monitoring(testing): memoize gate check health

scripts/dashboards_render_atlas.py

@@ -553,6 +553,9 @@ PLATFORM_TEST_SUITE_NAMES = [
 ]
 PLATFORM_TEST_SUCCESS_STATUS = "ok|passed|success"
 PLATFORM_TEST_NON_FAILURE_STATUS = f"{PLATFORM_TEST_SUCCESS_STATUS}|not_applicable|skipped|na|n/a"
+PLATFORM_TEST_STANDARD_CHECK_REGEX = (
+    "tests|coverage|loc|style|docs_naming|gate_glue|sonarqube|supply_chain"
+)
 PLATFORM_TEST_CI_JOB = "platform-quality-ci"
 PLATFORM_TEST_EXPORT_FILTER = f'exported_job="{PLATFORM_TEST_CI_JOB}"'
 PLATFORM_TEST_SUITE_VALUE_BY_NAME = {
@@ -698,6 +701,22 @@ def platform_check_status_expr(
     )
 
 
+def platform_check_rollup_status_expr(
+    suite_matcher: str,
+    *,
+    branch_matcher: str = 'branch!=""',
+    check_matcher: str = 'check!=""',
+    status_matcher: str = 'status!=""',
+) -> str:
+    """Return memoized hourly check gauges normalized by vmalert."""
+    return (
+        "sum by (suite, branch, check, status) ("
+        f"platform_quality:check_status:present_1h{{{suite_matcher},{branch_matcher},"
+        f"{check_matcher},{status_matcher}}}"
+        ")"
+    )
+
+
 PLATFORM_TEST_SUCCESS_RATE_SUITE_TARGETS = [
     {
         "refId": chr(ord("A") + index),
@@ -752,25 +771,25 @@ PLATFORM_TEST_CHECK_ROLLUP_MATCHERS = (
     f'suite=~"{PLATFORM_TEST_SUITE_CANONICAL_MATCHER}"'
 )
 PLATFORM_TEST_PRIMARY_BRANCH_MATCHER = 'branch=~"main|master|origin/main|origin/master"'
-PLATFORM_TEST_CHECK_ROLLUP_SELECTOR = platform_check_status_expr(
+PLATFORM_TEST_CHECK_ROLLUP_SELECTOR = platform_check_rollup_status_expr(
     PLATFORM_TEST_CHECK_ROLLUP_MATCHERS,
     branch_matcher=PLATFORM_TEST_PRIMARY_BRANCH_MATCHER,
-    window=None,
+    check_matcher=f'check=~"{PLATFORM_TEST_STANDARD_CHECK_REGEX}"',
 )
 PLATFORM_TEST_CHECK_ROLLUP_OK_SELECTOR = (
-    platform_check_status_expr(
+    platform_check_rollup_status_expr(
         PLATFORM_TEST_CHECK_ROLLUP_MATCHERS,
         branch_matcher=PLATFORM_TEST_PRIMARY_BRANCH_MATCHER,
+        check_matcher=f'check=~"{PLATFORM_TEST_STANDARD_CHECK_REGEX}"',
         status_matcher=f'status=~"{PLATFORM_TEST_NON_FAILURE_STATUS}"',
-        window=None,
     )
 )
 PLATFORM_TEST_CHECK_ROLLUP_FAILED_SELECTOR = (
-    platform_check_status_expr(
+    platform_check_rollup_status_expr(
         PLATFORM_TEST_CHECK_ROLLUP_MATCHERS,
         branch_matcher=PLATFORM_TEST_PRIMARY_BRANCH_MATCHER,
+        check_matcher=f'check=~"{PLATFORM_TEST_STANDARD_CHECK_REGEX}"',
         status_matcher=f'status!~"{PLATFORM_TEST_NON_FAILURE_STATUS}"',
-        window=None,
     )
 )
 PLATFORM_TEST_CHECK_ROLLUP_SEEN_FLAGS = (
@@ -3856,12 +3875,11 @@ def build_jobs_dashboard():
         check_matcher: str = 'check!=""',
         status_matcher: str = 'status!=""',
     ) -> str:
-        selector = platform_check_status_expr(
+        selector = platform_check_rollup_status_expr(
             f'suite=~"{suite_var}"',
             branch_matcher=f'branch!="",branch=~"{branch_var}"',
             check_matcher=check_matcher,
             status_matcher=status_matcher,
-            window=None,
         )
         return selector
 
@@ -3905,8 +3923,9 @@ def build_jobs_dashboard():
         f'or on(suite) ({selected_suite_missing})'
     )
     non_failure = PLATFORM_TEST_NON_FAILURE_STATUS
-    current_gate_seen_vector = _latest_check_flags()
-    current_gate_ok_vector = _latest_healthy_check_flags()
+    standard_check_matcher = f'check=~"{PLATFORM_TEST_STANDARD_CHECK_REGEX}"'
+    current_gate_seen_vector = _latest_check_flags(standard_check_matcher)
+    current_gate_ok_vector = _latest_healthy_check_flags(standard_check_matcher)
     current_gate_ok_checks = (
         f"sum by (suite) ({current_gate_ok_vector})"
     )