…ations#193

FastAPI + SQLAlchemy 2.x + psycopg2 + Postgres 13 sample built to
exercise the v3 dispatcher's simple-query ClassCatalog branch. The
sample's init.sql pre-creates the `project` table so SQLAlchemy's
Base.metadata.create_all skips CREATE TABLE at record time -- the
shape the dispatcher bug requires.

Used by the keploy/integrations Woodpecker lane
sqlalchemy-pg-catalog-postgres to assert that recorded `type: query`
mocks with `class: CATALOG` are consulted by the simple-query path,
not just the extended-query path.

Signed-off-by: Akash Kumar <meakash7902@gmail.com>

… review

flow.sh:
- set -Eeuo pipefail so curl/other failures actually fail the script.
- Track readiness explicitly; exit 1 with a clear message if the app
  never reaches /health within READY_TIMEOUT_S (default 60s) instead
  of silently falling through and proceeding against a dead app.
- curl -fsS for the readiness probe and the endpoint calls so HTTP
  failures propagate as exit codes (the previous `curl -sS ... && echo`
  shape silenced non-2xx responses).

init.sql:
- Replace `INSERT ... ON CONFLICT DO NOTHING` with `INSERT ... SELECT
  WHERE NOT EXISTS`. The model has no UNIQUE constraint on name, so
  ON CONFLICT had nothing to fire on; this form is genuinely idempotent
  on re-runs against an existing volume.

Refs Copilot review on #102.

Signed-off-by: Akash Kumar <meakash7902@gmail.com>

Addresses two Copilot review nits on #102:

* README.md: the local-repro snippet hardcoded
  `--container-name pg-catalog-repro-app`, but the compose file uses
  `${APP_CONTAINER:-pg-catalog-repro-app}` so a user who overrides
  APP_CONTAINER (e.g. to isolate concurrent runs) would have keploy
  point at a non-existent container. Both keploy invocations now
  thread `${APP_CONTAINER:-pg-catalog-repro-app}` so they pick up
  the same env override compose sees. Also switched the deprecated
  camelCase `--apiTimeout`/`--disableMockUpload` to the kebab forms
  the v3 CLI actually registers (`--api-timeout`; mock-upload flag
  dropped — not registered on v3-dev `test`).

* main.py: SQL echo was unconditional. It is INTENTIONALLY on by
  default for this sample (the whole point is to surface SQLAlchemy's
  pg_class probe + create_all behaviour in the app log so a reader
  can correlate it with the keploy agent log), but the noise is real
  for unrelated investigations. Gated behind SQL_ECHO env var with a
  comment explaining why it's on by default.

Signed-off-by: Akash Kumar <meakash7902@gmail.com>

…-safe create_all

Two nits from the Copilot review:

* DATABASE_URL: `os.environ["DATABASE_URL"]` raised a bare KeyError
  at import time, which surfaces as a noisy traceback in container
  logs with no hint about what's missing. Switched to os.getenv with
  an explicit RuntimeError that names the env var and gives a sample
  URL format.

* lifespan: `Base.metadata.create_all(engine)` is synchronous
  psycopg2 I/O running inside an async lifespan, blocking uvicorn's
  event loop until pg_class probe + any CREATE TABLE round-trips
  complete. Switched to `await asyncio.to_thread(...)` so the loop
  stays responsive. For this minimal repro the difference is small,
  but the pattern is the right FastAPI shape for any startup that
  touches a sync DB driver.

Signed-off-by: Akash Kumar <meakash7902@gmail.com>

…meterized SQL

Copilot review noted (correctly) that the README + main.py docstring
described the pg_class probe as "parameter-less SQL", which conflicts
with the recorded mock carrying 7 bind values. psycopg2 in fact uses
the simple-Query protocol even when the source SQL is parameterized:
it substitutes %(param)s placeholders client-side and emits the
resulting inlined SQL as a single Q packet (no Bind/Execute frames).
The bind values exist at the application layer; the wire shape is
simple-Query.

Both docs now say so explicitly so a reader doesn't see the recorded
bind-value list and conclude the wire path must be extended-Query.

Signed-off-by: Akash Kumar <meakash7902@gmail.com>

…te_engine

SQLAlchemy 2.x defaults to the future-2.0 behaviour; passing
`future=True` is redundant and can trip a deprecation warning in
some 2.x point releases. Dropped to keep the sample free of incidental
warning noise that would distract from the dispatcher-bug repro.

Refs Copilot review on #102.

Signed-off-by: Akash Kumar <meakash7902@gmail.com>

…utdown

Add try/finally around the lifespan yield so engine.dispose() runs
at shutdown. Releases pooled psycopg2 connections cleanly across
repeated start/stop cycles (local repro loops, CI lane reruns),
which otherwise leak half-open postgres connections.

Refs Copilot review on #102.

Signed-off-by: Akash Kumar <meakash7902@gmail.com>

…y/finally

Copilot noted that the previous try/finally only covered the yield,
so a failure in create_all (the *exact* failure mode this repro
demonstrates: pre-fix keploy makes create_all raise
psycopg2.DatabaseError) wouldn't reach the finally and the engine
pool would leak. Moved the startup logging + create_all call inside
the try block; finally now runs regardless of where the failure
occurs.

Signed-off-by: Akash Kumar <meakash7902@gmail.com>

…lly matters

Copilot noted /docker-entrypoint-initdb.d scripts only run on first
database initialization, so the previous comment's framing ('re-run
against an existing volume') was misleading. Reworded to call out
the actual scenario: this is a single-shot insert on a clean volume,
and NOT EXISTS is defensive coverage for stale-volume reuse.

Signed-off-by: Akash Kumar <meakash7902@gmail.com>

… runs on first init

Copilot's repeated note: /docker-entrypoint-initdb.d scripts run only
on first database init. On a stale data volume the script doesn't
run at all, so the NOT EXISTS guard couldn't help anyway. Simplified
to a plain INSERT and updated the comment to tell readers to recreate
the volume (docker compose down -v) if they want a deterministic
repro.

Signed-off-by: Akash Kumar <meakash7902@gmail.com>