Show accessible tables only in New Query view for PostgreSQL

[dairyo]

What type of PR is this? (check all applicable)

• Bug Fix

Description

Background

Left pane of New Query view shows table list. The list includes tables which user can not access.
Tables which user can not access should be hidden.

The bug

Function redash.query_runner.pg.PostgreSQL._get_tables does not consider privileges of a user for tables.

The Fix

Get tables for which current_user of PostgreSQL have privileges.

Related Tickets & Documents

--

Mobile & Desktop Screenshots/Recordings (if there are UI changes)

--

[ranbena]

@arikfr is omitting inaccessible tables the desired user experience?
An alternative would be to show these in disabled mode with a tooltip explanation, but I'm not knowledgable enough to know if this has any value.

Also, should this type of filtering be replicated across all query runners?

[arikfr]

@ranbena

is omitting inaccessible tables the desired user experience?

Yes.

Also, should this type of filtering be replicated across all query runners?

In theory query runners should be implemented in this way already. But there is no single recipe that will work for all of them anyway.

[arikfr]

Thanks, @shinsuke-nara! This was introduced in #2549 when the support for partitioned tables was added.

While your fix makes sure we load tables from schemas the user has access to, it still might show tables/columns the user doesn't have access to:

1. When the user doesn't have access to all tables in a schema.
2. When the user doesn't have access to all columns in a table.

Some possible solutions:

We could fix the above by using information_schema.column_privileges and making sure the schema/table/columns are referenced there, but I'm concerned this might break support for the special kinds (materialized views, foreign tables and partitioned tables).

Use the has_column_privilege function, but it has the same concern as above (does it apply to the special kinds?). Also it was introduced in Postgres 8.4 only.

Use information_schema.columns, which already applies ACL. For the special kinds it does not include, we can augment with information from the current query while ignoring ACL.

I think option 3 is the safest way forward which for sure improves the current situation without requiring much research.

What do you think?

[dairyo]

@arikfr

Thank you for checking my PR.
I agree with you. I will fix this PR with option 3.

[dairyo]

@arikfr

I fixed this PR according to your advice.
Could you check it if you have time?

[arikfr]

Thank you for the update, but we still need the old query for the "special types", so it should be something like:

query = """
        SELECT s.nspname as table_schema,
               c.relname as table_name,
               a.attname as column_name
        FROM pg_class c
        JOIN pg_namespace s
        ON c.relnamespace = s.oid
        AND s.nspname NOT IN ('pg_catalog', 'information_schema')
        JOIN pg_attribute a
        ON a.attrelid = c.oid
        AND a.attnum > 0
        AND NOT a.attisdropped
        WHERE c.relkind IN ('m', 'f', 'p')

        UNION 

        SELECT table_schema,
               table_name,
               column_name
        FROM information_schema.columns
        WHERE table_schema NOT IN ('pg_catalog', 'information_schema')
"""

Thanks :)

[dairyo]

@akariv

Thank you for the advice. I have fixed again.

[arikfr]

Thanks!

[dairyo]

@arikfr

Thank you for accepting my PR :)