Skip to content

Troubleshooting

Find the words you see, then read the fix.

Starting the portal

You see Cause Fix
The frontend has not been built. Run make web. (503) Fresh clone or clean checkout make web-build
Unauthorized. Open the URL printed by praxis. (401) No token, or a token from before a restart Use the URL printed at the last start
An empty thinking panel REASONING_SUMMARY=auto, or API version earlier than 2025-03-01-preview Set REASONING_SUMMARY=detailed and a newer API version

Connecting

You see Cause Fix
pick the kind of database this is No warehouse type chosen Choose one
still needs a host, a database, … A required setting is blank Fill it in — see Warehouses
'X' is not a connection setting A setting without the CUBEJS_DB_/CUBEJS_AWS_ prefix Use the warehouse's Cube setting name
no database is connected yet Downloading a compose file before connecting Connect first
no matching manifest at docker compose up Cube Store image for the wrong CPU architecture Add or remove the -arm64v8 suffix on the cubestore image line
nothing listening on … The container is not running, or the port is different docker compose ps; check the address
no response from … in Ns Firewall, or the service is hung docker compose logs
ReadError reaching … The container is still starting Wait a few seconds and try again
… answered, but not as the model store (or sandbox) Another program is on that port, or the two addresses are swapped Check which address is which
the store rejected the portal's token — this deployment was run from a file rendered for a different connection The services were started from an old compose file Download the file again and docker compose up -d
Cube is not ready behind the store Cube is still compiling, or cannot reach Cube Store or the warehouse docker compose logs cube-api
the model store has nothing published to main yet Normal before the first model is built Run the four build steps
this sandbox speaks contract … and the portal expects 1 Sandbox image does not match the portal Re-run it from a freshly downloaded file

A wrong warehouse password passes every connection check. It shows up at Read tables, the first step that connects to the warehouse. Fix it with Edit connection, then download the semantic compose file again and re-run it.

A database on the same machine as Docker is host.docker.internal, not localhost.

Building and editing the model

You see Cause Fix
a job is already running (409) Only one build step runs at a time, across the whole machine Wait for it, or cancel it
schema '…' has not been read Building cubes before reading tables Run Read tables first
Cube reports zero cubes A stray .yml outside cubes//views/ — Cube compiles every YAML under the model The editor refuses such paths; remove any added by hand
ambiguous column once a join is involved A multi-column sql: expression without {CUBE}. Qualify each column with {CUBE}.
A merge refused because it would break dashboard tiles The change removes or renames members a saved tile uses Fix the tiles, or merge with force

Asking questions

You see Cause Fix
no Cube deployment is configured (503) No model store address stored Finish connecting
Cube is not answering (502) The semantic deployment is down docker compose ps in its folder
A query rejected with "closest matches" A member name that does not exist The agent normally retries on its own
A slow query arrives as a 502 It ran past the store's 120 s limit Narrow the question, or add a filter
Upload refused with 413 Over 100 MB Split the file

Flows

See What fails a Phase for every run-time message. The common ones:

You see Fix
input missing: 02_x/output/… A Phase folder was renamed; update the working: path
declared output has no rows: … Find out why nothing matched, or add the file to may_be_empty
This query returns more than 5,000,000 rows, so nothing was written Narrow the query, or extract it in parts
the phase stopped before it finished The portal restarted or the stream dropped — re-run the Phase
No streaming chunk received for 120.0s The LLM stream stalled — re-run the Phase
409 when starting a run A run of this Flow is already going
A Flow missing from the list Its folder has no flow.yml, or FLOWS_DIR points elsewhere — Flows are found relative to where praxis was started