Skip to main content
Question

Partitioning problem + question

  • April 12, 2021
  • 3 replies
  • 13 views

kfruchtman
Forum|alt.badge.img+2

Hi Experts!
I have been trying to partition a billion rows table according to time by CALENDAR_HIERARCHY_DAY algorithm.
when I run the statement:
ALTER TABLE public.spins
PARTITION BY time::DATE
GROUP BY CALENDAR_HIERARCHY_DAY(time::DATE, 2, 1) REORGANIZE;

I get the error:
ROLLBACK 2648: Column time in PARTITION BY expression is not allowed, since it is not present in some projections
HINT: Projection "spins_pk" does not contain column "time"

When I try to drop the projection (I don't really need it..) I get:

WARNING 7000: The projection 'spins_pk' was used to enforce the enabled key constraint 'pk', and may be regenerated to validate a DML statement on the base table

how can I workaround that issue?should I add the column time to the pk projection?

And a general partitioning question
If I want my partitions to be as follows: 24 hours - by hours, 7 days , 1 month, 3 month, 1 year..
what is the group by clause needed for the
GROUP BY CALENDAR_HIERARCHY algorithm.

Many thanks!!
Keren

3 replies

Jim_Knicely
Forum|alt.badge.img+2
  • Participating Frequently
  • April 14, 2021

This smells llike a bug... But, if you disable the PK you should be able to add the PARTITIONing...

Example:

verticademos=> CREATE TABLE spins(c INT NOT NULL, time TIMESTAMP) ORDER BY time;
CREATE TABLE

verticademos=> ALTER TABLE spins ADD CONSTRAINT spins_pk PRIMARY KEY (c) ENABLED;
ALTER TABLE

verticademos=> INSERT INTO spins SELECT 1, SYSDATE;
 OUTPUT
--------
      1
(1 row)

verticademos=> ALTER TABLE spins PARTITION BY time::DATE GROUP BY CALENDAR_HIERARCHY_DAY(time::DATE, 2, 1) REORGANIZE;
ROLLBACK 2648:  Column time in PARTITION BY expression is not allowed, since it is not present in some projections
HINT:  Projection "spins_pk" does not contain column "time"

verticademos=> DROP PROJECTION public.spins_pk;
WARNING 7000:  The projection 'spins_pk' was used to enforce the enabled key constraint 'spins_pk', and may be regenerated to validate a DML statement on the base table
DROP PROJECTION

verticademos=> ALTER TABLE spins ALTER CONSTRAINT spins_pk DISABLED;
ALTER TABLE

verticademos=> ALTER TABLE spins PARTITION BY time::DATE GROUP BY CALENDAR_HIERARCHY_DAY(time::DATE, 2, 1) REORGANIZE;
NOTICE 8364:  The new partitioning scheme will produce partitions in 1 physical storage containers per projection
WARNING 9249:  Using PARTITION expression that may result in NULL values
HINT:  DML statements which produce NULL values for partition expressions will roll back
NOTICE 4785:  Started background repartition table task
ALTER TABLE

verticademos=> ALTER TABLE spins ALTER CONSTRAINT spins_pk ENABLED;
ALTER TABLE


Jim_Knicely
Forum|alt.badge.img+2
  • Participating Frequently
  • April 14, 2021

And for your general partitioning question... You can get pretty creative with the PARTITION GROUP BY clause!

Example:

verticademos=> SELECT c FROM t;
             c
----------------------------
 2021-08-14 00:00:00
 2023-04-14 00:00:00
 2024-04-15 00:00:00
 2021-06-14 00:00:00
 2021-04-29 02:46:58.528717
 2021-04-17 02:46:58.516394
 2021-04-14 02:46:58.470838
 2021-04-14 04:46:58.488554
 2021-04-14 04:46:58.500997
 2021-04-15 02:46:58.510026
(10 rows)

verticademos=> SELECT c,
verticademos->        CASE
verticademos->          WHEN DATEDIFF('HOUR', NOW(), c) <= 24
verticademos->            THEN 1
verticademos->          WHEN DATEDIFF('DAY', NOW(), c) <= 7
verticademos->            THEN 2
verticademos->          WHEN DATEDIFF('MONTH', NOW(), c) <= 1
verticademos->            THEN 3
verticademos->          WHEN DATEDIFF('MONTH', NOW(), c) <= 3
verticademos->            THEN 4
verticademos->          ELSE 5
verticademos->        END
verticademos-> FROM t;
             c              | case
----------------------------+------
 2021-08-14 00:00:00        |    5
 2023-04-14 00:00:00        |    5
 2024-04-15 00:00:00        |    5
 2021-06-14 00:00:00        |    4
 2021-04-29 02:46:58.528717 |    3
 2021-04-17 02:46:58.516394 |    2
 2021-04-14 02:46:58.470838 |    1
 2021-04-14 04:46:58.488554 |    1
 2021-04-14 04:46:58.500997 |    1
 2021-04-15 02:46:58.510026 |    1
(10 rows)

verticademos=> ALTER TABLE t PARTITION BY (c)
verticademos->   GROUP BY (
verticademos(>     CASE WHEN DATEDIFF('HOUR', NOW(), c) <= 24
verticademos(>            THEN 1
verticademos(>          WHEN DATEDIFF('DAY', NOW(), c) <= 7
verticademos(>            THEN 2
verticademos(>          WHEN DATEDIFF('MONTH', NOW(), c) <= 1
verticademos(>            THEN 3
verticademos(>          WHEN DATEDIFF('MONTH', NOW(), c) <= 3
verticademos(>            THEN 4
verticademos(>          ELSE 5
verticademos(>      END) REORGANIZE;
WARNING 6100:  Using PARTITION expression that returns a Timestamp value
HINT:  This PARTITION expression may cause too many data partitions.  Use of an expression that returns a more accurate value, such as a regular VARCHAR or INT, is encouraged
NOTICE 4785:  Started background repartition table task
ALTER TABLE

kfruchtman
Forum|alt.badge.img+2
  • Author
  • Participating Frequently
  • April 14, 2021

Trying that now.Thank you so much!