Server table rows permanently fixed (cannot be deleted or truncated)

On Thu, Nov 17, 2022 at 8:37 AM Fred Williams <fredmw7@gmail.com> wrote:

PostgreSQL version 12.2

*Problem*
I am unable to delete or edit a specific range of table rows (I'll call
them corrupted rows) for a specific primary key. If I truncate the entire
table, the corrupted rows remain. In the corrupted rows, I can successfully
update non-primary key fields, but not primary-key fields.

*Miscellaneous*

1. The problem does not occur on a test server using the same exact
table.
2. If I rename the table, I can remove the corrupted rows and/or
update the primary keys (such as changing the DateTime). However, when I
rename it back, the corrupted rows reappear!

*Table info*

CREATE TABLE public."NCAASchedule" (
"GameDateTime" timestamp NOT NULL,
"Week" int2 NOT NULL,
"HomeTeam" int2 NOT NULL,
"AwayTeam" int2 NOT NULL,
"Line" numeric(5, 1) NULL,
"OverUnder" numeric(5, 1) NULL,
"HomeTeamScore" int2 NULL,
"AwayTeamScore" int2 NULL,
"OpeningLine" numeric(5, 1) NULL,
"BowlID" int2 NULL,
"Control" int2 NULL,
CONSTRAINT ncaaschedule_pk PRIMARY KEY ("GameDateTime", "Week",
"HomeTeam", "AwayTeam")
);

Thanks for any help you can provide.

Some suggestions of things you can do (not mutually exclusive):
Update PostgreSQL to a supported version, v14.6
Post the output of running: \d+ "NCAASchedule"

Rename the table to something else. Create a new table with this name.
Populate it with the data you want it to have. Recreate objects depending
on the old table to instead depend on the newly created one. Drop the
problematic table. (If you aren't married to the table name, you might
consider creating the new table with a new name so that nothing existing,
internal or external, is capable of referencing it).

I seriously doubt this is a PostgreSQL bug; some trigger or external
process messing with the table seems much more plausible.

Turning on statement logging could help.

David J.

Fred Williams <fredmw7@gmail.com> writes:

I am unable to delete or edit a specific range of table rows (I'll call
them corrupted rows) for a specific primary key. If I truncate the entire
table, the corrupted rows remain. In the corrupted rows, I can successfully
update non-primary key fields, but not primary-key fields.

*Miscellaneous*

1. The problem does not occur on a test server using the same exact
table.
2. If I rename the table, I can remove the corrupted rows and/or update
the primary keys (such as changing the DateTime). However, when I rename it
back, the corrupted rows reappear!

TBH, I'm suspecting pilot error. I wonder whether you have another
table by the same name in a different schema, which is the one
containing the "corrupted" rows, and after you rename this table
out of the way you're unintentionally accessing the other one.

That theory doesn't explain the rows surviving TRUNCATE, but
maybe inheritance could --- IIRC, TRUNCATE will not touch child
tables. So maybe the alias table is also an inheritance child
of the one you are modifying?

This command in psql would clarify much:

postgres=# \d+ *."NCAASchedule"

regards, tom lane