Hi everyone,
Consider the example below:
-- thread 1:
\c db1;
CREATE TABLE t1(c0 int);
INSERT INTO t1(c0) VALUES(2147483647);
UPDATE t1 SET c0 = 0;
CREATE INDEX i0 ON t1((1 + t1.c0));
VACUUM FULL; -- unexpected: ERROR: integer out of range
-- thread 2:
DROP DATABASE db2;
CREATE DATABASE db2;
I would expect that the VACUUM FULL executes without error. However,
it shows an error "integer out of range", which would be expected for
the CREATE INDEX, had the UPDATE not been executed. I can reliably
reproduce this on my machine on the first execution of thread 1's
statements when having a second thread that (repeatedly) drops and
creates a database. I've attached a Java program that should make it
possible to reproduce this by throwing an exception
"org.postgresql.util.PSQLException: ERROR: integer out of range"
immediately after starting the program.
This bug report is similar to a previous case that I reported, where
executing VACUUM FULL on distinct databases can result in a deadlock
(see https://www.postgresql.org/message-id/CA%2Bu7OA6pL%2B7Xm_NXHLenxffe3tCr3gTamVdr7zPjcWqW0RFM-A%40mail.gmail.com),
which might also be unexpected from a user perspective, so I am not
sure if this case bug report is considered relevant.
Best,
Manuel