Troubleshooting

Find the message or symptom, then follow the Do column. Admin means a workspace admin or platform operator must act. When you contact them, include:

the time (with timezone),
the pipeline, job or table name,
the run ID,
the full error text or stack trace.

Symptom	Likely cause	Do	Admin?
Account locked	5 failed logins, often a tool retrying an old password	Wait (up to about 15 min) and update saved passwords	If it persists
JDBC / DBeaver gets 401 but the browser works	Pending account action (password change, profile)	Sign in to the portal in a browser and complete it	No
Page looks broken after an upgrade	Browser cached the old UI	Hard refresh (Ctrl+Shift+R)	No
Pages visible but every save fails with read / write errors	You have a viewer role	Ask for the right role	Yes
403 on Semantic Models	Page access is View	Ask for Use	Yes
Tool tile stuck on "Starting up…"	Tool still starting, or health reporting is stale	Wait 5 min	If it persists
Whole portal and tools down	Platform or network outage	Wait; nothing is lost	Yes

Ingestion (Connect)

Symptom	Likely cause	Do	Admin?
…still being deployed — its DAG has not reached Airflow yet	Workflow not published yet	Retry in 1–2 min	If over 10 min
…DAG that failed to parse. Re-deploy it	Broken generated workflow	Force delete and recreate	Often
already has a run in flight	One run per pipeline	Wait	No
Deploy spins at "Enable services", then fails	Catalog or storage credentials problem	Retry deploy once	If it repeats
Run fails after 30–60 min with `NotAuthorizedException` on the write	Ingestion's catalog session expired	Re-trigger after the admin recycles it	Yes
SUCCESS but 0 rows / table short	Empty window, sub-day chunk on a YYYYMMDD column, or failed batch	Check the window and bookmark; use Reconciliation	If repair is refused
Transfer data shows no work due; two runs per trigger	Backfill in progress	Expected; watch the backfill run	No
Watermark doesn't advance	Runs failing	Open the failed step's log	Maybe
Run "running" for a long time	Stalled write or queue	Wait for the 5-min reconciler; don't re-trigger	If over 1 h
`ClassCastException` / `NullPointerException` on write	Binary column (old pipeline), precision over 38, legacy required column	Recreate the pipeline; cast in a source view	Often
`ORA-17002 … Could not initialize class oracle.net.nt.Clock`	Oracle driver stuck on one ingestion node	nothing	Yes (restart that node)
Ingestion UI unreachable; loads crashing	Out of memory from day-partition fan-out	Switch to month or no partitioning; one big table at a time	Yes
Duplicated rows after a historical load	Overlapping range (append-only)	Reconciliation → Delete + Reload for the window	Maybe
Old table still exists after a rename	Rename orphans the old Iceberg table	Drop the old table, or ask	Maybe
Runs & logs shows Internal server error	Service version mismatch	nothing	Yes

dbt & Jobs

Symptom	Likely cause	Do	Admin?
Model files vanished	Committed, not pushed; workspace reset	Push immediately from now on	Yes (possible recovery)
`[ahead N]` in `git status`	Silent push failure	`git push`	If it recurs
Git prompts for username	Repository credential rotated	Push what you can	Yes
`Cannot set database in spark!`	Wrong engine for the model	`--target trino`	No
`INVALID_TABLE_PROPERTY`	Spark config on Trino	Engine → Spark	No
`'dbt_expectations' is undefined`	Packages not committed	Commit `dbt_packages/`, re-save the job	No
not a valid dbt error name	Version-specific flags	Remove `flags.warn_error_options`	No
Job "succeeds" instantly with no tasks	Stale project copy	Re-save the job	If it persists
Job fires at the wrong hour	Saved before the timezone fix	Re-save	No
`409` saving a job	Model or pipeline already in another job	Remove it from the other job	No
Many jobs fail at once with git-sync errors	Version-control service hiccup	Retry after about 5 min	If it doesn't recover

Notebooks & Spark

Symptom	Likely cause	Do	Admin?
Spawn failed: (empty reason)	Hub lost track of servers	Stop after 2 tries	Yes (restart the hub)
`NotAuthorized` / `S3V4RestSignerClient.sign`	Shared Spark server's catalog session went stale	Send the stack trace	Yes (restart Spark)
`INVALID_HANDLE`	Spark server restarted	Kernel → Restart	No
First query takes 20–40 s	Executors starting	Normal	No
Kernel out of memory	Too much collected into pandas	Aggregate in Spark	No

SQL, cubes, BI, AI

Symptom	Likely cause	Do	Admin?
Only 10,000 rows	Interactive cap	Export CSV	No
Export CSV greyed out	Warehouse tab, or no Lakehouse query run	Switch tab and run	No
`Access Denied: Cannot select from…`	No policy grants you that table	Request access	Yes
Schema list shows only `default`	Over-broad policy	nothing	Yes
Exceeded per-node memory limit	Query too heavy	Filter earlier, reorder joins, pre-aggregate	No
`$partitions` empty or one NULL partition	Known partition case issue	Filter on real columns	FYI only
Cube: Unknown database	`lakehouse.` name in cube `table`	Use `<safe>_iceberg.<schema>.<table>`	No
Cube: Select database (0)	Warehouse grants	nothing	Yes
Materialized cube stale	MV not refreshed	Refresh / Refresh cubes node	No
`joins_not_materializable`	One-to-many join / no dimension	Pre-join in gold / add a dimension	No
Superset ignores your row filters	Hand-made connection without impersonation, or a cached result	Use the pre-configured connection; tick Impersonate logged-in user	Maybe
Exported CSV is exactly about 1 GiB	Hit the export cap; the file is cut off	Filter or split the export	No
Cube query returns only 100 rows	Default semantic row limit	Set `limit` (max 1,000), or query gold in the Workbench	No
Superset Could not load database driver: OracleEngineSpec	Older BI image	Use the `oracle+cx_oracle://` URI after the upgrade	Yes
Voir: empty Catalog dropdown / Could not reach dn-lakehouse	Connector configuration	nothing	Yes
AI Gateway Not configured	No model endpoint yet	nothing	Yes