fix(server): ensure graph clear API does not clear meta table for hbase

[LYD031106]

Purpose of the PR

• close [Bug] The backend store version is inconsistent #2209

Main Changes

The PR fixes the issue where the graph clear API mistakenly clears the meta table in Hbase backend storage.

Verifying these changes

• Trivial rework / code cleanup without any test coverage. (No Need)
• Already covered by existing tests, such as (please modify tests here).
• Need tests and can be verified as follows:
  • xxx

Does this PR potentially affect the following parts?

• Dependencies (add/update license info & regenerate_known_dependencies.sh)
• Modify configurations
• The public API
• Other affects (typed here)
• Nope

Documentation Status

• Doc - TODO
• Doc - Done
• Doc - No Need

[imbajin]

⚠️ Test coverage insufficient

The test only verifies version equality but doesn't validate:

1. Meta table content: Check that specific meta entries (like backend version, graph name, etc.) are preserved
2. Data table clearing: Verify that actual graph data (vertices, edges) are properly cleared
3. Concurrent operations: Test behavior when truncate is called during other operations

Suggestion:

@Test
public void testHbaseMetaVersion(){
    // Insert some test data
    // ... add vertices/edges ...
    
    String beforeVersion = this.store.storedVersion();
    // Get other meta entries
    
    this.store.truncate();
    
    String afterVersion = this.store.storedVersion();
    Assert.assertEquals(beforeVersion, afterVersion);
    
    // Verify data is cleared but meta is intact
    // ... check vertices/edges are gone ...
    // ... check other meta entries preserved ...
}

[imbajin]

⚠️ Test setup/teardown issue

Both BaseHbaseUnitTest and HbaseUnitTest have setup/teardown methods that may cause issues:

1. Duplicate cleanup: @After in both base and child class will close store/provider twice
2. Setup order: Child's @Before runs after parent's, but calls provider.open() which may conflict

Suggestion:

Suggested change

	@Before
	@Before
	public void setUp(){
	super.setup(); // Call parent setup first
	this.provider.open(GRAPH_NAME);
	this.provider.init();
	}
	@After
	public void teardown(){
	// Remove duplicate cleanup - handled by parent
	// Only add child-specific cleanup here if needed
	}

[imbajin]

🧹 Code quality: Silent exception swallowing

Using empty catch blocks makes debugging difficult. Consider:

Suggested change

	public void down(){
	@After
	public void down(){
	if (this.store != null) {
	try {
	this.store.close();
	} catch (Exception e) {
	LOG.warn("Failed to close store", e);
	}
	}
	if (this.provider != null) {
	try {
	this.provider.close();
	} catch (Exception e) {
	LOG.warn("Failed to close provider", e);
	}
	}
	}

[imbajin]

🧹 Naming: Space missing in class declaration

Minor style issue:

Suggested change

	public class BaseHbaseUnitTest extends BaseUnitTest {

[imbajin]

‼️ Root cause analysis: Design issue in truncate implementation

After reviewing the code, I found the real problem:

Current Issue:

• HbaseSystemStore.tableNames() (line 572-576) includes the meta table in the list
• truncate() clears ALL tables returned by tableNames(), including the meta table
• Adding init() is a workaround, but meta table should never be cleared in the first place

Proper Fix:
Override truncate() in HbaseSystemStore to exclude meta table:

@Override
public void truncate() {
    // Save meta table before truncate
    List<String> originalTables = this.tableNames();
    
    // Temporarily remove meta table from truncation
    // Then call super.truncate() on data tables only
    // Or better: add a separate method tableNamesToTruncate()
}

Alternative approach (cleaner):

protected List<String> tableNamesToTruncate() {
    // Only return data tables, not meta/system tables
    return super.tableNames(); // Don't include meta.table()
}

Then use tableNamesToTruncate() in the truncate() method instead of tableNames().

This matches how MySQL backend works - it only truncates data tables, never system tables.

[LYD031106]

Your solution is indeed a superior approach. By overriding the truncate() method to exclude metadata tables, we are solving the problem at the design level, which is far more elegant and robust than relying on a temporary solution involving initialization calls.

I will proceed with the modifications as suggested, implementing this fix for the HBase backend. For consistency, I will also apply the same change to the MySQL backend in the corresponding PR (#2888).

Regarding the current implementation for the RocksDB backend, which uses the direct init() call workaround, I believe it should also be optimized. Adopting this unified strategy of overriding the truncate() method will enhance code consistency and maintainability across all backends.Here is the implementation of the truncate() method in RocksStore.

    @Override
    public synchronized void truncate() {
        Lock writeLock = this.storeLock.writeLock();
        writeLock.lock();
        try {
            this.checkOpened();

            this.clear(false);
            this.init();
            // Clear write-batch
            this.dbs.values().forEach(BackendSessionPool::forceResetSessions);
            LOG.debug("Store truncated: {}", this.store);
        } finally {
            writeLock.unlock();
        }
    }

Thank you again for your valuable feedback. It has truly helped me arrive at a much more elegant solution!

[imbajin]

‼️ Critical: Logic Error - Counters Table Should Be Truncated

The Counters table contains graph-specific data like vertex/edge ID sequences. When truncating a graph, this data should be cleared to avoid ID conflicts when repopulating the graph.

Only the Meta table (which stores backend version metadata) should be preserved during truncation.

Suggested change

	.map(BackendTable::table)
	protected List<String> truncatedTableNames() {
	return this.tables.values().stream()
	.filter(table -> !(table instanceof HbaseTables.Meta))
	.map(BackendTable::table)
	.collect(Collectors.toList());
	}

[imbajin]

⚠️ Test Method Name Mismatch

The test method is named testMysqlMetaVersion() but it's testing HBase functionality. This appears to be a copy-paste error from MySQL tests.

Suggested change

	public void testMysqlMetaVersion(){
	@Test
	public void testHbaseMetaVersionAfterTruncate() {

This makes the test's purpose clearer and matches the actual functionality being tested.

[imbajin]

⚠️ Incomplete Test Coverage

The test only verifies that meta version is preserved, but doesn't verify the actual truncation behavior. Consider adding assertions to verify:

1. Data tables are actually truncated (e.g., insert some data, truncate, verify data is gone)
2. Meta table content remains intact
3. The graph can be used normally after truncation

Example enhancement:

@Test
public void testHbaseMetaVersionAfterTruncate() {
    BackendStore systemStore = this.provider.loadSystemStore(config);
    BackendStore graphStore = this.provider.loadGraphStore(config);
    
    // Record initial version
    String beforeVersion = systemStore.storedVersion();
    
    // Insert some test data to verify truncation
    // ... add test data insertion code ...
    
    // Perform truncation
    this.provider.truncate();
    
    // Verify version preserved
    String afterVersion = systemStore.storedVersion();
    Assert.assertEquals(beforeVersion, afterVersion);
    
    // Verify data tables are empty
    // ... add verification code ...
}

[imbajin]

🧹 Code Style: Formatting

Missing space after class declaration:

Suggested change

	public class BaseHbaseUnitTest extends BaseUnitTest{
	public class BaseHbaseUnitTest extends BaseUnitTest {

[imbajin]

🧹 Minor: Unnecessary Whitespace Change

Removing this blank line is cosmetic and adds noise to the diff without functional benefit. Consider keeping the original formatting to make the diff cleaner and focus on actual logic changes.

[imbajin]

wrong header format

[imbajin]

‼️ Critical: Potential NullPointerException

The filter condition HugeType.META == e.getKey() may not correctly handle null keys, though this is unlikely in practice. However, the main concern is whether HugeType.META is actually the correct type to filter out.

Consider adding a comment explaining why META type should be preserved during truncation to improve code maintainability.

Suggested change

	.filter(e -> !(HugeType.META == e.getKey()))
	protected List<String> truncatedTableNames() {
	// Exclude META table to preserve system metadata (e.g., version info) during graph clear
	return this.tables.entrySet().stream()
	.filter(e -> !(HugeType.META == e.getKey()))
	.map(e -> e.getValue().table())
	.collect(Collectors.toList());
	}

[imbajin]

⚠️ Missing @Before annotation

The method is named setup() but the annotation @Before is missing. JUnit won't automatically call this method before each test. This could cause tests to fail or behave unexpectedly.

Suggested change

	protected HbaseSessions sessions;
	@Before
	public void setup() throws IOException {

[imbajin]

⚠️ Naming inconsistency: down() should be tearDown()

The cleanup method should follow JUnit naming conventions and be named tearDown() to match the common pattern.

Suggested change

	this.sessions = new HbaseSessions(config,GRAPH_NAME, this.provider.loadGraphStore(config).store());
	@After
	public void tearDown() {

[imbajin]

⚠️ Missing space after comma - code style

For code consistency, add a space after commas in method calls.

Suggested change

	this.provider.loadSystemStore(config).open(config);
	this.sessions = new HbaseSessions(config, GRAPH_NAME, this.provider.loadGraphStore(config).store());

[imbajin]

⚠️ Missing space after comma in log statement

Suggested change

	public void down(){
	LOG.warn("Failed to close provider", e);

[imbajin]

⚠️ Missing space after class declaration

Suggested change

	public class BaseHbaseUnitTest extends BaseUnitTest {

[imbajin]

⚠️ Missing space after method opening brace

Suggested change

	this.sessions = new HbaseSessions(config,GRAPH_NAME, this.provider.loadGraphStore(config).store());
	public void tearDown() {

[imbajin]

‼️ Critical: Resource leak - HbaseSessions not closed

The sessions field is opened in setup() but never closed in the cleanup method. This could lead to HBase connection leaks.

Suggested change

	this.sessions = new HbaseSessions(config,GRAPH_NAME, this.provider.loadGraphStore(config).store());
	@After
	public void tearDown() {
	if (this.sessions != null) {
	try {
	this.sessions.close();
	} catch (Exception e) {
	LOG.warn("Failed to close sessions", e);
	}
	}
	if (this.provider != null) {

[imbajin]

⚠️ Potential resource leak with iterators

If an assertion fails, the subsequent close() calls won't execute. Consider using try-with-resources or try-finally blocks.

Suggested change

	testsession.put("g_oe", "f".getBytes(), "row_trunc_oe".getBytes(), StringEncoding.encode("q"), StringEncoding.encode("v"));
	// Verify data insertion success
	try (BackendIterator<Result> vIterator = testsession.get("g_v", "f".getBytes(), "row_trunc_v".getBytes());
	BackendIterator<Result> oeIterator = testsession.get("g_oe", "f".getBytes(), "row_trunc_oe".getBytes());
	BackendIterator<Result> ieIterator = testsession.get("g_ie", "f".getBytes(), "row_trunc_ie".getBytes())) {
	Assert.assertTrue("data should exist", vIterator.hasNext());
	Assert.assertTrue("data should exist", oeIterator.hasNext());
	Assert.assertTrue("data should exist", ieIterator.hasNext());
	}

[imbajin]

⚠️ Missing space after comma in Assert.assertNotNull

Suggested change

	String afterVersion = systemStore.storedVersion();
	Assert.assertNotNull("System metadata version should exist", afterVersion);

[imbajin]

‼️ Critical: Method name doesn't follow Java naming conventions

The method name truncatedTableNames() is grammatically incorrect and misleading. It suggests tables that have already been truncated, when it actually returns tables that SHOULD be truncated.

Suggested change

	protected List<String> truncatedTableNames() {
	protected List<String> getNonMetaTableNames() {
	// Exclude meta table to preserve system metadata during graph clear
	return this.tables.entrySet().stream()
	.filter(e -> !(HugeType.META == e.getKey()))
	.map(e -> e.getValue().table())
	.collect(Collectors.toList());
	}

Alternative names: getClearableTableNames(), getDataTableNames(), or getTableNamesExcludingMeta()

[imbajin]

⚠️ Consider defensive programming: check for null HugeType

While unlikely, if e.getKey() is null, this comparison would fail with NPE. Consider using Objects.equals() or put the enum constant on the left side.

Suggested change

	return this.tables.entrySet().stream()
	.filter(e -> !HugeType.META.equals(e.getKey()))

[LYD031106]

The current code 'HugeType.META == e.getKey()' is already NPE-safe regardless of which operand is null, as the '==' operator never throws NullPointerException.

[imbajin]

‼️ Critical: Potential resource leak in setup()

If any exception occurs after opening stores but before sessions.open(), the opened stores won't be properly closed. This can lead to resource leaks in test execution.

Suggested change

	this.provider = new HbaseStoreProvider();
	@Before
	public void setup() throws IOException {
	Configuration conf = Utils.getConf();
	this.config = new HugeConfig(conf);
	this.provider = new HbaseStoreProvider();
	try {
	this.provider.open(GRAPH_NAME);
	this.provider.loadSystemStore(config).open(config);
	this.provider.loadGraphStore(config).open(config);
	this.provider.loadSchemaStore(config).open(config);
	this.provider.init();
	this.sessions = new HbaseSessions(config, GRAPH_NAME,
	this.provider.loadGraphStore(config).store());
	this.sessions.open();
	} catch (Exception e) {
	tearDown();
	throw e;
	}
	}

[imbajin]

🧹 Code style: inconsistent spacing

Missing space after if keyword. Per project code style guidelines, control flow statements should have space before parentheses.

Suggested change

	if (this.provider != null){
	if (this.provider != null) {

[imbajin]

🧹 Code style: inconsistent spacing

Missing space after catch keyword.

Suggested change

	this.provider.close();
	} catch (Exception e) {

[imbajin]

🧹 Code style: inconsistent spacing

Missing space after closing brace and before catch. Should be } catch, not }catch.

Suggested change

	LOG.warn("Failed to close provider ",e);
	LOG.warn("Failed to close provider ", e);

[imbajin]

⚠️ Consider using try-with-resources for session management

The session testsession is not explicitly closed, which could lead to resource leaks. While it might be managed by the parent sessions object, it's better to be explicit.

Consider wrapping the session usage in a try-with-resources block or explicitly closing it after use.

[imbajin]

⚠️ Test isolation concern: verify test data cleanup

The test inserts data with specific row keys (row_trunc_v, row_trunc_oe, row_trunc_ie). If this test runs multiple times or fails mid-execution, residual data might affect subsequent runs.

Consider:

1. Using unique row keys per test run (e.g., append timestamp/UUID)
2. Adding explicit cleanup in @Before setup
3. Verifying the @After teardown properly cleans all test data

[Copilot]

Pull request overview

This PR fixes a critical bug where the HBase graph clear API (truncate()) was incorrectly clearing the system metadata table. The fix introduces a new truncatedTableNames() method that returns only the tables that should be cleared during truncation, while preserving system metadata like backend version information stored in the meta table.

Key Changes

• Added truncatedTableNames() method in HbaseStore to filter tables for truncation
• Modified truncate() to use truncatedTableNames() instead of tableNames()
• Added comprehensive unit tests to verify meta table preservation after truncation

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 5 comments.

File	Description
hugegraph-server/hugegraph-hbase/src/main/java/org/apache/hugegraph/backend/store/hbase/HbaseStore.java	Introduces truncatedTableNames() method to exclude meta table and updates truncate() to use it instead of tableNames()
hugegraph-server/hugegraph-test/src/main/java/org/apache/hugegraph/unit/hbase/BaseHbaseUnitTest.java	New base test class for HBase unit tests with setup/teardown infrastructure
hugegraph-server/hugegraph-test/src/main/java/org/apache/hugegraph/unit/hbase/HbaseUnitTest.java	New test case verifying meta table version persists after truncation while graph data is cleared

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

This filter condition !(HugeType.META == e.getKey()) is redundant because HugeType.META is never registered via registerTableManager() and therefore will never exist in the this.tables map. The meta table is added separately in HbaseSystemStore.tableNames() (line 582) by calling this.meta.table().

The fix works correctly because truncatedTableNames() simply doesn't include the meta table at all (since it's not in the map), but the explicit filter is misleading and suggests META might be in the map. Consider removing the filter:

protected List<String> truncatedTableNames() {
    // Exclude meta table to preserve system metadata during graph clear
    return this.tables.values().stream()
                      .map(BackendTable::table)
                      .collect(Collectors.toList());
}

Suggested change

	return this.tables.entrySet().stream()
	.filter(e -> !(HugeType.META == e.getKey()))
	.map(e -> e.getValue().table())
	return this.tables.values().stream()
	.map(BackendTable::table)

[Copilot]

Missing space after if keyword. Should be if (this.provider != null) { for consistency with the code style used elsewhere in the method (line 56).

Suggested change

	if (this.provider != null){
	if (this.provider != null) {

[Copilot]

Missing space after catch keyword. Should be } catch (Exception e) { for consistency with the code style used on line 59.

Suggested change

	}catch (Exception e){
	} catch (Exception e) {

[Copilot]

Missing space after catch keyword. Should be } catch (Exception e) { for consistency with the code style used on line 59.

Suggested change

	}catch (Exception e){
	} catch (Exception e) {

[Copilot]

Missing space after comma in log message. Should be "Failed to close provider ", e for consistency with line 60.

Suggested change

	LOG.warn("Failed to close provider ",e);
	LOG.warn("Failed to close provider ", e);