fix(provider-msp): leave first-run platform running

Start the provider services after licensed compose validation so the printed bootstrap and portal link have a live control plane. Correct the first-run prompt to name only operator-provided values and retain an install contract regression test.

Change-source: pulse-maintainer

deploy/provider-msp/setup.sh

@@ -529,26 +529,16 @@ ensure_env_file() {
 
 Created ${env_path} from .env.example.
 
-Edit it now and set required values:
-  - DOMAIN
+Edit it now and set the three values only you can supply:
+  - DOMAIN (client workspaces are served at https://<client-id>.DOMAIN)
   - ACME_EMAIL
   - CF_DNS_API_TOKEN (with the default ACME_DNS_PROVIDER=cloudflare; for any
     other Traefik dnsChallenge provider, set ACME_DNS_PROVIDER and put that
     provider's credential variables in dns-credentials.env)
-  - TRAEFIK_IMAGE (digest pinned)
-  - DOCKER_SOCKET_PROXY_IMAGE (digest pinned)
-  - CONTROL_PLANE_IMAGE (digest pinned)
-  - CP_PULSE_IMAGE (digest pinned)
-  - PULSE_PROVIDER_MSP_DATA_DIR
-  - PULSE_PROVIDER_MSP_DOCKER_NETWORK
-  - PULSE_PROVIDER_MSP_DOCKER_SUBNET
-  - PULSE_PROVIDER_MSP_DOCKER_SOCKET
-  - PULSE_PROVIDER_MSP_ROOT_SPACECHECK_DIR
-  - PULSE_PROVIDER_MSP_DOCKER_SPACECHECK_DIR
-  - CP_TRUSTED_PROXY_CIDRS
 
-setup.sh will generate CP_ADMIN_KEY and CP_ENTITLEMENT_SIGNING_PRIVATE_KEY if they
-are still blank.
+Everything else has a working default. setup.sh resolves the image pins to
+digests and generates CP_ADMIN_KEY and CP_ENTITLEMENT_SIGNING_PRIVATE_KEY while
+they are blank.
 
 EOF
 
@@ -750,6 +740,24 @@ pull_provider_images() {
   done
 }
 
+start_provider_services() {
+  # Leave the platform running. The next steps setup prints (bootstrap, then
+  # the portal sign-in link) only work against a running control plane; a
+  # setup that stopped at "prepared" handed a first-time provider a sign-in
+  # link that answered 404 until something else happened to start it.
+  log "starting provider MSP services"
+  (cd "${PULSE_PROVIDER_MSP_INSTALL_DIR}" && docker compose up -d traefik docker-socket-proxy control-plane)
+
+  local attempt
+  for attempt in $(seq 1 30); do
+    if (cd "${PULSE_PROVIDER_MSP_INSTALL_DIR}" && docker compose ps --services --status running 2>/dev/null) | grep -qx control-plane; then
+      return 0
+    fi
+    sleep 2
+  done
+  die "control plane did not reach running state; inspect: cd ${PULSE_PROVIDER_MSP_INSTALL_DIR} && docker compose logs control-plane"
+}
+
 run_install_proof_if_requested() {
   local mode
   mode="$(echo "${PULSE_PROVIDER_MSP_RUN_INSTALL_PROOF}" | tr '[:upper:]' '[:lower:]')"
@@ -786,7 +794,7 @@ print_summary() {
 
   cat <<EOF
 
-Pulse Provider MSP setup prepared.
+Pulse Provider MSP is running.
 
 Paths:
   - Deploy dir: ${PULSE_PROVIDER_MSP_INSTALL_DIR}
@@ -850,6 +858,7 @@ main() {
   # signal rather than a record created before setup can succeed.
   ensure_eval_license
   validate_compose_config
+  start_provider_services
   run_install_proof_if_requested
   print_summary
 }

docs/release-control/v6/internal/subsystems/deployment-installability.md

@@ -2416,6 +2416,20 @@ artifact-selection behaviour.
 
 ## Current State
 
+### Provider MSP setup leaves the platform running
+
+`deploy/provider-msp/setup.sh` now ends by starting `traefik`,
+`docker-socket-proxy` and `control-plane` and waits for the control plane to
+run before printing its summary. It previously stopped at "setup prepared"
+with only the bootstrap command as the next step, so a first-time provider's
+sign-in link answered `404` until `run-install-proof.sh` or a manual
+`docker compose up -d` happened to start the control plane. The first-run
+message also names only the three values a provider must supply (`DOMAIN`,
+`ACME_EMAIL`, and the DNS-01 credential) instead of listing fourteen, most of
+which have working defaults or are generated. PR #2212 reports a v6.4.1
+first-run 404 and a 200 after starting the platform; this assigned candidate
+has not repeated that live-host proof.
+
 ### Provider MSP operations accept a renewed licence
 
 `provider-msp preflight`, `proof`, `recover` and `backup` treat a licence

scripts/installtests/provider_msp_deploy_test.go

@@ -593,3 +593,42 @@ func TestProviderMSPTraefikEnvIsMinimalAndDNSProviderOverridable(t *testing.T) {
 		t.Fatal("CF_DNS_API_TOKEN is back in the unconditional required-env list; it must be required only when ACME_DNS_PROVIDER is cloudflare")
 	}
 }
+
+// setup.sh used to finish at "setup prepared" without starting the control
+// plane, so the bootstrap sign-in link it printed as the next step answered
+// 404 until the install proof or a manual compose up happened to start it.
+// Setup must bring the provider services up after the final compose
+// validation, wait for the control plane, and only then run the optional
+// install proof and print the summary.
+func TestProviderMSPSetupLeavesPlatformRunning(t *testing.T) {
+	scriptBytes, err := os.ReadFile(repoFile("deploy", "provider-msp", "setup.sh"))
+	if err != nil {
+		t.Fatalf("read provider MSP setup: %v", err)
+	}
+	script := string(scriptBytes)
+
+	assertContainsAll(t, script,
+		"start_provider_services() {",
+		"docker compose up -d traefik docker-socket-proxy control-plane",
+		"docker compose ps --services --status running",
+		"grep -qx control-plane",
+		"control plane did not reach running state",
+		"Pulse Provider MSP is running.",
+	)
+	assertNotContainsAny(t, script, "Pulse Provider MSP setup prepared.")
+
+	sequence := "  ensure_eval_license\n  validate_compose_config\n  start_provider_services\n  run_install_proof_if_requested\n  print_summary\n"
+	if !strings.Contains(script, sequence) {
+		t.Fatal("setup must start provider services after the licensed compose validation and before the install proof and summary")
+	}
+
+	// The first-run prompt names only what a provider must supply.
+	first := script[strings.Index(script, "Edit it now and set"):]
+	first = first[:strings.Index(first, "EOF")]
+	for _, generatedOrDefaulted := range []string{"TRAEFIK_IMAGE", "CP_PULSE_IMAGE", "PULSE_PROVIDER_MSP_DOCKER_SUBNET", "CP_TRUSTED_PROXY_CIDRS"} {
+		if strings.Contains(first, "  - "+generatedOrDefaulted) {
+			t.Fatalf("first-run prompt still lists %s as a value the provider must set", generatedOrDefaulted)
+		}
+	}
+	assertContainsAll(t, first, "DOMAIN", "ACME_EMAIL", "CF_DNS_API_TOKEN")
+}