use userdata only to seed ami bootstrap env

agent/nix_builder_autoscaler/bootstrap/userdata.py

@@ -1,71 +1,27 @@
-"""EC2 user-data template rendering for builder instance bootstrap.
-
-The generated script follows the NixOS AMI pattern: write config files
-that existing systemd services (tailscale-autoconnect, nix-daemon) consume,
-rather than calling ``tailscale up`` directly.
-"""
+"""EC2 user-data template rendering for builder instance bootstrap."""
 
 from __future__ import annotations
 
 import textwrap
 
 
-def render_userdata(slot_id: str, region: str, ssm_param: str = "/nix-builder/ts-authkey") -> str:
-    """Render a bash user-data script for builder instance bootstrap.
+def render_userdata(slot_id: str, ssm_param: str = "/nix-builder/ts-authkey") -> str:
+    """Render user-data that seeds AMI bootstrap inputs only.
 
-    The returned string is a complete shell script.  On NixOS AMIs the script
-    is executed by ``amazon-init.service``.  The caller (EC2Runtime) passes it
-    to ``run_instances`` as ``UserData``; boto3 base64-encodes automatically.
-
-    Args:
-        slot_id: Autoscaler slot identifier (used as Tailscale hostname suffix).
-        region: AWS region for SSM parameter lookup.
-        ssm_param: SSM parameter path containing the Tailscale auth key.
+    The AMI's buildbot-ami-bootstrap service consumes this env file and handles
+    SSM fetch + tailscale-autoconnect config generation.
     """
     return textwrap.dedent(f"""\
         #!/usr/bin/env bash
         set -euo pipefail
 
         SLOT_ID="{slot_id}"
-        REGION="{region}"
         SSM_PARAM="{ssm_param}"
 
-        # --- Fetch Tailscale auth key from SSM Parameter Store ---
-        mkdir -p /run/credentials
-        TS_AUTHKEY=$(aws ssm get-parameter \\
-            --region "$REGION" \\
-            --with-decryption \\
-            --name "$SSM_PARAM" \\
-            --query 'Parameter.Value' \\
-            --output text)
-        printf '%s' "$TS_AUTHKEY" > /run/credentials/tailscale-auth-key
-        chmod 600 /run/credentials/tailscale-auth-key
-
-        # --- Resolve instance identity from IMDSv2 for unique hostname ---
-        IMDS_TOKEN=$(curl -fsS -X PUT "http://169.254.169.254/latest/api/token" \\
-            -H "X-aws-ec2-metadata-token-ttl-seconds: 21600" || true)
-        INSTANCE_ID=$(curl -fsS -H "X-aws-ec2-metadata-token: $IMDS_TOKEN" \\
-            "http://169.254.169.254/latest/meta-data/instance-id" || true)
-        if [ -z "$INSTANCE_ID" ]; then
-          INSTANCE_ID="unknown"
-        fi
-
-        # --- Write tailscale-autoconnect config ---
-        mkdir -p /etc/tailscale
-        cat > /etc/tailscale/autoconnect.conf <<TSCONF
-        TS_AUTHKEY_FILE=/run/credentials/tailscale-auth-key
-        TS_AUTHKEY_EPHEMERAL=true
-        TS_AUTHKEY_PREAUTHORIZED=true
-        TS_HOSTNAME=nix-builder-$SLOT_ID-$INSTANCE_ID
-        TS_EXTRA_ARGS="--ssh --advertise-tags=tag:nix-builder"
-        TSCONF
-
-        # --- Start/restart tailscale-autoconnect so it picks up the config ---
-        systemctl restart tailscale-autoconnect.service || true
-
-        # --- Ensure nix-daemon is running ---
-        systemctl start nix-daemon.service || true
-
-        # --- Signal readiness ---
-        echo "ready" > /run/nix-builder-ready
+        # Seed AMI bootstrap inputs only; buildbot-ami-bootstrap reads this file.
+        cat > /etc/nix-builder-bootstrap-env <<EOF
+        SLOT_ID="$SLOT_ID"
+        TS_SSM_PARAM="$SSM_PARAM"
+        EOF
+        chmod 600 /etc/nix-builder-bootstrap-env
     """)

agent/nix_builder_autoscaler/scheduler.py

@@ -216,7 +216,7 @@ def _launch_slot(
 ) -> None:
     """Launch a single slot. Transition to LAUNCHING on success, ERROR on failure."""
     slot_id = slot["slot_id"]
-    user_data = render_userdata(slot_id, config.aws.region)
+    user_data = render_userdata(slot_id)
     try:
         instance_id = runtime.launch_spot(slot_id, user_data)
         metrics.counter("autoscaler_ec2_launch_total", {"result": "success"}, 1.0)

agent/nix_builder_autoscaler/tests/test_bootstrap_userdata.py

@@ -0,0 +1,18 @@
+"""Unit tests for builder bootstrap user-data rendering."""
+
+from nix_builder_autoscaler.bootstrap.userdata import render_userdata
+
+
+def test_render_userdata_writes_bootstrap_env_inputs() -> None:
+    script = render_userdata("slot001", ssm_param="/nix-builder/ts-authkey")
+    assert "cat > /etc/nix-builder-bootstrap-env <<EOF" in script
+    assert 'SLOT_ID="$SLOT_ID"' in script
+    assert 'TS_SSM_PARAM="$SSM_PARAM"' in script
+    assert "chmod 600 /etc/nix-builder-bootstrap-env" in script
+
+
+def test_render_userdata_does_not_inline_tailscale_setup() -> None:
+    script = render_userdata("slot001", ssm_param="/nix-builder/ts-authkey")
+    assert "aws ssm get-parameter" not in script
+    assert "/etc/tailscale/autoconnect.conf" not in script
+    assert "TS_EXTRA_ARGS=" not in script