Amazon Redshift

The redshift driver connects to Amazon Redshift, which speaks the PostgreSQL wire protocol. It is built into the tobira binary and shares the PostgreSQL driver’s code. It is for reading: you can query, browse, explain and use completion. Editing, transactions, COPY export and LISTEN are not available, because Redshift lacks what they need.

Quick start

tobira query "redshift://analyst@cluster.abc123.eu-west-1.redshift.amazonaws.com:5439/dev?sslmode=verify-full" \
  "select count(*) from sales" --read-only

connections:
  - name: warehouse
    dsn: redshift://analyst@cluster.abc123.eu-west-1.redshift.amazonaws.com:5439/dev?sslmode=verify-full
    password: env:REDSHIFT_PASSWORD
    read_only: true

Connecting

DSN	Result
`redshift://user@cluster:5439/dev?sslmode=verify-full`	Redshift, through the `redshift` driver
`postgres://user@cluster:5439/dev?flavor=redshift`	the same, through the `postgres` driver
`postgres://user@cluster:5439/dev`	tobira asks the server when it connects and detects Redshift

Use redshift:// or flavor=redshift for a read-only connection. With a plain postgres:// DSN, tobira only learns it is Redshift after connecting, and by then it has already asked for a read-only session setting that Redshift does not document.

The DSN, password, TLS (sslmode) and options: rules are those of the PostgreSQL driver.

Capabilities

Capability	Supported
Query	yes
Browse	yes
Explain	yes (Redshift’s text plan, shown as a tree; no `EXPLAIN ANALYZE`)
Completion	yes
Edit rows, transactions	no (Redshift has no savepoints and no `RETURNING`)
Native csv export	no (no `COPY ... TO STDOUT`); `--output` uses tobira’s own csv
Watch	no (no `LISTEN`/`NOTIFY`)
Vector search	no
Read-only guard	yes

Browsing

Row counts and sizes come from SVV_TABLE_INFO, which lists only tables that hold data. Other tables show an unknown count.
Redshift has no indexes. The description shows the primary key and foreign keys (Redshift’s informational constraints), plus the distribution key as distkey and the sort key as compound sortkey or interleaved sortkey.
Tables in schemas outside your search_path are described too.

Queries and paging work as described in PostgreSQL queries, browsing and editing.

Read-only guard

Each read-only call runs in BEGIN READ ONLY ... ROLLBACK, one statement at a time, with the refusals of the PostgreSQL read-only guard (COPY, DO, server-file functions).

Redshift’s UNLOAD writes query results to S3 without writing to the database, so a read-only transaction does not stop it. On a read-only connection tobira therefore only runs statements that start with SELECT, WITH, VALUES, TABLE, SHOW or (, or EXPLAIN [VERBOSE] of one of those. UNLOAD, COPY, CALL, SET, SELECT ... INTO and every write are refused before they reach the server.

What it cannot stop. Redshift has no DISCARD, so tobira resets a reused connection with RESET ALL. That restores settings but not temp tables or SET SESSION AUTHORIZATION made by an earlier query on a writable connection. Use a user with SELECT only as the real boundary.

Compatibility

Tested by running the PostgreSQL 17 suite in Redshift mode, with Redshift’s system views (SVV_TABLE_INFO, PG_TABLE_DEF) added. Not yet tested against a real Redshift cluster. If something fails there, please open an issue.