Appearance
Limits & known issues
Limits
| Area | Limit |
|---|---|
| Ingestion sources | Oracle, SQL Server, MySQL, PostgreSQL. Others are preview only. |
| YYYYMMDD text bookmarks | Oracle only; day granularity |
| Backfill / historical ranges | Whole days, midnight bounds, end-exclusive |
| Historical loads | Append-only, no de-duplication |
| Runs per pipeline | 1 at a time |
| Concurrent tasks per pipeline | 1–8 (default 1) |
| Columns | Binary/LOB columns dropped; numeric precision over 38 unsupported |
| Deploy-time settings | Partition spec, column types, excluded columns, tuning. Changing them requires delete + recreate. |
| Pipeline membership | A pipeline or dbt model belongs to one job |
| Interactive SQL | 10,000 rows |
| CSV export | 1 GiB, 20 min, 5/min, 50/h, Lakehouse (Trino) only |
| Semantic queries | 100 rows by default, 1,000 max |
| Semantic MVs | No one-to-many joins, need at least one dimension, manual/job refresh over Iceberg |
| Notebooks | Small / Medium / Large (up to 4 CPU / 8 GiB), 10 GiB home, 30 min idle shutdown |
| Spark | One shared server per workspace; executor size fixed by the platform |
| ML tools | In-cluster only, single instance |
Known issues
| Issue | Impact | Workaround |
|---|---|---|
| Partition case mismatch. Mixed-case (CamelCase / upper-case) bookmark or partition columns can produce a partition spec that doesn't bind. | Data correct; all rows in one NULL partition; no pruning; OPTIMIZE does nothing | Use no partitioning for such tables; filter on data columns. A platform fix is pending. |
| Ingestion catalog session expiry on very long runs | Runs over 30–60 min can fail with NotAuthorized | Keep chunks small; re-trigger after the admin recycles the service |
| Shared Spark catalog session | A stale session affects all notebook and dbt-spark users at once | Admin restarts Spark; users restart kernels |
| Rename orphans the old table | Old bronze table left behind silently | Drop it manually |
| No Redeploy / Undeploy button | Deploy-time changes need a recreate | Delete (Force) and recreate; history is kept |
| Silent dbt artifact push failures | Repository falls behind ([ahead N]) | git push manually |
| Shared identities in Voir, the warehouse tab, notebooks and hand-made BI connections | Per-user row filters don't apply there | Use governed surfaces for sensitive data |
| CSV export cut off silently at 1 GiB | File looks complete but isn't | Keep exports well under 1 GiB; split by date |
| Notifications capped at 20 | Bell shows "20+" | Use Atlas pages for full history |
| Lakebase credentials panel may say not available on self-hosted installs | Can't view Postgres credentials there | Ask your admin. Don't press Rotate. |
| Preview connectors in the catalog | Can't be saved | Use the four supported databases |
| "Soon" menu items (Streaming, ODS, Version Control, Secrets) | Not available | none |