This is an automated email from the ASF dual-hosted git repository.
dongjoon pushed a commit to branch branch-1.9
in repository https://gitbox.apache.org/repos/asf/orc.git
The following commit(s) were added to refs/heads/branch-1.9 by this push:
new 97867e7da ORC-1578: Fix `SparkBenchmark` on `sales` data according to
SPARK-40918
97867e7da is described below
commit 97867e7da3208f69c03c0aa190ee3d6386c8c631
Author: Dongjoon Hyun <[email protected]>
AuthorDate: Mon Jan 8 19:20:28 2024 -0800
ORC-1578: Fix `SparkBenchmark` on `sales` data according to SPARK-40918
### What changes were proposed in this pull request?
This PR aims to fix `SparkBenchmark` according to the requirement of
SPARK-40918.
Note that this fixes the synthetic benchmark on `Sales` data. For the other
real-life dataset (`github` and `taxi`), we will revisit.
### Why are the changes needed?
1. Generate `Sales` data
```
$ java -jar core/target/orc-benchmarks-core-*-uber.jar generate data -f orc
-d sales -s 1000000
```
2. Run Spark Benchmark
```
$ java -jar spark/target/orc-benchmarks-spark-2.1.0-SNAPSHOT.jar spark data
-d sales -f orc
# Run complete. Total time: 00:10:45
Benchmark (compression) (dataset)
(format) Mode Cnt Score Error Units
SparkBenchmark.fullRead gz sales
orc avgt 5 686792.235 ± 4398.971 us/op
SparkBenchmark.fullRead:bytesPerRecord gz sales
orc avgt 5 0.192 #
SparkBenchmark.fullRead:ops gz sales
orc avgt 5 40.000 #
SparkBenchmark.fullRead:perRecord gz sales
orc avgt 5 0.687 ± 0.004 us/op
SparkBenchmark.fullRead:records gz sales
orc avgt 5 5000000.000 #
SparkBenchmark.fullRead snappy sales
orc avgt 5 286166.380 ± 19864.429 us/op
SparkBenchmark.fullRead:bytesPerRecord snappy sales
orc avgt 5 0.201 #
SparkBenchmark.fullRead:ops snappy sales
orc avgt 5 40.000 #
SparkBenchmark.fullRead:perRecord snappy sales
orc avgt 5 0.286 ± 0.020 us/op
SparkBenchmark.fullRead:records snappy sales
orc avgt 5 5000000.000 #
SparkBenchmark.fullRead zstd sales
orc avgt 5 384394.233 ± 10057.315 us/op
SparkBenchmark.fullRead:bytesPerRecord zstd sales
orc avgt 5 0.192 #
SparkBenchmark.fullRead:ops zstd sales
orc avgt 5 40.000 #
SparkBenchmark.fullRead:perRecord zstd sales
orc avgt 5 0.384 ± 0.010 us/op
SparkBenchmark.fullRead:records zstd sales
orc avgt 5 5000000.000 #
SparkBenchmark.partialRead gz sales
orc avgt 5 41683.914 ± 4046.077 us/op
SparkBenchmark.partialRead:bytesPerRecord gz sales
orc avgt 5 0.192 #
SparkBenchmark.partialRead:ops gz sales
orc avgt 5 40.000 #
SparkBenchmark.partialRead:perRecord gz sales
orc avgt 5 0.042 ± 0.004 us/op
SparkBenchmark.partialRead:records gz sales
orc avgt 5 5000000.000 #
SparkBenchmark.partialRead snappy sales
orc avgt 5 23981.054 ± 17874.229 us/op
SparkBenchmark.partialRead:bytesPerRecord snappy sales
orc avgt 5 0.201 #
SparkBenchmark.partialRead:ops snappy sales
orc avgt 5 40.000 #
SparkBenchmark.partialRead:perRecord snappy sales
orc avgt 5 0.024 ± 0.018 us/op
SparkBenchmark.partialRead:records snappy sales
orc avgt 5 5000000.000 #
SparkBenchmark.partialRead zstd sales
orc avgt 5 41433.277 ± 25110.021 us/op
SparkBenchmark.partialRead:bytesPerRecord zstd sales
orc avgt 5 0.192 #
SparkBenchmark.partialRead:ops zstd sales
orc avgt 5 40.000 #
SparkBenchmark.partialRead:perRecord zstd sales
orc avgt 5 0.041 ± 0.025 us/op
SparkBenchmark.partialRead:records zstd sales
orc avgt 5 5000000.000 #
SparkBenchmark.pushDown gz sales
orc avgt 5 23760.997 ± 833.034 us/op
SparkBenchmark.pushDown:bytesPerRecord gz sales
orc avgt 5 19.153 #
SparkBenchmark.pushDown:ops gz sales
orc avgt 5 40.000 #
SparkBenchmark.pushDown:perRecord gz sales
orc avgt 5 2.376 ± 0.083 us/op
SparkBenchmark.pushDown:records gz sales
orc avgt 5 50000.000 #
SparkBenchmark.pushDown snappy sales
orc avgt 5 14062.508 ± 1793.691 us/op
SparkBenchmark.pushDown:bytesPerRecord snappy sales
orc avgt 5 20.105 #
SparkBenchmark.pushDown:ops snappy sales
orc avgt 5 40.000 #
SparkBenchmark.pushDown:perRecord snappy sales
orc avgt 5 1.406 ± 0.179 us/op
SparkBenchmark.pushDown:records snappy sales
orc avgt 5 50000.000 #
SparkBenchmark.pushDown zstd sales
orc avgt 5 15597.651 ± 1307.246 us/op
SparkBenchmark.pushDown:bytesPerRecord zstd sales
orc avgt 5 19.213 #
SparkBenchmark.pushDown:ops zstd sales
orc avgt 5 40.000 #
SparkBenchmark.pushDown:perRecord zstd sales
orc avgt 5 1.560 ± 0.131 us/op
SparkBenchmark.pushDown:records zstd sales
orc avgt 5 50000.000 #
```
### How was this patch tested?
Pass the CIs.
Closes #1734 from dongjoon-hyun/ORC-1578.
Authored-by: Dongjoon Hyun <[email protected]>
Signed-off-by: Dongjoon Hyun <[email protected]>
(cherry picked from commit fbe49d71e0e66903508a52952be1cb0ca9d09ac1)
Signed-off-by: Dongjoon Hyun <[email protected]>
---
.../src/java/org/apache/orc/bench/spark/SparkBenchmark.java | 10 ++++++++++
1 file changed, 10 insertions(+)
diff --git
a/java/bench/spark/src/java/org/apache/orc/bench/spark/SparkBenchmark.java
b/java/bench/spark/src/java/org/apache/orc/bench/spark/SparkBenchmark.java
index dc1dcf6f1..360c4de92 100644
--- a/java/bench/spark/src/java/org/apache/orc/bench/spark/SparkBenchmark.java
+++ b/java/bench/spark/src/java/org/apache/orc/bench/spark/SparkBenchmark.java
@@ -122,6 +122,7 @@ public class SparkBenchmark implements OrcBenchmark {
public void setup() {
session = SparkSession.builder().appName("benchmark")
.config("spark.master", "local[4]")
+ .config("spark.log.level", "ERROR")
.config("spark.sql.orc.filterPushdown", true)
.config("spark.sql.orc.impl", "native")
.getOrCreate();
@@ -189,6 +190,9 @@ public class SparkBenchmark implements OrcBenchmark {
case "json":
options.add(new Tuple2<>("timestampFormat", "yyyy-MM-dd
HH:mm:ss.SSS"));
break;
+ case "orc":
+ options.add(new Tuple2<>("returning_batch", "true")); // SPARK-40918
+ break;
default:
break;
}
@@ -219,6 +223,9 @@ public class SparkBenchmark implements OrcBenchmark {
case "json":
case "avro":
throw new IllegalArgumentException(source.format + " can't handle
projection");
+ case "orc":
+ options.add(new Tuple2<>("returning_batch", "true")); // SPARK-40918
+ break;
default:
break;
}
@@ -291,6 +298,9 @@ public class SparkBenchmark implements OrcBenchmark {
case "json":
case "avro":
throw new IllegalArgumentException(source.format + " can't handle
pushdown");
+ case "orc":
+ options.add(new Tuple2<>("returning_batch", "true")); // SPARK-40918
+ break;
default:
break;
}
