[YSQL] Pre-split multi-tablet range tables SPLIT AT VALUES clause split rows ordering issue

[yifanguan]

Jira Link: DB-1282

Description

CREATE TABLE tbl_floating_point_nan (
  a DOUBLE PRECISION,
  PRIMARY KEY(a ASC)
) SPLIT AT VALUES (('NaN'), (100), (200), ('NaN'));

The above CREATE TABLE statement hangs and generates the error in ysqlsh.

ERROR:  Timed out: Read(tablet: 00000000000000000000000000000000, num_ops: 1, num_attempts: 3, txn: 00000000-0000-0000-0000-000000000000, subtxn: [none]) passed its deadline 494430.885s (passed: 121.139s): Network error (yb/util/net/socket.cc:540): recvmsg error: Connection refused (system error 61)

The master error log is:

F0319 23:36:50.679088 109367296 catalog_entity_info.cc:419] Two tablets with the same partition key start and split depth: state: PREPARING table_id: "000033e1000030008000000000004000" partition { partition_key_start: "D\377\370\000\000\000\000\000\000!" partition_key_end: "" } table_ids: "000033e1000030008000000000004000" and 

On the other hand,

CREATE TABLE tbl_floating_point_nan (
  a DOUBLE PRECISION,
  PRIMARY KEY(a ASC)
) SPLIT AT VALUES (('NaN'), (100), (200));

and

CREATE TABLE tbl_floating_point_nan (
  a DOUBLE PRECISION,
  PRIMARY KEY(a ASC)
) SPLIT AT VALUES ((100), (200), ('NaN'));

work fine.

CREATE TABLE tbl_floating_point_nan (
  a DOUBLE PRECISION,
  PRIMARY KEY(a ASC)
) SPLIT AT VALUES ((100), (200), ('NaN'), ('NaN'));
ERROR:  Invalid argument: Cannot have duplicate split rows

works as expected.

[yifanguan]

This example seems to be related to this issue.

CREATE TABLE tbl_floating_point_nan (
  a DOUBLE PRECISION,
  PRIMARY KEY(a ASC)
) SPLIT AT VALUES (('NaN'), (100), (200));

CREATE TABLE tbl_has_nan_value (
  a DOUBLE PRECISION
);
INSERT INTO tbl_has_nan_value VALUES ('NAN'), (100), (200);
SELECT * FROM tbl_has_nan_value ORDER BY a ASC;
  a
-----
 100
 200
 NaN
(3 rows)

Table tbl_floating_point_nan specifies ASC for column a. The SPLIT AT clause allows 'NaN' to be present before 100 and 200, but in reality, it is greater than 100 and 200.

[yifanguan]

There are split rows ordering issues with MINVALUE and MAXVALUE as well.
Example:

CREATE TABLE tbl(
  a int,
  primary key(a asc)
) SPLIT AT VALUES((MINVALUE), (100));
CREATE TABLE

CREATE TABLE tbl2(
  a int,
  primary key(a desc)
) SPLIT AT VALUES((MINVALUE), (100));
CREATE TABLE

CREATE TABLE tbl3(
  a int,
  primary key(a desc)
) SPLIT AT VALUES((MAXVALUE), (100));
ERROR:  Invalid argument: Split rows ordering does not match column ordering

The creation of tbl2 should fail, and the creation of tbl3 should succeed based on column ordering of column a.