Skip to content

Apache Doris connector

Apache Doris is reached through its FE's MySQL-protocol port with the same adapter as MySQL, and compiled SQL is generated for the Doris dialect, which keeps FULL OUTER JOIN (unlike MySQL and TiDB).

Doris also offers Arrow Flight SQL: add one key and result sets are pulled as Arrow batches straight from the BEs instead of being serialized row by row through the FE.

Gated by exec-mysql, plus exec-flightsql for the Arrow path.

Connection profile

datasources:
  doris:
    type: doris
    host: doris.internal
    port: 9030               # FE MySQL-protocol port
    username: dosi
    password: ${DORIS_PASSWORD}
    database: analytics

With Arrow Flight SQL:

    arrow_flight_port: 8070  # the FE's arrow_flight_sql_port

Parameters

Key Type Required Default Notes
type string yes — doris
host string yes* — *Unless uri: is given. Also the Flight SQL host.
port int no 3306 (driver default) The FE MySQL port — normally 9030; set it explicitly.
username string no — Also the Flight SQL handshake user, where it defaults to root.
password string no — Prefer ${VAR}.
database string no — On the Flight path it is applied as USE <database> after the handshake.
arrow_flight_port int no — Presence switches the executor. Doris calls this arrow_flight_sql_port in fe.conf (Doris ≥ 2.1.5 recommended); Dosi uses one key name for both engines.
uri string no — A mysql:// DSN. It replaces the discrete host/port/username/password/database keys.
default bool no false See connection profiles.

Parsed and ignored on this connector: sslmode, sslrootcert, schema, catalog, account, role, warehouse, compat_mode.

Arrow-native results

The key is the only configuration change; see Arrow. What to watch for:

  • BE reachability. The FE hands the client each BE's own address for the data pull, so that address must be routable from wherever Dosi runs. When it is not, use the FE proxy settings (public_host + arrow_flight_sql_proxy_port).
  • Seed over the MySQL wire. Keep port: in the profile so schema management and seeding have a path that does not depend on the Flight endpoint.

Without exec-flightsql in the build, the key errors rather than silently falling back.

Limitations

  • No NTH_VALUE — an nth_value window is refused at compile time for this dialect. FIRST_VALUE/LAST_VALUE are unaffected.
  • Degenerate CORR/COVAR_* windows differ. Over a single-row frame Doris returns 0 where the standard result is NULL, so those two-argument statistical windows are refused rather than returning a value that disagrees with every other engine.
  • No 3-argument DATEDIFF.
  • COUNT(*) is not accepted under OVER in Doris 2.1 — the compiler emits COUNT(1) for you, so this one needs no action.
  • No TLS on either path.

Verify the connection

$ dosi query --model model.yaml \
    --metrics revenue --group-by orders.status --execute --connection doris

Troubleshooting

Message Cause and fix
datasource "x" sets arrow_flight_port but this build lacks the exec-flightsql feature Rebuild with --features exec-flightsql (or exec-all), or drop the key.
flight sql connect: <e> The FE Flight port is unreachable, or a BE address is not routable from this client — see the proxy settings above.
flight sql handshake: <e> Credentials rejected by the Flight service.
cannot reach server: <io error> MySQL-wire path: host, port or network. The FE port is 9030.
server rejected SQL (<code>): <message> Doris refused the compiled statement.

Reference