Troubleshooting
This page lists common problems and their usual causes. Start by checking the logs and the healthchecks of the components.
The components do not start
MongoDB connection errors. Tock requires a MongoDB replica set, even with a single node:
tock_mongo_url must contain the replicaSet parameter and the replica set must be initialized
(see Installation).
no 'tock_encrypt_pass' set error. Outside the dev environment (tock_env different from dev), the
tock_encrypt_pass property is required as soon as data is encrypted (see Security).
I cannot log in to Tock Studio
The credentials are refused. By default, the credentials are admin@app.com / password. If tock_users
and tock_passwords are set, they replace this default user (see Tock Studio authentication).
The login succeeds, but the session is immediately lost. Outside the dev environment, the session cookie
is only sent over HTTPS. Serve Tock Studio over HTTPS, or set tock_https_env=false if TLS is handled by a
proxy that forwards HTTP requests. With several bot_admin instances, the sessions are not shared: use a single
instance, or sticky sessions on the load balancer.
The bot does not answer
In Test > Test, the bot answers with an error. The test connector calls the bot at the Application base url
of its configuration, in Settings > Configurations: this URL must be reachable from the bot_admin service
(for instance http://bot_api:8080 with the tock-docker descriptors, see Configuration).
In Bot API mode, nothing happens. Check the logs of your bot: the API key must be the one of the
configuration, and the bot must reach the bot_api service (WebSocket mode) or be reachable by it
(WebHook mode, see Bot API).
The bot answers with another story. The sentence is not recognized as the expected intent: qualify it in Language Understanding > Inbox sentences (see Language understanding), and check that the intent is the main intent of the story, in Stories & Answers > All stories.
The new sentences are not taken into account. The model is rebuilt by the build_worker service shortly
after a sentence is validated: check that this service is running, and see the builds in
Model Quality > Model Builds.
The RAG does not answer
The bot answers with a technical error. The Gen AI orchestrator is not reachable: check
tock_gen_ai_orchestrator_server_url in the bot_api and bot_admin services, and the orchestrator
healthcheck (/health-check). Since Tock 26.3.0, the Question answering prompt must also produce a JSON
object (see Upgrading Tock).
The answers never cite any document. Check that an indexing session is selected in the RAG settings, and that the vector store configured in Tock Studio or by the environment variables is the one where the documents were indexed: the default vector store is not the same in Tock Studio and the bots, and in the orchestrator (see Vector DB settings). The retrieval diagnostic shows the documents found for a question.
The model building fails
OutOfMemoryError in the build_worker. The memory needed grows with the number of sentences, intents and
entities: see the memory recommendations.
Chat with Tock