Skip to content

Troubleshooting ​

Start with dak doctor --profile <name>. It runs the checks in order and prints the next step for each failure. The table below lists the messages you are most likely to see.

Authentication and access ​

SymptomFix
No Databricks credentials foundFollow one of the authentication options, then run dak doctor
Could not get an OAuth token from the Databricks CLIdatabricks auth login --profile <name>
Databricks CLI not found on PATHInstall the CLI, or use a personal access token
[PERMISSION_DENIED] ...Ask an admin for the privilege named in the message. See privileges
No SQL warehouse is accessibleSet DATABRICKS_WAREHOUSE_ID, or get CAN USE on a warehouse

Statements ​

SymptomFix
Read-only mode: ...Expected. Restart with --allow-writes only if writes are intended. If the statement starts with a comment, remove it. Pass literals that contain ; as :name parameters. See read-only mode
Read-only check did not finish ... or Statement did not finish in time ...The warehouse is starting. Retry in a moment
Statement still RUNNINGThe warehouse is starting. Call get_statement with wait_seconds

Exports ​

SymptomFix
Export <path> already exists; choose another name.Choose another name, or delete the old export. The new statement was cancelled
Export FAILED: chunk N download failed (...)The result links could not be downloaded. Run dak doctor and check Result download. Allow the cloud-storage host in the proxy or firewall
Export FAILED: Parquet export: chunk N could not be converted ...The chunk could not be converted. Export as csv or sqlite instead
Export FAILED: SQLite export: the result has N columns ...SQLite allows 2000 columns. Select fewer columns, or export as csv or parquet
Export SUCCEEDED with truncated: trueThe result exceeded DAK_EXPORT_MAX_BYTES. Raise it, or filter the query
Export FAILED with a file system error (EACCES, ENOSPC)Point DAK_EXPORT_DIR to a writable directory with free space. dak doctor checks it
An export vanished after a restartExports are not resumed. Run the statement again

Where to look next ​

  • The operational logs hold the failed calls with their error entries. Read resource://logs/errors/recent from your agent, or open logs.db in the state directory.
  • The configuration page lists every setting and its default.
  • If the problem is not listed, open an issue in the repository.

Released under the MIT License.