Question

问题设置（Postgesql 9.6）

我有一个表，其中分区按我的意图工作（1），一个没有（2）：

CASE（1）

表，在没有时区的s_date时间戳上进行分区NOT NULL

CREATE TABLE "diagnoseAW"."AWIORECORDERAWCOMMAND"
(
    "ebpZone" text COLLATE pg_catalog."default" NOT NULL,
    "elementName" text COLLATE pg_catalog."default" NOT NULL,
    command text COLLATE pg_catalog."default" NOT NULL,
    s_date timestamp without time zone NOT NULL,
    i_io_value text COLLATE pg_catalog."default" NOT NULL,
    source text COLLATE pg_catalog."default",
    time_created timestamp with time zone,
    user_created text COLLATE pg_catalog."default",
    CONSTRAINT "AWIORECORDERAWCOMMAND_pkey" PRIMARY KEY ("ebpZone", "elementName", command, s_date, i_io_value)
        USING INDEX TABLESPACE s1297_index
)
WITH (
    OIDS = FALSE
)
TABLESPACE s1297_data;

分区看起来像这样：

CREATE TABLE partitions."AWIORECORDERAWCOMMAND_2018_02_20"
(
    "ebpZone" ,
    "elementName" ,
    command ,
    s_date ,
    i_io_value ,
    source ,
    time_created ,
    user_created ,
    CONSTRAINT "AWIORECORDERAWCOMMAND_2018_02_20_NEW_pkey" PRIMARY KEY ("ebpZone", "elementName", command, s_date, i_io_value)
        USING INDEX TABLESPACE s1297_data,
    CONSTRAINT "AWIORECORDERAWCOMMAND_2018_02_20_s_date_check" CHECK (s_date::date = '2018-02-20'::date)
)
    INHERITS ("diagnoseAW"."AWIORECORDERAWCOMMAND")
WITH (
    OIDS = FALSE
)
TABLESPACE s1297_data;

执行此查询时：

SELECT "ebpZone", "elementName", command, s_date, i_io_value, source, time_created, user_created
                FROM "diagnoseAW"."AWIORECORDERAWCOMMAND"
                WHERE s_date::date = '2018-02-20'::date;

它工作正常，它只扫描所需的分区：

"Append  (cost=0.00..3083.05 rows=610 width=48)"
"  ->  Seq Scan on "AWIORECORDERAWCOMMAND"  (cost=0.00..0.00 rows=1 width=208)"
"        Filter: ((s_date)::date = '2018-02-20'::date)"
"  ->  Seq Scan on "AWIORECORDERAWCOMMAND_2018_02_20"  (cost=0.00..3083.05 rows=609 width=48)"
"        Filter: ((s_date)::date = '2018-02-20'::date)"

情况（2）：

表格，＆＃34; startDate＆＃34;带时区的时间戳NOT NULL

CREATE TABLE "diagnoseAW"."AWIORECORDERMOVEMENT"
(
    event text COLLATE pg_catalog."default",
    "ebpZone" text COLLATE pg_catalog."default" NOT NULL,
    "awName" text COLLATE pg_catalog."default" NOT NULL,
    "startDate" timestamp with time zone NOT NULL,
    "lossDate" timestamp with time zone NOT NULL,
    "controlDate" timestamp with time zone NOT NULL,
    "startDuration" numeric,
    "totalDuration" numeric,
    time_created timestamp with time zone,
    user_created text COLLATE pg_catalog."default",
    "lossEnd" boolean,
    "controlEnd" boolean,
    railtemp double precision,
    airtemp double precision,
    CONSTRAINT "AWIORECORDERMOVEMENT_pkey" PRIMARY KEY ("ebpZone", "awName", "startDate", "lossDate", "controlDate")
        USING INDEX TABLESPACE s1297_index
)
WITH (
    OIDS = FALSE
)
TABLESPACE s1297_data;

分区看起来像这样：

CREATE TABLE partitions."AWIORECORDERMOVEMENT_2018_02_20"
(
    event ,
    "ebpZone" ,
    "awName" ,
    "startDate" ,
    "lossDate" ,
    "controlDate" ,
    "startDuration" ,
    "totalDuration" ,
    time_created ,
    user_created ,
    "lossEnd" ,
    "controlEnd" ,
    railtemp ,
    airtemp ,
    CONSTRAINT "AWIORECORDERMOVEMENT_2018_02_20_NEW_pkey" PRIMARY KEY ("ebpZone", "awName", "startDate", "lossDate", "controlDate")
        USING INDEX TABLESPACE s1297_data,
    CONSTRAINT "AWIORECORDERMOVEMENT_2018_02_20_startDate_check" CHECK ("startDate"::date = '2018-02-20'::date)
)
    INHERITS ("diagnoseAW"."AWIORECORDERMOVEMENT")
WITH (
    OIDS = FALSE
)
TABLESPACE s1297_data;

执行此查询时：

SELECT event, "ebpZone", "awName", "startDate", "lossDate", "controlDate", "startDuration", "totalDuration", time_created, user_created, "lossEnd", "controlEnd", railtemp, airtemp
FROM "diagnoseAW"."AWIORECORDERMOVEMENT"
WHERE "startDate"::date='2018-02-20'::date;

它不能正常工作，它会扫描所有分区：

"Append  (cost=0.00..1386483.44 rows=241434 width=80)"
"  ->  Seq Scan on "AWIORECORDERMOVEMENT"  (cost=0.00..0.00 rows=1 width=242)"
"        Filter: (("startDate")::date = '2018-02-20'::date)"
"  ->  Seq Scan on "AWIORECORDERMOVEMENT_2017_01_02"  (cost=0.00..2566.48 rows=446 width=80)"
"        Filter: (("startDate")::date = '2018-02-20'::date)"
"  ->  Seq Scan on "AWIORECORDERMOVEMENT_2017_01_07"  (cost=0.00..1588.55 rows=276 width=80)"
"        Filter: (("startDate")::date = '2018-02-20'::date)"
"  ->  Seq Scan on "AWIORECORDERMOVEMENT_2017_01_09"  (cost=0.00..3137.61 rows=546 width=80)"
"        Filter: (("startDate")::date = '2018-02-20'::date)"
"  ->  Seq Scan on "AWIORECORDERMOVEMENT_2017_01_12"  (cost=0.00..3279.47 rows=569 width=80)"
"        Filter: (("startDate")::date = '2018-02-20'::date)"
"  ->  Seq Scan on "AWIORECORDERMOVEMENT_2017_01_13"  (cost=0.00..3390.38 rows=589 width=80)"
"        Filter: (("startDate")::date = '2018-02-20'::date)"
"  ->  Seq Scan on "AWIORECORDERMOVEMENT_2017_01_16"  (cost=0.00..3139.07 rows=546 width=80)"
"        Filter: (("startDate")::date = '2018-02-20'::date)"

...

问题

我看到的唯一区别是我用于分区的时间戳中有或没有时间戳。（另外在第二种情况下，分区有索引，不是第一种情况，但我认为它不相关）任何想法如何我不能扫描所有我的分区CASE（2）的查询？我最好不会重新加载所有数十亿行。

timestamp上的分区表::日期查询扫描所有分区

问题设置（Postgesql 9.6）

CASE（1）

情况（2）：

问题

0 个答案: