You are viewing a plain text version of this content. The canonical link for it is here.
Posted to reviews@spark.apache.org by yhuai <gi...@git.apache.org> on 2016/07/06 06:49:59 UTC
[GitHub] spark pull request #14064: [SPARK-15968][SQL] Nonempty partitioned metastore...
GitHub user yhuai opened a pull request:
https://github.com/apache/spark/pull/14064
[SPARK-15968][SQL] Nonempty partitioned metastore tables are not cached
This PR backports your fix (https://github.com/apache/spark/pull/13818) to branch 2.0.
This PR addresses [SPARK-15968](https://issues.apache.org/jira/browse/SPARK-15968).
## What changes were proposed in this pull request?
The `getCached` method of [HiveMetastoreCatalog](https://github.com/apache/spark/blob/master/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveMetastoreCatalog.scala) computes `pathsInMetastore` from the metastore relation's catalog table. This only returns the table base path, which is incomplete/inaccurate for a nonempty partitioned table. As a result, cached lookups on nonempty partitioned tables always miss.
Rather than get `pathsInMetastore` from
metastoreRelation.catalogTable.storage.locationUri.toSeq
I modified the `getCached` method to take a `pathsInMetastore` argument. Calls to this method pass in the paths computed from calls to the Hive metastore. This is how `getCached` was implemented in Spark 1.5:
https://github.com/apache/spark/blob/e0c3212a9b42e3e704b070da4ac25b68c584427f/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveMetastoreCatalog.scala#L444.
I also added a call in `InsertIntoHiveTable.scala` to invalidate the table from the SQL session catalog.
## How was this patch tested?
I've added a new unit test to `parquetSuites.scala`:
SPARK-15968: nonempty partitioned metastore Parquet table lookup should use cached relation
Note that the only difference between this new test and the one above it in the file is that the new test populates its partitioned table with a single value, while the existing test leaves the table empty. This reveals a subtle, unexpected hole in test coverage present before this patch.
Note I also modified a different but related unit test in `parquetSuites.scala`:
SPARK-15248: explicitly added partitions should be readable
This unit test asserts that Spark SQL should return data from a table partition which has been placed there outside a metastore query immediately after it is added. I changed the test so that, instead of adding the data as a parquet file saved in the partition's location, the data is added through a SQL `INSERT` query. I made this change because I could find no way to efficiently support partitioned table caching without failing that test.
In addition to my primary motivation, I can offer a few reasons I believe this is an acceptable weakening of that test. First, it still validates a fix for [SPARK-15248](https://issues.apache.org/jira/browse/SPARK-15248), the issue for which it was written. Second, the assertion made is stronger than that required for non-partitioned tables. If you write data to the storage location of a non-partitioned metastore table without using a proper SQL DML query, a subsequent call to show that data will not return it. I believe this is an intentional limitation put in place to make table caching feasible, but I'm only speculating.
Building a large `HadoopFsRelation` requires `stat`-ing all of its data files. In our environment, where we have tables with 10's of thousands of partitions, the difference between using a cached relation versus a new one is a matter of seconds versus minutes. Caching partitioned table metadata vastly improves the usability of Spark SQL for these cases.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/yhuai/spark spark-15968-branch-2.0
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/spark/pull/14064.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #14064
----
commit 02aa39cf56d37f800803c5d8f0e92510d21e8af3
Author: Michael Allman <mi...@videoamp.com>
Date: 2016-07-05T01:49:25Z
[SPARK-15968][SQL] Nonempty partitioned metastore tables are not cached
(Please note this is a revision of PR #13686, which has been closed in favor of this PR.)
This PR addresses [SPARK-15968](https://issues.apache.org/jira/browse/SPARK-15968).
## What changes were proposed in this pull request?
The `getCached` method of [HiveMetastoreCatalog](https://github.com/apache/spark/blob/master/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveMetastoreCatalog.scala) computes `pathsInMetastore` from the metastore relation's catalog table. This only returns the table base path, which is incomplete/inaccurate for a nonempty partitioned table. As a result, cached lookups on nonempty partitioned tables always miss.
Rather than get `pathsInMetastore` from
metastoreRelation.catalogTable.storage.locationUri.toSeq
I modified the `getCached` method to take a `pathsInMetastore` argument. Calls to this method pass in the paths computed from calls to the Hive metastore. This is how `getCached` was implemented in Spark 1.5:
https://github.com/apache/spark/blob/e0c3212a9b42e3e704b070da4ac25b68c584427f/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveMetastoreCatalog.scala#L444.
I also added a call in `InsertIntoHiveTable.scala` to invalidate the table from the SQL session catalog.
## How was this patch tested?
I've added a new unit test to `parquetSuites.scala`:
SPARK-15968: nonempty partitioned metastore Parquet table lookup should use cached relation
Note that the only difference between this new test and the one above it in the file is that the new test populates its partitioned table with a single value, while the existing test leaves the table empty. This reveals a subtle, unexpected hole in test coverage present before this patch.
Note I also modified a different but related unit test in `parquetSuites.scala`:
SPARK-15248: explicitly added partitions should be readable
This unit test asserts that Spark SQL should return data from a table partition which has been placed there outside a metastore query immediately after it is added. I changed the test so that, instead of adding the data as a parquet file saved in the partition's location, the data is added through a SQL `INSERT` query. I made this change because I could find no way to efficiently support partitioned table caching without failing that test.
In addition to my primary motivation, I can offer a few reasons I believe this is an acceptable weakening of that test. First, it still validates a fix for [SPARK-15248](https://issues.apache.org/jira/browse/SPARK-15248), the issue for which it was written. Second, the assertion made is stronger than that required for non-partitioned tables. If you write data to the storage location of a non-partitioned metastore table without using a proper SQL DML query, a subsequent call to show that data will not return it. I believe this is an intentional limitation put in place to make table caching feasible, but I'm only speculating.
Building a large `HadoopFsRelation` requires `stat`-ing all of its data files. In our environment, where we have tables with 10's of thousands of partitions, the difference between using a cached relation versus a new one is a matter of seconds versus minutes. Caching partitioned table metadata vastly improves the usability of Spark SQL for these cases.
Thanks.
Author: Michael Allman <mi...@videoamp.com>
Closes #13818 from mallman/spark-15968.
commit d53afa6b22325fe5d9ce0d526714f56347ccc054
Author: Reynold Xin <rx...@databricks.com>
Date: 2016-07-05T19:06:41Z
[HOTFIX] Fix build break.
----
---
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.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
[GitHub] spark pull request #14064: [SPARK-15968][SQL] Nonempty partitioned metastore...
Posted by yhuai <gi...@git.apache.org>.
Github user yhuai closed the pull request at:
https://github.com/apache/spark/pull/14064
---
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.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
[GitHub] spark issue #14064: [SPARK-15968][SQL] Nonempty partitioned metastore tables...
Posted by AmplabJenkins <gi...@git.apache.org>.
Github user AmplabJenkins commented on the issue:
https://github.com/apache/spark/pull/14064
Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/SparkPullRequestBuilder/61823/
Test PASSed.
---
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.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
[GitHub] spark issue #14064: [SPARK-15968][SQL] Nonempty partitioned metastore tables...
Posted by cloud-fan <gi...@git.apache.org>.
Github user cloud-fan commented on the issue:
https://github.com/apache/spark/pull/14064
LGTM
---
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.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
[GitHub] spark issue #14064: [SPARK-15968][SQL] Nonempty partitioned metastore tables...
Posted by liancheng <gi...@git.apache.org>.
Github user liancheng commented on the issue:
https://github.com/apache/spark/pull/14064
LGTM pending Jenkins.
---
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.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
[GitHub] spark issue #14064: [SPARK-15968][SQL] Nonempty partitioned metastore tables...
Posted by cloud-fan <gi...@git.apache.org>.
Github user cloud-fan commented on the issue:
https://github.com/apache/spark/pull/14064
thanks, merging to 2.0!
---
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.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
[GitHub] spark issue #14064: [SPARK-15968][SQL] Nonempty partitioned metastore tables...
Posted by yhuai <gi...@git.apache.org>.
Github user yhuai commented on the issue:
https://github.com/apache/spark/pull/14064
@mallman this backports your fix to branch 2.0.
---
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.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
[GitHub] spark issue #14064: [SPARK-15968][SQL] Nonempty partitioned metastore tables...
Posted by SparkQA <gi...@git.apache.org>.
Github user SparkQA commented on the issue:
https://github.com/apache/spark/pull/14064
**[Test build #61823 has finished](https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/61823/consoleFull)** for PR 14064 at commit [`d53afa6`](https://github.com/apache/spark/commit/d53afa6b22325fe5d9ce0d526714f56347ccc054).
* This patch passes all tests.
* This patch merges cleanly.
* This patch adds no public classes.
---
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.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
[GitHub] spark issue #14064: [SPARK-15968][SQL] Nonempty partitioned metastore tables...
Posted by AmplabJenkins <gi...@git.apache.org>.
Github user AmplabJenkins commented on the issue:
https://github.com/apache/spark/pull/14064
Merged build finished. Test PASSed.
---
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.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
[GitHub] spark issue #14064: [SPARK-15968][SQL] Nonempty partitioned metastore tables...
Posted by mallman <gi...@git.apache.org>.
Github user mallman commented on the issue:
https://github.com/apache/spark/pull/14064
@cloud-fan Muchas gracias!
---
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.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
[GitHub] spark issue #14064: [SPARK-15968][SQL] Nonempty partitioned metastore tables...
Posted by SparkQA <gi...@git.apache.org>.
Github user SparkQA commented on the issue:
https://github.com/apache/spark/pull/14064
**[Test build #61823 has started](https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/61823/consoleFull)** for PR 14064 at commit [`d53afa6`](https://github.com/apache/spark/commit/d53afa6b22325fe5d9ce0d526714f56347ccc054).
---
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.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org