Trino Lakehouse

Trino Lakehouse (CLI)

Query the Cruise production Trino cluster against Iceberg lakehouse tables (interactive / analytical SQL).

Pre-flight: Authentication (two-tier)

Tier 1 (try first — usually instant, no browser MFA):

authcli refresh
authcli app get spidercat -out stdout | head -c 30

Tier 2 (if Tier 1 fails or Trino still returns auth errors): follow ~/.cursor/skills/cruise-auth-refresh/SKILL.md for the full SSO flow.

Re-fetch the spidercat token after any refresh.

CLI invocation

The Trino CLI ships as a local JAR (typically $HOME/trino.jar). Prefer $HOME over ~ inside nested command substitutions — tilde expansion is unreliable there.

"$HOME/trino.jar" \
  --server https://trino.prd.paasapps.robot.car \
  --access-token "$(authcli app get spidercat -out stdout)" \
  --execute "YOUR SQL HERE" 2>&1

Catalog / schema

• A common NVIDIA workspace is lakehouse.nvidia_prod. Other catalogs and schemas exist — use SHOW SCHEMAS FROM lakehouse (or equivalent) to discover.

Struct / nested fields

Use dotted paths for struct columns, e.g. row_key.label_class_id (exact paths depend on the table DDL).

Gotchas

• Stderr noise: Java “illegal reflective access” / module warnings are usually harmless; focus on the query result or Trino error text.
• Runtime: Wide scans or large joins can take minutes; add LIMIT while exploring.
• Auth expiry: On 401 / access denied, run Tier 1 auth again, then Tier 2 if needed.

When to use this skill

Use when you need ad-hoc SQL across VINs or global aggregates where the Polars client is too constrained, or for quick validation counts against Iceberg tables.