You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues-all@impala.apache.org by "Quanlong Huang (Jira)" <ji...@apache.org> on 2020/10/14 12:20:00 UTC
[jira] [Commented] (IMPALA-9733) catalog oom with partition table
x(91partitions,200k+files per partition) query like 'describe default.x'
[ https://issues.apache.org/jira/browse/IMPALA-9733?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17213863#comment-17213863 ]
Quanlong Huang commented on IMPALA-9733:
----------------------------------------
Hi [~h.s], thanks for filing this issue! Could you try this again on the master branch? We have some improvements (IMPALA-3127, IMPALA-9778, IMPALA-10076) on this part. Hope the metadata of each partition can fit into the 2GB limit.
BTW, it's not the best practise to have so many small files in a partition. Even if the DESCRIBE statement works, you will also hit performance issues due to too many small files.
> catalog oom with partition table x(91partitions,200k+files per partition) query like 'describe default.x'
> ---------------------------------------------------------------------------------------------------------
>
> Key: IMPALA-9733
> URL: https://issues.apache.org/jira/browse/IMPALA-9733
> Project: IMPALA
> Issue Type: Bug
> Components: Catalog
> Affects Versions: Impala 3.2.0
> Environment: jdk version: openjdk1.8.0_172
> os: centos 7.6.1810
> Reporter: h.s
> Priority: Major
>
> a table default.test (5 column, 91 partitions, about 256k files per partitions, 3.2k per file, totally 23340537 files)
> when query 'describe default.test' catalog log throw oom like
> {quote}I0507 21:03:10.629343 4618 HdfsTable.java:1235] Fetching partition metadata from the Metastore: default.test
> I0507 21:03:10.811630 4618 HdfsTable.java:1239] Fetched partition metadata from the Metastore: default.test
> I0507 21:03:11.136696 4618 HdfsTable.java:862] Loading file and block metadata for 91 paths for table default.test using a thread pool of size 50
> I0507 21:05:03.368173 4618 HdfsTable.java:903] Loaded file and block metadata for default.test
> I0507 21:05:03.391232 4618 TableLoader.java:97] Loaded metadata for: default.test
> E0507 21:05:16.206063 51284 catalog-server.cc:377] OutOfMemoryError: null
> E0507 21:05:28.912606 51284 catalog-server.cc:377] OutOfMemoryError: null
> {quote}
> i do debug catalog use idea trace the oom stack like
> {quote}java.io.ByteArrayOutputStream.hugeCapacity(ByteArrayOutputStream.java:123)
> java.io.ByteArrayOutputStream.grow(ByteArrayOutputStream.java:117)
> java.io.ByteArrayOutputStream.ensureCapacity(ByteArrayOutputStream.java:93)
> java.io.ByteArrayOutputStream.write(ByteArrayOutputStream.java:153)
> org.apache.thrift.transport.TIOStreamTransport.write(TIOStreamTransport.java:145)
> org.apache.thrift.protocol.TBinaryProtocol.writeBinary(TBinaryProtocol.java:211)
> org.apache.impala.thrift.THdfsFileDesc$THdfsFileDescStandardScheme.write(THdfsFileDesc.java:369)
> org.apache.impala.thrift.THdfsFileDesc$THdfsFileDescStandardScheme.write(THdfsFileDesc.java:332)
> org.apache.impala.thrift.THdfsFileDesc.write(THdfsFileDesc.java:283)
> org.apache.impala.thrift.THdfsPartition$THdfsPartitionStandardScheme.write(THdfsPartition.java:2137)
> org.apache.impala.thrift.THdfsPartition$THdfsPartitionStandardScheme.write(THdfsPartition.java:1862)
> org.apache.impala.thrift.THdfsPartition.write(THdfsPartition.java:1677)
> org.apache.impala.thrift.THdfsTable$THdfsTableStandardScheme.write(THdfsTable.java:1249)
> org.apache.impala.thrift.THdfsTable$THdfsTableStandardScheme.write(THdfsTable.java:1076)
> org.apache.impala.thrift.THdfsTable.write(THdfsTable.java:939)
> org.apache.impala.thrift.TTable$TTableStandardScheme.write(TTable.java:1632)
> org.apache.impala.thrift.TTable$TTableStandardScheme.write(TTable.java:1403)
> org.apache.impala.thrift.TTable.write(TTable.java:1212)
> org.apache.impala.thrift.TCatalogObject$TCatalogObjectStandardScheme.write(TCatalogObject.java:1245)
> org.apache.impala.thrift.TCatalogObject$TCatalogObjectStandardScheme.write(TCatalogObject.java:1102)
> org.apache.impala.thrift.TCatalogObject.write(TCatalogObject.java:942)
> org.apache.thrift.TSerializer.serialize(TSerializer.java:79)
> org.apache.impala.catalog.CatalogServiceCatalog$GetCatalogDeltaContext.addCatalogObject(CatalogServiceCatalog.java:591)
> org.apache.impala.catalog.CatalogServiceCatalog.addTableToCatalogDeltaHelper(CatalogServiceCatalog.java:1028)
> org.apache.impala.catalog.CatalogServiceCatalog.addTableToCatalogDelta(CatalogServiceCatalog.java:974)
> org.apache.impala.catalog.CatalogServiceCatalog.addDatabaseToCatalogDelta(CatalogServiceCatalog.java:880)
> org.apache.impala.catalog.CatalogServiceCatalog.getCatalogDelta(CatalogServiceCatalog.java:671)
> org.apache.impala.service.JniCatalog.getCatalogDelta(JniCatalog.java:135)
> {quote}
> and the code stack
> {quote}grow:119, ByteArrayOutputStream (java.io)
> ensureCapacity:93, ByteArrayOutputStream (java.io)
> write:153, ByteArrayOutputStream (java.io)
> write:145, TIOStreamTransport (org.apache.thrift.transport)
> writeBinary:211, TBinaryProtocol (org.apache.thrift.protocol)
> write:369, THdfsFileDesc$THdfsFileDescStandardScheme (org.apache.impala.thrift)
> write:332, THdfsFileDesc$THdfsFileDescStandardScheme (org.apache.impala.thrift)
> write:283, THdfsFileDesc (org.apache.impala.thrift)
> write:2137, THdfsPartition$THdfsPartitionStandardScheme (org.apache.impala.thrift)
> write:1862, THdfsPartition$THdfsPartitionStandardScheme (org.apache.impala.thrift)
> write:1677, THdfsPartition (org.apache.impala.thrift)
> write:1249, THdfsTable$THdfsTableStandardScheme (org.apache.impala.thrift)
> write:1076, THdfsTable$THdfsTableStandardScheme (org.apache.impala.thrift)
> write:939, THdfsTable (org.apache.impala.thrift)
> write:1632, TTable$TTableStandardScheme (org.apache.impala.thrift)
> write:1403, TTable$TTableStandardScheme (org.apache.impala.thrift)
> write:1212, TTable (org.apache.impala.thrift)
> write:1245, TCatalogObject$TCatalogObjectStandardScheme (org.apache.impala.thrift)
> write:1102, TCatalogObject$TCatalogObjectStandardScheme (org.apache.impala.thrift)
> write:942, TCatalogObject (org.apache.impala.thrift)
> serialize:79, TSerializer (org.apache.thrift)
> addCatalogObject:591, CatalogServiceCatalog$GetCatalogDeltaContext (org.apache.impala.catalog)
> addTableToCatalogDeltaHelper:1028, CatalogServiceCatalog (org.apache.impala.catalog)
> addTableToCatalogDelta:974, CatalogServiceCatalog (org.apache.impala.catalog)
> addDatabaseToCatalogDelta:880, CatalogServiceCatalog (org.apache.impala.catalog)
> getCatalogDelta:671, CatalogServiceCatalog (org.apache.impala.catalog)
> getCatalogDelta:135, JniCatalog (org.apache.impala.service)
> {quote}
> {{and in file ByteArrayOutputStream.java (Integer.MAX_VALUE=2147483647) code like}}
> {code:java}
> private static int hugeCapacity(int minCapacity)
> { if (minCapacity < 0) // overflow
> throw new OutOfMemoryError();
> return (minCapacity > MAX_ARRAY_SIZE) ? Integer.MAX_VALUE : MAX_ARRAY_SIZE; }
> {code}
> and in ByteArrayOutputStream.java per len=176, and count+ (totally file number times)up to oom
> {code:java}
> public synchronized void write(byte b[], int off, int len) {
> if ((off < 0) || (off > b.length) || (len < 0) || ((off + len) - b.length > 0))
> { throw new IndexOutOfBoundsException(); }
> ensureCapacity(count + len);
> System.arraycopy(b, off, buf, count, len);
> count += len; }
> {code}
> catalog setting
> {quote}-default_pool_mem_limit=280g \
> -mem_limit=330g \
> -mem_limit_includes_jvm=true \
> -buffer_pool_limit=180g \
> -max_hdfs_partitions_parallel_load=50 \
> -non_impala_java_vlog=0 \
> -inc_stats_size_limit_bytes=4400000000 \
> -pull_incremental_statistics=false \
> -num_threads_per_core=8 \
> -num_metadata_loading_threads=50 \
> -num_hdfs_worker_threads=50 \
> -num_acceptor_threads=20 \
> -accepted_cnxn_setup_thread_pool_size=10 \
> -max_row_batches=4000 \
> -max_queued_row_batch_bytes=100000000 \
> -compact_catalog_topic=true \
> {quote}
> statestore setting
> {quote}-inc_stats_size_limit_bytes=4400000000 \
> -pull_incremental_statistics=false \
> -compact_catalog_topic=true \
> {quote}
> impalad setting
> {quote}-mem_limit=250g \
> -buffer_pool_limit=60g \
> -mem_limit_includes_jvm=true \
> -default_pool_mem_limit=180g \
> -num_metadata_loading_threads=50 \
> -num_threads_per_core=8 \
> -num_hdfs_worker_threads=50 \
> -max_hdfs_partitions_parallel_load=50 \
> -num_acceptor_threads=20 \
> -accepted_cnxn_setup_thread_pool_size=5 \
> -pull_incremental_statistics=false \
> -inc_stats_size_limit_bytes=4400000000 \
> -compact_catalog_topic=true \
> {quote}
> noticed https://issues.apache.org/jira/browse/IMPALA-8667
> set -pull_incremental_statistics=false but true is the same
> and https://issues.apache.org/jira/browse/IMPALA-5990
> [https://github.com/cloudera/Impala/commit/c2184e56ae91bc9c6f04ac0dae1b75261d6d20a2]
> like catalog object over the limit and not compressed, is there any solutions can query 'describe default.test'
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-all-unsubscribe@impala.apache.org
For additional commands, e-mail: issues-all-help@impala.apache.org