Mark Dilger <mark.dilger@enterprisedb.com> writes:
> The pg_amcheck patch Peter committed for me adds a new test, src/bin/pg_amcheck/t/006_bad_targets.pl, which creates
twoPostgresNode objects (a primary and a standby) and uses PostgresNode::background_psql(). It doesn't bother to
"finish"the returned harness, which may be the cause of an installation hanging around long enough to be in the way
whenanother test tries to start.
(a) Isn't that just holding open one connection, not the whole instance?
(b) Wouldn't finish()ing that connection cause the temp tables to be
dropped, negating the entire point of the test?
TBH, I seriously doubt this test case is worth expending buildfarm
cycles on forevermore. I'm more than a bit tempted to just drop
it, rather than also expending developer time figuring out why it's
not as portable as it looks.
regards, tom lane