drill-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From paul-rogers <...@git.apache.org>
Subject [GitHub] drill pull request #846: DRILL-5544: Out of heap running CTAS against text d...
Date Tue, 06 Jun 2017 19:16:44 GMT
Github user paul-rogers commented on a diff in the pull request:

    https://github.com/apache/drill/pull/846#discussion_r120453820
  
    --- 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 --
    
    I'm not sure that 20 GB is "OK" to write to a file: whether on the heap or otherwise.
But, if we accept the poor design that needs this much, then I agree it is better to have
uncontrolled use of direct memory than the heap. But, in either case, uncontrolled memory
use is a very bad idea.
    
    Note that the only buffering needed is 4K: the disk block size. Experiments with spilling
show that there is zero benefit to buffering above 32K.
    
    So, for 10 runs * 32K = 320K. This is FAR less than the 20 GB proposed, and plenty for
heap, to avoid the cost of copying data into, then out of, direct memory.
    
    Further, what is wrong with Java's own `BufferedOutputStream`? Was this tried? It works
well for simple cases, not sure about more complex cases.
    
    Is the problem that Parquet must somehow buffer an entire page so it can "backpatch" the
first bytes after writing all the rest? If so, that part of Parquet design is incompatible
with a write-once file system such as HDFS: it forces unnecessary load on the memory system.
(Note, however, than an updatable file system, such as Linux or MFS, would allow us to write
data to disk, then go back and write the header, without buffering.)


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastructure@apache.org or file a JIRA ticket
with INFRA.
---

Mime
View raw message