fix(infra): gg tls certs timer & precreate tls dir

Author: NathanFlurry

svc/pkg/cluster/worker/src/workers/server_install/install_scripts/components/rivet.rs

@@ -26,17 +26,25 @@ pub fn fetch_tls(
 	initialize_immediately: bool,
 	server_token: &str,
 	traefik_instance_name: &str,
+	datacenter_id: Uuid,
 ) -> GlobalResult<String> {
 	let mut script = include_str!("../files/rivet_fetch_tls.sh")
 		.replace("__NAME__", traefik_instance_name)
 		.replace("__SERVER_TOKEN__", server_token)
 		.replace(
 			"__TUNNEL_API_INTERNAL_PORT__",
 			&TUNNEL_API_INTERNAL_PORT.to_string(),
-		);
+		)
+		.replace("__DATACENTER_ID__", &datacenter_id.to_string());
 
 	if initialize_immediately {
-		script.push_str("systemctl start rivet_fetch_tls.timer\n");
+		// Start timer & run script immediately
+		script.push_str(indoc!(
+			"
+			systemctl start rivet_fetch_tls.timer
+			systemctl start --no-block rivet_fetch_tls.service
+			"
+		));
 	}
 
 	Ok(script)

svc/pkg/cluster/worker/src/workers/server_install/install_scripts/files/rivet_fetch_tls.sh

@@ -1,3 +1,10 @@
+# Create dir to hold TLS certs
+#
+# The Traefik install script also creates these directories (and chown them),
+# but we need the dirs to exist for the rivet_fetch_tls.sh script to run before
+# Traefik is installed when using initialize_immediately.
+mkdir -p /etc/__NAME__/dynamic/tls /etc/__NAME__/tls
+
 # Write script
 cat << 'EOF' > /usr/bin/rivet_fetch_tls.sh
 #!/usr/bin/env bash
@@ -6,12 +13,13 @@ set -eu -o pipefail
 CERT_ID="job"
 STUB="/etc/__NAME__/tls/$CERT_ID"
 
+
 # Retry script every 5 seconds
 while true; do
   response=$(
     curl -f \
       -H "Authorization: Bearer __SERVER_TOKEN__" \
-      "http://127.0.0.1:__TUNNEL_API_INTERNAL_PORT__/provision/datacenters/___DATACENTER_ID___/tls"
+      "http://127.0.0.1:__TUNNEL_API_INTERNAL_PORT__/provision/datacenters/__DATACENTER_ID__/tls"
   ) && break || sleep 5
 done
 
@@ -59,7 +67,12 @@ Requires=network-online.target
 After=network-online.target
 
 [Timer]
-OnUnitInactiveSec=1h
+# Run immediately on startup
+OnBootSec=0
+# Trigger every hour
+OnCalendar=*:0
+# Prevent stampeding herd
+RandomizedDelaySec=60
 Unit=rivet_fetch_tls.service
 
 [Install]

svc/pkg/cluster/worker/src/workers/server_install/install_scripts/mod.rs

@@ -14,6 +14,7 @@ pub async fn gen_install(
 	pool_type: backend::cluster::PoolType,
 	initialize_immediately: bool,
 	server_token: &str,
+	datacenter_id: Uuid,
 ) -> GlobalResult<String> {
 	// MARK: Common (pre)
 	let mut script = vec![
@@ -41,6 +42,7 @@ pub async fn gen_install(
 				initialize_immediately,
 				server_token,
 				GG_TRAEFIK_INSTANCE_NAME,
+				datacenter_id,
 			)?);
 		}
 		backend::cluster::PoolType::Ats => {

svc/pkg/cluster/worker/src/workers/server_install/mod.rs

@@ -13,6 +13,8 @@ mod install_scripts;
 
 #[worker(name = "cluster-server-install", timeout = 200)]
 async fn worker(ctx: &OperationContext<cluster::msg::server_install::Message>) -> GlobalResult<()> {
+	let datacenter_id = unwrap!(ctx.datacenter_id).as_uuid();
+
 	// Check for stale message
 	if ctx.req_dt() > util::duration::hours(1) {
 		tracing::warn!("discarding stale message");
@@ -71,8 +73,13 @@ async fn worker(ctx: &OperationContext<cluster::msg::server_install::Message>) -
 	.await?;
 	let server_token = &unwrap_ref!(token_res.token).token;
 
-	let install_script =
-		install_scripts::gen_install(pool_type, ctx.initialize_immediately, server_token).await?;
+	let install_script = install_scripts::gen_install(
+		pool_type,
+		ctx.initialize_immediately,
+		server_token,
+		datacenter_id,
+	)
+	.await?;
 	let hook_script = install_scripts::gen_hook(server_token).await?;
 	let initialize_script = install_scripts::gen_initialize(pool_type).await?;
 
@@ -161,7 +168,6 @@ async fn worker(ctx: &OperationContext<cluster::msg::server_install::Message>) -
 		.await?;
 
 		// Scale to get rid of tainted servers
-		let datacenter_id = unwrap_ref!(ctx.datacenter_id).as_uuid();
 		msg!([ctx] @recursive cluster::msg::datacenter_scale(datacenter_id) {
 			datacenter_id: ctx.datacenter_id,
 		})