chore(tra-963): remove dead Redpanda Connect ingester source artifacts

docker-compose.yaml

@@ -41,21 +41,3 @@ services:
       timescaledb:
         condition: service_healthy
     restart: unless-stopped
-
-  ingester:
-    image: docker.redpanda.com/redpandadata/connect
-    container_name: ingester
-    command: ["/redpanda-connect", "run", "/connect.yaml"]
-    volumes:
-      - ./ingester/connect.yaml:/connect.yaml
-    ports:
-      - "4195:4195"
-    environment:
-      MQTT_TOPIC: ${MQTT_TOPIC}
-      MQTT_URL: ${MQTT_URL}
-      MQTT_CLIENT_ID: ${MQTT_CLIENT_ID:-trakrf-platform-ingester}
-      PG_URL: ${PG_URL}
-    depends_on:
-      timescaledb:
-        condition: service_healthy
-    restart: unless-stopped

justfile

@@ -21,17 +21,13 @@ backend *args:
 database *args:
     cd database && just {{args}}
 
-ingester *args:
-    cd ingester && just {{args}}
-
 # ============================================================================
 # Lazy Dev Aliases
 # ============================================================================
 
 alias db := database
 alias fe := frontend
 alias be := backend
-alias ing := ingester
 
 # ============================================================================
 # Combined Validation Commands

ingester/acceptance/README.md → test/acceptance/cs463-replay/README.md

@@ -2,26 +2,26 @@
 
 TRA-834. Pipeline gate for the GKE Mosquitto broker (TRA-828): publish a
 corpus of real CS463 MQTT messages at `mqtt.{env}.gke.trakrf.id` and assert
-the rows land in `trakrf.tag_scans`, exercising the Redpanda Connect transform
-and DB write end to end.
+the rows land in `trakrf.tag_scans`, exercising the in-backend Go subscriber
+(TRA-900) and DB write end to end.
 
 Companion to TRA-835 (broker-liveness pub/sub ping, infra side). This deck
 deliberately doesn't test broker-liveness on its own — that's TRA-835's job.
 
 ## Layout
 
 ```
-ingester/acceptance/
+test/acceptance/cs463-replay/
 ├── README.md
-├── capture-cs463.sh      # one-shot corpus capture from live EMQX
 ├── replay-cs463.sh       # replay + DB assertion against the GKE broker
 └── corpus/
     └── cs463.tsv         # 2521 messages, 4 capture points, 71 EPCs
 ```
 
 `cs463.tsv` is tab-separated `topic<TAB>payload-json`, captured 2026-05-25
 from EMQX Cloud after the cs463-214 capture-point rename. Frozen as a
-fixture — EMQX Cloud is on the teardown list.
+fixture — EMQX Cloud has since been decommissioned (TRA-828), so this corpus
+is permanent and not recapturable.
 
 ## Run
 
@@ -35,7 +35,7 @@ Required env (load from `.env.local`):
 ```sh
 set -a; source .env.local; set +a
 MQTT_GKE_HOST=mqtt.preview.gke.trakrf.id \
-  ingester/acceptance/replay-cs463.sh
+  test/acceptance/cs463-replay/replay-cs463.sh
 ```
 
 Assertion runs via `kubectl exec` into the per-env CNPG primary (TRA-823 +
@@ -62,28 +62,24 @@ defaulting to `CURRENT_TIMESTAMP` (microsecond resolution). Real device
 traffic at ~1 msg/s/topic is fine. This script replays the 2521-message
 corpus in ~30s (~85 msg/s sustained), which clusters multiple same-topic
 messages into the same microsecond — the second one violates the PK and is
-dropped, with a corresponding broker backpressure event on the ingester's
-loopback subscriber. Expect ~5-10% PARTIAL on a clean run.
+dropped, with a corresponding broker backpressure event on the Go
+subscriber. Expect ~5-10% PARTIAL on a clean run.
 
 The actual schema fix lives in a separate platform ticket. Until then,
 `PARTIAL` is the steady-state expected outcome and is treated as PASS.
 
 ## Dependencies
 
 - TRA-828 broker deployed (`mqtt.{env}.gke.trakrf.id:8883` reachable, TLS 1.2)
-- Ingester subscribed to the broker, writing to `trakrf_preview` / `trakrf_prod`
+- In-backend Go subscriber (TRA-900) running against the broker, writing to `trakrf_preview` / `trakrf_prod`
 - `kubectl` configured for the target cluster (default access path)
 - `mosquitto_pub`, `jq`, `awk` on `PATH`
 
-## Re-capture (only while EMQX is still live)
+## Re-capture
 
-```sh
-set -a; source .env.local; set +a
-ingester/acceptance/capture-cs463.sh
-```
-
-Replaces `corpus/cs463.tsv`. Don't re-run unless you have a reason — the
-checked-in corpus is the artifact this ticket exists to preserve.
+Not possible — EMQX Cloud was decommissioned (TRA-828) and the
+`capture-cs463.sh` script was retired with it (TRA-963). The checked-in
+`corpus/cs463.tsv` is the permanent artifact this deck exists to preserve.
 
 ## Note on table name
 

ingester/acceptance/replay-cs463.sh → test/acceptance/cs463-replay/replay-cs463.sh

@@ -4,7 +4,7 @@
 #
 # Pipeline under test:
 #   mosquitto_pub -> mqtt.{env}.gke.trakrf.id:8883 (Mosquitto, TLS 1.2)
-#     -> ingester (Redpanda Connect, MQTT input -> sql_raw output)
+#     -> in-backend Go MQTT subscriber (TRA-900, raw insert)
 #     -> trakrf.tag_scans (TimescaleDB hypertable in the per-env CNPG cluster)
 #
 # Each replayed payload gets a unique top-level tra834_replay_id field; the
@@ -19,12 +19,12 @@
 #
 # Env (required):
 #   MQTT_GKE_HOST       broker hostname, e.g. mqtt.preview.gke.trakrf.id
-#   MOSQUITTO_USER      broker username (matches helm/trakrf-ingester auth secret)
+#   MOSQUITTO_USER      broker username (broker auth secret, TRA-828)
 #   MOSQUITTO_PASSWORD  broker password
 #
 # Env (optional):
 #   MQTT_GKE_PORT       default 8883
-#   CORPUS              default ingester/acceptance/corpus/cs463.tsv
+#   CORPUS              default test/acceptance/cs463-replay/corpus/cs463.tsv
 #   INGEST_WAIT_SECONDS default 15
 #   ASSERT_PSQL_CMD     SQL-from-stdin invocation; defaults to:
 #                       kubectl exec -i -n trakrf-system trakrf-db-1 -c postgres \
@@ -77,15 +77,15 @@ done < <(cut -f1 "$CORPUS" | sort -u)
 
 echo "published: $published"
 
-echo "waiting ${INGEST_WAIT_SECONDS}s for ingester to drain..."
+echo "waiting ${INGEST_WAIT_SECONDS}s for the Go subscriber to drain..."
 sleep "$INGEST_WAIT_SECONDS"
 
 LANDED=$(printf "SELECT count(*) FROM trakrf.tag_scans WHERE created_at >= '%s' AND message_data ->> 'tra834_replay_id' = '%s';\n" "$T_START" "$MARKER" | assert_psql)
 TOPICS=$(printf "SELECT string_agg(message_topic || ':' || c, ' ' ORDER BY message_topic) FROM (SELECT message_topic, count(*) AS c FROM trakrf.tag_scans WHERE created_at >= '%s' AND message_data ->> 'tra834_replay_id' = '%s' GROUP BY message_topic) t;\n" "$T_START" "$MARKER" | assert_psql)
 echo "landed:    $LANDED / $published"
 echo "by topic:  $TOPICS"
 
-# Pipeline acceptance: landed > 0 proves broker -> ingester -> DB works end
+# Pipeline acceptance: landed > 0 proves broker -> Go subscriber -> DB works end
 # to end. A gap between landed and published is a known schema behaviour
 # under burst replay rate (tag_scans PK = (created_at, message_topic) with
 # microsecond-resolution CURRENT_TIMESTAMP — same-topic messages within one