creating a postgresql database back end for a new Label Studio project

Question

I am creating a local Label Studio server to host images to annotate in our office. I would like the database back end to be postgresql and not sqlite and be located in a particular directory, not the default and not the same as the 'data-dir'. I have got a test server working across the network with various machines annotating images on the server, but the backend was sqlite for this test.

Everything I've tried to get a postgresql backend db has failed for various reasons. Some commands result in a sqlite db (occasionally with the name 'postgresql') located in my required directory; others create postgres/pyscopg2 errors but I think they're up a garden path.

The host machine is running Ubuntu 20.04 LTS. And serves another postgresql db over the network using other APIs. Postgresql version running is 12.9.

I have created a conda environment and pip installed Label Studio as the documentation suggested.

Here's what I've tried:

Start the conda environment. Follow instructions to assign environment variables from https://labelstud.io/guide/storedata.html#PostgreSQL-database which at time of writing is:

DJANGO_DB=default
POSTGRE_NAME=postgres
POSTGRE_USER=postgres
POSTGRE_PASSWORD=
POSTGRE_PORT=5432
POSTGRE_HOST=db

Then a few variations on the start command (I didn't include the backslashes, just put here for readability/comparability):

label-studio start --init \
-db postgresql \
--database /path/to/label-studio/databases/newdb \
--data-dir /path/to/label-studio/media_dirs/test_proj

result: db is where expected, but:

file newdb

gives "newdb: SQLite 3.x database, last written using SQLite version 3038002"

label-studio start --init \
--database /path/to/label-studio/databases/newdb \
-db postgresql \
--data-dir /path/to/label-studio/media_dirs/test_proj

result: a db at specified path named 'postgresql' and still an sqlite db. This seems to mirror the mistake mentioned at: https://github.com/heartexlabs/label-studio/issues/1660

I have also tried the above two commands with the '--init' argument omitted with same results.

Then I tried adding something on the front of the command suggested at the same link above:

DJANGO_DB=default label-studio start \
--database /path/to/label-studio/databases/newdb \
--data-dir /path/to/label-studio/media_dirs/test_proj

result: psycopg2.OperationalError: FATAL: password authentication failed for user "postgres" FATAL: password authentication failed for user "postgres"

DJANGO_DB=default POSTGRE_PASSWORD= label-studio start \
--database /path/to/label-studio/databases/newdb \
--data-dir /path/to/label-studio/media_dirs/test_proj

result: psycopg2.OperationalError: fe_sendauth: no password supplied

Any help and resolution would be highly appreciated.

Also, I can't tag this with 'label-studio' because I'm not quite at the required reputation to create a new tag, so if anyone who can feels like doing so, pleaseandthankyou!

Answer 1

Your last option was closer than all the others. Have you tried to run LS using this:

DJANGO_DB=default POSTGRE_NAME=<postgres_name> POSTGRE_USER=<postgres_user> POSTGRE_PASSWORD=<password> POSTGRE_PORT=<db_port> POSTGRE_HOST=<db_host> label-studio

Sure, you have to run postgres service by yourself, configure it properly, create the DB <postgres_name>, the user <postgres_user> and set the password <password>, grant access rights to this user. Also don't forget to specify <db_host> (localhost?), <db_port> (5432?)