Add deltalake source and sink for topsql

[yibin87]

TopSQL v2 Source and Delta Lake Sinks Implementation

Overview

This PR implements a complete TopSQL data collection and storage solution, including:

1. topsql_v2 source - Next-generation TopSQL data source that collects TopSQL data from TiDB and TiKV clusters
2. topsql_data_deltalake sink - Writes TopSQL data (tidb_topsql, tikv_topsql, tikv_topregion) to Delta Lake
3. topsql_meta_deltalake sink - Writes TopSQL metadata (SQL meta and Plan meta) to Delta Lake with deduplication

Key Features

1. TopSQL v2 Source (topsql_v2)

Data Output

• TiDB TopSQL Data: tidb_topsql table containing SQL execution statistics
• TiKV TopSQL Data: tikv_topsql table containing TiKV-level SQL statistics
• TiKV TopRegion Data: tikv_topregion table containing Region-level statistics
• SQL Meta: topsql_sql_meta table containing SQL text and metadata
• Plan Meta: topsql_plan_meta table containing execution plan information

Configuration Options

sources:
  topsql:
    type: "topsql_v2"
    pd_address: "127.0.0.1:2379"
    tidb_group: "tidb-cluster"           # Optional: Filter TiDB instance group
    label_k8s_instance: "tidb"            # Optional: Kubernetes label filter
    downsampling_interval: 60             # Downsampling interval (seconds)
    top_n: 100                            # Keep Top N SQL statements
    topology_fetch_interval_seconds: 30   # Topology fetch interval
    init_retry_delay_seconds: 1.0         # Initial retry delay

2. TopSQL Data Delta Lake Sink (topsql_data_deltalake)

Core Features

• Multi-Table Support: Automatically identifies source_table field and routes data to corresponding Delta Lake tables
• Date Partitioning: Supports automatic date partitioning for easier data management and query optimization
• Batch Processing: Supports batch writes to improve write performance
• Time-Based Flush: Supports automatic flush based on time intervals

Supported Data Types

• tidb_topsql: TiDB TopSQL data
• tikv_topsql: TiKV TopSQL data
• tikv_topregion: TiKV TopRegion data

Schema Definition

Contains complete TopSQL metric fields:

• Timestamp and date fields
• SQL/Plan digest
• Database and table information
• Performance metrics: CPU time, execution count, network traffic, logical reads/writes, etc.
• TiKV-specific fields: Region ID, read/write keys, logical read/write bytes, etc.

3. TopSQL Meta Delta Lake Sink (topsql_meta_deltalake)

Core Features

• Deduplication Mechanism: LRU Cache-based deduplication to avoid writing duplicate metadata
  • Maintains separate caches for SQL Meta and Plan Meta
  • Deduplication key format: {digest}_{date} (e.g., sql_digest_2024-01-01)
  • Supports configurable cache capacity (default 10000)
• Event Buffering: Implements event buffering mechanism for batch writes
• Auto Flush:
  • Based on time interval (max_delay_secs, default 180 seconds)
  • Based on buffer size (EVENT_BUFFER_MAX_SIZE, default 1000)

Supported Data Types

• topsql_sql_meta: SQL metadata (normalized_sql, sql_digest, etc.)
• topsql_plan_meta: Plan metadata (normalized_plan, encoded_normalized_plan, etc.)

Configuration Examples

Complete Configuration Example

sources:
  topsql:
    type: "topsql_v2"
    pd_address: "127.0.0.1:2379"
    downsampling_interval: 30
    top_n: 80

transforms:
  route_by_type:
    type: "route"
    inputs:
      - "topsql"
    route:
      non_meta_logs: '.source_table == "tidb_topsql" || .source_table == "tikv_topsql" || .source_table == "tikv_topregion"'
      meta_logs: '.source_table == "topsql_sql_meta" || .source_table == "topsql_plan_meta"'

sinks:
  # TopSQL data write
  topsql_data:
    type: "topsql_data_deltalake"
    inputs:
      - "route_by_type.non_meta_logs"
    base_path: "s3://bucket-name/topsql-data"
    batch_size: 10000
    max_delay_secs: 60
    timeout_secs: 30
    bucket: "bucket-name"
    region: "us-east-1"
    auth:
      assume_role: "arn:aws:iam::123456789012:role/vector-role"
    acknowledgements:
      enabled: true

  # TopSQL metadata write (with deduplication)
  topsql_meta:
    type: "topsql_meta_deltalake"
    inputs:
      - "route_by_type.meta_logs"
    base_path: "s3://bucket-name/topsql-meta"
    batch_size: 1000
    max_delay_secs: 180
    meta_cache_capacity: 10000
    bucket: "bucket-name"
    region: "us-east-1"
    endpoint: "https://oss-cn-hangzhou.aliyuncs.com"  # Alibaba Cloud OSS
    force_path_style: false
    auth:
      # Use environment variables to configure Alibaba Cloud RRSA
      # ALIBABA_CLOUD_OIDC_TOKEN_FILE=/var/run/secrets/tokens/oidc-token
      # ALIBABA_CLOUD_ROLE_ARN=acs:ram::123456789012:role/vector-role
    acknowledgements:
      enabled: true

[pingcap-cla-assistant]

All committers have signed the CLA.

[zhoucai-pingcap]

我们现在只支持aws和alicloud？

[yibin87]

嗯 是的

[zhoucai-pingcap]

meta信息和data信息只是存储格式不同，但是mod.rs大部分是重复的，有必要分开吗？

[yibin87]

嗯 好观点。目前分开的话好处在于，两部分的数据完全分开，这样两部分的 batch 不会相互影响，同时两部分实际处理逻辑还是不一样的，后续可以考虑把 不同云的 适配这部分逻辑单独提取出来，像 deltalake_writer 这样，这样两部分 mod 其实就基本是个壳子了

[zhoucai-pingcap]

当前版本的格式应该不用改吧，为什么要加入新的字段呢？

[yibin87]

因为变更了协议，所以一些代码路径需要补全，另外加字段其实不影响最后存 vm 数据，因为 metrics 是0 的情况下，是会被过滤掉的。我把不影响编译部分的代码更新逻辑去掉吧

[yibin87]

更新了下，目前只保留了一些为了通过编译的改动，还有一些测试中的变动，原来 topsql 到 vm 的数据不会有任何变化