pg_upgrade errors out if distributed deadlock detector fires on new server

[velioglu]

Within our upgrade documentation we stated 'stop old or new servers' as the third step. If we run the new server before running the pg_upgrade, distributed deadlock daemon also starts. Even if we stop the new server before running pg_upgrade, upgrade process errors out since the daemon cannot read pg_index table. So, we may add a warning about this issue similar to the warning about creating the extension on step 3 of upgrading coordinator node.

[sumedhpathak]

@velioglu Could you provide an amended set of instructions? Does following the instructions on docs right now cause an issue?

[velioglu]

@sumedhpathak Instructions given in the documentation is correct but we may change the position of NEW_PG_DATA contains an empty database initialized by new PostgreSQL version caution to the third step (for usability). We may also warn user about not starting the new server at all before calling the pg_upgrade. Users should know that any operation on the new server may wake up the daemon which cause problems with pg_upgrade.

[begriffs]

Given step 2 warns, "do not start the new server," does that remove the danger that the deadlock daemon may be active when the user runs pg_upgrade?

(Also the instructions may have changed since this issue was opened. I'm looking at http://docs.citusdata.com/en/v8.1/admin_guide/upgrading_citus.html#upgrading-postgresql-version-from-10-to-11 )

[jonels-msft]

@velioglu does the latest version of our upgrade page still have the problem?