Compare commits

...

3 commits

Author SHA1 Message Date
2f421d7439 Merge pull request 'infra(deploy): serve MCP on multiple domains + SKIP_BUILD' (#4) from infra/deploy-multidomain into main 2026-06-26 13:56:53 +00:00
kiania
af6fdbcd3f fix(logging): attribute each query to its analyst caller
The BearerAuth middleware matched a per-analyst token but only stashed it on
request.state, which the FastMCP tools never see — so the query log line logged
rows/sql with no caller, defeating the per-token attribution the auth design
promises. Bridge the caller name through a ContextVar (anyio propagates it into
the worker thread that runs each sync tool) and include it in the query() log.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-26 16:54:07 +03:00
kiania
fae40942a2 infra(deploy): serve MCP on multiple domains + SKIP_BUILD for label-only redeploys
The prod connector domain (fleetmcp.rahamafresh.com) had no Traefik router —
deploy.sh only ever set one HOST_DOMAIN (defaulting to fleetmcp.fivetitude.com),
so requests to the prod domain returned 503 "no available server" even with the
container healthy.

- HOST_DOMAINS: comma-separated list folded into one Traefik router rule
  (Host(`a`) || Host(`b`)). One LE cert covers all names (SANs), so connectors
  on either domain keep working. Defaults to HOST_DOMAIN (back-compatible).
- SKIP_BUILD=1: reuse the existing image for a labels/env-only redeploy, so a
  routing change can't accidentally bake in new/stale code.

Deployed to prod with HOST_DOMAINS="fleetmcp.rahamafresh.com,fleetmcp.fivetitude.com";
both domains verified (healthz 200, /mcp 401, valid SAN cert).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-23 21:35:31 +03:00
2 changed files with 37 additions and 6 deletions

View file

@ -26,6 +26,7 @@ Env:
"""
from __future__ import annotations
import contextvars
import hmac
import logging
import os
@ -65,6 +66,12 @@ def _get_logger(name: str) -> logging.Logger:
log = _get_logger("server")
# Per-request caller name, set by BearerAuth from the matched token so the tools can
# attribute each query to an analyst in the logs. A ContextVar (not a tool arg) because
# FastMCP tools never receive the HTTP request; anyio propagates the context into the
# worker thread that runs each sync tool. Defaults to "?" if auth ever didn't run.
_caller_var: contextvars.ContextVar[str] = contextvars.ContextVar("caller", default="?")
DATABASE_URL = os.environ["DATABASE_URL"] # analytics_ro DSN (set by deploy)
MAX_ROWS_CEIL = int(os.getenv("MCP_MAX_ROWS", "10000"))
# Schemas the introspection helpers (list_tables/describe_table/sample_table) expose.
@ -284,7 +291,10 @@ def query(sql: str, max_rows: int = 1000) -> dict:
truncated = len(rows) > cap
rows = rows[:cap]
dur_ms = int((time.monotonic() - t0) * 1000)
log.info("query rows=%d trunc=%s %dms :: %s", len(rows), truncated, dur_ms, sql[:200])
log.info(
"query caller=%s rows=%d trunc=%s %dms :: %s",
_caller_var.get(), len(rows), truncated, dur_ms, sql[:200],
)
return {"row_count": len(rows), "truncated": truncated, "rows": rows}
@ -393,6 +403,7 @@ class BearerAuth(BaseHTTPMiddleware):
if caller is None:
return JSONResponse({"error": "unauthorized"}, status_code=401)
request.state.caller = caller
_caller_var.set(caller) # so the tools can attribute the query in the logs
return await call_next(request)

View file

@ -24,6 +24,20 @@ set -euo pipefail
NAME=analytics_mcp
PORT=8892
HOST_DOMAIN="${HOST_DOMAIN:-fleetmcp.fivetitude.com}" # prod: fleetmcp.rahamafresh.com
# Comma-separated list of every domain this service answers on (defaults to
# HOST_DOMAIN). All are folded into ONE Traefik router rule so a single cert
# covers them and connectors on either domain keep working.
HOST_DOMAINS="${HOST_DOMAINS:-$HOST_DOMAIN}"
BT='`'
RULE=""
IFS=',' read -ra _DOMS <<< "$HOST_DOMAINS"
for _d in "${_DOMS[@]}"; do
_d="${_d// /}"
if [ -n "$_d" ]; then
seg="Host(${BT}${_d}${BT})"
if [ -z "$RULE" ]; then RULE="$seg"; else RULE="$RULE || $seg"; fi
fi
done
IMAGE="fleetanalytics-mcp:latest"
ENV_FILE="$(pwd)/.deploy.env"
@ -53,9 +67,15 @@ RO_PW=$(cat "${ANALYTICS_RO_PW_FILE:-$HOME/.analytics_ro.pw}" 2>/dev/null || tru
HOSTPART="${SRC_DB_URL#*@}" # host:port/dbname[?params]
RO_DB_URL="postgresql://analytics_ro:${RO_PW}@${HOSTPART}"
# Build the image from this repo.
# Build the image from this repo (SKIP_BUILD=1 reuses the existing image for a
# labels/env-only change — no new code is pulled in).
if [ "${SKIP_BUILD:-0}" = "1" ]; then
echo "SKIP_BUILD=1 — reusing existing $IMAGE (no rebuild)."
docker image inspect "$IMAGE" >/dev/null 2>&1 || { echo "ERROR: $IMAGE not present"; exit 1; }
else
echo "Building $IMAGE ..."
docker build -t "$IMAGE" .
fi
# Minimal env (read-only DSN + auth only — no Tracksolid ingestion secrets).
{ echo "DATABASE_URL=${RO_DB_URL}"; echo "MCP_AUTH_TOKENS=${MCP_AUTH_TOKENS}"; } > "$ENV_FILE"
@ -73,9 +93,9 @@ docker run -d --name "$NAME" --restart unless-stopped \
--label 'traefik.http.middlewares.fleetmcp-ratelimit.ratelimit.burst=60' \
--label "traefik.http.routers.http-0-fleetmcp.entryPoints=http" \
--label "traefik.http.routers.http-0-fleetmcp.middlewares=redirect-to-https" \
--label "traefik.http.routers.http-0-fleetmcp.rule=Host(\`${HOST_DOMAIN}\`)" \
--label "traefik.http.routers.http-0-fleetmcp.rule=${RULE}" \
--label "traefik.http.routers.https-0-fleetmcp.entryPoints=https" \
--label "traefik.http.routers.https-0-fleetmcp.rule=Host(\`${HOST_DOMAIN}\`)" \
--label "traefik.http.routers.https-0-fleetmcp.rule=${RULE}" \
--label "traefik.http.routers.https-0-fleetmcp.middlewares=fleetmcp-ratelimit" \
--label "traefik.http.routers.https-0-fleetmcp.tls=true" \
--label "traefik.http.routers.https-0-fleetmcp.tls.certresolver=letsencrypt" \