Light, fluffy, and always free - The AWS Local Emulator alternative
test(lambda): add tests for partial batch failure handling with filtered records (#3524)
Matheus Andréae7548a70cc2
diff --git a/src/test/java/io/github/hectorvent/floci/services/lambda/DynamoDbStreamsEventSourcePollerTest.java b/src/test/java/io/github/hectorvent/floci/services/lambda/DynamoDbStreamsEventSourcePollerTest.javaindex 44ab2ecf9d..5871070fa3 100644--- a/src/test/java/io/github/hectorvent/floci/services/lambda/DynamoDbStreamsEventSourcePollerTest.java+++ b/src/test/java/io/github/hectorvent/floci/services/lambda/DynamoDbStreamsEventSourcePollerTest.java@@ -417,6 +417,83 @@ class DynamoDbStreamsEventSourcePollerTest {assertEquals("s2", esm.getShardSequenceNumbers().get(DynamoDbStreamService.SHARD_ID));}+ // ──────────────────── FilterCriteria + partial batch failure ────────────────────++ /**+ * The retry boundary is computed against the full FETCHED batch, while the function only ever+ * sees the post-filter records. A filtered-out record sitting immediately before the lowest+ * failed record is therefore the checkpoint: it is consumed (never re-read), and the failed+ * record plus everything after it is retried.+ */+ @Test+ void partialFailureWithFilterCheckpointsAtFilteredOutRecordBeforeLowestFailure() {+ stubTrimHorizon(List.of(+ ddbRecord("s1", "INSERT", "{\"status\":{\"S\":\"active\"}}"),+ ddbRecord("s2", "INSERT", "{\"status\":{\"S\":\"inactive\"}}"), // filtered out+ ddbRecord("s3", "INSERT", "{\"status\":{\"S\":\"active\"}}"),+ ddbRecord("s4", "INSERT", "{\"status\":{\"S\":\"active\"}}")));+ InvokeResult result = new InvokeResult();+ result.setPayload("{\"batchItemFailures\":[{\"itemIdentifier\":\"s4\"},{\"itemIdentifier\":\"s3\"}]}".getBytes());+ when(executorService.invoke(any(), any(byte[].class), eq(InvocationType.RequestResponse)))+ .thenReturn(result);+ EsmStore store = mock(EsmStore.class);+ EventSourceMapping esm = filterEsm(PATTERN);+ esm.setFunctionResponseTypes(List.of("ReportBatchItemFailures"));++ pollerWith(store).pollAndInvoke(esm);++ ArgumentCaptor<byte[]> payload = ArgumentCaptor.forClass(byte[].class);+ verify(executorService, timeout(2000)).invoke(any(), payload.capture(), eq(InvocationType.RequestResponse));+ JsonNode records = readRecords(payload.getValue());+ assertEquals(3, records.size(), "only the post-filter records are delivered");+ assertEquals("s1", records.get(0).path("dynamodb").path("SequenceNumber").asText());+ assertEquals("s3", records.get(1).path("dynamodb").path("SequenceNumber").asText());+ assertEquals("s4", records.get(2).path("dynamodb").path("SequenceNumber").asText());++ verify(store, timeout(2000)).saveForAccount(eq(ACCOUNT_ID), any());+ // s2 was never sent to the function, yet it is the record immediately before the lowest+ // failure (s3) in the fetched batch — so it is the checkpoint: s1 and s2 are consumed and+ // the next poll resumes AFTER s2, re-delivering s3 and s4.+ assertEquals("s2", esm.getShardSequenceNumbers().get(DynamoDbStreamService.SHARD_ID));+ }++ /**+ * A reported identifier that names a record the filter removed is a record the function never+ * received. It cannot be a legitimate failure, so it is treated as malformed and the whole batch+ * is retried instead of silently advancing past it.+ */+ @Test+ void partialFailureReportingFilteredOutRecordRetriesWholeBatch() {+ stubTrimHorizon(List.of(+ ddbRecord("s1", "INSERT", "{\"status\":{\"S\":\"active\"}}"),+ ddbRecord("s2", "INSERT", "{\"status\":{\"S\":\"inactive\"}}"), // filtered out+ ddbRecord("s3", "INSERT", "{\"status\":{\"S\":\"active\"}}")));+ InvokeResult result = new InvokeResult();+ // s2 is in the fetched batch but was never delivered to the function.+ result.setPayload("{\"batchItemFailures\":[{\"itemIdentifier\":\"s2\"}]}".getBytes());+ when(executorService.invoke(any(), any(byte[].class), eq(InvocationType.RequestResponse)))+ .thenReturn(result);+ EsmStore store = mock(EsmStore.class);+ EventSourceMapping esm = filterEsm(PATTERN);+ esm.setFunctionResponseTypes(List.of("ReportBatchItemFailures"));+ esm.getShardSequenceNumbers().put(DynamoDbStreamService.SHARD_ID, "s0");+ DynamoDbStreamsEventSourcePoller p = pollerWith(store);++ p.pollAndInvoke(esm);+ ArgumentCaptor<byte[]> payload = ArgumentCaptor.forClass(byte[].class);+ verify(executorService, timeout(2000)).invoke(any(), payload.capture(), eq(InvocationType.RequestResponse));+ JsonNode records = readRecords(payload.getValue());+ assertEquals(2, records.size());+ assertEquals("s1", records.get(0).path("dynamodb").path("SequenceNumber").asText());+ assertEquals("s3", records.get(1).path("dynamodb").path("SequenceNumber").asText());+ awaitPollCompletedViaSecondFetch(p, esm);++ // Had the identifier been accepted, the checkpoint would have landed on s1 (the record+ // before s2). Instead the prior checkpoint is kept and the entire window is retried.+ verify(store, never()).saveForAccount(anyString(), any());+ assertEquals("s0", esm.getShardSequenceNumbers().get(DynamoDbStreamService.SHARD_ID));+ }+/*** A checkpoint that has aged out of the retained window names a cursor that can never succeed.* Retrying it wedges the ESM for good: later writes keep reaching the stream and none is ever