[
https://issues.apache.org/jira/browse/DRILL-5544?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16030203#comment-16030203
]
ASF GitHub Bot commented on DRILL-5544:
---------------------------------------
Github user paul-rogers commented on a diff in the pull request:
https://github.com/apache/drill/pull/846#discussion_r119181468
--- Diff:
exec/java-exec/src/main/java/org/apache/parquet/hadoop/ParquetColumnChunkPageWriteStore.java
---
@@ -0,0 +1,269 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+package org.apache.parquet.hadoop;
+
+import static
org.apache.parquet.column.statistics.Statistics.getStatsBasedOnType;
+
+import java.io.Closeable;
+import java.io.IOException;
+import java.util.List;
+import java.util.Map;
+import java.util.Set;
+
+import com.google.common.collect.Lists;
+import com.google.common.collect.Maps;
+import com.google.common.collect.Sets;
+import
org.apache.drill.exec.store.parquet.ParquetDirectByteBufferAllocator;
+import org.apache.parquet.bytes.BytesInput;
+import org.apache.parquet.bytes.CapacityByteArrayOutputStream;
+import org.apache.parquet.column.ColumnDescriptor;
+import org.apache.parquet.column.Encoding;
+import org.apache.parquet.column.page.DictionaryPage;
+import org.apache.parquet.column.page.PageWriteStore;
+import org.apache.parquet.column.page.PageWriter;
+import org.apache.parquet.column.statistics.Statistics;
+import org.apache.parquet.format.converter.ParquetMetadataConverter;
+import org.apache.parquet.hadoop.CodecFactory.BytesCompressor;
+import org.apache.parquet.io.ParquetEncodingException;
+import org.apache.parquet.schema.MessageType;
+import org.apache.parquet.bytes.ByteBufferAllocator;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+/**
+ * This is a copy of ColumnChunkPageWriteStore from parquet library except
of OutputStream that is used here.
+ * Using of CapacityByteArrayOutputStream allows to use different
ByteBuffer allocators.
--- End diff --
Please explain, perhaps in the JIRA or in a code comment, the benefit of
using a ByteBuffer here. Here is my question. Data comes from value vectors,
where it is stored as direct memory. But, individual values have to be
retrieved from the vector and converted to some other form: perhaps an int, a
double or however Parquet represents strings.
Parquet will then work its magic on the values: compressing them,
dictionary encoding them, etc.
Finally, the resulting buffers are written to disk. If I/O is synchronous,
data comes from a heap buffer. If async, then from direct.
What is the benefit of introducing ByteBuffer allocators here? Are we doing
that to use direct memory? What is the advantage of copying Parquet's likely
heap buffers into direct memory for buffering prior to writing to a file?
Said another way, why can't we just use the original Parquet version of
this code?
Some bit of explanation would be very helpful for us reviewers...
> Out of heap running CTAS against text delimited
> -----------------------------------------------
>
> Key: DRILL-5544
> URL: https://issues.apache.org/jira/browse/DRILL-5544
> Project: Apache Drill
> Issue Type: Bug
> Components: Storage - Parquet
> Affects Versions: 1.10.0
> Environment: - 2 or 4 nodes cluster
> - 4G or 8G of Java heap and more than 8G of direct memory
> - planner.width.max_per_node = 40
> - store.parquet.compression = none
> To generate lineitem.tbl file unzip dbgen.tgz archive and run:
> {code}dbgen -TL -s 500{code}
> Reporter: Vitalii Diravka
> Assignee: Vitalii Diravka
> Fix For: 1.11.0
>
> Attachments: dbgen.tgz
>
>
> This query causes the drillbit to hang:
> {code}
> create table xyz as
> select
> cast(columns[0] as bigint) l_orderkey,
> cast(columns[1] as integer) l_poartkey,
> cast(columns[2] as integer) l_suppkey,
> cast(columns[3] as integer) l_linenumber,
> cast(columns[4] as double) l_quantity,
> cast(columns[5] as double) l_extendedprice,
> cast(columns[6] as double) l_discount,
> cast(columns[7] as double) l_tax,
> cast(columns[8] as char(1)) l_returnflag,
> cast(columns[9] as char(1)) l_linestatus,
> cast(columns[10] as date) l_shipdate,
> cast(columns[11] as date) l_commitdate,
> cast(columns[12] as date) l_receiptdate,
> cast(columns[13] as char(25)) l_shipinstruct,
> cast(columns[14] as char(10)) l_shipmode,
> cast(columns[15] as varchar(44)) l_comment
> from
> `lineitem.tbl`;
> {code}
> OOM "Java heap space" from the drillbit.log:
> {code:title=drillbit.log|borderStyle=solid}
> ...
> 2017-02-07 22:38:11,031 [2765b496-0b5b-a3df-c252-a8bb9cd2e52f:frag:1:53]
> DEBUG o.a.d.e.s.p.ParquetDirectByteBufferAllocator -
> ParquetDirectByteBufferAllocator: Allocated 209715 bytes. Allocated
> ByteBuffer id: 1563631814
> 2017-02-07 22:38:16,478 [2765b496-0b5b-a3df-c252-a8bb9cd2e52f:frag:1:1] ERROR
> o.a.d.exec.server.BootStrapContext -
> org.apache.drill.exec.work.WorkManager$WorkerBee$1.run() leaked an exception.
> java.lang.OutOfMemoryError: Java heap space
> 2017-02-07 22:38:17,391 [2765b496-0b5b-a3df-c252-a8bb9cd2e52f:frag:1:13]
> ERROR o.a.drill.common.CatastrophicFailure - Catastrophic Failure Occurred,
> exiting. Information message: Unable to handle out of memory condition in
> FragmentExecutor.
> ...
> {code}
> To reproduce the issue please see environment details.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
