transactions, use explicit transaction and combine SQLs | ||
| tx, err := e.db.Begin() | ||
| if err != nil { | ||
| return errors.Trace(err) | ||
| } | ||
| defer func() { | ||
| if err != nil { | ||
| tx.Rollback() | ||
| } | ||
| }() | ||
|
|
||
| // Build combined SQL from all transactions | ||
| var combinedSQL strings.Builder | ||
|
|
||
| // Collect all SQL statements from all transactions | ||
| for _, txnSQL := range batch { | ||
| for _, sql := range txnSQL.SQLs { | ||
| // Keep the original SQL with BEGIN/COMMIT | ||
| cleanSQL := strings.TrimSpace(sql) | ||
| if len(cleanSQL) > 0 { | ||
| combinedSQL.WriteString(cleanSQL) | ||
| if !strings.HasSuffix(cleanSQL, ";") { | ||
| combinedSQL.WriteString(";") | ||
| } | ||
| } | ||
| } | ||
| } | ||
|
|
||
| finalSQL := combinedSQL.String() | ||
|
|
||
| log.Debug("txnSink: executing combined SQL batch with explicit transaction", | ||
| zap.String("sql", finalSQL), | ||
| zap.Int("batchSize", len(batch))) | ||
|
|
||
| // Execute the combined SQL within transaction | ||
| ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second) | ||
| _, execErr := tx.ExecContext(ctx, finalSQL) | ||
| cancel() | ||
|
|
||
| if execErr != nil { | ||
| log.Error("txnSink: failed to execute SQL batch", | ||
| zap.String("sql", finalSQL), | ||
| zap.Int("batchSize", len(batch)), | ||
| zap.Error(execErr)) | ||
| return errors.Trace(execErr) | ||
| } | ||
|
|
||
| // Commit the transaction | ||
| if err = tx.Commit(); err != nil { | ||
| log.Error("txnSink: failed to commit batch transaction", | ||
| zap.Int("batchSize", len(batch)), | ||
| zap.Error(err)) | ||
| return errors.Trace(err) | ||
| } | ||
|
|
||
| log.Debug("txnSink: successfully executed SQL batch", | ||
| zap.Int("batchSize", len(batch)), | ||
| zap.Int("sqlLength", len(finalSQL))) | ||
|
|
||
| return nil |
There was a problem hiding this comment.
The current implementation of ExecuteSQLBatch has two critical issues. First, the logic for batching multiple transactions (len(batch) > 1) is incorrect because it wraps already-transactional SQL strings (with BEGIN/COMMIT) inside another transaction, which will fail. Second, the error handling for len(batch) > 1 is buggy and does not correctly roll back the transaction on ExecContext failure.
The function should be simplified to iterate through the batch and execute each TxnSQL as a separate, self-contained transaction, as prepared by the sql_generator. This resolves both issues and also correctly handles the case where a single TxnSQL might contain multiple statements that need to be executed atomically.
if len(batch) == 0 {
return nil
}
log.Debug("txnSink: executing SQL batch",
zap.Int("batchSize", len(batch)))
for _, txnSQL := range batch {
// Each TxnSQL is a self-contained transaction with BEGIN/COMMIT.
if len(txnSQL.SQLs) == 0 {
continue
}
// sql_generator is expected to generate a single SQL string.
sql := txnSQL.SQLs[0]
ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
_, execErr := e.db.ExecContext(ctx, sql)
cancel()
if execErr != nil {
log.Error("txnSink: failed to execute transaction SQL",
zap.String("sql", sql),
zap.Uint64("commitTs", txnSQL.TxnGroup.CommitTs),
zap.Uint64("startTs", txnSQL.TxnGroup.StartTs),
zap.Error(execErr))
return errors.Trace(execErr)
}
log.Debug("txnSink: successfully executed transaction",
zap.Uint64("commitTs", txnSQL.TxnGroup.CommitTs),
zap.Uint64("startTs", txnSQL.TxnGroup.StartTs))
}
return nil| txnConfig := &txnsink.TxnSinkConfig{ | ||
| MaxConcurrentTxns: 16, | ||
| BatchSize: 16, | ||
| FlushInterval: 100, | ||
| MaxSQLBatchSize: 1024 * 16, // 1MB | ||
| } |
There was a problem hiding this comment.
The configuration for txnSink is hardcoded. These values should be made configurable, for example, through the sink URI, to allow for tuning in different environments.
Also, the comment for MaxSQLBatchSize is incorrect. 1024 * 16 is 16KB, not 1MB. 1MB would be 1024 * 1024.
| txnConfig := &txnsink.TxnSinkConfig{ | |
| MaxConcurrentTxns: 16, | |
| BatchSize: 16, | |
| FlushInterval: 100, | |
| MaxSQLBatchSize: 1024 * 16, // 1MB | |
| } | |
| txnConfig := &txnsink.TxnSinkConfig{ | |
| MaxConcurrentTxns: 16, // TODO: make this configurable | |
| BatchSize: 16, // TODO: make this configurable | |
| FlushInterval: 100, // TODO: make this configurable | |
| MaxSQLBatchSize: 1024 * 1024, // 1MB | |
| } |
| if len(batch) == 1 { | ||
| txnSQL := batch[0] | ||
| for _, sql := range txnSQL.SQLs { | ||
| ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second) |
There was a problem hiding this comment.
The 30-second timeout for ExecContext is hardcoded. This should be configurable to allow users to adjust it based on their downstream database performance and workload.
| dmlEventChan: make(chan *commonEvent.DMLEvent, 10000), | ||
| checkpointChan: make(chan uint64, 100), | ||
| txnChan: make(chan *TxnGroup, 10000), | ||
| sqlChan: make(chan *TxnSQL, 10000), |
There was a problem hiding this comment.
The channel buffer sizes are hardcoded. These values can have a significant impact on performance and memory usage. It would be better to make them configurable.
| select { | ||
| case s.sqlChan <- txnSQL: | ||
| return nil | ||
| default: | ||
| return errors.New("SQL channel is full") | ||
| } |
There was a problem hiding this comment.
When the sqlChan is full, processTxnGroup returns an error, which will cause the entire sink to shut down. This seems too aggressive. A full channel might be a temporary condition due to a burst of data or a slow downstream. Consider blocking until there is space in the channel, or implementing a retry mechanism with backoff.
| opt := TxnCacheOption{ | ||
| Count: 10, // Default worker count | ||
| Size: 1024, | ||
| BlockStrategy: BlockStrategyWaitEmpty, | ||
| } |
There was a problem hiding this comment.
The TxnCacheOption for the ConflictDetector is hardcoded. These values, especially Count (number of workers) and Size (cache size), should be configurable to allow tuning for different workloads.
| func ConflictKeysForTxnGroup(txnGroup *TxnGroup) []uint64 { | ||
| if len(txnGroup.Events) == 0 { | ||
| return nil | ||
| } | ||
|
|
||
| hashRes := make(map[uint64]struct{}) | ||
| hasher := fnv.New32a() | ||
|
|
||
| // Iterate through all events in the transaction group | ||
| for _, event := range txnGroup.Events { | ||
| // Iterate through all rows in the event | ||
| event.Rewind() | ||
| for { | ||
| rowChange, ok := event.GetNextRow() | ||
| if !ok { | ||
| break | ||
| } | ||
|
|
||
| // Generate keys for each row | ||
| keys := genRowKeysForTxnGroup(rowChange, event.TableInfo, event.DispatcherID) | ||
| for _, key := range keys { | ||
| if n, err := hasher.Write(key); n != len(key) || err != nil { | ||
| log.Panic("transaction key hash fail") | ||
| } | ||
| hashRes[uint64(hasher.Sum32())] = struct{}{} | ||
| hasher.Reset() | ||
| } | ||
| } | ||
| event.Rewind() | ||
| } | ||
|
|
||
| keys := make([]uint64, 0, len(hashRes)) | ||
| for key := range hashRes { | ||
| keys = append(keys, key) | ||
| } | ||
| return keys | ||
| } | ||
|
|
||
| // genRowKeysForTxnGroup generates row keys for a row change in transaction group | ||
| func genRowKeysForTxnGroup(rowChange commonEvent.RowChange, tableInfo *common.TableInfo, dispatcherID common.DispatcherID) [][]byte { | ||
| var keys [][]byte | ||
|
|
||
| if !rowChange.Row.IsEmpty() { | ||
| for iIdx, idxColID := range tableInfo.GetIndexColumns() { | ||
| key := genKeyListForTxnGroup(&rowChange.Row, iIdx, idxColID, dispatcherID, tableInfo) | ||
| if len(key) == 0 { | ||
| continue | ||
| } | ||
| keys = append(keys, key) | ||
| } | ||
| } | ||
| if !rowChange.PreRow.IsEmpty() { | ||
| for iIdx, idxColID := range tableInfo.GetIndexColumns() { | ||
| key := genKeyListForTxnGroup(&rowChange.PreRow, iIdx, idxColID, dispatcherID, tableInfo) | ||
| if len(key) == 0 { | ||
| continue | ||
| } | ||
| keys = append(keys, key) | ||
| } | ||
| } | ||
| if len(keys) == 0 { | ||
| // use dispatcherID as key if no key generated (no PK/UK), | ||
| // no concurrence for rows in the same dispatcher. | ||
| log.Debug("Use dispatcherID as the key", zap.Any("dispatcherID", dispatcherID)) | ||
| tableKey := make([]byte, 8) | ||
| binary.BigEndian.PutUint64(tableKey, uint64(dispatcherID.GetLow())) | ||
| keys = [][]byte{tableKey} | ||
| } | ||
| return keys | ||
| } | ||
|
|
||
| // genKeyListForTxnGroup generates a key list for a row in transaction group | ||
| func genKeyListForTxnGroup( | ||
| row *chunk.Row, iIdx int, idxColID []int64, dispatcherID common.DispatcherID, tableInfo *common.TableInfo, | ||
| ) []byte { | ||
| var key []byte | ||
| for _, colID := range idxColID { | ||
| info, ok := tableInfo.GetColumnInfo(colID) | ||
| // If the index contain generated column, we can't use this key to detect conflict with other DML, | ||
| if !ok || info == nil || info.IsGenerated() { | ||
| return nil | ||
| } | ||
| offset, ok := tableInfo.GetRowColumnsOffset()[colID] | ||
| if !ok { | ||
| log.Warn("can't find column offset", zap.Int64("colID", colID), zap.String("colName", info.Name.O)) | ||
| return nil | ||
| } | ||
| value := common.ExtractColVal(row, info, offset) | ||
| // if a column value is null, we can ignore this index | ||
| if value == nil { | ||
| return nil | ||
| } | ||
|
|
||
| val := common.ColumnValueString(value) | ||
| if columnNeeds2LowerCase(info.GetType(), info.GetCollate()) { | ||
| val = strings.ToLower(val) | ||
| } | ||
|
|
||
| key = append(key, []byte(val)...) | ||
| key = append(key, 0) | ||
| } | ||
| if len(key) == 0 { | ||
| return nil | ||
| } | ||
| tableKey := make([]byte, 16) | ||
| binary.BigEndian.PutUint64(tableKey[:8], uint64(iIdx)) | ||
| binary.BigEndian.PutUint64(tableKey[8:], dispatcherID.GetLow()) | ||
| key = append(key, tableKey...) | ||
| return key | ||
| } |
There was a problem hiding this comment.
The functions ConflictKeysForTxnGroup, genRowKeysForTxnGroup, and genKeyListForTxnGroup appear to be duplicating logic that exists in pkg/sink/mysql/txn.go for conflict key generation. Duplicating code makes maintenance harder. It would be better to refactor the existing logic into a shared package so it can be reused by both the mysql sink and this new txn sink.
|
[FORMAT CHECKER NOTIFICATION] Notice: To remove the 📖 For more info, you can check the "Contribute Code" section in the development guide. |
|
@hongyunyan: The following tests failed, say
Full PR test history. Your PR dashboard. Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes-sigs/prow repository. I understand the commands that are listed here. |
1 participant
What problem does this PR solve?
Issue Number: close #xxx
What is changed and how it works?
Check List
Tests
Questions
Will it cause performance regression or break compatibility?
Do you need to update user documentation, design documentation or monitoring documentation?
Release note