[
https://issues.apache.org/jira/browse/DRILL-8484?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17828185#comment-17828185
]
ASF GitHub Bot commented on DRILL-8484:
---------------------------------------
shfshihuafeng commented on code in PR #2889:
URL: https://github.com/apache/drill/pull/2889#discussion_r1529743261
##########
exec/java-exec/src/main/java/org/apache/drill/exec/cache/VectorAccessibleSerializable.java:
##########
@@ -155,12 +157,18 @@ public void readFromStreamWithContainer(VectorContainer
myContainer, InputStream
for (SerializedField metaData : fieldList) {
final int dataLength = metaData.getBufferLength();
final MaterializedField field = MaterializedField.create(metaData);
- final DrillBuf buf = allocator.buffer(dataLength);
- final ValueVector vector;
+ DrillBuf buf = null;
+ ValueVector vector = null;
try {
+ buf = allocator.buffer(dataLength);
buf.writeBytes(input, dataLength);
vector = TypeHelper.getNewVector(field, allocator);
vector.load(metaData, buf);
+ } catch (OutOfMemoryException oom) {
+ for (ValueVector valueVector : vectorList) {
+ valueVector.clear();
+ }
+ throw UserException.memoryError(oom).message("Allocator memory
failed").build(logger);
Review Comment:
when we prepare to allocator memory using "allocator.buffer(dataLength)"
for hashjoinPop allocator, if actual memory > maxAllocation(The parameter is
calculated by call computeOperatorMemory) ,then it throw exception, like
following my test。
user can adjust directMemory parameters (DRILL_MAX_DIRECT_MEMORY) or
reduce concurrency based on actual conditions.
**throw exception code**
```
public DrillBuf buffer(final int initialRequestSize, BufferManager manager) {
assertOpen();
AllocationOutcome outcome = allocateBytes(actualRequestSize);
if (!outcome.isOk()) {
throw new OutOfMemoryException(createErrorMsg(this, actualRequestSize,
initialRequestSize));
}
```
**my test scenario**
```
Caused by: org.apache.drill.exec.exception.OutOfMemoryException: Unable to
allocate buffer of size 16384 (rounded from 14359) due to memory limit
(41943040). Current allocation: 22583616
at
org.apache.drill.exec.memory.BaseAllocator.buffer(BaseAllocator.java:241)
at
org.apache.drill.exec.memory.BaseAllocator.buffer(BaseAllocator.java:216)
at
org.apache.drill.exec.cache.VectorAccessibleSerializable.readFromStreamWithContainer(VectorAccessibleSerializable.java:172)
```
> HashJoinPOP memory leak is caused by an oom exception when read data from
> Stream with container
> -------------------------------------------------------------------------------------------------
>
> Key: DRILL-8484
> URL: https://issues.apache.org/jira/browse/DRILL-8484
> Project: Apache Drill
> Issue Type: Bug
> Components: Server
> Affects Versions: 1.21.1
> Reporter: shihuafeng
> Priority: Major
> Fix For: 1.22.0
>
>
> *Describe the bug*
> An oom exception occurred When read data from Stream with container
> ,resulting in hashJoinPOP memory leak
> *To Reproduce*
> prepare data for tpch 1s
> # 30 concurrent for tpch sql8
> # set direct memory 5g
> # when it had OutOfMemoryException , stopped all sql.
> # finding memory leak
> *leak info*
> {code:java}
> Allocator(frag:5:0) 5000000/1000000/31067136/40041943040
> (res/actual/peak/limit)
> child allocators: 1
> Allocator(op:5:0:1:HashJoinPOP) 1000000/16384/22822912/41943040
> (res/actual/peak/limit)
> child allocators: 0
> ledgers: 2
> ledger[1882757] allocator: op:5:0:1:HashJoinPOP), isOwning: true,
> size: 8192, references: 2, life: 16936270178816167..0, allocatorManager:
> [1703465, life: 16936270178813617..0] holds 4 buffers.
> DrillBuf[2041995], udle: [1703441 0..957]{code}
> {code:java}
> {code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
