自我管理PostgreSQL分区表

Question

我正在尝试使用Postgres进行自我管理的分区表设置。这一切都围绕着这个功能，但我似乎无法让Postgres接受我的表名。自我管理分区表触发函数的任何想法或例子？

我目前的职能：

DECLARE
day integer;
year integer;
tablename text;
startdate text;
enddate text;
BEGIN
day:=date_part('doy',to_timestamp(NEW.date));
year:=date_part('year',to_timestamp(NEW.date));
tablename:='pings_'||year||'_'||day||'_'||NEW.id;
-- RAISE EXCEPTION 'tablename=%',tablename;
PERFORM 'tablename' FROM pg_tables WHERE 'schemaname'=tablename;
-- RAISE EXCEPTION 'found=%',FOUND;
IF FOUND <> TRUE THEN
    startdate:=date_part('year',to_timestamp(NEW.date))||'-'||date_part('month',to_timestamp(NEW.date))||'-'||date_part('day',to_timestamp(NEW.date));
    enddate:=startdate::timestamp + INTERVAL '1 day';
    EXECUTE 'CREATE TABLE $1 (
        CHECK ( date >= DATE $2 AND date < DATE $3 )
    ) INHERITS (pings)' USING quote_ident(tablename),startdate,enddate;
END IF;
EXECUTE 'INSERT INTO $1 VALUES (NEW.*)' USING quote_ident(tablename);
RETURN NULL;
END;

我希望它能够自动创建一个名为pings_YEAR_DOY_ID的表，但它始终会失败：

2011-10-24 13:39:04 CDT [15804]: [1-1] ERROR:  invalid input syntax for type double precision: "-" at character 45
2011-10-24 13:39:04 CDT [15804]: [2-1] QUERY:  SELECT date_part('year',to_timestamp( $1 ))+'-'+date_part('month',to_timestamp( $2 ))+'-'+date_part('day',to_timestamp( $3 ))
2011-10-24 13:39:04 CDT [15804]: [3-1] CONTEXT:  PL/pgSQL function "ping_partition" line 15 at assignment
2011-10-24 13:39:04 CDT [15804]: [4-1] STATEMENT:  INSERT INTO pings VALUES (0,0,5);

尝试2

应用更改后再修改它（date是unixtimestamp列，我的想法是选择时整数列比时间戳列快）。我收到以下错误，不确定我是否使用USING NEW的正确语法？

更新功能：

CREATE FUNCTION ping_partition() RETURNS trigger
    LANGUAGE plpgsql
    AS $_$DECLARE
day integer;
year integer;
tablename text;
startdate text;
enddate text;
BEGIN
day:=date_part('doy',to_timestamp(NEW.date));
year:=date_part('year',to_timestamp(NEW.date));
tablename:='pings_'||year||'_'||day||'_'||NEW.id;
-- RAISE EXCEPTION 'tablename=%',tablename;
PERFORM 'tablename' FROM pg_tables WHERE 'schemaname'=tablename;
-- RAISE EXCEPTION 'found=%',FOUND;
IF FOUND <> TRUE THEN
    startdate := to_char(to_timestamp(NEW.date), 'YYYY-MM-DD');
    enddate:=startdate::timestamp + INTERVAL '1 day';
    EXECUTE 'CREATE TABLE ' || quote_ident(tablename) || ' (
        CHECK ( date >= EXTRACT(EPOCH FROM DATE ' || quote_literal(startdate) || ')
            AND date < EXTRACT(EPOCH FROM DATE ' || quote_literal(enddate) || ') )
    ) INHERITS (pings)';
END IF;
EXECUTE 'INSERT INTO ' || quote_ident(tablename) || ' SELECT $1' USING NEW; 
RETURN NULL;
END;
$_$;

我的陈述：

INSERT INTO pings VALUES (0,0,5);

SQL错误：

ERROR:  column "date" is of type integer but expression is of type pings
LINE 1: INSERT INTO pings_1969_365_0 SELECT $1
                                            ^
HINT:  You will need to rewrite or cast the expression.
QUERY:  INSERT INTO pings_1969_365_0 SELECT $1
CONTEXT:  PL/pgSQL function "ping_partition" line 22 at EXECUTE statement

Answer 1

您正在将date_part()的double precision输出与text '-'混合。这对Po​​stgreSQL没有意义。您需要明确转换为text。但是有一种更简单的方法可以做到这一切：

startdate:=date_part('year',to_timestamp(NEW.date))
||'-'||date_part('month',to_timestamp(NEW.date))
||'-'||date_part('day',to_timestamp(NEW.date));

改为使用：

startdate := to_char(NEW.date, 'YYYY-MM-DD');

这也没有意义：

EXECUTE 'CREATE TABLE $1 (
        CHECK (date >= DATE $2 AND date < DATE $3 )
    ) INHERITS (pings)' USING quote_ident(tablename),startdate,enddate;

您只能使用USING子句提供值。 Read the manual here。请尝试改为：

EXECUTE 'CREATE TABLE ' || quote_ident(tablename) || ' (
            CHECK ("date" >= ''' || startdate || ''' AND
                   "date" <  ''' || enddate   || '''))
            INHERITS (ping)';

或者更好的是，使用format()。见下文。

此外，与@a_horse answered类似：您需要将文字值放在单引号中。

在此类似：

EXECUTE 'INSERT INTO $1 VALUES (NEW.*)' USING quote_ident(tablename);

相反：

EXECUTE 'INSERT INTO ' || quote_ident(tablename) || ' VALUES ($1.*)'
USING NEW;

相关答案：

• How to dynamically use TG_TABLE_NAME in PostgreSQL 8.2?

旁白：虽然PostgreSQL中的列名允许“date”，但它是reserved word in every SQL standard。不要将列命名为“日期”，否则会导致语法错误。

完整的工作演示

CREATE TABLE ping (ping_id integer, the_date date);

CREATE OR REPLACE FUNCTION trg_ping_partition()
  RETURNS trigger AS
$func$
DECLARE
   _tbl text := to_char(NEW.the_date, '"ping_"YYYY_DDD_') || NEW.ping_id;
BEGIN
   IF NOT EXISTS (
      SELECT 1
      FROM   pg_catalog.pg_class c
      JOIN   pg_catalog.pg_namespace n ON n.oid = c.relnamespace
      WHERE  n.nspname = 'public'  -- your schema
      AND    c.relname = _tbl
      AND    c.relkind = 'r') THEN

      EXECUTE format('CREATE TABLE %I (CHECK (the_date >= %L AND
                                              the_date <  %L)) INHERITS (ping)'
              , _tbl
              , to_char(NEW.the_date,     'YYYY-MM-DD')
              , to_char(NEW.the_date + 1, 'YYYY-MM-DD')
              );
   END IF;

   EXECUTE 'INSERT INTO ' || quote_ident(_tbl) || ' VALUES ($1.*)'
   USING NEW; 

   RETURN NULL;
END
$func$ LANGUAGE plpgsql SET search_path = public;

CREATE TRIGGER insbef
BEFORE INSERT ON ping
FOR EACH ROW EXECUTE PROCEDURE trg_ping_partition();

• 更新： Postgres的更高版本有更优雅的方法来检查表是否存在：

  • How to check if a table exists in a given schema

• to_char()可以date作为$1。这会自动转换为timestamp The manual on date / time functions

• （可选）SET the search_path了解您的功能范围，以避免更改search_path设置的不当行为。

• 多种其他简化和改进。比较代码。

试验：

INSERT INTO ping VALUES (1, now()::date);
INSERT INTO ping VALUES (2, now()::date);
INSERT INTO ping VALUES (2, now()::date + 1);
INSERT INTO ping VALUES (2, now()::date + 1);

SQL Fiddle.

Answer 2

PostgreSQL中的动态分区只是一个坏主意。您的代码在多用户环境中不安全。为了安全起见，您必须使用锁定，这会降低执行速度。最佳分区数约为一百。您可以提前轻松创建很多，以大大简化分区所需的逻辑。

Answer 3

您需要将日期文字放在单引号中。目前你正在执行这样的事情：

 CHECK ( date >= DATE 2011-10-25 AND date < DATE 2011-11-25 )

无效。在这种情况下，2011-10-25被解释为 2011减去10减25

您的代码需要使用日期文字周围的单引号创建SQL：

CHECK ( date >= DATE '2011-10-25' AND date < DATE '2011-11-25' )

Answer 4

我想出了整体而且效果很好，甚至在30天后自动删除。我希望这有助于未来的人们寻找自动分配触发功能。

CREATE FUNCTION ping_partition() RETURNS trigger
    LANGUAGE plpgsql
    AS $_$
DECLARE
_keepdate text;
_tablename text;
_startdate text;
_enddate text;
_result record;
BEGIN
_keepdate:=to_char(to_timestamp(NEW.date) - interval '30 days', 'YYYY-MM-DD');
_startdate := to_char(to_timestamp(NEW.date), 'YYYY-MM-DD');
_tablename:='pings_'||NEW.id||'_'||_startdate;
PERFORM 1
FROM   pg_catalog.pg_class c
JOIN   pg_catalog.pg_namespace n ON n.oid = c.relnamespace
WHERE  c.relkind = 'r'
AND    c.relname = _tablename
AND    n.nspname = 'pinglog';
IF NOT FOUND THEN
    _enddate:=_startdate::timestamp + INTERVAL '1 day';
    EXECUTE 'CREATE TABLE pinglog.' || quote_ident(_tablename) || ' (
        CHECK ( date >= EXTRACT(EPOCH FROM DATE ' || quote_literal(_startdate) || ')
            AND date < EXTRACT(EPOCH FROM DATE ' || quote_literal(_enddate) || ')
            AND id = ' || quote_literal(NEW.id) || '
        )
    ) INHERITS (pinglog.pings)';
    EXECUTE 'CREATE INDEX ' || quote_ident(_tablename||'_indx1') || ' ON pinglog.' || quote_ident(_tablename) || ' USING btree (microseconds) WHERE microseconds IS NULL';
    EXECUTE 'CREATE INDEX ' || quote_ident(_tablename||'_indx2') || ' ON pinglog.' || quote_ident(_tablename) || ' USING btree (date, id)';
    EXECUTE 'CREATE INDEX ' || quote_ident(_tablename||'_indx3') || ' ON pinglog.' || quote_ident(_tablename) || ' USING btree (date, id, microseconds) WHERE microseconds IS NULL';
END IF;
EXECUTE 'INSERT INTO ' || quote_ident(_tablename) || ' VALUES ($1.*)' USING NEW;
FOR _result IN SELECT * FROM pg_tables WHERE schemaname='pinglog' LOOP
    IF char_length(substring(_result.tablename from '[0-9-]*$')) <> 0 AND (to_timestamp(NEW.date) - interval '30 days') > to_timestamp(substring(_result.tablename from '[0-9-]*$'),'YYYY-MM-DD') THEN
        -- RAISE EXCEPTION 'timestamp=%,table=%,found=%',to_timestamp(substring(_result.tablename from '[0-9-]*$'),'YYYY-MM-DD'),_result.tablename,char_length(substring(_result.tablename from '[0-9-]*$'));
        -- could have it check for non-existant ids as well, or for archive bit and only delete if the archive bit is not set
        EXECUTE 'DROP TABLE ' || quote_ident(_result.tablename);
    END IF;
END LOOP;
RETURN NULL;
END;
$_$;