iemejia commented on code in PR #12400:
URL: https://github.com/apache/gluten/pull/12400#discussion_r3630595489
##########
backends-velox/src/main/scala/org/apache/gluten/config/VeloxConfig.scala:
##########
@@ -534,10 +535,35 @@ object VeloxConfig extends ConfigRegistry {
val COLUMNAR_VELOX_FILE_HANDLE_CACHE_ENABLED =
buildStaticConf("spark.gluten.sql.columnar.backend.velox.fileHandleCacheEnabled")
.doc(
- "Disables caching if false. File handle cache should be disabled " +
- "if files are mutable, i.e. file content may change while file path
stays the same.")
+ "Enables caching of file handles to avoid repeated open/close overhead
on remote " +
+ "filesystems. Should be disabled if files are mutable, i.e. file
content may " +
+ "change while file path stays the same.")
Review Comment:
Good catch. Broadened the docstring to make clear the cache benefits both
local filesystems (fewer open/close syscalls and file descriptor churn) and
remote filesystems/object stores (reused connection state). Regenerated
`docs/velox-configuration.md` accordingly.
##########
docs/velox-configuration.md:
##########
@@ -30,7 +30,8 @@ nav_order: 16
| spark.gluten.sql.columnar.backend.velox.directorySizeGuess
| ⚓ Static | 32KB | Deprecated, rename to
spark.gluten.sql.columnar.backend.velox.footerEstimatedSize
|
| spark.gluten.sql.columnar.backend.velox.driverSideBroadcastHashTableBuild
| 🔄 Dynamic | false | Enable driver-side broadcast hash
table build. When enabled, the hash table is built and serialized on the
driver, then broadcast to executors. When disabled, each executor builds its
own hash table from the broadcast data.
|
| spark.gluten.sql.columnar.backend.velox.enableTimestampNtzValidation
| 🔄 Dynamic | false | Enable validation fallback for
TimestampNTZ type. When true, any plan containing TimestampNTZ will fall back
to Spark execution. When false, allows native execution for TimestampNTZ scan.
|
-| spark.gluten.sql.columnar.backend.velox.fileHandleCacheEnabled
| ⚓ Static | false | Disables caching if false. File
handle cache should be disabled if files are mutable, i.e. file content may
change while file path stays the same.
|
+| spark.gluten.sql.columnar.backend.velox.fileHandleCacheEnabled
| ⚓ Static | true | Enables caching of file handles to
avoid repeated open/close overhead on remote filesystems. Should be disabled if
files are mutable, i.e. file content may change while file path stays the same.
|
+| spark.gluten.sql.columnar.backend.velox.fileHandleExpirationDurationMs
| ⚓ Static | 600000ms | Expiration time in milliseconds for
cached file handles. Handles not accessed within this duration are evicted from
the cache. This prevents stale handles from accumulating (e.g., expired HDFS
leases, closed remote connections). A value of 0 disables TTL-based eviction.
|
Review Comment:
Changed the default to a Spark-style duration literal
(`createWithDefaultString("10m")`), so the docs now render `10m` and the
docstring explicitly states both a duration string (`"10m"`, `"600s"`) and a
plain millisecond number are accepted.
On the cross-parsing concern: values are honored consistently.
`GlutenConfigUtil.parseConfig` reads every `spark.gluten.*` entry through its
`ConfigEntry` and emits the parsed value's `toString`, so any `timeConf` input
is normalized to plain milliseconds (`600000`) before it reaches native —
`ConfigExtractor.cc` always receives a numeric string for its `int64` parse.
##########
backends-velox/src/test/scala/org/apache/spark/sql/execution/VeloxFileHandleCacheSuite.scala:
##########
@@ -0,0 +1,324 @@
+/*
+ * 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
+
+import java.io.FileNotFoundException
+import java.nio.file.NoSuchFileException
+
+/**
+ * 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"
+
+ // TTL for file handle cache eviction (used in sparkConf and sleep
calculations).
+ // Kept small to minimize CI time; the TTL test only asserts scan
correctness after
+ // the window elapses (it passes whether or not eviction has occurred), so a
short
+ // wait is sufficient and does not introduce flakiness.
+ private val ttlMs = 500
+ private val ttlWaitMs = ttlMs + 500 // TTL + buffer for lazy eviction on
next access
+
+ /** Walks the exception cause chain looking for an instance of the given
type. */
+ private def hasCauseOfType(e: Throwable, cls: Class[_ <: Throwable]):
Boolean = {
+ var cause = e.getCause
+ while (cause != null) {
+ if (cls.isInstance(cause)) return true
+ cause = cause.getCause
+ }
+ false
+ }
+
+ 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,
ttlMs.toString)
+ .set(VeloxConfig.COLUMNAR_VELOX_NUM_CACHE_FILE_HANDLES.key, "10000")
+ }
+
+ test("basic scan correctness with file handle cache enabled") {
+ // 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]
+ }
+ }
+ }
+
+ test("repeated scans produce consistent results") {
+ // 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")
+ }
+ }
+ }
+
+ test("many small files do not cause errors with file handle cache") {
+ // 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 - results must remain consistent
+ val count2 = spark.read.parquet(dir.getCanonicalPath).count()
+ assert(count2 == 20000)
+ }
+ }
+
+ test("filtered scan correctness with file handle cache") {
+ // 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 - results must remain consistent
+ val filtered2 = spark.read.parquet(path).where("partition_key =
5").count()
+ assert(filtered2 == filtered, "Filtered count mismatch on repeated
scan")
+ }
+ }
+
+ test("scan after file deletion does not silently return wrong data") {
+ // If a file is deleted between scans, the next scan should either:
+ // - Succeed with the original count (cached FD keeps inode alive on Linux)
+ // - Succeed with a reduced count (deleted file not accessible)
+ // - Throw a file-not-found error
+ // The key invariant: it must NOT silently return incorrect data.
+ withTempPath {
+ dir =>
+ spark
+ .range(1000)
+ .selectExpr("id")
+ .repartition(5)
+ .write
+ .parquet(dir.getCanonicalPath)
+
+ val path = dir.getCanonicalPath
+ // First scan populates the cache
+ val count1 = spark.read.parquet(path).count()
+ assert(count1 == 1000)
+
+ // Verify scans go through Gluten/Velox
+ checkGlutenPlan[BasicScanExecTransformer](spark.read.parquet(path))
+
+ // Delete one parquet file
+ val parquetFiles =
dir.listFiles().filter(_.getName.endsWith(".parquet"))
+ assert(parquetFiles.nonEmpty)
+ val deletedFile = parquetFiles.head
+ val deletedRows =
spark.read.parquet(deletedFile.getCanonicalPath).count()
+ assert(deletedFile.delete(), s"Failed to delete
${deletedFile.getCanonicalPath}")
+
+ // On Linux, the cached FD to the deleted file may still work
(unlinked inode).
+ // Either way, the remaining files should be readable.
+ // The scan may also throw if the FS detects the missing file.
+ try {
+ val count2 = spark.read.parquet(path).count()
+ // The count should be either (count1 - deletedRows) or count1
+ // depending on whether the OS kept the inode accessible
+ assert(
+ count2 == count1 || count2 == count1 - deletedRows,
+ s"Unexpected count after deletion: $count2 (original: $count1,
deleted: $deletedRows)")
+ } catch {
+ case e: FileNotFoundException =>
+ // Direct file-not-found exception.
+ case e: NoSuchFileException =>
+ // NIO equivalent of FileNotFoundException.
+ case e: Exception
+ if hasCauseOfType(e, classOf[FileNotFoundException]) ||
+ hasCauseOfType(e, classOf[NoSuchFileException]) =>
+ // Wrapped file-not-found in the cause chain (e.g., SparkException
wrapping).
+ case e: Exception
+ if e.getMessage != null &&
+ (e.getMessage.contains("FileNotFoundException") ||
+ e.getMessage.contains("No such file") ||
+ e.getMessage.contains("Path does not exist") ||
+ e.getMessage.contains("does not exist")) =>
+ // Fallback: message-based matching for FS implementations that use
+ // custom exception types (e.g., Hadoop, Velox native errors).
+ }
+ }
+ }
+
+ test("scans remain correct after TTL expiration window") {
+ // Correctness guard: verify that scans produce correct results after the
+ // configured TTL (set in sparkConf) has elapsed and cached handles may
+ // have been evicted. This does NOT directly assert that eviction occurred
+ // (Velox exposes no JVM-visible eviction counter), but it exercises the
+ // re-open path: if a handle was evicted, the scan must transparently
+ // re-open the file and return the same data. Combined with the "scan after
+ // file deletion" test -- which proves cached handles keep the inode alive
--
+ // this gives reasonable coverage that the TTL wiring works end-to-end.
+ withTempPath {
+ dir =>
+ spark
+ .range(5000)
+ .selectExpr("id", "id * 2 as doubled")
+ .repartition(20)
+ .write
+ .parquet(dir.getCanonicalPath)
+
+ val path = dir.getCanonicalPath
+
+ // First scan populates the cache
+ val count1 = spark.read.parquet(path).count()
+ assert(count1 == 5000)
+
+ // Verify scans go through Gluten/Velox
+ checkGlutenPlan[BasicScanExecTransformer](spark.read.parquet(path))
+
+ val sum1 =
spark.read.parquet(path).selectExpr("sum(id)").collect()(0).getLong(0)
+
+ // Wait for TTL to expire
+ Thread.sleep(ttlWaitMs)
+
+ // Scan after TTL expiration: verify results remain correct
+ // (handles may have been evicted and transparently re-opened)
+ val count2 = spark.read.parquet(path).count()
+ assert(count2 == 5000, s"Count mismatch after TTL expiration: expected
5000, got $count2")
+ val sum2 =
spark.read.parquet(path).selectExpr("sum(id)").collect()(0).getLong(0)
+ assert(sum2 == sum1, s"Sum mismatch after TTL expiration: expected
$sum1, got $sum2")
+ }
+ }
Review Comment:
Added a best-effort eviction probe to this test: after the TTL window, it
deletes a cached file, waits past the TTL again, and re-scans. This drives the
"cached handle expired → re-open" path for a file that no longer exists.
I intentionally kept the assertion tolerant rather than requiring the count
to drop. Velox exposes no JVM-visible eviction counter, and on Linux a
still-cached FD keeps the unlinked inode readable, so requiring a reduced count
would be platform/timing dependent and flaky. The assertion enforces the
invariant that must always hold: the post-TTL scan returns either the full
count, a reduced count, or a file-not-found error — never silently corrupted
data.
##########
docs/velox-configuration.md:
##########
@@ -72,7 +74,7 @@ nav_order: 16
| spark.gluten.sql.columnar.backend.velox.showTaskMetricsWhenFinished
| 🔄 Dynamic | false | Show velox full task metrics when
finished.
|
| spark.gluten.sql.columnar.backend.velox.spillFileSystem
| 🔄 Dynamic | local | The filesystem used to store spill
data. local: The local file system. heap-over-local: Write file to JVM heap if
having extra heap space. Otherwise write to local file system.
|
| spark.gluten.sql.columnar.backend.velox.spillStrategy
| 🔄 Dynamic | auto | none: Disable spill on Velox backend;
auto: Let Spark memory manager manage Velox's spilling
|
-| spark.gluten.sql.columnar.backend.velox.ssdCacheIOThreads
| ⚓ Static | 1 | The IO threads for cache promoting
|
+| spark.gluten.sql.columnar.backend.velox.ssdCacheIOThreads
| ⚓ Static | 4 | The number of IO threads for SSD
cache read/write operations
|
Review Comment:
Already addressed — `docs/get-started/VeloxLocalCache.md` was updated in
this PR to `default is 4` with the new description ("the number of IO threads
for SSD cache read/write operations"), so it's consistent with this table.
##########
backends-velox/src/test/scala/org/apache/spark/sql/execution/VeloxFileHandleCacheSuite.scala:
##########
@@ -0,0 +1,250 @@
+/*
+ * 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,
"600000")
+ .set(VeloxConfig.COLUMNAR_VELOX_NUM_CACHE_FILE_HANDLES.key, "10000")
+ }
Review Comment:
Already addressed — the suite now includes the `"scans remain correct after
TTL expiration window"` test, which sets a short TTL, waits past it, and
exercises the eviction/re-open path end-to-end (now also with a best-effort
deletion probe after the TTL window).
--
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]
