Skip to content

fix: report closed failed-stage cause signals #351

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Pigbibi merged 1 commit into from
Oct 2, 2026
Merged

fix: report closed failed-stage cause signals #351

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
87 changes: 81 additions & 6 deletions scripts/diagnose_cached_stage_failure.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,10 @@
"port",
"startup",
"missing_secret",
"update_mask",
"environment",
"network",
"quota",
)
_REASON_PATTERNS = {
"act_as": re.compile(r"iam\.serviceaccounts\.actas|actas|service account user", re.I),
Expand All @@ -39,6 +43,41 @@
"port": re.compile(r"port[^\n]*(listen|start|bind|set)|listen on the port", re.I),
"startup": re.compile(r"startup probe|container failed to start|startup[^\n]*failed", re.I),
"permission": re.compile(r"permission denied|permission_denied|not authorized|forbidden|\b403\b", re.I),
"update_mask": re.compile(r"update[ _-]?mask|field mask", re.I),
"environment": re.compile(r"environment variable|environment configuration|env var", re.I),
"network": re.compile(r"\bvpc\b|network|subnet|connector", re.I),
"quota": re.compile(r"quota|resource[_ ]exhausted|limit exceeded", re.I),
}
_KNOWN_RPC_STATUS_CODES = frozenset(range(17))
_AUDIT_MESSAGE_TERMS = (
"container",
"name",
"image",
"traffic",
"tag",
"revision",
"env",
"secret",
"update_mask",
"port",
"vpc",
"quota",
"immutable",
)
_AUDIT_MESSAGE_TERM_PATTERNS = {
"container": re.compile(r"\bcontainers?\b", re.I),
"name": re.compile(r"\bname\b", re.I),
"image": re.compile(r"\bimages?\b", re.I),
"traffic": re.compile(r"\btraffic\b", re.I),
"tag": re.compile(r"\btags?\b", re.I),
"revision": re.compile(r"\brevisions?\b", re.I),
"env": re.compile(r"\benv(?:ironment)?\b", re.I),
"secret": re.compile(r"\bsecret(?:keyref)?\b", re.I),
"update_mask": _REASON_PATTERNS["update_mask"],
"port": re.compile(r"\bports?\b", re.I),
"vpc": re.compile(r"\bvpc\b", re.I),
"quota": _REASON_PATTERNS["quota"],
"immutable": re.compile(r"\bimmutab(?:le|ility)\b", re.I),
}
_COMBINED_TAG_NAME_LIMIT = re.compile(
r"traffic[^\n]*tag[^\n]*too long|combined traffic tag and service name cannot exceed 46",
Expand Down Expand Up @@ -116,6 +155,24 @@ def _condition_error_texts(resource: Any) -> list[str]:
return texts


def _condition_error_status_codes(resource: Any) -> list[int]:
status = resource.get("status") if isinstance(resource, dict) else None
conditions = status.get("conditions") if isinstance(status, dict) else None
if not isinstance(conditions, list):
return []
codes = set()
for condition in conditions:
if not isinstance(condition, dict):
continue
state = condition.get("state", condition.get("status"))
if state not in ("CONDITION_FAILED", "False", False):
continue
code = condition.get("code")
if isinstance(code, int) and not isinstance(code, bool) and code in _KNOWN_RPC_STATUS_CODES:
codes.add(code)
return sorted(codes)


def _audit_identity_matches(
entry: dict[str, Any], *, project: str, region: str, service: str
) -> bool:
Expand Down Expand Up @@ -149,20 +206,20 @@ def _audit_error_texts(
service: str,
window_start: str,
window_end: str,
) -> tuple[int | None, int | None, list[str]]:
) -> tuple[int | None, int | None, list[str], list[int] | None]:
if not isinstance(entries, list):
return None, None, []
return None, None, [], None
try:
start_time = datetime.fromisoformat(window_start.replace("Z", "+00:00"))
end_time = datetime.fromisoformat(window_end.replace("Z", "+00:00"))
except (TypeError, ValueError):
return None, None, []
return None, None, [], None
if (
start_time.tzinfo is None
or end_time.tzinfo is None
or end_time < start_time
):
return None, None, []
return None, None, [], None
failed = []
matching_count = 0
for entry in entries:
Expand Down Expand Up @@ -199,7 +256,11 @@ def _audit_error_texts(
texts = [message for _, message in failed if message]
if any(code == 7 for code, _ in failed):
texts.append("permission_denied")
return matching_count, len(failed), texts
status_codes = sorted({
code for code, _ in failed
if isinstance(code, int) and not isinstance(code, bool) and code in _KNOWN_RPC_STATUS_CODES
})
return matching_count, len(failed), texts, status_codes


def _control_summary(policy: Any, jobs: Any) -> dict[str, Any]:
Expand Down Expand Up @@ -234,6 +295,13 @@ def _control_summary(policy: Any, jobs: Any) -> dict[str, Any]:
}


def _audit_message_terms(texts: list[str], *, available: bool) -> dict[str, bool | None]:
if not available:
return {term: None for term in _AUDIT_MESSAGE_TERMS}
joined = "\n".join(texts)
return {term: bool(_AUDIT_MESSAGE_TERM_PATTERNS[term].search(joined)) for term in _AUDIT_MESSAGE_TERMS}


def summarize(
*,
service: Any,
Expand Down Expand Up @@ -277,7 +345,7 @@ def summarize(
if latest_ready is True
else ["unknown"]
)
audit_entries_count, audit_error_count, audit_texts = _audit_error_texts(
audit_entries_count, audit_error_count, audit_texts, audit_status_codes = _audit_error_texts(
audit_entries,
project=expected_project,
region=expected_region,
Expand All @@ -301,6 +369,7 @@ def summarize(
failure_texts = []
combined_tag_name_error = _has_combined_tag_name_limit(failure_texts)
target_matches = bool(isinstance(metadata, dict) and metadata.get("name") == expected_service)
revision_status_codes = _condition_error_status_codes(latest)
tag_budget_ok = bool(
target_matches
and isinstance(metadata, dict)
Expand All @@ -323,11 +392,17 @@ def summarize(
and all(isinstance(item, dict) for item in revisions),
"latest_created_revision_ready": latest_ready,
"latest_revision_error_categories": revision_categories,
"latest_revision_error_status_codes": revision_status_codes,
"audit_error_message_terms": _audit_message_terms(
audit_texts,
available=audit_status == "ok" and audit_entries_count is not None,
),
**_control_summary(policy, jobs),
"audit_query_status": audit_status,
"audit_entry_count": audit_entries_count if audit_status == "ok" else None,
"audit_error_count": audit_error_count if audit_status == "ok" else None,
"audit_error_categories": audit_categories,
"audit_error_status_codes": audit_status_codes if audit_status == "ok" else None,
"failure_source": failure_source,
"failure_categories": failure_categories,
}
Expand Down
122 changes: 122 additions & 0 deletions tests/test_cached_stage_failure_diagnostic.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -99,6 +99,22 @@ def test_summary_exposes_only_closed_statuses_counts_and_categories():
"revision_list_readable": True,
"latest_created_revision_ready": False,
"latest_revision_error_categories": ["image"],
"latest_revision_error_status_codes": [],
"audit_error_message_terms": {
"container": False,
"name": False,
"image": True,
"traffic": False,
"tag": False,
"revision": False,
"env": False,
"secret": True,
"update_mask": False,
"port": False,
"vpc": False,
"quota": False,
"immutable": False,
},
"iam_policy_readable": True,
"iam_binding_count": 1,
"scheduler_readable": True,
Expand All @@ -108,6 +124,7 @@ def test_summary_exposes_only_closed_statuses_counts_and_categories():
"audit_entry_count": 1,
"audit_error_count": 1,
"audit_error_categories": ["image"],
"audit_error_status_codes": [13],
"failure_source": "audit",
"failure_categories": ["image"],
}
Expand All @@ -126,6 +143,10 @@ def test_summary_exposes_only_closed_statuses_counts_and_categories():
("container failed to listen on the port", ["port"]),
("startup probe failed", ["startup"]),
("secret version was not found", ["missing_secret"]),
("updateMask contains a private field", ["update_mask"]),
("environment variable is invalid", ["environment"]),
("VPC connector could not be resolved", ["network"]),
("RESOURCE_EXHAUSTED: quota exceeded", ["quota"]),
(
"traffic[].tag: traffic tag [TAG] and service name [SERVICE] together are too long. "
"Combined traffic tag and service name cannot exceed 46 characters.",
Expand Down Expand Up @@ -185,6 +206,107 @@ def test_combined_traffic_tag_length_is_closed_subcategory_and_budget_stays_priv
assert PRIVATE_SERVICE not in json.dumps(summary)


def test_status_codes_and_message_terms_are_numeric_and_closed():
message = (
"permission denied: updateMask for VPC connector private-vpc "
"failed on immutable revision private-revision-placeholder"
)
service, revisions, policy, jobs, audit = _evidence(message)
revisions[0]["status"]["conditions"][0]["code"] = 14
audit[0]["protoPayload"]["status"]["code"] = 7
summary = diagnostic.summarize(
service=service,
revisions=revisions,
expected_service=PRIVATE_SERVICE,
expected_project=PRIVATE_PROJECT,
expected_region=PRIVATE_REGION,
audit_window_start=WINDOW_START,
audit_window_end=WINDOW_END,
policy=policy,
jobs=jobs,
audit_entries=audit,
audit_status="ok",
)
assert summary["latest_revision_error_status_codes"] == [14]
assert summary["audit_error_status_codes"] == [7]
assert summary["failure_categories"] == ["permission", "update_mask", "network"]
assert summary["audit_error_message_terms"] == {
"container": False,
"name": False,
"image": False,
"traffic": False,
"tag": False,
"revision": True,
"env": False,
"secret": False,
"update_mask": True,
"port": False,
"vpc": True,
"quota": False,
"immutable": True,
}
serialized = json.dumps(summary)
for private_value in (
PRIVATE_SERVICE,
PRIVATE_NAME,
PRIVATE_MESSAGE,
"private-vpc",
"private-revision-placeholder",
"updateMask for VPC connector private-vpc",
):
assert private_value not in serialized


@pytest.mark.parametrize(
("message", "present"),
[
("multiple containers are unsupported on private-revision-placeholder", {"container", "revision"}),
("revision private-revision-placeholder is immutable", {"revision", "immutable"}),
("traffic target references tag private-tag on revision private-revision-placeholder", {"traffic", "tag", "revision"}),
("cannot update container name or image private-image-reference", {"container", "name", "image"}),
],
)
def test_audit_message_terms_are_closed_and_private_values_are_not_returned(message, present):
service, revisions, policy, jobs, audit = _evidence(message)
summary = diagnostic.summarize(
service=service,
revisions=revisions,
expected_service=PRIVATE_SERVICE,
expected_project=PRIVATE_PROJECT,
expected_region=PRIVATE_REGION,
audit_window_start=WINDOW_START,
audit_window_end=WINDOW_END,
policy=policy,
jobs=jobs,
audit_entries=audit,
audit_status="ok",
)
terms = summary["audit_error_message_terms"]
assert {term for term, value in terms.items() if value} == present
assert all(type(value) is bool for value in terms.values())
for private_value in (PRIVATE_SERVICE, PRIVATE_NAME, PRIVATE_MESSAGE, "private-image-reference", "private-tag"):
assert private_value not in json.dumps(summary)


def test_audit_message_terms_ignore_events_outside_target():
service, revisions, policy, jobs, audit = _evidence()
summary = diagnostic.summarize(
service=service,
revisions=revisions,
expected_service="different-service-placeholder",
expected_project=PRIVATE_PROJECT,
expected_region=PRIVATE_REGION,
audit_window_start=WINDOW_START,
audit_window_end=WINDOW_END,
policy=policy,
jobs=jobs,
audit_entries=audit,
audit_status="ok",
)
assert summary["target_matches"] is False
assert not any(summary["audit_error_message_terms"].values())


def test_tag_budget_false_is_reported_as_boolean_without_service_name():
service, revisions, policy, jobs, audit = _evidence()
long_name = "s" * 45
Expand Down
Loading