# Limits (/reference/limits)

What Databaas does not do today and the boundaries to plan around.

## Pipeline visibility is all or nothing

Pipeline access is fleet-wide: open Pipelines and you see every pipeline in your instance, or none.
There is no per-pipeline or per-team scope, and that includes task logs.

## No column masking or row-level security

Data access is granted per warehouse, namespace or table. There is no column masking and no
row-level filtering, in the catalog, in Trino or in Superset. To show a group only some columns or
rows, write them to a separate table (for example in a pipeline) and grant that table.

## No groups in Access control

Feature permissions are given to people directly or through roles. There are no groups, and roles
cannot contain other roles.

## Lineage is limited to table history

Each Iceberg table keeps its own snapshot history, but there is no lineage view across tables,
pipelines and dashboards.

## No external SQL clients

Query the lakehouse from notebooks, SQL Lab, dashboards and pipelines. Connecting external BI tools
or SQL clients directly is not supported.

## Dashboards are a door, not a filter

Everyone who can open Superset sees the same charts, dashboards and datasets; only which queries
return rows differs, decided by data grants, so a chart on data you have no grant for errors. See
[how access works](/access/how-access-works).

## The catalog preview is a sample, not a query

The **Preview** tab shows the first 20 rows of a table's current snapshot, so it cannot
filter, join or aggregate and never shows the whole table. To read for real, use a notebook or SQL
Lab. See [table details](/catalog/table-details).

## Generic tables have no preview or schema

A non-Iceberg table has no snapshots, so it shows no preview and no schema — only its format,
location and permissions. See [Iceberg tables](/catalog/iceberg).

## A notebook scan uses session memory

A notebook server has 4 GB of memory and 2 CPUs by default, and a scan loads everything it reads into
that memory, so an unbounded scan of a large table exhausts the session. Narrow it — pick columns, filter rows, cap the size — or run it in
[SQL Lab](/dashboards/sql-lab), where Trino returns only the result. See [querying data](/notebooks/querying-data).

If you need more, ask your platform administrator.

## Notebook storage is not backed up

Your home (10 GB by default) and the shared folder persist across restarts but are not backed up,
and `/tmp` is cleared each start. Treat the git remote as source of truth and push often. See
[notebook storage](/notebooks/storage) and [working with Git](/notebooks/git).

## SQL Lab queries stop after 300 seconds

A SQL Lab query that runs longer than 300 seconds is cancelled. Narrow it, or move the work into a
pipeline.

## Sign-ins expire

A sign-in and its query token have a limited lifetime, so a long or idle session ages out and queries
stop authenticating. Sign in again — in a notebook, that shows up as an authentication error saying
your login could not be refreshed; sign out of Notebooks and back in (the catalog itself does not need
re-creating, see [your catalog does not go stale](/notebooks/querying-data#your-catalog-does-not-go-stale)).
In Superset the token lasts about 15 minutes and is not refreshed in the background, so queries there
can fail sooner; sign out and back in.