Storage probe failures
What the registration probe's three errors mean - auth, not-found, unreachable - and managed-identity gotchas.
Registering a storage connection probes the store before saving anything, so a misconfiguration fails fast with a specific reason. Here's what each means.
The three probe errors
| Error | Cause | Fix |
|---|---|---|
authentication failed … | Credentials rejected | Check the access key / SAS / service-principal secret; confirm it's not expired or scoped too narrowly. |
storage path or bucket not found … | Wrong bucket/container or base_path | Verify the exact bucket/container name and that any base_path prefix exists. |
could not reach storage … | Network / endpoint problem | Check the endpoint URL, region, firewall/egress, and (S3-compatible) that --endpoint is correct. |
Good to know
A reachable but empty bucket is not an error -
registration succeeds and reports
0 tables. You don't need existing tables to
connect.Azure: managed identity unavailable
managed identity unavailable - not running on Azure infrastructure…use_msi only works when Eddytor runs on Azure infra (AKS/VM/ACI) with an
identity assigned - it fetches tokens from the Azure IMDS endpoint. If IMDS is
unreachable (e.g. running off-Azure), use a service principal
(client_id+client_secret+tenant_id) or an access key instead. See
Azure Blob.
Azure: also check
- The conn path/endpoint and that the AKS egress IP / VNet can reach the storage account.
- For OneLake,
use_fabric_endpoint=true.
Provider-link variant
If you registered via a linked Azure/Google account
and later see provider_reauth_required, the delegated credential lapsed -
re-link it (it's not a retry).