Hello,

I’m trying to restore a PostgreSQL custom-format dump into a Railway Postgres database, but pg_restore repeatedly hangs while streaming data through the public proxy.

Environment: staging

Project/service context:

• Target: Postgres database
• Connection path: public proxy, centerbeam.proxy.rlwy.net:50330
• Client: local macOS terminal
• Restore command shape:

pg_restore --verbose --no-owner --no-acl \
  --dbname "$TARGET_STAGING_DATABASE_URL" \
  production.dump

Earlier attempts also used:

pg_restore --verbose --clean --if-exists --no-owner --no-acl \
  --dbname "$TARGET_STAGING_DATABASE_URL" \
  production.dump

What happens:

• The restore starts, but eventually stops making progress.
• From pg_stat_activity, the restore backend sits on ClientRead, waiting for the client.
• Example:

pid  | application_name | state  | wait_event_type | wait_event | duration | query
3875 | pg_restore       | active | Client          | ClientRead | 00:04:23 | COPY public.leagues_masterplayer (...) FROM stdin;

• pg_stat_progress_copy stops changing:

pid  | table_name            | command   | type | bytes_processed | bytes_total | tuples_processed
3875 | leagues_masterplayer  | COPY FROM | PIPE | 103521          | 0           | 643

• Running the progress query again later shows the exact same bytes_processed and tuples_processed.
• Locally, the pg_restore process remains alive but shows essentially no CPU activity.

It does not hang after a long-running restore. It usually hangs very quickly — sometimes within a minute or less. In the latest attempt, it stalled during COPY on public.leagues_masterplayer after only 643 tuples / 103,521 bytes processed.

This has happened multiple times. Earlier attempts also left idle restore sessions like:

pg_restore | idle | Client | ClientRead | DROP INDEX IF EXISTS ...
pg_restore | idle | Client | ClientRead | CREATE INDEX ...

What I’ve tried:

• Killed the local pg_restore process.
• Terminated stale pg_restore backends with pg_terminate_backend.
• Retried without --clean.
• Wiped the target schema manually first using:

DROP SCHEMA public CASCADE;
CREATE SCHEMA public;

• Verified from Postgres that it is not blocked on locks; the server appears to be waiting for the client stream.

An issue related to Railway public TCP proxy?

Is there a recommended way to run pg_restore from inside Railway, or to use an internal/private database URL instead of the public proxy?

Thanks!