fix(gateway): persist git-gate state across gateway restarts
tracker-policy-pr / check-pr (pull_request) Successful in 12s
lint / lint (push) Successful in 59s
test / unit (pull_request) Successful in 53s
test / integration-docker (pull_request) Successful in 1m7s
test / coverage (pull_request) Successful in 19s

Per-bottle git-gate state (bare repos under /git/<id>, deploy creds under
/git-gate/creds/<id>) was provisioned once at bottle launch and lived only
in the gateway's ephemeral storage. A gateway rebuild/restart wiped it and
nothing re-provisioned already-running bottles, so their agents 404'd on
fetch/push. Same class of bug as the CA (#510); the orchestrator restores
only egress tokens, not git-gate declarations.

Persist the state on both backends, mirroring the CA-persistence approach:

- firecracker: attach a second persistent data drive (/dev/vdc) to the
  gateway VM and bind-mount its git/ + creds/ subdirs onto /git and
  /git-gate/creds in the gateway guest init, before the data plane starts.
  Generalize the VM config to a stable-ordered data_drives tuple (CA=vdb,
  git=vdc; orchestrator registry stays vdb).
- docker: bind-mount host dirs (host_gateway_git_dir / creds_dir, under the
  never-pruned app-data root) onto /git and /git-gate/creds, with
  BOT_BOTTLE_DOCKER_GIT_MOUNT / _CREDS_MOUNT env overrides so CI isolates
  them to per-run volumes it cleans up.

Teardown already rm -rf's /git/<id> + creds, so the persistent store
self-cleans over the normal lifecycle.

Closes #512

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
2026-07-26 17:34:22 -04:00
parent 0ddaa95c14
commit dff811f190
13 changed files with 213 additions and 27 deletions
+7 -4
View File
@@ -404,16 +404,19 @@ class TestBootArgs(unittest.TestCase):
self.assertEqual("bbfc0", cfg["network-interfaces"][0]["host_dev_name"])
self.assertEqual(1, len(cfg["drives"])) # no data drive by default
def test_config_adds_data_drive(self):
def test_config_adds_data_drives_in_order(self):
cfg = cast(Any, firecracker_vm._config(
rootfs=Path("/run/rootfs.ext4"), tap="bbfc0",
guest_ip="100.64.0.1", host_ip="100.64.0.0", pubkey="k",
vcpus=2, mem_mib=2048, guest_mac="06:00:AC:10:00:02",
data_drive=Path("/run/registry.ext4"),
data_drives=(Path("/run/ca.ext4"), Path("/run/git.ext4")),
))
self.assertEqual(2, len(cfg["drives"]))
# rootfs (vda) + two data drives, attached in list order so the guest
# sees them as /dev/vdb, /dev/vdc — an order callers depend on.
self.assertEqual(3, len(cfg["drives"]))
self.assertFalse(cfg["drives"][1]["is_root_device"])
self.assertEqual("/run/registry.ext4", cfg["drives"][1]["path_on_host"])
self.assertEqual("/run/ca.ext4", cfg["drives"][1]["path_on_host"])
self.assertEqual("/run/git.ext4", cfg["drives"][2]["path_on_host"])
class TestBottleExecClose(unittest.TestCase):
+34 -3
View File
@@ -60,9 +60,12 @@ class TestFirecrackerGatewayConnect(unittest.TestCase):
gw = FirecrackerGateway()
booted = infra_vm.InfraVm(guest_ip="10.243.255.3", private_key=Path("/k"))
ca_vol = Path("/gw/gateway-ca.ext4")
git_vol = Path("/gw/gateway-git.ext4")
with patch.object(infra_vm, "boot_vm", return_value=booted) as boot, \
patch.object(FirecrackerGateway, "_ensure_ca_volume",
return_value=ca_vol), \
patch.object(FirecrackerGateway, "_ensure_git_volume",
return_value=git_vol), \
patch.object(infra_vm, "push_secret") as push:
gw.connect_to_orchestrator(_ORCH_URL, _TOKEN)
# Booted on the gateway link with the gateway role; the orchestrator's
@@ -70,9 +73,9 @@ class TestFirecrackerGatewayConnect(unittest.TestCase):
kw = boot.call_args.kwargs
self.assertEqual("gateway", kw["role"])
self.assertIn("bb_orch=10.243.255.1", kw["extra_boot_args"])
# The persistent CA volume rides as /dev/vdb so the mitmproxy CA
# survives a gateway-VM rebuild (issue #450).
self.assertEqual(ca_vol, kw.get("data_drive"))
# The persistent volumes ride in a FIXED order — CA as /dev/vdb, git-gate
# as /dev/vdc — so both survive a gateway-VM rebuild (issues #450, #512).
self.assertEqual((ca_vol, git_vol), kw.get("data_drives"))
# The host-minted token (never the key — #469) is pushed to the guest.
push.assert_called_once()
self.assertEqual(_TOKEN, push.call_args.args[1])
@@ -107,6 +110,34 @@ class TestGatewayCaVolume(unittest.TestCase):
self.assertIn(str(out), argv)
class TestGatewayGitVolume(unittest.TestCase):
"""The persistent git-gate volume that keeps per-bottle bare repos + creds
stable across a gateway-VM rebuild (issue #512)."""
def test_reuses_existing_volume(self) -> None:
gw = FirecrackerGateway()
with tempfile.TemporaryDirectory() as td:
vol = Path(td) / "gateway-git.ext4"
vol.write_bytes(b"") # already present
with patch.object(infra_vm, "_gw_dir", return_value=Path(td)), \
patch(f"{_GW}.subprocess.run") as run:
out = gw._ensure_git_volume()
run.assert_not_called() # no mke2fs when it exists — the repos survive
self.assertEqual(vol, out)
def test_creates_volume_when_missing(self) -> None:
gw = FirecrackerGateway()
with tempfile.TemporaryDirectory() as td:
with patch.object(infra_vm, "_gw_dir", return_value=Path(td)), \
patch(f"{_GW}.subprocess.run",
return_value=CompletedProcess([], 0)) as run:
out = gw._ensure_git_volume()
argv = run.call_args.args[0]
self.assertIn("mke2fs", argv)
self.assertEqual(str(Path(td) / "gateway-git.ext4"), out.__fspath__())
self.assertIn(str(out), argv)
class TestFirecrackerGatewaySurface(unittest.TestCase):
def test_is_running_reads_the_gateway_pidfile(self) -> None:
gw = FirecrackerGateway()
+12
View File
@@ -66,6 +66,18 @@ class TestRoleInits(unittest.TestCase):
self.assertIn(f"mount -t ext4 /dev/vdb {infra_vm._GATEWAY_CA_MOUNT}", init)
self.assertLess(init.index("/dev/vdb"),
init.index("bot_bottle.gateway.bootstrap"))
# Persistent git-gate volume (/dev/vdc) bind-mounted onto /git and
# /git-gate/creds so per-bottle repos + creds survive a rebuild (#512),
# also before the data plane (and any provisioning writes).
self.assertIn(f"mount -t ext4 /dev/vdc {infra_vm._GATEWAY_GIT_MOUNT}", init)
self.assertIn(
f"mount --bind {infra_vm._GATEWAY_GIT_MOUNT}/git "
f"{infra_vm._GATEWAY_GIT_REPO_ROOT}", init)
self.assertIn(
f"mount --bind {infra_vm._GATEWAY_GIT_MOUNT}/creds "
f"{infra_vm._GATEWAY_GIT_CREDS_DIR}", init)
self.assertLess(init.index("/dev/vdc"),
init.index("bot_bottle.gateway.bootstrap"))
self.assertIn("export PATH=", init) # shared preamble
self.assertIn("BOT_BOTTLE_GATEWAY_DAEMONS=egress,git-http,supervise", init)
self.assertIn(f"cat {infra_vm._GUEST_GATEWAY_JWT_PATH}", init) # host-minted JWT
+1 -1
View File
@@ -52,7 +52,7 @@ class TestEnsureRunning(unittest.TestCase):
kw = boot.call_args.kwargs
self.assertEqual("orchestrator", kw["role"])
self.assertEqual(4096, kw["mem_mib"]) # orchestrator keeps build headroom
self.assertEqual(Path("/reg"), kw["data_drive"]) # DB volume on the CP
self.assertEqual((Path("/reg"),), kw["data_drives"]) # DB volume on the CP
# The host-canonical signing key is pushed to the guest signing-key path.
self.assertEqual("host-key", push.call_args.args[1])
self.assertEqual(infra_vm._GUEST_SIGNING_KEY_PATH, push.call_args.args[2])
+20
View File
@@ -161,6 +161,26 @@ class TestDockerGateway(unittest.TestCase):
"ci-ca-volume:/home/mitmproxy/.mitmproxy", argv
)
def test_persists_git_gate_state_across_recreation(self) -> None:
# Per-bottle bare repos + deploy creds are bind-mounted from the host so
# a gateway restart doesn't drop already-running bottles' git-gate state
# (issue #512). Named sources here stand in for CI's per-run volumes.
sc = DockerGateway(
"bot-bottle-gateway:latest",
git_mount_source="ci-git-volume",
creds_mount_source="ci-creds-volume",
)
def fake(argv: list[str], **_kw: object) -> Mock:
return _proc(stdout="") if argv[:2] == ["docker", "ps"] else _proc()
with patch(_RUN_DOCKER, side_effect=fake) as run:
sc.connect_to_orchestrator(_ORCH_URL, _TOKEN)
argv = next(c.args[0] for c in run.call_args_list
if c.args[0][:2] == ["docker", "run"])
self.assertIn("ci-git-volume:/git", argv)
self.assertIn("ci-creds-volume:/git-gate/creds", argv)
def test_connect_injects_the_pre_minted_gateway_token(self) -> None:
# The gateway presents the token the orchestrator handed it — it never
# mints (holds no signing key). The value rides the env (bare `--env