[
https://issues.apache.org/jira/browse/HBASE-26384?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
chenglei updated HBASE-26384:
-----------------------------
Description:
When {{CompactingMemStore}} prepares flushing, {{CompactingMemStore.snapshot}}
invokes following {{CompactingMemStore.pushPipelineToSnapshot}} method to get
{{Snapshot}}, following line 570 and line 575 uses
{{CompactionPipeline#version}} to track whether the Segments in
{{CompactionPipeline#pipeline}} has changed since it gets
{{VersionedSegmentsList}} in line 570 before emptying
{{CompactionPipeline#pipeline}} in line 575.
{code:java}
565 private void pushPipelineToSnapshot() {
566 int iterationsCnt = 0;
567 boolean done = false;
568 while (!done) {
569 iterationsCnt++;
570 VersionedSegmentsList segments = pipeline.getVersionedList();
571 pushToSnapshot(segments.getStoreSegments());
572 // swap can return false in case the pipeline was updated by
ongoing compaction
573 // and the version increase, the chance of it happenning is
very low
574 // In Swap: don't close segments (they are in snapshot now)
and don't update the region size
575 done = pipeline.swap(segments, null, false, false);
.......
}
{code}
However, when {{CompactingMemStore#inMemoryCompaction}} executes
{{CompactionPipeline#flattenOneSegment}}, it does not change
{{CompactionPipeline#version}} , if there is a in memeory compaction which
executes {{CompactingMemStore#flattenOneSegment}} between above line 570 and
line 575, the {{CompactionPipeline#version}} not change, but the
{{ImmutableSegment}} in {{CompactionPipeline}} has changed.
{{pipeline.swap(segments, null, false, false)}} in above line 575 could
successfully invokes following {{CompactionPipeline#swapSuffix}} method to
remove {{Segment}} in {{CompactionPipeline}} bcause
{{CompactionPipeline#version}} not change, but the {{Segment}} in
{{CompactionPipeline}} has changed because
{{CompactingMemStore#flattenOneSegment}} is completed , so the {{Segment}} not
removed in following line 295 and still remaining in {{CompactionPipeline}}.
{code:java}
293 private void swapSuffix(List<? extends Segment> suffix, ImmutableSegment
segment,
294 boolean closeSegmentsInSuffix) {
295 pipeline.removeAll(suffix);
296 if(segment != null) pipeline.addLast(segment);
....
{code}
However {{CompactingMemStore.snapshot}} think it is successful and continues to
flush the {{Segment}} got by {{CompactingMemStore.snapshot}} as normal, but
the {{Segment}} with the same cells still be left in {{CompactingMemStore}}.
Leaving {{Segment}} which already flushed in {{MemStore}} is dangerous: if a
Major Compaction before the left {{Segment}} flushing, there may be data
erroneous.
My Fix in the PR is as following:
# Increasing the {{CompactionPipeline#version}} in
{{CompactingMemStore#flattenOneSegment}}
# For {{CompactionPipeline#swapSuffix}} in above step 6 , explicitly checking
that the Segments in {{suffix}} input parameter is same as the Segments in
{{pipeline}} one by one from
the last element to the first element of {{suffix}} .
was:
When {{CompactingMemStore}} prepares flushing, {{CompactingMemStore.snapshot}}
uses {{CompactionPipeline#version}} to track whether the Segments in
{{CompactionPipeline#pipeline}} has changed since it gets
{{VersionedSegmentsList}} last time before emptying
{{CompactionPipeline#pipeline}}. However, when
{{CompactingMemStore#inMemoryCompaction}} executes
{{CompactionPipeline#flattenOneSegment}}, it does not change
{{CompactionPipeline#version}} , which could causes the segment which has be
collected to {{MemStoreSnapshot}} may still be remained in
{{CompactingMemStore}}.
Considing following thread sequence:
1. When {{CompactingMemStore}} size exceeds
{{CompactingMemStore#getInmemoryFlushSize}}, the write thread adds a new
{{ImmutableSegment}} to the head of
{{CompactingMemStore#pipeline}}, and start a in memory compact thread to
execute{{CompactingMemStore#inMemoryCompaction}} .
2. The in memory compact thread starts and then stopping before
{{CompactingMemStore#flattenOneSegment}}.
3. The snapshot thread starts {{CompactingMemStore#snapshot}} concurrently,
after the snapshot thread executing following line 570
{{pipeline.getVersionedList}}, the in memory
compact thread continues. Assuming {{VersionedSegmentsList#version}}
returned from {{pipeline.getVersionedList}} is v.
{code:java}
565 private void pushPipelineToSnapshot() {
566 int iterationsCnt = 0;
567 boolean done = false;
568 while (!done) {
569 iterationsCnt++;
570 VersionedSegmentsList segments = pipeline.getVersionedList();
571 pushToSnapshot(segments.getStoreSegments());
572 // swap can return false in case the pipeline was updated by
ongoing compaction
573 // and the version increase, the chance of it happenning is
very low
574 // In Swap: don't close segments (they are in snapshot now)
and don't update the region size
575 done = pipeline.swap(segments, null, false, false);
.......
}
{code}
4. The snapshot thread stopping before {{pipeline.swap(segments, null, false,
false)}} in above line 575.
5. The in memory compact thread completes
{{CompactingMemStore#flattenOneSegment}},{{CompactingMemStore#flattenOneSegment}}
does not change
{{CompactionPipeline#version}}, {{CompactionPipeline#version}} is still v.
6. The snapshot thread continues {{pipeline.swap(segments, null, false,
false)}} in above line 575, and because {{CompactionPipeline#version}}
is still v, {{pipeline.swap(segments, null, false, false)}} successfully
invokes following {{swapSuffix}} to remove {{ImmutableSegment}} in
{{CompactionPipeline}} ,
{code:java}
293 private void swapSuffix(List<? extends Segment> suffix, ImmutableSegment
segment,
294 boolean closeSegmentsInSuffix) {
295 pipeline.removeAll(suffix);
296 if(segment != null) pipeline.addLast(segment);
....
{code}
but the {{ImmutableSegment}} in {{CompactionPipeline}} has changed because
{{CompactingMemStore#flattenOneSegment}} is completed in step 5, so the
{{ImmutableSegment}}
in {{CompactionPipeline}} is not removed in above line 295 and still
remaining in {{CompactionPipeline}}.However the snapshot thread think
{{CompactingMemStore.snapshot}} is successful and continues to flush the
{{Segment}} got by {{pipeline.getVersionedList}} in step 3 as normal, but the
{{Segment}} with the same
cells still be left in {{CompactingMemStore}}.
Leaving {{Segment}} which already flushed in {{MemStore}} is dangerous: if a
Major Compaction before the left {{Segment}} flushing, there may be data
erroneous.
My Fix in the PR is as following:
# Increasing the {{CompactionPipeline#version}} in
{{CompactingMemStore#flattenOneSegment}}
# For {{CompactionPipeline#swapSuffix}} in above step 6 , explicitly checking
that the Segments in {{suffix}} input parameter is same as the Segments in
{{pipeline}} one by one from
the last element to the first element of {{suffix}} .
> Segment which already flushed to hfile may still be remained in
> CompactingMemStore
> ------------------------------------------------------------------------------------
>
> Key: HBASE-26384
> URL: https://issues.apache.org/jira/browse/HBASE-26384
> Project: HBase
> Issue Type: Bug
> Components: in-memory-compaction
> Affects Versions: 3.0.0-alpha-1, 2.4.7
> Reporter: chenglei
> Priority: Major
>
> When {{CompactingMemStore}} prepares flushing,
> {{CompactingMemStore.snapshot}} invokes following
> {{CompactingMemStore.pushPipelineToSnapshot}} method to get {{Snapshot}},
> following line 570 and line 575 uses {{CompactionPipeline#version}} to track
> whether the Segments in {{CompactionPipeline#pipeline}} has changed since it
> gets {{VersionedSegmentsList}} in line 570 before emptying
> {{CompactionPipeline#pipeline}} in line 575.
> {code:java}
> 565 private void pushPipelineToSnapshot() {
> 566 int iterationsCnt = 0;
> 567 boolean done = false;
> 568 while (!done) {
> 569 iterationsCnt++;
> 570 VersionedSegmentsList segments =
> pipeline.getVersionedList();
> 571 pushToSnapshot(segments.getStoreSegments());
> 572 // swap can return false in case the pipeline was updated
> by ongoing compaction
> 573 // and the version increase, the chance of it happenning is
> very low
> 574 // In Swap: don't close segments (they are in snapshot now)
> and don't update the region size
> 575 done = pipeline.swap(segments, null, false, false);
> .......
> }
> {code}
> However, when {{CompactingMemStore#inMemoryCompaction}} executes
> {{CompactionPipeline#flattenOneSegment}}, it does not change
> {{CompactionPipeline#version}} , if there is a in memeory compaction which
> executes {{CompactingMemStore#flattenOneSegment}} between above line 570 and
> line 575, the {{CompactionPipeline#version}} not change, but the
> {{ImmutableSegment}} in {{CompactionPipeline}} has changed.
> {{pipeline.swap(segments, null, false, false)}} in above line 575 could
> successfully invokes following {{CompactionPipeline#swapSuffix}} method to
> remove {{Segment}} in {{CompactionPipeline}} bcause
> {{CompactionPipeline#version}} not change, but the {{Segment}} in
> {{CompactionPipeline}} has changed because
> {{CompactingMemStore#flattenOneSegment}} is completed , so the {{Segment}}
> not removed in following line 295 and still remaining in
> {{CompactionPipeline}}.
> {code:java}
> 293 private void swapSuffix(List<? extends Segment> suffix,
> ImmutableSegment segment,
> 294 boolean closeSegmentsInSuffix) {
> 295 pipeline.removeAll(suffix);
> 296 if(segment != null) pipeline.addLast(segment);
> ....
> {code}
> However {{CompactingMemStore.snapshot}} think it is successful and continues
> to flush the {{Segment}} got by {{CompactingMemStore.snapshot}} as normal,
> but the {{Segment}} with the same cells still be left in
> {{CompactingMemStore}}. Leaving {{Segment}} which already flushed in
> {{MemStore}} is dangerous: if a Major Compaction before the left {{Segment}}
> flushing, there may be data erroneous.
> My Fix in the PR is as following:
> # Increasing the {{CompactionPipeline#version}} in
> {{CompactingMemStore#flattenOneSegment}}
> # For {{CompactionPipeline#swapSuffix}} in above step 6 , explicitly checking
> that the Segments in {{suffix}} input parameter is same as the Segments in
> {{pipeline}} one by one from
> the last element to the first element of {{suffix}} .
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
