fix(cloudwatch): publish SNS actions on alarm state transitions (#5918)

(cherry picked from commit f5811c3a0735be634b6122503237330826e0ba4a)
2026-10-18 10:15:34 -05:00
parent ccdf32335d
commit 2e0a438d70
4 changed files with 266 additions and 1 deletions
@@ -0,0 +1,127 @@
+package io.github.hectorvent.floci.services.cloudwatch.metrics;
+
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import io.github.hectorvent.floci.core.common.RegionResolver;
+import io.github.hectorvent.floci.core.storage.InMemoryStorage;
+import io.github.hectorvent.floci.services.cloudwatch.metrics.model.MetricAlarm;
+import io.github.hectorvent.floci.services.sns.SnsService;
+import org.junit.jupiter.api.BeforeEach;
+import org.junit.jupiter.api.Test;
+import org.mockito.ArgumentCaptor;
+
+import java.time.Instant;
+import java.util.List;
+
+import static org.junit.jupiter.api.Assertions.*;
+import static org.mockito.ArgumentMatchers.*;
+import static org.mockito.Mockito.*;
+
+class AlarmSnsNotifierTest {
+    private static final String REGION = "us-east-1";
+    private static final String TOPIC = "arn:aws:sns:us-east-1:000000000000:alarms";
+    private final ObjectMapper mapper = new ObjectMapper();
+    private SnsService sns;
+    private CloudWatchMetricsService service;
+    private MetricAlarm alarm;
+
+    @BeforeEach
+    void setUp() {
+        sns = mock(SnsService.class);
+        service = new CloudWatchMetricsService(new InMemoryStorage<>(), new InMemoryStorage<>(),
+                new RegionResolver(REGION, "000000000000"), new AlarmSnsNotifier(sns, mapper));
+        alarm = new MetricAlarm();
+        alarm.setAlarmName("test-alarm");
+        alarm.setActionsEnabled(true);
+        alarm.setAlarmActions(List.of(TOPIC));
+        alarm.setOkActions(List.of(TOPIC));
+        alarm.setInsufficientDataActions(List.of(TOPIC));
+        alarm.setEvaluationPeriods(3);
+        alarm.setStatistic("Average");
+        alarm.setMetricName("Errors");
+        alarm.setNamespace("Local/PoC");
+        service.putMetricAlarm(alarm, REGION);
+    }
+
+    @Test
+    void publishesAwsPayloadForEachTransitionButNotRepeatedState() throws Exception {
+        service.setAlarmState("test-alarm", "ALARM", "breached", null, REGION);
+        service.setAlarmState("test-alarm", "ALARM", "still breached", null, REGION);
+        service.setAlarmState("test-alarm", "OK", "recovered", null, REGION);
+        service.setAlarmState("test-alarm", "INSUFFICIENT_DATA", "missing", null, REGION);
+        ArgumentCaptor<String> messages = ArgumentCaptor.forClass(String.class);
+        verify(sns, times(3)).publish(eq(TOPIC), isNull(), messages.capture(), anyString(), eq(REGION));
+        JsonNode first = mapper.readTree(messages.getAllValues().getFirst());
+        assertEquals("test-alarm", first.path("AlarmName").asText());
+        assertEquals("INSUFFICIENT_DATA", first.path("OldStateValue").asText());
+        assertEquals("ALARM", first.path("NewStateValue").asText());
+        assertEquals("000000000000", first.path("AWSAccountId").asText());
+        assertEquals(alarm.getAlarmArn(), first.path("AlarmArn").asText());
+        assertEquals("Errors", first.path("Trigger").path("MetricName").asText());
+        assertFalse(first.path("Trigger").has("DatapointsToAlarm"));
+        assertEquals("AVERAGE", first.path("Trigger").path("Statistic").asText());
+        assertEquals("", first.path("Trigger").path("EvaluateLowSampleCountPercentile").asText());
+        assertTrue(first.path("Trigger").path("EvaluateLowSampleCountPercentile").isTextual());
+        assertNull(alarm.getDatapointsToAlarm());
+        assertEquals("OK", mapper.readTree(messages.getAllValues().get(2)).path("OldStateValue").asText());
+    }
+
+    @Test
+    void formatsBothTimestampsWithMillisAndUtcOffset() throws Exception {
+        alarm.setAlarmConfigurationUpdatedTimestamp(1716280510L);
+        service.setAlarmState("test-alarm", "ALARM", "breached", null, REGION);
+        ArgumentCaptor<String> message = ArgumentCaptor.forClass(String.class);
+        verify(sns).publish(eq(TOPIC), isNull(), message.capture(), anyString(), eq(REGION));
+        JsonNode payload = mapper.readTree(message.getValue());
+        assertEquals("2024-05-21T08:35:10.000+0000",
+                payload.path("AlarmConfigurationUpdatedTimestamp").asText());
+        String stateChangeTime = payload.path("StateChangeTime").asText();
+        assertTrue(stateChangeTime
+                .matches("[0-9]{4}-[0-9]{2}-[0-9]{2}T[0-9]{2}:[0-9]{2}:[0-9]{2}\\.[0-9]{3}\\+0000"));
+        assertEquals(alarm.getStateUpdatedTimestamp(),
+                Instant.parse(stateChangeTime.replace("+0000", "Z")).getEpochSecond());
+    }
+
+    @Test
+    void preservesExplicitDatapointsToAlarmInPayload() throws Exception {
+        alarm.setDatapointsToAlarm(2);
+        alarm.setStatistic("Maximum");
+        alarm.setEvaluateLowSampleCountPercentile("ignore");
+        service.setAlarmState("test-alarm", "ALARM", "breached", null, REGION);
+        ArgumentCaptor<String> message = ArgumentCaptor.forClass(String.class);
+        verify(sns).publish(eq(TOPIC), isNull(), message.capture(), anyString(), eq(REGION));
+        assertEquals(2, mapper.readTree(message.getValue()).path("Trigger").path("DatapointsToAlarm").intValue());
+        JsonNode trigger = mapper.readTree(message.getValue()).path("Trigger");
+        assertEquals("MAXIMUM", trigger.path("Statistic").asText());
+        assertEquals("ignore", trigger.path("EvaluateLowSampleCountPercentile").asText());
+        assertEquals(2, alarm.getDatapointsToAlarm());
+    }
+
+    @Test
+    void disabledActionsDoNotPublish() {
+        alarm.setActionsEnabled(false);
+        service.setAlarmState("test-alarm", "ALARM", "breached", null, REGION);
+        verifyNoInteractions(sns);
+        assertEquals("ALARM", alarm.getStateValue());
+    }
+
+    @Test
+    void failedTargetDoesNotPreventOtherTargetsOrStateChange() {
+        String missing = "arn:aws:sns:us-east-1:000000000000:missing";
+        alarm.setAlarmActions(List.of(missing, "arn:aws:autoscaling:us-east-1:000000000000:scalingPolicy:x", TOPIC));
+        doThrow(new IllegalStateException("missing topic"))
+                .when(sns).publish(eq(missing), isNull(), anyString(), anyString(), eq(REGION));
+        assertDoesNotThrow(() -> service.setAlarmState("test-alarm", "ALARM", "breached", null, REGION));
+        verify(sns).publish(eq(TOPIC), isNull(), anyString(), anyString(), eq(REGION));
+        assertEquals("ALARM", alarm.getStateValue());
+    }
+
+    @Test
+    void usesTopicRegionAndPreservesChinaPartition() {
+        String topic = "arn:aws-cn:sns:cn-north-1:000000000000:alarms";
+        alarm.setAlarmArn("arn:aws-cn:cloudwatch:cn-north-1:000000000000:alarm:test-alarm");
+        alarm.setAlarmActions(List.of(topic));
+        service.setAlarmState("test-alarm", "ALARM", "breached", null, REGION);
+        verify(sns).publish(eq(topic), isNull(), contains("arn:aws-cn:cloudwatch:"), anyString(), eq("cn-north-1"));
+    }
+}