Skip to content

Ministack: Free, open-source local AWS emulator - 60+ services, Terraform compatible, real databases. Free forever. MIT licensed.

Release/release v1.3.0 (#375)

Nahuel Nucerae39bc55b650f
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 6d718f97..777530bd 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -17,6 +17,8 @@ Versioning follows [Semantic Versioning](https://semver.org/).
### Fixed
- **Glue `GetPartitionIndexes` Keys format** — service returned Keys as flat strings (`["year"]`) instead of KeySchemaElement objects (`[{"Name": "year"}]`), causing boto3 deserialization failures
- **RDS `LatestRestorableTime` empty timestamp** — `DescribeDBInstances` rendered `<LatestRestorableTime></LatestRestorableTime>` (empty string) which boto3 couldn't parse as a timestamp. Now defaults to current time
+- **EKS graceful fallback when k3s fails** — if Docker is unavailable or k3s container fails to start (e.g. privileged containers blocked), `CreateCluster` now returns ACTIVE with a mock endpoint and CA certificate instead of FAILED. The EKS API works identically regardless of Docker availability; real k3s is used when possible
+- **EKS state persistence** — restored clusters stay ACTIVE instead of being marked FAILED on restart
- **EKS Docker tests flaky in parallel** — k3s containers interfere with each other under pytest-xdist. Added both EKS Docker tests to `_SERIAL_TESTS`
- **EKS CFN test CI failure** — k3s can't start on CI (no Docker), cluster stays in CREATING. Test now polls and accepts CREATING status
diff --git a/ministack/app.py b/ministack/app.py
index 15d20337..99be428c 100644
--- a/ministack/app.py
+++ b/ministack/app.py
@@ -221,6 +221,16 @@ BANNER = r"""
"""
+_reset_lock: "asyncio.Lock | None" = None
+
+
+def _get_reset_lock() -> asyncio.Lock:
+ global _reset_lock
+ if _reset_lock is None:
+ _reset_lock = asyncio.Lock()
+ return _reset_lock
+
+
async def app(scope, receive, send):
"""ASGI application entry point."""
if scope["type"] == "lifespan":
@@ -286,6 +296,14 @@ async def app(scope, receive, send):
request_id = str(uuid.uuid4())
+ # If a /_ministack/reset is in flight, wait for it to finish before
+ # serving this request. The lock is uncontended in steady state
+ # (acquire/release is near-free); during a reset, new requests block
+ # until state-wipe completes so no test can observe a half-reset server.
+ if path != "/_ministack/reset":
+ async with _get_reset_lock():
+ pass
+
# Set per-request account ID from credentials (multi-tenancy support).
# If the access key is a 12-digit number, it becomes the account ID.
_access_key = extract_access_key_id(headers)
@@ -342,7 +360,10 @@ async def app(scope, receive, send):
# Admin endpoints — no wildcard CORS headers (return early, before CORS block)
if path == "/_ministack/reset" and method == "POST":
- await asyncio.to_thread(_reset_all_state)
+ # Hold the reset lock exclusively so in-flight requests drain first
+ # and no new requests can interleave with service state wipes.
+ async with _get_reset_lock():
+ await asyncio.to_thread(_reset_all_state)
run_init = query_params.get("init", [""])[0] == "1"
if run_init:
_run_init_scripts()
diff --git a/ministack/services/eks.py b/ministack/services/eks.py
index 3c51d9db..d424bab9 100644
--- a/ministack/services/eks.py
+++ b/ministack/services/eks.py
@@ -78,18 +78,14 @@ def restore_state(data):
_tags.update(data.get("tags", {}))
if "port_counter" in data:
_port_counter[0] = data["port_counter"]
- # Restored clusters have no running k3s container — mark status accordingly
+ # Restored clusters have no running k3s container — keep ACTIVE with mock endpoint
if isinstance(_clusters, AccountScopedDict):
for key in list(_clusters._data):
c = _clusters._data[key]
- if c.get("status") == "ACTIVE":
- c["status"] = "FAILED"
- c["_docker_id"] = None
+ c["_docker_id"] = None
else:
for c in _clusters.values():
- if c.get("status") == "ACTIVE":
- c["status"] = "FAILED"
- c["_docker_id"] = None
+ c["_docker_id"] = None
# ---------------------------------------------------------------------------
@@ -267,12 +263,22 @@ def _create_cluster(body):
"--tls-san=0.0.0.0",
f"--https-listen-port=6443"],
detach=True,
- privileged=True,
+ cap_add=[
+ "SYS_ADMIN", "NET_ADMIN", "NET_RAW", "NET_BIND_SERVICE",
+ "SYS_PTRACE", "SYS_RESOURCE", "SYS_CHROOT",
+ "DAC_OVERRIDE", "DAC_READ_SEARCH",
+ "FOWNER", "FSETID", "CHOWN", "MKNOD",
+ "KILL", "SETGID", "SETUID", "SETPCAP", "SETFCAP",
+ "AUDIT_WRITE",
+ ],
+ security_opt=["seccomp=unconfined", "apparmor=unconfined"],
+ devices=["/dev/fuse"],
ports={"6443/tcp": port},
name=f"ministack-eks-{name}",
labels={"ministack": "eks", "cluster_name": name},
environment={"K3S_KUBECONFIG_MODE": "644"},
- tmpfs={"/run": "", "/var/run": ""},
+ volumes={"/lib/modules": {"bind": "/lib/modules", "mode": "ro"}},
+ tmpfs={"/run": "", "/var/run": "", "/tmp": ""},
)
if ms_network:
run_kwargs["network"] = ms_network
@@ -300,8 +306,10 @@ def _create_cluster(body):
cluster["certificateAuthority"]["data"] = _extract_ca_cert(container)
cluster["status"] = "ACTIVE"
except Exception as e:
- logger.warning("EKS: failed to start k3s for %s: %s", name, e)
- cluster["status"] = "FAILED"
+ logger.warning("EKS: failed to start k3s for %s — falling back to mock: %s", name, e)
+ cluster["status"] = "ACTIVE"
+ cluster["certificateAuthority"]["data"] = base64.b64encode(b"MOCK-CA-CERTIFICATE").decode()
+ cluster["endpoint"] = f"https://localhost:{port}"
threading.Thread(target=_bg_start, daemon=True, name=f"eks-{name}").start()
return _json_resp(200, {"cluster": _sanitize(cluster)})
diff --git a/tests/test_eks.py b/tests/test_eks.py
index d535b4da..4f74f9ea 100644
--- a/tests/test_eks.py
+++ b/tests/test_eks.py
@@ -10,6 +10,7 @@ import pytest
import boto3
from botocore.exceptions import ClientError
+
ENDPOINT = "http://localhost:4566"
REGION = "us-east-1"
@@ -37,6 +38,7 @@ def _uid():
# ---------------------------------------------------------------------------
def test_eks_create_describe_delete_cluster(eks):
+ """Test EKS API contract: create → describe → delete → gone."""
name = f"test-cluster-{_uid()}"
resp = eks.create_cluster(
name=name,
@@ -55,14 +57,21 @@ def test_eks_create_describe_delete_cluster(eks):
assert "identity" in cluster
assert "oidc" in cluster["identity"]
- # Describe — poll until background k3s startup completes
- for _ in range(30):
- resp = eks.describe_cluster(name=name)
- if resp["cluster"]["status"] == "ACTIVE":
- break
- time.sleep(1)
+ # Describe — wait for background thread to finish.
+ # In CI the first describe can transiently fail; retry with backoff.
+ resp = None
+ for attempt in range(60):
+ try:
+ resp = eks.describe_cluster(name=name)
+ if resp["cluster"]["status"] == "ACTIVE":
+ break
+ except ClientError as e:
+ if e.response["Error"]["Code"] != "ResourceNotFoundException":
+ raise
+ time.sleep(0.5)
+ assert resp is not None, f"Cluster {name} never became describable after 30s"
assert resp["cluster"]["name"] == name
- assert resp["cluster"]["status"] == "ACTIVE"
+ assert resp["cluster"]["status"] in ("ACTIVE", "CREATING")
# Delete
resp = eks.delete_cluster(name=name)
@@ -224,9 +233,7 @@ def test_eks_cfn_cluster(cfn, eks):
stack_name = f"eks-stack-{uid}"
cfn.create_stack(StackName=stack_name, TemplateBody=template)
- # Poll for stack creation — stack is stored synchronously but deploy
- # runs as an async task, so describe_stacks may briefly not find it
- # if the event loop hasn't yielded yet.
+ # Poll for stack — deploy runs as an async task
stack = None
for _ in range(30):
try:
diff --git a/tests/test_sfn.py b/tests/test_sfn.py
index 82d9740d..52669813 100644
--- a/tests/test_sfn.py
+++ b/tests/test_sfn.py
@@ -42,7 +42,7 @@ def test_sfn_create_execute(sfn):
exec_resp = sfn.start_execution(stateMachineArn=sm_arn, input=json.dumps({"key": "value"}))
exec_arn = exec_resp["executionArn"]
desc = sfn.describe_execution(executionArn=exec_arn)
- assert desc["status"] == "RUNNING"
+ assert desc["status"] in ("RUNNING", "SUCCEEDED")
def test_sfn_list(sfn):
machines = sfn.list_state_machines()