Skip to content

armgit / Pulse

Public Imported from GitHub
0

Real-time monitoring dashboard for Proxmox VE, PBS, Docker, Kubernetes, TrueNAS and vSphere. Self-hosted, with alerts for failed backups, full disks, unhealthy containers and offline nodes.

Require successful npm audit completion before a clean verdict

pulse-triage[bot]056ada528228
diff --git a/docs/release-control/v6/internal/subsystems/deployment-installability.md b/docs/release-control/v6/internal/subsystems/deployment-installability.md
index 1cb24b2952..6a47430081 100644
--- a/docs/release-control/v6/internal/subsystems/deployment-installability.md
+++ b/docs/release-control/v6/internal/subsystems/deployment-installability.md
@@ -5033,8 +5033,13 @@ must never be introduced; only an unreachable endpoint is
retried. A nonempty package-finding map or any positive integer
summary count must fail immediately, even when the total is missing or
contradicts that evidence; a later response must not replace a known finding.
-A clean verdict requires all six known summary counts to be integer zero,
-no endpoint error and an absent or empty package-finding map. Booleans,
+A clean verdict requires the same audit command to exit zero, all six known
+summary counts to be integer zero, no endpoint error and an absent or empty
+package-finding map. A zero summary written before a nonzero exit or signal
+is unavailable evidence, not a clean audit. Positive findings still take
+precedence even when the command exits zero or is stopped; they must not be
+retried in search of a different verdict. Diagnostics retain the observed
+command status without attributing an unknown signal initiator. Booleans,
strings, nulls, negative counts and partial summaries are not zero-finding
evidence. Without positive evidence, malformed reports retain the existing
bounded outage policy, never a clean verdict. Summary diagnostics may print
@@ -5050,8 +5055,11 @@ it must not trigger another request. `scripts/tests/test_npm_audit_retry.py`
executes changed-second-response, missing-detail, escaping and single-request
fixtures, including production argument forwarding and findings accompanied by
transport errors, positive package/severity evidence with missing or zero
-totals, malformed zero summaries under both outage modes, and metadata
-annotation injection. Retrying is bounded by wall clock and not by attempt count alone,
+totals, malformed zero summaries under both outage modes, failed/signalled
+zero summaries, zero and positive reports written before a real watchdog stop,
+and metadata annotation injection. The shell smoke provider models a clean
+report with exit zero and findings with exit one, like npm's terminal contract.
+Retrying is bounded by wall clock and not by attempt count alone,
because npm's own `fetch-timeout` defaults to five minutes and it retries
internally: on 2026-09-04 three attempts against a hanging endpoint ran for
10m56s and cancelled the Frontend job at its own timeout with every test
diff --git a/scripts/npm-audit-retry.sh b/scripts/npm-audit-retry.sh
index c80d6529f8..1a8e15bec8 100755
--- a/scripts/npm-audit-retry.sh
+++ b/scripts/npm-audit-retry.sh
@@ -122,17 +122,15 @@ run_audit() {
kill -TERM "${killer_pid}" 2>/dev/null
wait "${killer_pid}" 2>/dev/null
- # 143 = SIGTERM, 137 = SIGKILL: the watchdog fired. The report is then
- # empty or truncated, which classify_report already reads as unreachable.
- if [ "${status}" -eq 143 ] || [ "${status}" -eq 137 ]; then
- return 124
- fi
- return 0
+ # Keep the observed command status. Even a complete-looking zero summary
+ # cannot establish a clean audit when npm failed or was stopped afterwards.
+ # Positive findings in that same response still take precedence.
+ return "${status}"
}
# Classify one audit run. Prints a verdict word, summary and (for a finding)
# allowlisted, JSON-escaped package/advisory details from that same response:
-# clean — audit completed, no vulnerabilities
+# clean — audit exited zero, complete report, no vulnerabilities
# vulnerable — audit completed, vulnerabilities present
# unreachable — npm could not get an answer from the advisory endpoint
classify_report() {
@@ -210,18 +208,20 @@ if has_findings or any(count > 0 for count in counts.values()):
sys.exit(0)
if (
- isinstance(report, dict) and not report.get("error")
+ sys.argv[1] == "0"
+ and isinstance(report, dict) and not report.get("error")
and len(counts) == len(count_names) and all(count == 0 for count in counts.values())
and ("vulnerabilities" not in report or isinstance(findings, dict))
):
print("clean")
print(" ".join(f"{name}=0" for name in count_names))
else:
- # A malformed/partial zero summary or endpoint error is not a clean verdict.
+ # A failed/stopped command, partial zero summary or endpoint error is not
+ # a clean verdict, even if npm wrote zero counts before it ended.
# With no positive finding, retain the existing bounded outage policy.
print("unreachable")
-'
+' "$1"
}
report_file="$(mktemp)"
@@ -245,10 +245,18 @@ while [ "${attempt}" -le "${ATTEMPTS}" ]; do
fi
echo "npm audit (${SCOPE}) attempt ${attempt}/${ATTEMPTS} (limit ${attempt_limit}s, ${remaining}s of budget left)"
- if ! run_audit "${attempt_limit}" "${report_file}"; then
- echo "npm audit (${SCOPE}): attempt ${attempt} exceeded ${attempt_limit}s and was stopped"
- fi
- verdict_output="$(classify_report <"${report_file}")"
+ audit_status=0
+ run_audit "${attempt_limit}" "${report_file}" || audit_status=$?
+ case "${audit_status}" in
+ 0) ;;
+ 143|137)
+ echo "npm audit (${SCOPE}): command was stopped (status ${audit_status}, attempt limit ${attempt_limit}s)"
+ ;;
+ *)
+ echo "npm audit (${SCOPE}): command exited with status ${audit_status}"
+ ;;
+ esac
+ verdict_output="$(classify_report "${audit_status}" <"${report_file}")"
verdict="$(printf '%s\n' "${verdict_output}" | head -1)"
summary="$(printf '%s\n' "${verdict_output}" | sed -n '2p')"
diff --git a/scripts/tests/test-npm-audit-retry.sh b/scripts/tests/test-npm-audit-retry.sh
index dc8706ed0e..860ab1d784 100755
--- a/scripts/tests/test-npm-audit-retry.sh
+++ b/scripts/tests/test-npm-audit-retry.sh
@@ -36,11 +36,16 @@ make_fake_npm() {
printf 'printf "%%s" "$n" > "$count_file"\n'
printf 'case "$n" in\n'
local i=1
+ local status
for payload in "$@"; do
- printf ' %d) cat <<'"'"'JSON'"'"'\n%s\nJSON\n ;;\n' "${i}" "${payload}"
+ status=1
+ if [ "${payload}" = "${CLEAN}" ]; then status=0; fi
+ printf ' %d) cat <<'"'"'JSON'"'"'\n%s\nJSON\n exit %d\n ;;\n' "${i}" "${payload}" "${status}"
i=$((i + 1))
done
- printf ' *) cat <<'"'"'JSON'"'"'\n%s\nJSON\n ;;\n' "${!#}"
+ status=1
+ if [ "${!#}" = "${CLEAN}" ]; then status=0; fi
+ printf ' *) cat <<'"'"'JSON'"'"'\n%s\nJSON\n exit %d\n ;;\n' "${!#}" "${status}"
printf 'esac\n'
printf 'exit 1\n'
} > "${path}"
diff --git a/scripts/tests/test_npm_audit_retry.py b/scripts/tests/test_npm_audit_retry.py
index 5f96e9699f..5b05d16f13 100755
--- a/scripts/tests/test_npm_audit_retry.py
+++ b/scripts/tests/test_npm_audit_retry.py
@@ -20,7 +20,8 @@ SCRIPT = ROOT / "scripts" / "npm-audit-retry.sh"
class NpmAuditRetryTest(unittest.TestCase):
def run_check(
- self, mode: str, *arguments: str, require: str = "true", report=None
+ self, mode: str, *arguments: str, require: str = "true", report=None,
+ attempts: int = 3, attempt_timeout: int = 60, max_seconds: int = 240,
):
with tempfile.TemporaryDirectory() as directory:
fake_bin = Path(directory)
@@ -47,6 +48,26 @@ class NpmAuditRetryTest(unittest.TestCase):
printf '%s\n' "$clean"
exit 0
;;
+ clean-failure)
+ printf '%s\n' "$clean"
+ exit 42
+ ;;
+ clean-signal)
+ printf '%s\n' "$clean"
+ kill -TERM "$$"
+ ;;
+ clean-hang)
+ printf '%s\n' "$clean"
+ exec sleep 300
+ ;;
+ vulnerability-hang)
+ printf '%s\n' "$vulnerable"
+ exec sleep 300
+ ;;
+ vulnerability-success)
+ printf '%s\n' "$vulnerable"
+ exit 0
+ ;;
transient-success)
if [ "$count" -eq 1 ]; then
printf '%s\n' "$unavailable"
@@ -81,6 +102,10 @@ class NpmAuditRetryTest(unittest.TestCase):
cat "$FAKE_NPM_REPORT"
exit 1
;;
+ captured-clean)
+ cat "$FAKE_NPM_REPORT"
+ exit 0
+ ;;
garbage)
echo 'not json'
exit 1
@@ -99,9 +124,9 @@ class NpmAuditRetryTest(unittest.TestCase):
"FAKE_NPM_COUNT": str(count),
"FAKE_NPM_MODE": mode,
"FAKE_NPM_REPORT": str(report_path),
- "NPM_AUDIT_ATTEMPTS": "3",
- "NPM_AUDIT_ATTEMPT_TIMEOUT": "60",
- "NPM_AUDIT_MAX_SECONDS": "240",
+ "NPM_AUDIT_ATTEMPTS": str(attempts),
+ "NPM_AUDIT_ATTEMPT_TIMEOUT": str(attempt_timeout),
+ "NPM_AUDIT_MAX_SECONDS": str(max_seconds),
"NPM_AUDIT_CMD": str(fake_npm),
"NPM_AUDIT_REQUIRE_RESULT": require,
"NPM_AUDIT_RETRY_DELAY": "0",
@@ -114,6 +139,7 @@ class NpmAuditRetryTest(unittest.TestCase):
text=True,
capture_output=True,
check=False,
+ timeout=30,
)
recorded_calls = (
calls.read_text(encoding="utf-8").splitlines()
@@ -179,6 +205,56 @@ class NpmAuditRetryTest(unittest.TestCase):
)
self.assertIn("retrying", result.stdout)
+ def test_zero_summary_from_a_failed_or_signalled_audit_is_not_clean(self) -> None:
+ for mode in ["clean-failure", "clean-signal"]:
+ for require in ["true", "false"]:
+ with self.subTest(mode=mode, require=require):
+ result, calls = self.run_check(mode, "all", require=require)
+ self.assertEqual(result.returncode, 1 if require == "true" else 0,
+ result.stdout)
+ self.assertEqual(calls, ["audit --json"] * 3, result.stdout)
+ self.assertIn("could not reach the advisory endpoint", result.stdout)
+ self.assertNotIn("no vulnerabilities", result.stdout)
+ self.assertNotIn("vulnerabilities present", result.stdout)
+ self.assertIn("::error::" if require == "true" else "::warning::",
+ result.stdout)
+ self.assertIn("status 42" if mode == "clean-failure" else "status 143",
+ result.stdout)
+
+ def test_zero_summary_written_before_timeout_is_not_clean(self) -> None:
+ for require in ["true", "false"]:
+ with self.subTest(require=require):
+ result, calls = self.run_check(
+ "clean-hang", "production", require=require,
+ attempts=1, attempt_timeout=1, max_seconds=5,
+ )
+ self.assertEqual(result.returncode, 1 if require == "true" else 0,
+ result.stdout)
+ self.assertEqual(calls, ["audit --json --omit=dev"], result.stdout)
+ self.assertIn("was stopped", result.stdout)
+ self.assertIn("could not reach the advisory endpoint", result.stdout)
+ self.assertNotIn("no vulnerabilities", result.stdout)
+ self.assertIn("::error::" if require == "true" else "::warning::",
+ result.stdout)
+
+ def test_positive_findings_survive_zero_exit_or_timeout_without_requery(self) -> None:
+ for mode in ["vulnerability-success", "vulnerability-hang"]:
+ for require in ["true", "false"]:
+ with self.subTest(mode=mode, require=require):
+ result, calls = self.run_check(
+ mode, "all", require=require,
+ attempts=3, attempt_timeout=1, max_seconds=5,
+ )
+ self.assertEqual(result.returncode, 1 if require == "true" else 0,
+ result.stdout)
+ self.assertEqual(calls, ["audit --json"], result.stdout)
+ self.assertIn("vulnerabilities present", result.stdout)
+ self.assertNotIn("no vulnerabilities", result.stdout)
+ self.assertNotIn("retrying", result.stdout)
+ self.assertNotIn("could not reach the advisory endpoint", result.stdout)
+ self.assertIn("::error::" if require == "true" else "::warning::",
+ result.stdout)
+
def test_does_not_retry_a_vulnerability_report(self) -> None:
result, calls = self.run_check("vulnerability", "all")
self.assertEqual(result.returncode, 1)
@@ -404,7 +480,7 @@ class NpmAuditRetryTest(unittest.TestCase):
},
"vulnerabilities": {},
}
- result, calls = self.run_check("captured-report", "all", report=report)
+ result, calls = self.run_check("captured-clean", "all", report=report)
self.assertEqual(result.returncode, 0, result.stdout)
self.assertEqual(calls, ["audit --json"])
self.assertIn("no vulnerabilities", result.stdout)