Postgres- pgsql花费更多时间从超过15亿行的表中检索数据

Question

如何优化表或查询以下pgsql查询（需要34分钟才能获得770条记录）？已经将索引添加到表中的少数列。不知道还有什么可以让这个查询

查询：

SELECT 
    min(p.start_timestamp AT TIME ZONE p.timezone AT TIME ZONE 'America/Los_Angeles') as Date, 
    'America/Los_Angeles' AS Timezone, 
    sum(GREATEST(0, p.value)) as Value, 
    p.uom as UnitOfMeasurement 
FROM 
    pv.bsa_vessel_vs p         
WHERE 
     p.start_timestamp AT TIME ZONE p.timezone >= '2017-01-01' 
    and p.start_timestamp AT TIME ZONE p.timezone < '2017-02-01' 
    and p.vessel_serial_number ='U57625059' 
GROUP BY 
    date_trunc('hour', p.start_timestamp AT TIME ZONE p.timezone AT TIME ZONE 'America/Los_Angeles'), p.uom 
ORDER BY 
    Date ; 

表：

CREATE TABLE pv.bsa_vessel_vs 
(
    bsa_vessel_vs_id bigserial NOT NULL, 
    data_source_id bigint NOT NULL, 
    start_timestamp timestamp without time zone NOT NULL, 
    end_timestamp timestamp without time zone NOT NULL, 
    value numeric(12,4) NOT NULL, 
    uom text NOT NULL, 
    timezone text NOT NULL, 
    created_timestamp timestamp without time zone DEFAULT now(), 
    updated_timestamp timestamp without time zone DEFAULT now(), 
    vessel_serial_number text NOT NULL, 
    CONSTRAINT bsa_vessel_vs_pkey PRIMARY KEY (bsa_vessel_vs_id), 
    CONSTRAINT bsa_vessel_vs_data_source_id_fkey FOREIGN KEY (data_source_id) 
     REFERENCES pv.data_source (data_source_id) MATCH SIMPLE 
     ON UPDATE NO ACTION ON DELETE RESTRICT 
) 
WITH (
    OIDS=FALSE 
); 

CREATE INDEX pm_start_timestamp_ndex 
    ON pv.bsa_vessel_vs 
    USING btree 
    (start_timestamp DESC NULLS LAST); 

CREATE INDEX bsa_vessel_vs_meter_ts_idx 
    ON pv.bsa_vessel_vs 
    USING btree 
    (vessel_serial_number COLLATE pg_catalog."default", start_timestamp, end_timestamp); 


CREATE UNIQUE INDEX bsa_vessel_vs_u_idx 
    ON pv.bsa_vessel_vs 
    USING btree 
    (data_source_id, vessel_serial_number COLLATE pg_catalog."default", start_timestamp, end_timestamp DESC); 

感谢 Karthey

Answer 1

更改索引，以便它包含相同表达你在你的WHERE条款中使用，即：

CREATE INDEX bsa_vessel_vs_meter_ts_2_idx 
    ON bsa_vessel_vs 
    USING btree 
    (vessel_serial_number COLLATE pg_catalog."default", 
    (start_timestamp AT TIME ZONE timezone), 
    (start_timestamp AT TIME ZONE timezone) 
); 

当您定义的索引，你会得到一个使用它的执行计划：

 
| QUERY PLAN                                                               | 
| :-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | 
| Sort (cost=69.60..69.70 rows=39 width=83)                                                       | 
| Sort Key: (min(timezone('America/Los_Angeles'::text, timezone(timezone, start_timestamp))))                                           | 
| -> HashAggregate (cost=67.79..68.57 rows=39 width=83)                                                    | 
|   Group Key: date_trunc('hour'::text, timezone('America/Los_Angeles'::text, timezone(timezone, start_timestamp))), uom                                   | 
|   -> Index Scan using bsa_vessel_vs_meter_ts_2_idx on bsa_vessel_vs p (cost=0.28..67.20 rows=39 width=44)                                      | 
|    Index Cond: ((vessel_serial_number = 'U57625059'::text) AND (timezone(timezone, start_timestamp) >= '2017-01-01 00:00:00+00'::timestamp with time zone) AND (timezone(timezone, start_timestamp) < '2017-02-01 00:00:00+00'::timestamp with time zone)) | 

然而，如果指数是不那里，PostgreSQL的度假胜地，以全表扫描：

 
| QUERY PLAN                                                                | 
| :---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | 
| Sort (cost=298.84..298.94 rows=39 width=83)                                                       | 
| Sort Key: (min(timezone('America/Los_Angeles'::text, timezone(timezone, start_timestamp))))                                           | 
| -> GroupAggregate (cost=296.35..297.81 rows=39 width=83)                                                   | 
|   Group Key: (date_trunc('hour'::text, timezone('America/Los_Angeles'::text, timezone(timezone, start_timestamp)))), uom                                   | 
|   -> Sort (cost=296.35..296.45 rows=39 width=44)                                                    | 
|    Sort Key: (date_trunc('hour'::text, timezone('America/Los_Angeles'::text, timezone(timezone, start_timestamp)))), uom                                  | 
|    -> Seq Scan on bsa_vessel_vs p (cost=0.00..295.32 rows=39 width=44)                                              | 
|      Filter: ((vessel_serial_number = 'U57625059'::text) AND (timezone(timezone, start_timestamp) >= '2017-01-01 00:00:00+00'::timestamp with time zone) AND (timezone(timezone, start_timestamp) < '2017-02-01 00:00:00+00'::timestamp with time zone)) | 

您可以在检查所有的设置dbfiddle here