Todd Lipcon has posted comments on this change. ( 
http://gerrit.cloudera.org:8080/11341 )

Change subject: IMPALA-7424: Reduce in-memory footprint of incremental stats
......................................................................


Patch Set 6:

(7 comments)

Looking pretty close, just a few small nits.

http://gerrit.cloudera.org:8080/#/c/11341/6/common/thrift/CatalogObjects.thrift
File common/thrift/CatalogObjects.thrift:

http://gerrit.cloudera.org:8080/#/c/11341/6/common/thrift/CatalogObjects.thrift@303
PS6, Line 303: required
shouldn't this be optional since we don't need to send this to backends as part 
of descriptors?


http://gerrit.cloudera.org:8080/#/c/11341/6/fe/src/main/java/org/apache/impala/catalog/FeCatalogUtils.java
File fe/src/main/java/org/apache/impala/catalog/FeCatalogUtils.java:

http://gerrit.cloudera.org:8080/#/c/11341/6/fe/src/main/java/org/apache/impala/catalog/FeCatalogUtils.java@322
PS6, Line 322:         part.hasIncrementalStats());
per comment elsewhere, I dont think you need this in the non-FULL (descriptor) 
object


http://gerrit.cloudera.org:8080/#/c/11341/6/fe/src/main/java/org/apache/impala/catalog/HdfsPartition.java
File fe/src/main/java/org/apache/impala/catalog/HdfsPartition.java:

http://gerrit.cloudera.org:8080/#/c/11341/6/fe/src/main/java/org/apache/impala/catalog/HdfsPartition.java@32
PS6, Line 32: import com.google.common.base.Predicate;
nit: this should sort down with the other guava stuff


http://gerrit.cloudera.org:8080/#/c/11341/6/fe/src/main/java/org/apache/impala/catalog/HdfsPartition.java@668
PS6, Line 668:       // Delete the incremental stats entries.
I think you could simplify this code a bit and write:

Maps.filterKeys(partitionStats, IS_INCREMENTAL_STATS_KEY).clear();

(the doc for filterKeys says:
> When methods such as removeAll() and clear() are called on the filtered map 
> or its views, only mappings whose keys satisfy the filter will be removed 
> from the underlying map.
)

In Java 8, you could do 
partitionStats.keySet().removeIf(IS_INCREMENTAL_STATS_KEY) I think, but maybe 
you want to retain the option to backport this to Impala 2.x


http://gerrit.cloudera.org:8080/#/c/11341/6/fe/src/main/java/org/apache/impala/catalog/HdfsPartition.java@693
PS6, Line 693: hmsParamters_
nit: typo


http://gerrit.cloudera.org:8080/#/c/11341/6/fe/src/main/java/org/apache/impala/catalog/HdfsTable.java
File fe/src/main/java/org/apache/impala/catalog/HdfsTable.java:

http://gerrit.cloudera.org:8080/#/c/11341/6/fe/src/main/java/org/apache/impala/catalog/HdfsTable.java@1712
PS6, Line 1712: part.hasIncrementalStats
it's somewhat weird that this one is required. I would include it optionally, 
and only include it when the caller requests "want_partition_metadata"


http://gerrit.cloudera.org:8080/#/c/11341/6/fe/src/main/java/org/apache/impala/util/CompressionUtil.java
File fe/src/main/java/org/apache/impala/util/CompressionUtil.java:

http://gerrit.cloudera.org:8080/#/c/11341/6/fe/src/main/java/org/apache/impala/util/CompressionUtil.java@43
PS6, Line 43: BEST_COMPRESSION
curious if you compared the compression ratio and performance of 
BEST_COMPRESSION vs other options like DEFAULT and even HUFFMAN_ONLY? Maybe add 
a TODO here.



--
To view, visit http://gerrit.cloudera.org:8080/11341
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: I39f02ebfa0c6e9b0baedd0d76058a1b34efb5a02
Gerrit-Change-Number: 11341
Gerrit-PatchSet: 6
Gerrit-Owner: Bharath Vissapragada <[email protected]>
Gerrit-Reviewer: Bharath Vissapragada <[email protected]>
Gerrit-Reviewer: Impala Public Jenkins <[email protected]>
Gerrit-Reviewer: Todd Lipcon <[email protected]>
Gerrit-Reviewer: Vuk Ercegovac <[email protected]>
Gerrit-Comment-Date: Fri, 31 Aug 2018 04:13:46 +0000
Gerrit-HasComments: Yes

Reply via email to