M55: restore private RAGcore routing
MobilityOps acceptance / backend (push) Failing after 18s
MobilityOps acceptance / frontend (push) Successful in 27s
MobilityOps acceptance / e2e (push) Skipped

This commit is contained in:
NuklearRabbit
2026-08-24 03:45:09 +02:00
parent 81e3fd63bd
commit 444e61253b
4 changed files with 70 additions and 1 deletions
+3
View File
@@ -96,6 +96,9 @@ MOBILITYOPS_WEBHOOK_TRIGGER_TOKEN=replace-me-n8n-webhook-trigger-token
# RAGcore integration
KNOWLEDGE_PROVIDER=demo
RAGCORE_BASE_URL=http://ragcore-api:8000
# Optional existing Docker network used by hosted deployments to reach RAGcore through
# its private service alias instead of exposing RAGcore on the LAN.
RAGCORE_DOCKER_NETWORK=
RAGCORE_TENANT=northstar-mobility-demo
RAGCORE_WORKSPACE=mobilityops
RAGCORE_COLLECTION=internal-procedures
+21
View File
@@ -1,5 +1,26 @@
# Project state
## M55 — restore private cross-project RAGcore routing (2026-08-24)
- M54 promoted successfully as `81e3fd63bdbcb2e9c4ae1d709ea46f40537b6f62`, with two
API and two web replicas, exact OCI revision labels, Alembic `4f2b9c8d7e61 (head)` and
healthy public/loopback readiness. The live browser canary then proved that all core routes
worked but RAGcore degraded honestly to unavailable in both browsers.
- The failure was infrastructure routing, not answer validation: the existing healthy
RAGcore app now publishes host port 1237 on loopback only, while Fleet Ops still targeted
the host LAN address and received TCP `connection refused` from inside its container.
- Extended the start-first deployer with a validated optional `RAGCORE_DOCKER_NETWORK`.
Candidate APIs join that pre-existing network and take the non-secret `RAGCORE_BASE_URL`
from the authoritative server `.env`; all secrets still come from the serving API's
resolved environment and are never printed. The hosted configuration can now use the
private `ragcore-app:8080` alias without exposing RAGcore on the LAN.
- Validation: Alpine `sh -n` passed for the deployer; base/Unraid Compose config, source
budgets and `git diff --check` pass. M54's complete **331 backend / 171 Playwright** gates
remain applicable because M55 changes deployment topology and documentation only.
- Exact next action: commit/push M55, safely update the two non-secret RAGcore routing keys
in the server `.env`, take a fresh verified backup, deploy the exact M55 archive and repeat
the four Chromium/Firefox live acceptance checks.
## M54 — full logic, resilience and recruiter upgrade (2026-08-23, local candidate)
- Replaced the static Engineering architecture row with an interactive, keyboard-operable
+12
View File
@@ -22,6 +22,18 @@ Create `.env` from `.env.example`, replace every placeholder secret, set
The internal `1236` listener is
an upstream for the TLS proxy, not a user-facing URL.
When RAGcore runs as a separate Compose project on the same host, keep its HTTP listener
private and configure Fleet Ops through the existing shared Docker network, for example:
```dotenv
RAGCORE_BASE_URL=http://ragcore-app:8080
RAGCORE_DOCKER_NETWORK=ragcore-proxy
```
The release deployer validates the network, attaches only the API replicas and takes the
non-secret RAGcore URL from the server `.env` on every promotion. Tokens and the remaining
resolved API environment continue to be inherited without being printed.
```bash
cd /mnt/user/appdata/mobilityops
./deploy/unraid/configure-env.sh \
+34 -1
View File
@@ -15,6 +15,33 @@ case "$revision" in *[!0-9a-f]*) echo "Revision must be lowercase hexadecimal" >
[ -n "$expected_checksum" ] || { echo "Expected SHA-256 is required" >&2; exit 1; }
[ -f "$root/.env" ] || { echo "Production .env is missing" >&2; exit 1; }
read_env_value() {
key="$1"
awk -v key="$key" '
index($0, key "=") == 1 {
sub("^[^=]*=", "")
print
exit
}
' "$root/.env"
}
# Stateless replicas normally inherit the complete, already-secret-resolved environment
# from the serving API. RAGcore routing is an explicit exception: its private endpoint and
# cross-project Docker network are non-secret deployment topology and the server .env is
# authoritative so infrastructure changes take effect on the next release.
ragcore_base_url="$(read_env_value RAGCORE_BASE_URL)"
ragcore_network="$(read_env_value RAGCORE_DOCKER_NETWORK)"
case "$ragcore_base_url" in
http://*|https://*) ;;
*) echo "RAGCORE_BASE_URL must be an absolute HTTP(S) URL" >&2; exit 1 ;;
esac
case "$ragcore_network" in
"") ;;
*[!a-zA-Z0-9_.-]*) echo "RAGCORE_DOCKER_NETWORK contains invalid characters" >&2; exit 1 ;;
*) docker network inspect "$ragcore_network" > /dev/null ;;
esac
actual_checksum="$(sha256sum "$archive" | awk '{print $1}')"
[ "$actual_checksum" = "$expected_checksum" ] || {
echo "Archive checksum mismatch" >&2
@@ -123,7 +150,13 @@ for replica in 1 2; do
--label "com.mobilityops.revision=$revision" \
--network "$network" --network-alias api --network-alias "api-$short_revision" \
--env-file "$api_environment" \
--env RUN_MIGRATIONS=false "$api_image")"
--env RUN_MIGRATIONS=false --env "RAGCORE_BASE_URL=$ragcore_base_url" "$api_image")"
if [ -n "$ragcore_network" ] && ! docker network connect "$ragcore_network" "$id"; then
# shellcheck disable=SC2086
docker rm -f $new_api_ids "$id" > /dev/null 2>&1 || true
echo "Could not attach API candidates to RAGcore network $ragcore_network" >&2
exit 1
fi
new_api_ids="$new_api_ids $id"
done