> One argument for keeping the tests would be that they nicely bring together > the side-effect (logging) and these use-cases, whereas in the code they're > pretty distant, making the connection not obvious.
Another argument is that we have no coverage on temp logging, which is why this bug was discovered. There are likely plenty of tools out there, for example pgbadger or manual log analysis, that have probably misinterpreted temporary usage all this time, because of the way the data was emitted. -- Sami
