iemejia commented on code in PR #12400:
URL: https://github.com/apache/gluten/pull/12400#discussion_r3549514723
##########
backends-velox/src/main/scala/org/apache/gluten/config/VeloxConfig.scala:
##########
@@ -176,9 +176,9 @@ object VeloxConfig extends ConfigRegistry {
val COLUMNAR_VELOX_SSD_CACHE_IO_THREADS =
buildStaticConf("spark.gluten.sql.columnar.backend.velox.ssdCacheIOThreads")
- .doc("The IO threads for cache promoting")
+ .doc("The number of IO threads for SSD cache read/write operations")
.intConf
- .createWithDefault(1)
+ .createWithDefault(4)
Review Comment:
Fixed. Added `.checkValue(_ > 0, "must be a positive number")` consistent
with the pattern used by `numCacheFileHandles` and other numeric configs in
this file.
##########
backends-velox/src/test/scala/org/apache/spark/sql/execution/VeloxFileHandleCacheSuite.scala:
##########
@@ -0,0 +1,310 @@
+/*
+ * 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.spark.sql.execution
+
+import org.apache.gluten.config.VeloxConfig
+import org.apache.gluten.execution.{BasicScanExecTransformer,
VeloxWholeStageTransformerSuite}
+
+import org.apache.spark.SparkConf
+
+/**
+ * Test suite for Velox file handle cache behavior.
+ *
+ * Tests correctness, config propagation, and edge cases for the file handle
cache which caches open
+ * file handles (descriptors) to avoid repeated open/close overhead.
+ */
+class VeloxFileHandleCacheSuite extends VeloxWholeStageTransformerSuite {
+ override protected val resourcePath: String = "/parquet-for-read"
+ override protected val fileFormat: String = "parquet"
+
+ override protected def sparkConf: SparkConf = {
+ super.sparkConf
+ .set(VeloxConfig.COLUMNAR_VELOX_FILE_HANDLE_CACHE_ENABLED.key, "true")
+ .set(VeloxConfig.COLUMNAR_VELOX_FILE_HANDLE_EXPIRATION_DURATION_MS.key,
"2000")
+ .set(VeloxConfig.COLUMNAR_VELOX_NUM_CACHE_FILE_HANDLES.key, "10000")
+ }
+
+ testWithSpecifiedSparkVersion(
+ "basic scan correctness with file handle cache enabled",
+ "3.5",
+ "3.5") {
+ // Verify that enabling file handle cache produces correct scan results
+ withTempPath {
+ dir =>
+ spark
+ .range(10000)
+ .selectExpr("id", "cast(id % 7 as int) as category", "id * 1.5 as
value")
+ .repartition(10)
+ .write
+ .parquet(dir.getCanonicalPath)
+
+ val df = spark.read.parquet(dir.getCanonicalPath)
+ df.createOrReplaceTempView("t")
+
+ runQueryAndCompare("SELECT count(*) FROM t") {
+ checkGlutenPlan[BasicScanExecTransformer]
+ }
+ runQueryAndCompare("SELECT sum(value) FROM t WHERE category = 3") {
+ checkGlutenPlan[BasicScanExecTransformer]
+ }
+ runQueryAndCompare("SELECT category, count(*) FROM t GROUP BY
category") {
+ checkGlutenPlan[BasicScanExecTransformer]
+ }
+ }
+ }
+
+ testWithSpecifiedSparkVersion(
+ "repeated scans produce consistent results",
+ "3.5",
+ "3.5") {
+ // Repeated scans of the same files must produce identical results
regardless
+ // of whether handles are served from cache or re-opened after TTL
eviction.
+ withTempPath {
+ dir =>
+ spark
+ .range(5000)
+ .selectExpr("id", "cast(id as string) as name")
+ .repartition(50) // 50 files to exercise many cache entries
+ .write
+ .parquet(dir.getCanonicalPath)
+
+ val path = dir.getCanonicalPath
+ val expected = spark.read.parquet(path).count()
+ assert(expected == 5000)
+
+ // Verify scans go through Gluten/Velox
+ checkGlutenPlan[BasicScanExecTransformer](spark.read.parquet(path))
+
+ // Scan the same files multiple times - results must be consistent
+ for (i <- 1 to 5) {
+ val count = spark.read.parquet(path).count()
+ assert(
+ count == expected,
+ s"Iteration $i: expected $expected rows but got $count")
+ }
+
+ // Verify aggregation consistency across repeated scans
+ val firstSum =
spark.read.parquet(path).selectExpr("sum(id)").collect()(0).getLong(0)
+ for (i <- 1 to 3) {
+ val sum =
spark.read.parquet(path).selectExpr("sum(id)").collect()(0).getLong(0)
+ assert(
+ sum == firstSum,
+ s"Iteration $i: sum mismatch, expected $firstSum but got $sum")
+ }
+ }
+ }
+
+ testWithSpecifiedSparkVersion(
+ "many small files do not cause errors with file handle cache",
+ "3.5",
+ "3.5") {
+ // Verify that scanning many small files with caching enabled does not
cause
+ // file descriptor exhaustion or other resource-related errors.
+ withTempPath {
+ dir =>
+ // Create 200 small parquet files
+ spark
+ .range(20000)
+ .selectExpr("id", "uuid() as payload")
+ .repartition(200)
+ .write
+ .parquet(dir.getCanonicalPath)
+
+ val fileCount = dir.listFiles().count(_.getName.endsWith(".parquet"))
+ assert(fileCount >= 200, s"Expected at least 200 files, got
$fileCount")
+
+ // Verify scans go through Gluten/Velox
+
checkGlutenPlan[BasicScanExecTransformer](spark.read.parquet(dir.getCanonicalPath))
+
+ // Scan all files - should work without resource errors
+ val count = spark.read.parquet(dir.getCanonicalPath).count()
+ assert(count == 20000)
+
+ // Scan again (cache hit path) - should also work
Review Comment:
Already fixed in the previous commit — removed "cache hit path" wording and
replaced with "results must remain consistent".
##########
backends-velox/src/test/scala/org/apache/spark/sql/execution/VeloxFileHandleCacheSuite.scala:
##########
@@ -0,0 +1,310 @@
+/*
+ * 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.spark.sql.execution
+
+import org.apache.gluten.config.VeloxConfig
+import org.apache.gluten.execution.{BasicScanExecTransformer,
VeloxWholeStageTransformerSuite}
+
+import org.apache.spark.SparkConf
+
+/**
+ * Test suite for Velox file handle cache behavior.
+ *
+ * Tests correctness, config propagation, and edge cases for the file handle
cache which caches open
+ * file handles (descriptors) to avoid repeated open/close overhead.
+ */
+class VeloxFileHandleCacheSuite extends VeloxWholeStageTransformerSuite {
+ override protected val resourcePath: String = "/parquet-for-read"
+ override protected val fileFormat: String = "parquet"
+
+ override protected def sparkConf: SparkConf = {
+ super.sparkConf
+ .set(VeloxConfig.COLUMNAR_VELOX_FILE_HANDLE_CACHE_ENABLED.key, "true")
+ .set(VeloxConfig.COLUMNAR_VELOX_FILE_HANDLE_EXPIRATION_DURATION_MS.key,
"2000")
+ .set(VeloxConfig.COLUMNAR_VELOX_NUM_CACHE_FILE_HANDLES.key, "10000")
+ }
+
+ testWithSpecifiedSparkVersion(
+ "basic scan correctness with file handle cache enabled",
+ "3.5",
+ "3.5") {
+ // Verify that enabling file handle cache produces correct scan results
+ withTempPath {
+ dir =>
+ spark
+ .range(10000)
+ .selectExpr("id", "cast(id % 7 as int) as category", "id * 1.5 as
value")
+ .repartition(10)
+ .write
+ .parquet(dir.getCanonicalPath)
+
+ val df = spark.read.parquet(dir.getCanonicalPath)
+ df.createOrReplaceTempView("t")
+
+ runQueryAndCompare("SELECT count(*) FROM t") {
+ checkGlutenPlan[BasicScanExecTransformer]
+ }
+ runQueryAndCompare("SELECT sum(value) FROM t WHERE category = 3") {
+ checkGlutenPlan[BasicScanExecTransformer]
+ }
+ runQueryAndCompare("SELECT category, count(*) FROM t GROUP BY
category") {
+ checkGlutenPlan[BasicScanExecTransformer]
+ }
+ }
+ }
+
+ testWithSpecifiedSparkVersion(
+ "repeated scans produce consistent results",
+ "3.5",
+ "3.5") {
+ // Repeated scans of the same files must produce identical results
regardless
+ // of whether handles are served from cache or re-opened after TTL
eviction.
+ withTempPath {
+ dir =>
+ spark
+ .range(5000)
+ .selectExpr("id", "cast(id as string) as name")
+ .repartition(50) // 50 files to exercise many cache entries
+ .write
+ .parquet(dir.getCanonicalPath)
+
+ val path = dir.getCanonicalPath
+ val expected = spark.read.parquet(path).count()
+ assert(expected == 5000)
+
+ // Verify scans go through Gluten/Velox
+ checkGlutenPlan[BasicScanExecTransformer](spark.read.parquet(path))
+
+ // Scan the same files multiple times - results must be consistent
+ for (i <- 1 to 5) {
+ val count = spark.read.parquet(path).count()
+ assert(
+ count == expected,
+ s"Iteration $i: expected $expected rows but got $count")
+ }
+
+ // Verify aggregation consistency across repeated scans
+ val firstSum =
spark.read.parquet(path).selectExpr("sum(id)").collect()(0).getLong(0)
+ for (i <- 1 to 3) {
+ val sum =
spark.read.parquet(path).selectExpr("sum(id)").collect()(0).getLong(0)
+ assert(
+ sum == firstSum,
+ s"Iteration $i: sum mismatch, expected $firstSum but got $sum")
+ }
+ }
+ }
+
+ testWithSpecifiedSparkVersion(
+ "many small files do not cause errors with file handle cache",
+ "3.5",
+ "3.5") {
+ // Verify that scanning many small files with caching enabled does not
cause
+ // file descriptor exhaustion or other resource-related errors.
+ withTempPath {
+ dir =>
+ // Create 200 small parquet files
+ spark
+ .range(20000)
+ .selectExpr("id", "uuid() as payload")
+ .repartition(200)
+ .write
+ .parquet(dir.getCanonicalPath)
+
+ val fileCount = dir.listFiles().count(_.getName.endsWith(".parquet"))
+ assert(fileCount >= 200, s"Expected at least 200 files, got
$fileCount")
+
+ // Verify scans go through Gluten/Velox
+
checkGlutenPlan[BasicScanExecTransformer](spark.read.parquet(dir.getCanonicalPath))
+
+ // Scan all files - should work without resource errors
+ val count = spark.read.parquet(dir.getCanonicalPath).count()
+ assert(count == 20000)
+
+ // Scan again (cache hit path) - should also work
+ val count2 = spark.read.parquet(dir.getCanonicalPath).count()
+ assert(count2 == 20000)
+ }
+ }
+
+ testWithSpecifiedSparkVersion(
+ "filtered scan correctness with file handle cache",
+ "3.5",
+ "3.5") {
+ // Verify that predicate pushdown works correctly with cached file handles.
+ // This exercises the row group skipping path through cached handles.
+ withTempPath {
+ dir =>
+ spark
+ .range(100000)
+ .selectExpr(
+ "id",
+ "cast(id % 10 as int) as partition_key",
+ "cast(id * 0.01 as double) as metric")
+ .repartition(20)
+ .write
+ .parquet(dir.getCanonicalPath)
+
+ val path = dir.getCanonicalPath
+
+ // Verify scans go through Gluten/Velox
+ checkGlutenPlan[BasicScanExecTransformer](
+ spark.read.parquet(path).where("partition_key = 5"))
+
+ // Filter that matches ~10% of rows
+ val filtered = spark.read.parquet(path).where("partition_key =
5").count()
+ assert(filtered == 10000, s"Expected 10000 filtered rows, got
$filtered")
+
+ // Range filter
+ val rangeFiltered = spark.read.parquet(path).where("id >=
50000").count()
+ assert(rangeFiltered == 50000, s"Expected 50000 range-filtered rows,
got $rangeFiltered")
+
+ // Re-run same filters (cache hit path)
Review Comment:
Already fixed in the previous commit — removed "cache hit path" wording and
replaced with "results must remain consistent".
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
