Redash fails to start via docker-compose

[ezkl]

Issue Summary

Redash cannot be started from scratch using docker-compose due to recent upstream changes to the postgres docker image. This issue was also discussed recently on the redash forums.

Steps to Reproduce

NOTE: These steps should be run on a "fresh" install without existing docker images or anonymous volumes. This can be reproduced by using docker-compose down with additional flags for removing anonymous volumes and referenced images. I've not included the full command here for fear people will copy and paste and destroy data unintentionally.

1. git clone git@github.com:getredash/redash.git
2. cd redash && docker-compose up -d

Technical details:

After running docker-compose up, the postgres container prints the following error to the logs and exits before being restarted and repeating the same process again:

postgres_1   | Error: Database is uninitialized and superuser password is not specified.
postgres_1   |        You must specify POSTGRES_PASSWORD to a non-empty value for the
postgres_1   |        superuser. For example, "-e POSTGRES_PASSWORD=password" on "docker run".
postgres_1   |
postgres_1   |        You may also use "POSTGRES_HOST_AUTH_METHOD=trust" to allow all
postgres_1   |        connections without a password. This is *not* recommended.
postgres_1   |
postgres_1   |        See PostgreSQL documentation about "trust":
postgres_1   |        https://www.postgresql.org/docs/current/auth-trust.html

Changes were recently made to the docker-entrypoint.sh file packaged with new postgres docker images (including postgresql:9.5-alpine) that will stop the database from being initialized unless a password is set (via $POSTGRES_PASSWORD) or by allowing connections from all hosts (by setting $POSTGRES_HOST_AUTH_METHOD=trust.)

You can see the relevant PR and subsequent discussion for specifics.

I forked redash and pushed a branch including a small change to the docker-compose.yml that sets POSTGRES_HOST_AUTH_METHOD to trust. I'll open a PR momentarily.

[shnayan]

Try setting the environment variable - 'POSTGRES_PASSWORD' under postgres. shouldn't be empty.

[ezkl]

Try setting the environment variable - 'POSTGRES_PASSWORD' under postgres. you can keep it empty as well.

This may have worked in previous builds of the image before the most recent set of upstream changes were made, but this line explicitly tests (via -z) whether the length of $POSTGRES_PASSWORD is zero.

EDIT

It looks like you edited your comment while I was writing my reply.

Setting POSTGRES_PASSWORD would absolutely work and is probably the right solution, but looks like it would require adding the password to the various postgres client configs and connection strings across the code base and docs. If that would be preferable to the maintainers, I'd be happy to update my PR.