WIP autoscaler agent

agent/nix_builder_autoscaler/tests/integration/test_end_to_end_fake.py

@@ -0,0 +1 @@
+"""End-to-end integration tests with FakeRuntime — Plan 05."""

agent/nix_builder_autoscaler/tests/test_haproxy_provider.py

@@ -0,0 +1 @@
+"""HAProxy provider unit tests — Plan 02."""

agent/nix_builder_autoscaler/tests/test_reservations_api.py

@@ -0,0 +1 @@
+"""Reservations API unit tests — Plan 04."""

agent/nix_builder_autoscaler/tests/test_runtime_ec2.py

@@ -0,0 +1 @@
+"""EC2 runtime unit tests — Plan 02."""

agent/nix_builder_autoscaler/tests/test_runtime_fake.py

@@ -0,0 +1,116 @@
+"""Unit tests for the FakeRuntime adapter."""
+
+import contextlib
+
+from nix_builder_autoscaler.runtime.base import RuntimeError as RuntimeAdapterError
+from nix_builder_autoscaler.runtime.fake import FakeRuntime
+
+
+class TestLaunchSpot:
+    def test_returns_synthetic_instance_id(self):
+        rt = FakeRuntime()
+        iid = rt.launch_spot("slot001", "#!/bin/bash\necho hello")
+        assert iid.startswith("i-fake-")
+        assert len(iid) > 10
+
+    def test_instance_starts_pending(self):
+        rt = FakeRuntime()
+        iid = rt.launch_spot("slot001", "")
+        info = rt.describe_instance(iid)
+        assert info["state"] == "pending"
+        assert info["tailscale_ip"] is None
+
+
+class TestTickProgression:
+    def test_transitions_to_running_after_configured_ticks(self):
+        rt = FakeRuntime(launch_latency_ticks=3, ip_delay_ticks=1)
+        iid = rt.launch_spot("slot001", "")
+
+        for _ in range(2):
+            rt.tick()
+        assert rt.describe_instance(iid)["state"] == "pending"
+
+        rt.tick()  # tick 3
+        assert rt.describe_instance(iid)["state"] == "running"
+
+    def test_tailscale_ip_appears_after_configured_delay(self):
+        rt = FakeRuntime(launch_latency_ticks=2, ip_delay_ticks=2)
+        iid = rt.launch_spot("slot001", "")
+
+        for _ in range(2):
+            rt.tick()
+        assert rt.describe_instance(iid)["state"] == "running"
+        assert rt.describe_instance(iid)["tailscale_ip"] is None
+
+        rt.tick()  # tick 3 — still no IP (need tick 4)
+        assert rt.describe_instance(iid)["tailscale_ip"] is None
+
+        rt.tick()  # tick 4
+        info = rt.describe_instance(iid)
+        assert info["tailscale_ip"] is not None
+        assert info["tailscale_ip"].startswith("100.64.0.")
+
+
+class TestInjectedFailure:
+    def test_launch_failure_raises(self):
+        rt = FakeRuntime()
+        rt.inject_launch_failure("slot001")
+        try:
+            rt.launch_spot("slot001", "")
+            raise AssertionError("Should have raised")
+        except RuntimeAdapterError as e:
+            assert e.category == "capacity_unavailable"
+
+    def test_failure_is_one_shot(self):
+        rt = FakeRuntime()
+        rt.inject_launch_failure("slot001")
+        with contextlib.suppress(RuntimeAdapterError):
+            rt.launch_spot("slot001", "")
+        # Second call should succeed
+        iid = rt.launch_spot("slot001", "")
+        assert iid.startswith("i-fake-")
+
+
+class TestInjectedInterruption:
+    def test_interruption_returns_terminated(self):
+        rt = FakeRuntime(launch_latency_ticks=1)
+        iid = rt.launch_spot("slot001", "")
+        rt.tick()
+        assert rt.describe_instance(iid)["state"] == "running"
+
+        rt.inject_interruption(iid)
+        info = rt.describe_instance(iid)
+        assert info["state"] == "terminated"
+
+    def test_interruption_is_one_shot(self):
+        """After the interruption fires, subsequent describes stay terminated."""
+        rt = FakeRuntime(launch_latency_ticks=1)
+        iid = rt.launch_spot("slot001", "")
+        rt.tick()
+        rt.inject_interruption(iid)
+        rt.describe_instance(iid)  # consumes the injection
+        info = rt.describe_instance(iid)
+        assert info["state"] == "terminated"
+
+
+class TestTerminate:
+    def test_terminate_marks_instance(self):
+        rt = FakeRuntime(launch_latency_ticks=1)
+        iid = rt.launch_spot("slot001", "")
+        rt.tick()
+        rt.terminate_instance(iid)
+        assert rt.describe_instance(iid)["state"] == "terminated"
+
+
+class TestListManaged:
+    def test_lists_non_terminated(self):
+        rt = FakeRuntime(launch_latency_ticks=1)
+        iid1 = rt.launch_spot("slot001", "")
+        iid2 = rt.launch_spot("slot002", "")
+        rt.tick()
+        rt.terminate_instance(iid1)
+
+        managed = rt.list_managed_instances()
+        ids = [m["instance_id"] for m in managed]
+        assert iid2 in ids
+        assert iid1 not in ids

agent/nix_builder_autoscaler/tests/test_scheduler.py

@@ -0,0 +1 @@
+"""Scheduler unit tests — Plan 03."""