server: implement spill disk for cursorFetch result

[YangKeao]

What problem does this PR solve?

Issue Number: close #43233

What is changed and how it works?

1. Use the rowContainer to store the rows after execution.
2. Reset the memory/disk tracker when the statement is closed.

TODO:

• Add tests
• Consider the behavior of Reset.
• Consider carefully about the server status. 56 cases of mysql_client_test failed #44672 seems to have shown some bugs about server status.

Check List

Tests

• Unit test
• Integration test (TODO)
• Manual test (add detailed scripts or steps below)
• No code

I used the following codes to do some basic tests. The preloaded data is 5M rows sysbench ./oltp_insert.lua :

public static void tidbBigSelectTest() throws SQLException, InterruptedException {
    // the following are test codes for MySQL
    ConnectionImpl conn = (ConnectionImpl) DriverManager.getConnection("jdbc:mysql://127.0.0.1:4000/test?useCursorFetch=true&defaultFetchSize=1000", "root", "");
    conn.prepareStatement("set @@tidb_mem_quota_query = 8 << 20\n").execute();
    conn.prepareStatement("begin").execute();

    String sql = "SELECT * FROM test.sbtest1";
    PreparedStatement stmt = conn.prepareStatement(sql, ResultSet.TYPE_FORWARD_ONLY, ResultSet.CONCUR_READ_ONLY);
    stmt.setFetchSize(1000);
    Instant beforeExecuteTime = Clock.systemUTC().instant();
    ResultSet rs = stmt.executeQuery();
    Instant afterExecuteTime = Clock.systemUTC().instant();
    System.out.println("It takes " + (Duration.between(beforeExecuteTime, afterExecuteTime).toString()) + " to EXECUTE QUERY");

    ResultSetMetaData metaData = rs.getMetaData();
    int columnCount = metaData.getColumnCount();
    int batchSize = stmt.getFetchSize();
    int rowCount = 0;
    int totalCount = 0;
    while (rs.next()) {
        rowCount++;
        totalCount++;
        Object[] columns = new Object[columnCount];
        for (int i = 1; i <= columnCount; i++) {
            columns[i - 1] = rs.getObject(i);
        }
        Row row = new Row(columns);
        if (rowCount == batchSize) {
            System.out.println("Fetched " + batchSize + " rows");
            System.out.println("allocated memory is " + (Runtime.getRuntime().totalMemory()-Runtime.getRuntime().freeMemory()));
            System.out.println("TotalCount " + totalCount);
            rowCount = 0;
        }
        if (totalCount == 4999999) {
            System.out.println("here");
        }
    }
    if (rowCount > 0) {
        System.out.println("allocated memory is " + (Runtime.getRuntime().totalMemory()-Runtime.getRuntime().freeMemory()));
        System.out.println("Fetched " + rowCount + " rows");
        System.out.println("TotalCount " + totalCount);
    }

    conn.prepareStatement("commit").execute();
}

Release note

Automatically spill the memory exceeding `tidb_mem_quota_query` into the disk

[YangKeao]

/test tiprow_fast_test

[ti-chi-bot]

@YangKeao: The specified target(s) for /test were not found.
The following commands are available to trigger required jobs:

• /test build
• /test canary-scan-security
• /test check-dev
• /test check-dev2
• /test mysql-test
• /test pingcap/tidb/canary_ghpr_unit_test
• /test pull-integration-br-test
• /test pull-integration-mysql-test
• /test unit-test

Use /test all to run the following jobs that were automatically triggered:

• pingcap/tidb/ghpr_build
• pingcap/tidb/ghpr_check
• pingcap/tidb/ghpr_check2
• pingcap/tidb/ghpr_mysql_test
• pingcap/tidb/ghpr_unit_test

In response to this:

/test tiprow_fast_test

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/test-infra repository.

[ti-chi-bot]

PR needs rebase.

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/test-infra repository.

[YangKeao]

Github doesn't update this PR after I rebase and solve the conflict. It's really weird. The content of this PR is different from the content in my branch: https://github.com/YangKeao/tidb/tree/cursor-fetch-persist (e.g. the file /util/chunk/row_container_reader.go).

I'll try to close and reopen this PR to see whether it'll update 🤦 .

[YangKeao]

Good. Then it doesn't allow me to reopen it 😮‍💨 . I have to create a new one.