verify

name	verify
description	Drive the built ablo CLI end-to-end against a real local sync-server (journey wiring) — use to verify CLI changes (connect/push/status/logs) at their real HTTP surface instead of unit tests.

Verify ablo CLI changes end-to-end

The CLI's surface is the terminal talking HTTP to the sync-server. Verify by running the built CLI against a real locally-booted server, never by importing CLI internals.

Build the CLI

cd packages/cli && npm run build   # → dist/cli.cjs

Boot a real server (journey wiring: real routes + real api-key auth)

Write a small CJS-safe launch script (async main(), NO top-level await — the harness files use __dirname and break as ESM) that uses the journey harness primitives, then run it with npx tsx --experimental-wasm-modules from apps/sync-server/:

• startEphemeralPostgres() (src/__journeys__/harness/ephemeralPostgres.ts) — control-plane PG, prisma-pushed
• startJourneyServer(pg.url, { tenantRouting: true }) (harness/journeyServer.ts) — real routes on an ephemeral port; tenantRouting wires the datasource-registration stack
• seedJourneyOrg(server.sql, label) + mintSecretKey(server.sql, orgId, 'sandbox', label) (harness/seed.ts) — a real sk_test_ key
• print/write server.baseURL + key, hold the process open; server.shutdown() + pg.stop() on SIGTERM

Point the CLI at it with ABLO_API_URL=<baseURL> ABLO_API_KEY=<minted>.

Customer database for connect flows

connect check/--register locally probes DATABASE_URL, which needs wal_level=logical. Don't touch the user's main Postgres — init a throwaway cluster in the scratchpad:

initdb -D $DIR -U cust_owner
pg_ctl -D $DIR -o "-p 55444 -c listen_addresses=127.0.0.1 -c wal_level=logical -c unix_socket_directories='' -F" start
#                                     unix sockets OFF: scratchpad paths exceed the 103-byte socket limit
createdb -h 127.0.0.1 -p 55444 -U cust_owner custdb
psql ... -c "CREATE TABLE t(id text PRIMARY KEY);" -c "CREATE PUBLICATION ablo_publication FOR ALL TABLES;"

Gotchas learned the hard way:

• Host must look public. Registration's zod shape-check rejects 127.0.0.1 (database_url host must be a public hostname or public IP). Use db.localtest.me — public DNS that resolves to 127.0.0.1, so the CLI's local preflight AND the server's shape check both pass.
• Connection string needs a password. A passwordless URL 500s in replicationSourceRegistry.register (empty connectionRef ZodError). Any password works under local trust auth: postgres://cust_owner:[e-mail]:55444/custdb.

Useful contrast probe

The server's global not-found envelope is {"code":"entity_not_found","message":"Not found"} — any bare "Not found" from a CLI command means it hit an unmatched route (all real routes are mounted under /api). Curl the suspect path directly to tell route-miss from data-miss.

Teardown

Stop the launch-script task (SIGTERM runs its shutdown), pg_ctl stop the customer cluster, delete any file holding the minted key.