[
https://issues.apache.org/jira/browse/IMPALA-12856?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Sai Hemanth Gantasala resolved IMPALA-12856.
--------------------------------------------
Fix Version/s: Impala 4.5.0
Resolution: Fixed
> IllegalStateException in processing RELOAD events due to malformed HMS
> Partition objects
> ----------------------------------------------------------------------------------------
>
> Key: IMPALA-12856
> URL: https://issues.apache.org/jira/browse/IMPALA-12856
> Project: IMPALA
> Issue Type: Bug
> Components: Catalog
> Reporter: Quanlong Huang
> Assignee: Sai Hemanth Gantasala
> Priority: Critical
> Labels: catalog-2024
> Fix For: Impala 4.5.0
>
>
> When processing RELOAD events on partitions, catalogd fetch the Partition
> objects from HMS. The returned Partition objects could be malformed which
> causes an IllegalStateException and stops the event-processor. This was
> observed when a partition is added and dropped in a loop.
> {noformat}
> E0229 15:19:27.945312 12668 MetastoreEventsProcessor.java:990] Unexpected
> exception received while processing event
> Java exception follows:
> java.lang.IllegalStateException
> at
> com.google.common.base.Preconditions.checkState(Preconditions.java:496)
> at
> org.apache.impala.catalog.HdfsTable.getTypeCompatiblePartValues(HdfsTable.java:2598)
> at
> org.apache.impala.catalog.HdfsTable.reloadPartitionsFromNames(HdfsTable.java:2856)
> at
> org.apache.impala.service.CatalogOpExecutor.reloadPartitionsFromNamesIfExists(CatalogOpExecutor.java:4805)
> at
> org.apache.impala.service.CatalogOpExecutor.reloadPartitionsIfExist(CatalogOpExecutor.java:4742)
> at
> org.apache.impala.catalog.events.MetastoreEvents$MetastoreTableEvent.reloadPartitions(MetastoreEvents.java:1050)
> at
> org.apache.impala.catalog.events.MetastoreEvents$ReloadEvent.processPartitionReload(MetastoreEvents.java:2941)
> at
> org.apache.impala.catalog.events.MetastoreEvents$ReloadEvent.processTableEvent(MetastoreEvents.java:2906)
> at
> org.apache.impala.catalog.events.MetastoreEvents$MetastoreTableEvent.process(MetastoreEvents.java:1248)
> at
> org.apache.impala.catalog.events.MetastoreEvents$MetastoreEvent.processIfEnabled(MetastoreEvents.java:672)
> at
> org.apache.impala.catalog.events.MetastoreEventsProcessor.processEvents(MetastoreEventsProcessor.java:1164)
> at
> org.apache.impala.catalog.events.MetastoreEventsProcessor.processEvents(MetastoreEventsProcessor.java:972)
> at
> java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
> at java.util.concurrent.FutureTask.runAndReset(FutureTask.java:308)
> at
> java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access$301(ScheduledThreadPoolExecutor.java:180)
> at
> java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:294)
> at
> java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
> at
> java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
> at java.lang.Thread.run(Thread.java:750)
> E0229 15:19:27.963455 12668 MetastoreEventsProcessor.java:1251] Event id:
> 8697728
> Event Type: RELOAD
> Event time: 1709191166
> Database name: default
> Table name: part_tbl
> Event message: H4s{noformat}
> The failed check is asserting the number of partition columns cached in
> catalogd matches the number of partition values from the HMS object:
> {code:java}
> public List<LiteralExpr> getTypeCompatiblePartValues(List<String> values) {
> List<LiteralExpr> result = new ArrayList<>();
> List<Column> partitionColumns = getClusteringColumns();
> Preconditions.checkState(partitionColumns.size() == values.size()); //
> This failed{code}
> After adding some debug logs, I found the Partition obejct got from HMS had
> an empty values list:
> {noformat}
> I0229 16:04:04.679625 25867 HdfsTable.java:2829] HMS Partition:
> Partition(values:[], dbName:default, tableName:part_tbl,
> createTime:1709193844, lastAccessTime:0,
> sd:StorageDescriptor(cols:[FieldSchema(name:i, type:int, comment:null)],
> location:hdf
> s://localhost:20500/test-warehouse/part_tbl/p=1,
> inputFormat:org.apache.hadoop.mapred.TextInputFormat,
> outputFormat:org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat,
> compressed:false, numBuckets:0, serdeInfo:SerDeInfo(name:null, serializ
> ationLib:org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe, parameters:{}),
> bucketCols:[], sortCols:[], parameters:{},
> skewedInfo:SkewedInfo(skewedColNames:[], skewedColValues:[],
> skewedColValueLocationMaps:{}), storedAsSubDirectories:false), par
> ameters:{}, catName:hive, writeId:0)
> I0229 16:04:04.680133 25867 MetastoreEventsProcessor.java:1189] Time elapsed
> in processing event batch: 17.145ms
> E0229 16:04:04.680475 25867 MetastoreEventsProcessor.java:990] Unexpected
> exception received while processing event
> Java exception follows:
> java.lang.IllegalStateException: partitionColumns.size(): 1, values.size():
> 0. partCols: Column{name_=p, type_=INT, comment_=null,
> stats=ColumnStats{avgSize_=4.0, avgSerializedSize_=4.0, maxSize_=4,
> numDistinct_=-1, numNulls_=-1, numTrues=-1, numFalses=-1, lowValue=-1,
> highValue=-1}, position_=0}, partValues: {noformat}
> EventProcessor should recover in this case.
> On the other hand, returning a Partition object with empty partition value
> list seems a bug of HMS. CC [~hemanth619], [~VenuReddy]
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
