Skip to content

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

release v1.3.40 (#654)

Nahuel Nucera2caacfffbf0a
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 9e6368d4..306535a9 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -7,6 +7,20 @@ Versioning follows [Semantic Versioning](https://semver.org/).
---
+## [1.3.40] — 2026-05-15
+
+### Added
+- **Cognito invitation and verification emails via SES** — `AdminCreateUser`, `SignUp`, `ResendConfirmationCode`, `ForgotPassword`, and `AdminResetUserPassword` now hand their welcome / temporary-password / verification mail to the in-process SES emulator, so simulated apps see the message in `/_ministack/ses/messages` and it relays via SMTP when `SMTP_HOST` is set. Mirrors AWS behaviour: `MessageAction=SUPPRESS` skips, `RESEND` re-sends, `DesiredDeliveryMediums=["SMS"]` excludes email, and template placeholders (`{username}`, `{####}`) expand. Sender resolves to `EmailConfiguration.From`, falling back to `[email protected]` (overridable via `COGNITO_DEFAULT_FROM`); set `COGNITO_EMAIL_ENABLED=false` to short-circuit globally. Adds the previously-missing `ResendConfirmationCode` action while wiring the delivery path. Contributed by @kjdev.
+- **Step Functions JSONata `Assign` + workflow variables** — state-level `Assign` fields now bind values into an execution-scoped variable store, and later JSONata expressions can reference them as `$name` (with dotted-path access like `$user.email`). Pass `$states.result` resolves to the computed Output; Task `$states.result` to the raw API result; Catch handlers expose `$states.errorOutput`. Undefined references surface as `States.QueryEvaluationError`, matching the AWS error code for JSONata evaluation failures. Reported by @youngkwangk.
+
+### Fixed
+- **Cognito alias-attribute user lookup** — `_resolve_user` now honors the pool's `AliasAttributes` and `UsernameAttributes`, so signing in by `email` / `phone_number` / `preferred_username` resolves correctly. Email and phone aliases require the corresponding `_verified` attribute to equal `"true"` (matches AWS); `preferred_username` has no verification gate. The change routes `AdminInitiateAuth`, `InitiateAuth`, `AdminRespondToAuthChallenge`, `RespondToAuthChallenge`, `ConfirmSignUp`, `ForgotPassword`, `ConfirmForgotPassword`, and the hosted-UI `/login` form through the resolver — internal call sites that need the canonical username (group iteration, create-time uniqueness, post-code token issuance) are left untouched. Contributed by @rjmackay.
+- **SQS `ReceiveMessage` `InternalError` on FIFO queues with `RedrivePolicy`** — a double-JSON-encoded `RedrivePolicy` value slipped past `CreateQueue` / `SetQueueAttributes`, then crashed `_dlq_sweep` on receive because `json.loads` returned a string (not a dict) and `.get()` raised `AttributeError`. MiniStack now validates `RedrivePolicy` at intake (parseable JSON object with non-empty `deadLetterTargetArn` and numeric `maxReceiveCount` between 1 and 1000) and rejects malformed values with `InvalidAttributeValue` (400), matching real AWS. Receive carries a defensive guard so legacy persisted state doesn't crash. Reported by @rbonestell.
+- **DynamoDB `if_not_exists` arithmetic in `SET` expressions** — `SET v = (if_not_exists(v, :d) - :amt)` previously dropped the arithmetic and assigned the resolved value directly: the outer parens kept every token at depth > 0, so the top-level operator scan never saw the `-` at depth 0. `_eval_set_value` now strips a single layer of matched outer parens before parsing, guarded so `(a) + (b)` (two adjacent groups) isn't accidentally flattened. Reported by @youngkwangk.
+- **S3 → Lambda notifications fire for non-boto3 SDK clients** — MiniStack's notification XML parser only recognised the legacy `<CloudFunction>` ARN tag, which is what botocore wire-serialises `LambdaFunctionArn` as. AWS SDK for Java v2, Go SDK, Terraform's `aws_s3_bucket_notification`, and any hand-crafted XML send the modern `<LambdaFunctionArn>` tag — MiniStack silently dropped those configs, so uploads succeeded but the Lambda never fired. Both shapes are now accepted, matching real S3. Reported by @michael-denyer.
+
+---
+
## [1.3.39] — 2026-05-15
### Added
diff --git a/ministack/services/batch.py b/ministack/services/batch.py
index 5e9cb813..107c3400 100644
--- a/ministack/services/batch.py
+++ b/ministack/services/batch.py
@@ -10,7 +10,6 @@ here, not a real container runner.
import copy
import json
import logging
-import os
import time
from ministack.core.responses import (
@@ -185,123 +184,11 @@ def _describe_job_definitions(p):
return _json(200, {"jobDefinitions": revs})
-def _detect_host_limits():
- """Return the host's installed CPU/memory ceiling MS can compare against.
-
- Reports *installed* capacity, not currently-available — the gate's purpose
- is "could this jobdef ever run on this host", which is a stable property,
- not a fluctuating one. A dimension that can't be confidently measured on
- this platform returns ``None`` and the corresponding check is skipped, so
- we don't false-positive on exotic environments.
- """
- vcpu = None
- try:
- vcpu = os.cpu_count()
- except Exception:
- vcpu = None
-
- memory_mib = None
- try:
- page = os.sysconf("SC_PAGE_SIZE")
- pages = os.sysconf("SC_PHYS_PAGES")
- if isinstance(page, int) and isinstance(pages, int) and page > 0 and pages > 0:
- memory_mib = (page * pages) // (1024 * 1024)
- except (AttributeError, ValueError, OSError):
- memory_mib = None
- return {"vcpu": vcpu, "memory_mib": memory_mib}
-
-
-def _parse_resource_requirements(container_props):
- """Pull VCPU + MEMORY out of a jobdef's ``containerProperties.resourceRequirements``.
- Returns ``(vcpu_or_None, memory_mib_or_None)``. Malformed entries are
- treated as absent so we never block on parser quirks.
- """
- reqs = (container_props or {}).get("resourceRequirements") or []
- vcpu = None
- memory_mib = None
- for r in reqs:
- if not isinstance(r, dict):
- continue
- kind = (r.get("type") or "").upper()
- raw = r.get("value")
- if raw is None:
- continue
- try:
- num = float(raw)
- except (TypeError, ValueError):
- continue
- if kind == "VCPU":
- vcpu = num
- elif kind == "MEMORY":
- try:
- memory_mib = int(num)
- except (TypeError, ValueError):
- continue
- return vcpu, memory_mib
-
-
-def _resolve_jobdef(ref):
- """Look up a jobdef record by ``name``, ``name:revision``, or full ARN.
- Returns the record or ``None`` — never raises so callers can fall through.
- """
- if not ref:
- return None
- short = ref.split("/")[-1]
- if ":" in short:
- name, _, rev = short.rpartition(":")
- try:
- target_rev = int(rev)
- except ValueError:
- return None
- for r in _job_definitions.get(name, []):
- if r.get("revision") == target_rev:
- return r
- return None
- revs = _job_definitions.get(short, [])
- return revs[-1] if revs else None
-
-
-def _check_host_fit(container_props):
- """Return an error message if the jobdef can't physically run on this host,
- else ``None``. Both dimensions must be measurable on both sides for the
- check to fire — anything ambiguous passes through.
- """
- limits = _detect_host_limits()
- req_vcpu, req_mem = _parse_resource_requirements(container_props)
-
- over = []
- if req_vcpu is not None and limits["vcpu"] is not None and req_vcpu > limits["vcpu"]:
- over.append(f"VCPU={req_vcpu:g} (host has {limits['vcpu']})")
- if req_mem is not None and limits["memory_mib"] is not None and req_mem > limits["memory_mib"]:
- over.append(f"MEMORY={req_mem} MiB (host has {limits['memory_mib']} MiB)")
- if not over:
- return None
- return (
- "SubmitJob refused: jobdef requests "
- + ", ".join(over)
- + ". Reduce resourceRequirements for local testing, "
- "or run this jobdef against real AWS Batch."
- )
-
-
# ─── jobs ───────────────────────────────────────────────────────
def _submit_job(p):
name = p.get("jobName")
if not name:
return error_response_json("ClientException", "jobName is required", 400)
-
- # Honest fail-fast for the local-emulator case where the jobdef can't
- # physically fit on this host. Real AWS Batch has unlimited compute so
- # this failure mode doesn't exist there; in MS we'd otherwise return
- # SUCCEEDED for a job that could never actually run, hiding the problem
- # behind a second describe_jobs call. Issue #650.
- jobdef = _resolve_jobdef(p.get("jobDefinition"))
- if jobdef is not None:
- fit_err = _check_host_fit(jobdef.get("containerProperties"))
- if fit_err is not None:
- logger.warning("Batch SubmitJob refused (%s): %s", name, fit_err)
- return error_response_json("ClientException", fit_err, 400)
-
job_id = new_uuid()
now = _now_ms()
rec = {
diff --git a/pyproject.toml b/pyproject.toml
index d737d358..f411c29d 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
[project]
name = "ministack"
-version = "1.3.39"
+version = "1.3.40"
description = "Free, open-source local AWS emulator — drop-in LocalStack replacement"
readme = "README.md"
license = { file = "LICENSE" }
diff --git a/tests/test_batch.py b/tests/test_batch.py
index fbb3781a..efe8392d 100644
--- a/tests/test_batch.py
+++ b/tests/test_batch.py
@@ -98,156 +98,3 @@ def test_batch_account_isolation():
b_qs = [q["jobQueueName"] for q in b.describe_job_queues()["jobQueues"]]
assert name in a_qs
assert name not in b_qs
-
-
-# ---------------------------------------------------------------------------
-# SubmitJob host-resource gate (issue #650)
-# ---------------------------------------------------------------------------
-
-def _register(batch, name, reqs=None):
- container = {"image": "alpine:3.20", "command": ["echo", "hi"]}
- if reqs is not None:
- container["resourceRequirements"] = reqs
- return batch.register_job_definition(
- jobDefinitionName=name, type="container", containerProperties=container,
- )
-
-
-def test_batch_submit_job_refused_when_jobdef_exceeds_host(batch):
- """Issue #650: jobdef requesting more resources than MS' host can provide
- must surface at the SubmitJob call (400 ClientException) — not silently
- return SUCCEEDED for a job that could never run.
- """
- from botocore.exceptions import ClientError
- name = f"oversize-{_uid()}"
- _register(batch, name, reqs=[
- {"type": "VCPU", "value": "256"}, # impossibly high
- {"type": "MEMORY", "value": "9999999"},
- ])
- with pytest.raises(ClientError) as exc:
- batch.submit_job(jobName=f"r-{_uid()}", jobQueue="q",
- jobDefinition=f"{name}:1")
- err = exc.value.response["Error"]
- assert err["Code"] == "ClientException"
- msg = err["Message"]
- assert "SubmitJob refused" in msg
- assert "VCPU=256" in msg
- assert "MEMORY=9999999" in msg
-
-
-def test_batch_submit_job_vcpu_only_over_fails(batch):
- """When only VCPU exceeds, the message reports only VCPU."""
- from botocore.exceptions import ClientError
- name = f"vcpu-{_uid()}"
- _register(batch, name, reqs=[
- {"type": "VCPU", "value": "256"},
- {"type": "MEMORY", "value": "64"},
- ])
- with pytest.raises(ClientError) as exc:
- batch.submit_job(jobName=f"r-{_uid()}", jobQueue="q",
- jobDefinition=f"{name}:1")
- msg = exc.value.response["Error"]["Message"]
- assert "VCPU=256" in msg
- assert "MEMORY" not in msg # only the over-capacity dimension surfaces
-
-
-def test_batch_submit_job_memory_only_over_fails(batch):
- from botocore.exceptions import ClientError
- name = f"mem-{_uid()}"
- _register(batch, name, reqs=[
- {"type": "VCPU", "value": "1"},
- {"type": "MEMORY", "value": "9999999"},
- ])
- with pytest.raises(ClientError) as exc:
- batch.submit_job(jobName=f"r-{_uid()}", jobQueue="q",
- jobDefinition=f"{name}:1")
- msg = exc.value.response["Error"]["Message"]
- assert "MEMORY=9999999" in msg
- assert "VCPU" not in msg
-
-
-def test_batch_submit_job_fitting_resources_succeeds(batch):
- """Regression: a jobdef well within the host's ceiling must still pass
- through to the stub-SUCCEEDED behavior MS has always had.
- """
- name = f"fit-{_uid()}"
- _register(batch, name, reqs=[
- {"type": "VCPU", "value": "1"},
- {"type": "MEMORY", "value": "128"},
- ])
- resp = batch.submit_job(jobName=f"r-{_uid()}", jobQueue="q",
- jobDefinition=f"{name}:1")
- assert "jobId" in resp
- desc = batch.describe_jobs(jobs=[resp["jobId"]])["jobs"][0]
- assert desc["status"] == "SUCCEEDED"
-
-
-def test_batch_submit_job_no_resource_requirements_succeeds(batch):
- """A jobdef without `resourceRequirements` has nothing to compare against;
- pass through unchanged.
- """
- name = f"noreqs-{_uid()}"
- _register(batch, name, reqs=None)
- resp = batch.submit_job(jobName=f"r-{_uid()}", jobQueue="q",
- jobDefinition=f"{name}:1")
- assert "jobId" in resp
-
-
-def test_batch_submit_job_unknown_jobdef_passes_through(batch):
- """When the `jobDefinition` reference can't be resolved, MS preserves its
- existing permissive behavior (don't introduce a *new* failure mode while
- fixing a different one).
- """
- resp = batch.submit_job(jobName=f"r-{_uid()}", jobQueue="q",
- jobDefinition="does-not-exist:99")
- assert "jobId" in resp
-
-
-def test_batch_submit_job_resolves_by_name_takes_latest_revision(batch):
- """`jobDefinition='name'` (no `:rev`) resolves to the latest revision."""
- name = f"byname-{_uid()}"
- _register(batch, name, reqs=[{"type": "VCPU", "value": "1"},
- {"type": "MEMORY", "value": "128"}])
- # Bump revision; the latest one is still small, gate must still pass.
- _register(batch, name, reqs=[{"type": "VCPU", "value": "1"},
- {"type": "MEMORY", "value": "64"}])
- resp = batch.submit_job(jobName=f"r-{_uid()}", jobQueue="q",
- jobDefinition=name) # no :rev
- assert "jobId" in resp
-
-
-def test_batch_resource_check_skipped_when_host_unmeasurable(monkeypatch):
- """If host detection returns None for every dimension (exotic platform),
- the gate must skip silently — no false-positive refusals.
- """
- from ministack.services import batch as batch_mod
- monkeypatch.setattr(
- batch_mod, "_detect_host_limits",
- lambda: {"vcpu": None, "memory_mib": None},
- )
- container = {
- "image": "alpine:3.20",
- "command": ["echo", "hi"],
- "resourceRequirements": [
- {"type": "VCPU", "value": "9999"},
- {"type": "MEMORY", "value": "9999999"},
- ],
- }
- assert batch_mod._check_host_fit(container) is None
-
-
-def test_batch_resource_check_skipped_when_jobdef_dimension_unmeasurable(monkeypatch):
- """If the jobdef declares one dimension but not the other, only the
- declared one is compared — no spurious failures from absent fields.
- """
- from ministack.services import batch as batch_mod
- monkeypatch.setattr(
- batch_mod, "_detect_host_limits",
- lambda: {"vcpu": 2, "memory_mib": 1024},
- )
- container = {
- "image": "alpine:3.20",
- "command": ["echo", "hi"],
- "resourceRequirements": [{"type": "VCPU", "value": "1"}], # only VCPU
- }
- assert batch_mod._check_host_fit(container) is None