跳到主要内容
跳到主要内容

avgMergeState

描述​

MergeState 组合器可以应用于 avg 函数，以合并类型为 AverageFunction(avg, T) 的部分聚合状态，并返回一个新的中间聚合状态。

示例用法​

MergeState 组合器在多层级聚合场景中特别有用，在这些场景中，您希望组合预聚合状态并将其维护为状态（而不是最终状态）以进行进一步处理。为了说明这一点，我们将看一个示例，其中将单个服务器性能指标转换为跨多个级别的层次聚合：服务器级别 → 区域级别 → 数据中心级别。

首先，我们创建一个表来存储原始数据：

CREATE TABLE raw_server_metrics
(
    timestamp DateTime DEFAULT now(),
    server_id UInt32,
    region String,
    datacenter String,
    response_time_ms UInt32
)
ENGINE = MergeTree()
ORDER BY (region, server_id, timestamp);

我们将创建一个服务器级别的聚合目标表，并定义一个增量物化视图作为其插入触发器：

CREATE TABLE server_performance
(
    server_id UInt32,
    region String,
    datacenter String,
    avg_response_time AggregateFunction(avg, UInt32)
)
ENGINE = AggregatingMergeTree()
ORDER BY (region, server_id);

CREATE MATERIALIZED VIEW server_performance_mv
TO server_performance
AS SELECT
    server_id,
    region,
    datacenter,
    avgState(response_time_ms) AS avg_response_time
FROM raw_server_metrics
GROUP BY server_id, region, datacenter;

我们将在区域和数据中心级别执行相同的操作：

CREATE TABLE region_performance
(
    region String,
    datacenter String,
    avg_response_time AggregateFunction(avg, UInt32)
)
ENGINE = AggregatingMergeTree()
ORDER BY (datacenter, region);

CREATE MATERIALIZED VIEW region_performance_mv
TO region_performance
AS SELECT
    region,
    datacenter,
    avgMergeState(avg_response_time) AS avg_response_time
FROM server_performance
GROUP BY region, datacenter;

-- datacenter level table and materialized view

CREATE TABLE datacenter_performance
(
    datacenter String,
    avg_response_time AggregateFunction(avg, UInt32)
)
ENGINE = AggregatingMergeTree()
ORDER BY datacenter;

CREATE MATERIALIZED VIEW datacenter_performance_mv
TO datacenter_performance
AS SELECT
      datacenter,
      avgMergeState(avg_response_time) AS avg_response_time
FROM region_performance
GROUP BY datacenter;

然后我们将向源表插入示例原始数据：

INSERT INTO raw_server_metrics (timestamp, server_id, region, datacenter, response_time_ms) VALUES
    (now(), 101, 'us-east', 'dc1', 120),
    (now(), 101, 'us-east', 'dc1', 130),
    (now(), 102, 'us-east', 'dc1', 115),
    (now(), 201, 'us-west', 'dc1', 95),
    (now(), 202, 'us-west', 'dc1', 105),
    (now(), 301, 'eu-central', 'dc2', 145),
    (now(), 302, 'eu-central', 'dc2', 155);

我们将为每个级别编写三个查询：

SELECT
    server_id,
    region,
    avgMerge(avg_response_time) AS avg_response_ms
FROM server_performance
GROUP BY server_id, region
ORDER BY region, server_id;

┌─server_id─┬─region─────┬─avg_response_ms─┐
│       301 │ eu-central │             145 │
│       302 │ eu-central │             155 │
│       101 │ us-east    │             125 │
│       102 │ us-east    │             115 │
│       201 │ us-west    │              95 │
│       202 │ us-west    │             105 │
└───────────┴────────────┴─────────────────┘

我们可以插入更多数据：

INSERT INTO raw_server_metrics (timestamp, server_id, region, datacenter, response_time_ms) VALUES
    (now(), 101, 'us-east', 'dc1', 140),
    (now(), 201, 'us-west', 'dc1', 85),
    (now(), 301, 'eu-central', 'dc2', 135);

让我们再次检查数据中心级别的性能。注意整个聚合链是如何自动更新的：

SELECT
    datacenter,
    avgMerge(avg_response_time) AS avg_response_ms
FROM datacenter_performance
GROUP BY datacenter
ORDER BY datacenter;

┌─datacenter─┬────avg_response_ms─┐
│ dc1        │ 112.85714285714286 │
│ dc2        │                145 │
└────────────┴────────────────────┘

另见​