huskies: merge 811

2026-05-14 18:24:25 +00:00
parent 03a0ca258a
commit bb5abcd042
2 changed files with 381 additions and 0 deletions
@@ -0,0 +1,101 @@
+#!/usr/bin/env bash
+# fly_multitenant_poc.sh — Proof of concept for Spike 811.
+#
+# Demonstrates the Fly.io Machines API calls that the huskies gateway
+# will eventually make to provision and tear down a per-tenant project
+# machine. Run against a real Fly org with FLY_API_TOKEN set, or read it
+# as a commented sketch — the calls are the contract.
+#
+# This is NOT production code. Production will issue these requests
+# from Rust (see server::service::cloud::fly) with retries, structured
+# errors, and CRDT writes to record machine_id/volume_id. The shell
+# script exists so the spec is verifiable end-to-end.
+#
+# Required env:
+#   FLY_API_TOKEN   - org-scoped Fly token
+#   FLY_APP         - name of the huskies-projects Fly app (must exist)
+#   TENANT_ID       - identifier used to tag and name the machine
+#   REGION          - Fly region code, e.g. "iad" (default: iad)
+
+set -euo pipefail
+
+: "${FLY_API_TOKEN:?FLY_API_TOKEN must be set}"
+: "${FLY_APP:?FLY_APP must be set}"
+: "${TENANT_ID:?TENANT_ID must be set}"
+REGION="${REGION:-iad}"
+IMAGE="registry.fly.io/huskies-projects:latest"
+
+API="https://api.machines.dev/v1"
+AUTH=(-H "Authorization: Bearer ${FLY_API_TOKEN}" -H "Content-Type: application/json")
+
+echo "==> 1. Create a 1 GiB persistent volume for tenant ${TENANT_ID}"
+VOLUME_JSON=$(curl -sS -X POST "${API}/apps/${FLY_APP}/volumes" "${AUTH[@]}" --data @- <<EOF
+{
+  "name":   "huskies_${TENANT_ID}",
+  "region": "${REGION}",
+  "size_gb": 1
+}
+EOF
+)
+VOLUME_ID=$(echo "${VOLUME_JSON}" | jq -r .id)
+echo "    volume_id = ${VOLUME_ID}"
+
+echo "==> 2. Create a machine attached to the volume, with auto-suspend"
+MACHINE_JSON=$(curl -sS -X POST "${API}/apps/${FLY_APP}/machines" "${AUTH[@]}" --data @- <<EOF
+{
+  "name":   "huskies-${TENANT_ID}",
+  "region": "${REGION}",
+  "config": {
+    "image": "${IMAGE}",
+    "env": {
+      "TENANT_ID":      "${TENANT_ID}",
+      "HUSKIES_PORT":   "3001",
+      "PRIMARY_REGION": "${REGION}"
+    },
+    "guest": { "cpu_kind": "shared", "cpus": 2, "memory_mb": 2048 },
+    "mounts": [ { "volume": "${VOLUME_ID}", "path": "/data" } ],
+    "services": [ {
+      "ports": [
+        { "port": 443, "handlers": ["tls","http"] },
+        { "port": 80,  "handlers": ["http"] }
+      ],
+      "protocol": "tcp",
+      "internal_port": 3001,
+      "auto_stop_machines":  "suspend",
+      "auto_start_machines": true,
+      "min_machines_running": 0
+    } ],
+    "metadata": { "tenant": "${TENANT_ID}", "managed_by": "huskies-gw" },
+    "restart": { "policy": "on-failure", "max_retries": 5 }
+  }
+}
+EOF
+)
+MACHINE_ID=$(echo "${MACHINE_JSON}" | jq -r .id)
+PRIVATE_IP=$(echo "${MACHINE_JSON}" | jq -r .private_ip)
+echo "    machine_id = ${MACHINE_ID}"
+echo "    private_ip = ${PRIVATE_IP}"
+
+echo "==> 3. Wait for the machine to reach 'started' (long-poll, 60s timeout)"
+curl -sS "${API}/apps/${FLY_APP}/machines/${MACHINE_ID}/wait?state=started&timeout=60" "${AUTH[@]}" \
+  | jq -r '"    state = " + .ok'
+
+echo "    machine reachable at ${MACHINE_ID}.vm.${FLY_APP}.internal:3001"
+
+# ----- At this point the gateway would record (tenant, machine_id, volume_id)
+# ----- into the CRDT and start proxying traffic. We pause here.
+sleep 2
+
+echo "==> 4. Graceful stop (lets sled flush; idle-suspend uses the same path)"
+curl -sS -X POST "${API}/apps/${FLY_APP}/machines/${MACHINE_ID}/stop" "${AUTH[@]}" \
+  --data '{"signal":"SIGTERM","timeout":"30s"}' > /dev/null
+
+echo "==> 5. Destroy the machine"
+curl -sS -X DELETE "${API}/apps/${FLY_APP}/machines/${MACHINE_ID}?force=true" "${AUTH[@]}" > /dev/null
+echo "    machine destroyed"
+
+echo "==> 6. Reclaim the volume (only when the tenant deletes the project)"
+curl -sS -X DELETE "${API}/apps/${FLY_APP}/volumes/${VOLUME_ID}" "${AUTH[@]}" > /dev/null
+echo "    volume reclaimed"
+
+echo "==> done."