xiangfu0 commented on code in PR #19770:
URL: https://github.com/apache/pinot/pull/19770#discussion_r4223073694
##########
pinot-controller/src/main/java/org/apache/pinot/controller/helix/core/retention/RetentionManager.java:
##########
@@ -122,6 +127,210 @@ protected void processTable(String tableNameWithType) {
// Delete segments based on segment lineage and clean up segment lineage
metadata.
manageSegmentLineageCleanupForTable(tableConfig);
+
+ // Re-read active segments after time and lineage cleanup so
already-deleted bytes do not cause extra eviction.
+ manageSizeBasedRetention(tableConfig);
+ }
+
+ /// Enforces the optional compressed-segment size limit on live data
independently of time retention. Replaced
+ /// segments are excluded from accounting. Eviction proceeds oldest first
and stops before the first active segment
+ /// listed on either side of any lineage entry, including replacement copies
excluded from accounting. The newest
+ /// OFFLINE segment and realtime recovery segments are preserved; hybrid
realtime eviction requires offline coverage.
+ /// Unknown sizes or missing metadata disable this pass; consuming segments
are excluded from accounting and eviction.
+ @VisibleForTesting
+ protected void manageSizeBasedRetention(TableConfig tableConfig) {
+ String tableNameWithType = tableConfig.getTableName();
+ String retentionSize =
tableConfig.getValidationConfig().getRetentionSize();
+ if (retentionSize == null || (tableConfig.getTableType() ==
TableType.OFFLINE
+ &&
!"APPEND".equalsIgnoreCase(IngestionConfigUtils.getBatchSegmentIngestionType(tableConfig))))
{
+ _controllerMetrics.removeTableGauge(tableNameWithType,
ControllerGauge.SIZE_RETENTION_BLOCKED);
+ return;
+ }
+ boolean blocked = true;
+ try {
+ blocked = !enforceSizeBasedRetention(tableConfig, retentionSize);
+ } finally {
+ // Publish only the completed pass's result, including a blocked result
if an exception interrupted it.
+ _controllerMetrics.setOrUpdateTableGauge(tableNameWithType,
ControllerGauge.SIZE_RETENTION_BLOCKED,
+ blocked ? 1 : 0);
+ }
+ }
+
+ /// Returns whether the table is within its size cap after this pass; unsafe
skips and failed evictions return false.
+ private boolean enforceSizeBasedRetention(TableConfig tableConfig, String
retentionSize) {
+ String tableNameWithType = tableConfig.getTableName();
+ long retentionSizeBytes;
+ try {
+ retentionSizeBytes = DataSizeUtils.toBytes(retentionSize);
+ Preconditions.checkArgument(retentionSizeBytes > 0, "Retention size must
be positive");
+ } catch (Exception e) {
+ LOGGER.warn("Invalid retention size: {} for table: {}, skip size
retention", retentionSize, tableNameWithType);
+ return false;
+ }
+
+ // Read IdealState before lineage, matching the existing live-segment
filtering order. A replacement starting
+ // after this snapshot cannot add both its source and destination to the
bytes counted by this pass.
+ Set<String> activeSegments = new
HashSet<>(_pinotHelixResourceManager.getSegmentsFor(tableNameWithType, false));
+ SegmentLineage segmentLineage =
+
SegmentLineageAccessHelper.getSegmentLineage(_pinotHelixResourceManager.getPropertyStore(),
tableNameWithType);
+ Map<String, LineageEntry> lineageEntries = segmentLineage == null ?
Map.of() : segmentLineage.getLineageEntries();
+ Set<String> lineageOwnedSegments = new HashSet<>();
+ // Size retention protects both sides of every retained entry, regardless
of state, age, or the deletion flag.
+ lineageEntries.values().forEach(entry -> {
+ lineageOwnedSegments.addAll(entry.getSegmentsFrom());
+ lineageOwnedSegments.addAll(entry.getSegmentsTo());
+ });
+ lineageOwnedSegments.retainAll(activeSegments);
+ // Replaced segments awaiting lineage cleanup are no longer queryable and
must not trigger live-data eviction.
+ SegmentLineageUtils.filterSegmentsBasedOnLineageInPlace(activeSegments,
segmentLineage);
+ if (activeSegments.isEmpty()) {
+ return true;
+ }
+ List<SegmentZKMetadata> metadataList =
_pinotHelixResourceManager.getSegmentsZKMetadata(tableNameWithType);
+ Map<String, SegmentZKMetadata> lineageOwnedMetadata = new HashMap<>();
+ Set<String> segmentsWithMetadata = new HashSet<>();
+ List<SegmentZKMetadata> completedSegments = new ArrayList<>();
+ Map<String, Long> completedSegmentSizes = new HashMap<>();
+ boolean realtime = tableConfig.getTableType() == TableType.REALTIME;
+ long retainedBytes = 0L;
+ for (SegmentZKMetadata metadata : metadataList) {
+ String segmentName = metadata.getSegmentName();
+ if (lineageOwnedSegments.contains(segmentName)) {
+ lineageOwnedMetadata.put(segmentName, metadata);
+ }
+ if (!activeSegments.contains(segmentName)) {
+ continue;
+ }
+ segmentsWithMetadata.add(segmentName);
+ if (realtime && !metadata.getStatus().isCompleted()) {
+ continue;
+ }
+ long size = getSizeRetentionSegmentSize(tableNameWithType, metadata);
+ if (size < 0) {
+ LOGGER.warn("Unknown size for segment: {} in table: {}, skip size
retention", segmentName, tableNameWithType);
+ return false;
+ }
+ retainedBytes += size;
Review Comment:
The `BigInteger`→`long` simplification is fine for every physically-possible
table, but it silently changes the failure mode for exactly the input class
this pass otherwise defends against: corrupt ZK metadata. A single segment
whose `sizeInBytes` was written as ~`Long.MAX_VALUE` (the case the now-deleted
`testSizeRetentionAccountsForTotalAboveLongMaximum` covered) wraps the sum
negative, so `retainedBytes <= retentionSizeBytes` passes and the method
returns *healthy* — gauge 0, no WARN — on a table whose size metadata is
garbage. Every other corrupt-metadata shape here (unknown size, missing
metadata, unlocatable barrier) warns and reports blocked; overflow is now the
one that reports success.
You can keep the `long` and the safety in one line:
```java
try {
retainedBytes = Math.addExact(retainedBytes, size);
} catch (ArithmeticException e) {
LOGGER.warn("Segment size overflow for table: {}, skip size retention",
tableNameWithType);
return false;
}
```
and restore the deleted test with the new expectation (skip + gauge 1
instead of evict). If you'd rather not guard it at all, at least a comment
noting the wraparound-reports-healthy trade-off would stop a future reader from
re-adding `BigInteger`.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
