TROUBLESHOOTING.md

Troubleshooting

Problems starting Elasticsearch

Most issues in Vectory with Elasticsearch are related to Docker. Vectory assumes you have Docker installed with Docker Compose version >=2.3.4 and that the Docker daemon is running.

Using an old version of Docker Compose

You can check your version of Docker Compose by running:

docker compose version

You can follow Docker's official guide to update Docker Compose or install it altogether.

Problems with the database

Locked database

You can't have more than one instance of Vectory running at the same time. If you try to start Vectory while it's already running, you'll get an error message like this:

sqlite3.OperationalError: database is locked

Just stop the running instance of Vectory and try again.

Index already exists

If adding an index to the database fails in a way we haven't anticipated, it might happen that the index is loaded in Elasticsearch but not in the database. This can be checked by comparing the indices on both places:

from vectory.db.models import ElasticSearchIndexModel, Query
from vectory.es.client import ElasticKNNClient

es = ElasticKNNClient()
es_indices = es.list_indices()

db_indices = Query(ElasticSearchIndexModel).get()

es_indices and db_indices should be the same.

If the index is on Elasticsearch but not in the database, delete it from Elasticsearch and try again.

from vectory.es.client import ElasticKNNClient

es = ElasticKNNClient()
es.delete_index(INDEX_NAME)

If the index is in the database but not on Elasticsearch, delete it from the database and try again.

from vectory.db.models import ElasticSearchIndexModel, Query

index = Query(ElasticSearchIndexModel).get(name=INDEX_NAME)[0]
index.delete_instance(recursive=False)

Import errors

No module pkg_resources

The Python module setuptools has errors if installed with pip or poetry. You can fix this with:

wget https://bootstrap.pypa.io/ez_setup.py -O - | python