Refactor LuceneFailreTest to use the DataModel, not LuceneTestBase #3010

ohadzeliger · 2024-12-20T22:31:45Z

No description provided.

…estBase

foundationdb-ci · 2024-12-20T23:08:20Z

Result of fdb-record-layer-pr on Linux CentOS 7

Commit ID: b591aa5
Duration 0:36:26
Result: ❌ FAILED
Error: Error while executing command: ./gradlew --no-daemon --console=plain -b ./build.gradle build destructiveTest -PcoreNotStrict -PreleaseBuild=false -PpublishBuild=false -PspotbugsEnableHtmlReport. Reason: exit status 1
Build Log terminal output (available for 30 days)
Build Workspace zip file of the working directory (available for 30 days)

foundationdb-ci · 2025-01-02T18:57:16Z

Result of fdb-record-layer-pr on Linux CentOS 7

Commit ID: e3cdd50
Duration 0:47:27
Result: ✅ SUCCEEDED
Error: N/A
Build Log terminal output (available for 30 days)
Build Workspace zip file of the working directory (available for 30 days)

ScottDugas · 2025-01-03T20:54:53Z

...yer-lucene/src/test/java/com/apple/foundationdb/record/lucene/FDBLuceneIndexFailureTest.java

-            recordStore.saveRecord(createComplexDocument(8888L, WAYLON, 2, Instant.now().plus(1, ChronoUnit.DAYS).toEpochMilli()));
-            recordStore.saveRecord(createComplexDocument(9999L, "hello world!", 1, Instant.now().plus(2, ChronoUnit.DAYS).toEpochMilli()));
+            injectedFailures.addFailure(LUCENE_GET_FILE_REFERENCE_CACHE_ASYNC,
+                    new FDBExceptions.FDBStoreTransactionIsTooOldException("Blah", new FDBException("Blah", 7)),


Why change the exception being thrown?

The new method replaces two test methods: One with AsyncToSyncTimeoutException and one with DBStoreTransactionIsTooOldException. The new test method has an additional parameter (isGrouped) and throws DBStoreTransactionIsTooOldException.

ScottDugas · 2025-01-03T20:58:08Z

...yer-lucene/src/test/java/com/apple/foundationdb/record/lucene/FDBLuceneIndexFailureTest.java

        try (FDBRecordContext context = openContext(contextProps)) {
-            rebuildIndexMetaData(context, COMPLEX_DOC, COMPLEX_PARTITIONED_NOGROUP);
-            final LuceneScanBounds scanBounds = fullTextSearch(COMPLEX_PARTITIONED_NOGROUP, "text:propose");
-            injectedFailures.addFailure(LUCENE_GET_FDB_LUCENE_FILE_REFERENCE_ASYNC,


Before this would inject the failure before saving any records, but the injected failure wouldn't fire until it did the search, is that because the fullTextSearch above would have already loaded the file reference, and it wouldn't need to do so again until it got to the search below?
(Just trying to make sure I understand what's going on, I think it's better to start injecting after the save)

Before, there were two test methods (one for Grouped and one for Ungrouped search). I don't remember exactly why the injection was added at the top and I believe your reasoning is correct - the bind call for the creation of the scan bounds loaded the cache.
Once refactored the code and consolidated the two method, it made sense to move the injection below the save as it offers more direct relationship to the action action being carried out.

ScottDugas · 2025-01-03T21:01:09Z

...yer-lucene/src/test/java/com/apple/foundationdb/record/lucene/FDBLuceneIndexFailureTest.java


        // create/save documents
+        long docGroupFieldValue = groupingKey.isEmpty() ? 0L : groupingKey.getLong(0);


groupingKey is always Tuple.from(1L), would it make sense to reorder the assignments:

long docGroupFieldValue = 1L; Tuple groupingKey = Tuple.from(docGroupFieldValue);

ScottDugas · 2025-01-03T21:02:26Z

...yer-lucene/src/test/java/com/apple/foundationdb/record/lucene/FDBLuceneIndexFailureTest.java

-        long docGroupFieldValue = groupingKey.isEmpty() ? 0L : groupingKey.getLong(0);
+        final LuceneIndexTestDataModel dataModel = new LuceneIndexTestDataModel.Builder(seed, this::getStoreBuilderWithRegistry, pathManager)
+                .setIsGrouped(true)
+                .setPartitionHighWatermark(10)


It might be worth extracting this into a variable, and making totalDocCount a function of it, e.g.:

final int partitionHighWatermark = 10; final int totalDocCount = partitionHighWatermark * 2;

ScottDugas · 2025-01-03T21:10:04Z

...yer-lucene/src/test/java/com/apple/foundationdb/record/lucene/FDBLuceneIndexFailureTest.java

@@ -273,106 +272,122 @@ void repartitionAndMerge(boolean useLegacyAsyncToSync) throws IOException {
                segmentCounts);

        try (FDBRecordContext context = openContext(contextProps)) {


You could probably just call:

dataModel.validate(() -> openContext(contextProps));

It will assert that the documents align, and that the partitions are all an acceptable size. It is less strict than the assertions here, but it is probably less brittle in the long run.

Do you mean to replace the entire try block with both validateDocsInPartition in it? In that case, it would probably also make sense to change the original in LuceneIndexTest?

ScottDugas · 2025-01-03T21:11:12Z

...yer-lucene/src/test/java/com/apple/foundationdb/record/lucene/FDBLuceneIndexFailureTest.java

+
+        final LuceneIndexTestDataModel dataModel = new LuceneIndexTestDataModel.Builder(seed, this::getStoreBuilderWithRegistry, pathManager)
+                .setIsGrouped(true)
+                .setPartitionHighWatermark(10)


Based on the comment above, should this be:

Suggested change

.setPartitionHighWatermark(10)

.setPartitionHighWatermark(totalDocCount)

ScottDugas · 2025-01-03T21:12:21Z

...yer-lucene/src/test/java/com/apple/foundationdb/record/lucene/FDBLuceneIndexFailureTest.java

+                .setPartitionHighWatermark(10)
+                .build();
+
+        int docGroupFieldValue = groupingKey.isEmpty() ? 0 : (int)groupingKey.getLong(0);


Here, also, would it be worth rearranging so that the groupingKey is derived from docGroupFieldValue

ScottDugas · 2025-01-03T21:15:02Z

...yer-lucene/src/test/java/com/apple/foundationdb/record/lucene/FDBLuceneIndexFailureTest.java

-            for (int i = 0; i < 20; i++) {
-                recordStore.saveRecord(createComplexDocument(1000L + totalDocCount + i, ENGINEER_JOKE, docGroupFieldValue, start - i - 1));
+            for (int i = 0; i < 10; i++) {
+                dataModel.saveRecord(start - i - 1, store, docGroupFieldValue);


Why are you decrementing the start here?

ScottDugas · 2025-01-03T21:17:47Z

...yer-lucene/src/test/java/com/apple/foundationdb/record/lucene/FDBLuceneIndexFailureTest.java

            // Inject failures (using partition 0)
            injectedFailures.addFailure(LUCENE_READ_BLOCK,
                    new LuceneConcurrency.AsyncToSyncTimeoutException("Blah", new TimeoutException("Blah")),
                    5);
            // this should fail with injected exception
-            recordStore.saveRecord(createComplexDocument(1000L , ENGINEER_JOKE, docGroupFieldValue, 2));
+            dataModel.saveRecord(2, store, docGroupFieldValue);


This is not updating a record, it is saving a new record. You'll want to use

dataModel.recordsUnderTest().forEach(record -> record.updateOtherValue(store));

ScottDugas · 2025-01-03T21:21:47Z

...yer-lucene/src/test/java/com/apple/foundationdb/record/lucene/FDBLuceneIndexFailureTest.java

+            context.getDatabase().setAsyncToSyncExceptionMapper(this::mapExceptions);
+        } else {
+            context.getDatabase().setAsyncToSyncExceptionMapper((ex, ev) -> FDBExceptions.wrapException(ex));
+        }
    }

    private void setupExceptionMapping(boolean useExceptionMapping) {


Why have the two different methods?

Resolve failures, convert to use DataModel, remove changes to LuceneT…

b591aa5

…estBase

ohadzeliger self-assigned this Dec 20, 2024

Unused inputs

e3cdd50

ohadzeliger marked this pull request as ready for review January 2, 2025 20:57

ohadzeliger requested a review from ScottDugas January 2, 2025 20:57

ScottDugas requested changes Jan 3, 2025

View reviewed changes

ScottDugas mentioned this pull request Jan 14, 2025

Resolve #3008: Support Lucene index scrubbing #3009

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Refactor LuceneFailreTest to use the DataModel, not LuceneTestBase #3010

Refactor LuceneFailreTest to use the DataModel, not LuceneTestBase #3010

ohadzeliger commented Dec 20, 2024

foundationdb-ci commented Dec 20, 2024

foundationdb-ci commented Jan 2, 2025

ScottDugas Jan 3, 2025

ohadzeliger Jan 13, 2025

ScottDugas Jan 3, 2025

ohadzeliger Jan 13, 2025

ScottDugas Jan 3, 2025

ScottDugas Jan 3, 2025

ScottDugas Jan 3, 2025

ohadzeliger Jan 13, 2025

ScottDugas Jan 3, 2025

ScottDugas Jan 3, 2025

ScottDugas Jan 3, 2025

ScottDugas Jan 3, 2025

ScottDugas Jan 3, 2025


		// create/save documents
		long docGroupFieldValue = groupingKey.isEmpty() ? 0L : groupingKey.getLong(0);

		@@ -273,106 +272,122 @@ void repartitionAndMerge(boolean useLegacyAsyncToSync) throws IOException {
		segmentCounts);

		try (FDBRecordContext context = openContext(contextProps)) {

	.setPartitionHighWatermark(10)
	.setPartitionHighWatermark(totalDocCount)

Refactor LuceneFailreTest to use the DataModel, not LuceneTestBase #3010

Are you sure you want to change the base?

Refactor LuceneFailreTest to use the DataModel, not LuceneTestBase #3010

Conversation

ohadzeliger commented Dec 20, 2024

foundationdb-ci commented Dec 20, 2024

Result of fdb-record-layer-pr on Linux CentOS 7

foundationdb-ci commented Jan 2, 2025

Result of fdb-record-layer-pr on Linux CentOS 7

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment