tiger db: no way to load a local file (no COPY ... FROM STDIN equivalent)

Summary

There's no way to load a local file into a service using the CLI alone. tiger db query can run SQL from a file, but it can't stream data, so COPY ... FROM STDIN isn't reachable and neither is anything equivalent. Today the only options are to install psql for \copy or to write a program against a Postgres driver.

Why this matters

The CLI's own description is "an MCP server for helping coding agents write production-level Postgres code." Getting data into a service is step one of most of that work, and it's currently the one step that requires stepping outside the CLI.

Concretely: I built a workshop where attendees point a coding agent at a Tiger Cloud service in a GitHub Codespace. Every other operation — DDL, queries, EXPLAIN, schema inspection, forking, teardown — is tiger and nothing else. Loading 1M rows of NYC 311 data is the sole reason a second tool has to exist in the image.

What I tried

1. COPY ... FROM STDIN — hangs indefinitely. This is arguably a bug on its own:

$ tiger db query <service-id> -c "COPY t FROM STDIN WITH (FORMAT csv, HEADER true)" < data.csv
# never returns; had to kill it

The server is waiting for CopyData frames the client never sends, and the client is waiting for a result set that will never arrive. An immediate "COPY FROM STDIN is not supported" would be much kinder than a hang.

2. -f with a file of INSERTs — works, but doesn't scale. tiger db query sends a file as one query string, so a 146 MB file of 1M rows gets:

Error: ERROR: out of memory (SQLSTATE 53200)

against a free service.

3. The workaround I shipped. Chunk the data into 10 gzipped files of 100k INSERTs each (~15 MB apiece) and stream them through stdin:

for f in data/load/*.sql.gz; do
  gunzip -c "$f" | tiger db query "$SERVICE_ID"
done

This works and needs nothing but tiger and gunzip. It takes 78 seconds for 1M rows against a free service. The same data via a real COPY ... FROM STDIN (Node, pg-copy-streams) takes 12–17 seconds, so the workaround is roughly 5x slower and ships the dataset in a format that's ~6x larger uncompressed than the CSV it came from.

What would help

Something like:

tiger db copy <service-id> --table service_requests --from data.csv --format csv --header
tiger db copy <service-id> --table service_requests --from -            # stdin

Reading from stdin matters as much as the file case — it keeps gunzip -c … | and other pipelines working without unpacking large files to disk.

Two smaller things that would each be worth having on their own, even without the above:

• Make COPY ... FROM STDIN via tiger db query fail fast with a clear message instead of hanging.
• Consider streaming -f files statement-by-statement rather than as one query string, so a large SQL file doesn't OOM the server.

Environment

Tiger CLI 0.24.0, darwin/arm64 and linux/arm64 (devcontainer). Reproduced against both a free (shared/shared) service and a paid 0.5 CPU / 2 GB service in us-east-1.

Start with the tiger db query command and reproduce the hanging COPY ... FROM STDIN case using the issue's example. Define the CLI entry point and data-streaming behavior for tiger db copy, including file and stdin input, then verify that large CSV loads complete without requiring psql or a driver. Also confirm unsupported COPY FROM STDIN queries fail promptly with a clear message.