Fail-fast setting for psql-client prevents automatic import on startup of full db dumps

[rgielen]

postgres/9.6/docker-entrypoint.sh

Line 104 in f7dc572

psql=( psql -v ON_ERROR_STOP=1 )

In this line psql is advised to fail fast on errors. This is a problem for importing any dump created by pg_dumpall, since there is a 99% chance it will include a CREATE ROLE postgres; statement, causing an error. In default mode psql keeps on executing the following statements, which is perfectly fine to restore a cluster.

My expectation would be that a pg_dumpall dump can be imported smoothly when mounted at /docker-entrypoint-initdb.d.

I suggest to remove the fail-on-error setting or at least make it configurable through an environment variable (IMHO with defaulting to standard psql behaviour)

[tianon]

(Sorry for the long tail.)

The goal with adding ON_ERROR_STOP=1 is to ensure that errors are not hidden from the operator -- typically the last thing you want is for something important to fail and the script to barrell onwards as if nothing is wrong (especially since that becomes harder to debug/diagnose after the fact). I think the best suggestion I can offer here is that if the default behavior we provide isn't sufficient, it should be pretty trivial to drop in a one-liner .sh script to have some custom behavior which ignores errors instead. 👍

[tianon]

(To be clear, we do the same thing with shell scripts too -- set -e is enabled for the very same set of reasons.)