This page is generated by
RapidsConf.helpAdvancedagainst Apache Spark 3.3.0. Most of the content in this page is applicable to other versions of Apache Spark with variations
NVIDIA cuDF plugin for Apache Spark Advanced Configuration
Most users will not need to modify the configuration options listed below. They are documented here for completeness and advanced usage.
The following configuration options are supported by the cuDF plugin.
The “Since Version” column shows the first cuDF plugin release known to support the config. “Unreleased” means the config exists on this branch but is not in a tagged release yet. “Unknown” means the metadata has not been recorded yet.
For commonly used configurations and examples of setting options, please refer to the NVIDIA cuDF plugin for Apache Spark Configuration page.
Advanced Configuration
| Name | Description | Default Value | Applicable at | Since Version |
|---|---|---|---|---|
| spark.rapids.cgroups.memory.limit.path | The filepath of the local file on host that stores the memory limit for the process. If omitted, attempts to detect the file from common locations. | None | Startup | 25.08.0 |
| spark.rapids.cgroups.memory.usage.path | The filepath of the local file on host that stores the memory usage for the process. If omitted, attempts to detect the file from common locations. | None | Startup | 25.08.0 |
| spark.rapids.filecache.allowPathRegexp | A regular expression to decide which paths will be cached when the file cache is enabled. If this is not set, then all paths are allowed to cache. If a path is allowed by this regexp but blocked by spark.rapids.filecache.blockPathRegexp, then the path is blocked to cache. | None | Startup | 23.08.1 |
| spark.rapids.filecache.blockPathRegexp | A regular expression to decide which paths will not be cached when the file cache is enabled. If a path is blocked by this regexp but is allowed by spark.rapids.filecache.allowPathRegexp, then the path is blocked. | None | Startup | 23.08.1 |
| spark.rapids.filecache.checkStale | Controls whether the cached is checked for being out of date with respect to the input file. When enabled, the data that has been cached locally for a file will be invalidated if the file is updated after being cached. This feature is only necessary if an input file for a Spark application can be changed during the lifetime of the application. If an individual input file will not be overwritten during the Spark application then performance may be improved by setting this to false. | true | Startup | 23.06.0 |
| spark.rapids.filecache.maxBytes | Controls the maximum amount of data that will be cached locally. If left unspecified, it will use half of the available disk space detected on startup for the configured Spark local disks. | None | Startup | 23.06.0 |
| spark.rapids.filecache.minimumFreeSpace.bytes | Specify the minimum amount of free space in the Spark local disks. When the amount of free space on the Spark local disks drops below this value, cache data will be removed automatically to free disk space. A zero or negative value will disable this feature. Note if multiple Spark applications running on the same node, or there are other applications running with heavy disk writing, the filecache may not drop caches in time and may cause full disk errors. Please increase this value for this case. | 32212254720 | Startup | 25.02.0 |
| spark.rapids.filecache.useChecksums | Whether to write out and verify checksums for the cached local files. | false | Startup | 24.02.0 |
| spark.rapids.flameGraph.asyncProfiler.options | The cuDF plugin uses the async profiler to generate flame graphs. You can specify profiler options via this property. The plugin supports all options except for the ‘file’ option listed in https://github.com/async-profiler/async-profiler/blob/b3f58429f5c0252e9ced3f0fcb444fed17671321/docs/ProfilerOptions.md#options-applicable-to-any-output-format .The default values is ‘jfr,event=cpu,wall=10ms’. | jfr,event=cpu,wall=10ms | Runtime | 25.10.0 |
| spark.rapids.flameGraph.executors | Comma-separated list of executors IDs and hyphenated ranges of executor IDs to profile when async-profiler (for flame graph) is enabled. The default value ‘*’ means all executors | * | Runtime | 25.10.0 |
| spark.rapids.flameGraph.jfr.compression | Enable compression for JFR files generated by async profiler. When enabled, JFR files will be compressed after generation to save disk space. | false | Runtime | 25.10.0 |
| spark.rapids.flameGraph.pathPrefix | Enables collecting flame graph (with async profiler) and specifies a file prefix to use when writing the JFR file by async-profiler. The async-profiler will write a flame graph file for each stage. It is strongly recommended to set ‘spark.scheduler.mode’ to ‘FIFO’ so that there is a clean boundary between stages, and then we can better understand each stage. | None | Runtime | 25.10.0 |
| spark.rapids.flameGraph.stageEpochInterval | Interval in seconds to determine the current stage epoch based on running task counts. The profiler will check which stage has the most running tasks and profile that stage during each epoch. This allows profiling when multiple stages run concurrently even if FIFO scheduling is already chosen. | 5 | Runtime | 25.10.0 |
| spark.rapids.gpu.resourceName | The name of the Spark resource that represents a GPU that you want the plugin to use if using custom resources with Spark. | gpu | Startup | 21.12.0 |
| spark.rapids.iceberg.s3.async.api-call-timeout-ms | Optional maximum total time (ms) for an Iceberg S3 byte-range request, including retries. This is a wall-clock limit for the complete streaming response, so it can abort a healthy but slow transfer. Zero (the default) disables the limit. | 0 | Startup | Unreleased |
| spark.rapids.iceberg.s3.async.connection-max-idle-time-ms | Connection-max-idle-time (ms) for the async HTTP client used by the cuDF plugin Iceberg S3 byte-range reader. No equivalent Iceberg property. | 300000 | Startup | 26.06.0 |
| spark.rapids.iceberg.s3.async.max-concurrency | Max concurrent connections for the async HTTP client used by the cuDF plugin Iceberg S3 byte-range reader. Used only when the Iceberg FileIO property s3.crt.max-concurrency is not set. | 200 | Startup | 26.06.0 |
| spark.rapids.iceberg.s3.async.socket-timeout-ms | Maximum time (ms) that an Iceberg S3 response may stop making progress. Applied as the Netty read timeout or the CRT connection-health timeout. The minimum is 2000 ms because that is the minimum supported by CRT. | 30000 | Startup | Unreleased |
| spark.rapids.iceberg.s3.async.tcp-keepalive-interval-ms | TCP keep-alive probe interval (ms) for the CRT async HTTP client used by the cuDF plugin Iceberg S3 byte-range reader. No equivalent Iceberg property. | 60000 | Startup | 26.06.0 |
| spark.rapids.iceberg.s3.async.tcp-keepalive-timeout-ms | TCP keep-alive probe timeout (ms) for the CRT async HTTP client used by the cuDF plugin Iceberg S3 byte-range reader. No equivalent Iceberg property. | 30000 | Startup | 26.06.0 |
| spark.rapids.memory.gpu.allocFraction | The fraction of available (free) GPU memory that should be allocated for pooled memory. This must be less than or equal to the maximum limit configured via spark.rapids.memory.gpu.maxAllocFraction, and greater than or equal to the minimum limit configured via spark.rapids.memory.gpu.minAllocFraction. | 1.0 | Startup | 0.1.0 |
| spark.rapids.memory.gpu.debug | Provides a log of GPU memory allocations and frees. If set to STDOUT or STDERR the logging will go there. Setting it to NONE disables logging. All other values are reserved for possible future expansion and in the mean time will disable logging. | NONE | Startup | 0.1.0 |
| spark.rapids.memory.gpu.oomDumpDir | The path to a local directory where a heap dump will be created if the GPU encounters an unrecoverable out-of-memory (OOM) error. The filename will be of the form: “gpu-oom- | None | Startup | 21.06.0 |
| spark.rapids.memory.gpu.pool | Select the RMM pooling allocator to use. Valid values are “DEFAULT”, “ARENA”, “ASYNC”, and “NONE”. With “DEFAULT”, the RMM pool allocator is used; with “ARENA”, the RMM arena allocator is used; with “ASYNC”, the new CUDA stream-ordered memory allocator in CUDA 11.2+ is used. If set to “NONE”, pooling is disabled and RMM just passes through to CUDA memory allocation directly. | ASYNC | Startup | 0.1.0 |
| spark.rapids.memory.gpu.reserve | The amount of GPU memory that should remain unallocated by RMM and left for system use such as memory needed for kernels and kernel launches. | 671088640 | Startup | 21.06.0 |
| spark.rapids.memory.gpu.state.debug | To better recover from out of memory errors, RMM will track several states for the threads that interact with the GPU. This provides a log of those state transitions to aid in debugging it. STDOUT or STDERR will have the logging go there empty string will disable logging and anything else will be treated as a file to write the logs to. | Startup | 23.04.0 | |
| spark.rapids.memory.gpu.unspill.enabled | When a spilled GPU buffer is needed again, should it be unspilled, or only copied back into GPU memory temporarily. Unspilling may be useful for GPU buffers that are needed frequently, for example, broadcast variables; however, it may also increase GPU memory usage | false | Startup | 21.06.0 |
| spark.rapids.memory.pinnedPool.parallelInit.threads | Number of CPU threads used to initialize the pinned pool’s backing memory, capped at the number of executor cores. Set to ‘all’ to use the number of executor cores. A value of 1 initializes the backing memory using cudaHostAlloc. Values greater than 1 instead pre-touch pages concurrently before pinning for faster initialization. This does not affect subsequent suballocator behavior. Note: on multi-NUMA systems, multithreaded initialization can scatter pages across nodes if you do not constrain placement in advance. Pages cannot be migrated once pinned. | all | Startup | Unreleased |
| spark.rapids.perfio.gcs.enabled | Controls the Google Cloud Storage reader for improved performance in certain queries. When true, enables it. When false, disables it unconditionally. When unset (default), enables it on Google Cloud Dataproc when compatible connector classes are available; outside Dataproc it remains disabled. Explicit enablement fails at startup if the required classes are unavailable. | None | Startup | Unreleased |
| spark.rapids.perfio.s3.enabled | Controls the AWS S3 reader for improved performance in certain queries. When true, enables it and throws at startup if no compatible HTTP client is on the classpath. When false, disables it unconditionally. When unset (default), enables it opportunistically if a compatible HTTP client is found, otherwise falls back to S3A with a warning. The presence of AWS SDK packages for Netty and/or CRT HTTP clients on the classpath is required. You can use Spark submit option --packages software.amazon.awssdk:s3:2.22.12,software.amazon.awssdk:aws-crt-client:2.22.12 to achieve this. See https://docs.aws.amazon.com/sdk-for-java/latest/developer-guide/crt-based-s3-client.html#crt-based-s3-client-depend | None | Startup | 24.04.0 |
| spark.rapids.perfio.s3.httpClient | Selects the async HTTP client backend for the PerfIO S3 reader. Supported values: NETTY, CRT | NETTY | Startup | Unreleased |
| spark.rapids.python.concurrentPythonWorkers | Set the number of Python worker processes that can execute concurrently per GPU. Python worker processes may temporarily block when the number of concurrent Python worker processes started by the same executor exceeds this amount. Allowing too many concurrent tasks on the same GPU may lead to GPU out of memory errors. >0 means enabled, while <=0 means unlimited | 0 | Runtime | 21.06.0 |
| spark.rapids.python.memory.gpu.allocFraction | The fraction of total GPU memory that should be initially allocated for pooled memory for all the Python workers. It supposes to be less than (1 - $(spark.rapids.memory.gpu.allocFraction)), since the executor will share the GPU with its owning Python workers. Half of the rest will be used if not specified | None | Runtime | 21.06.0 |
| spark.rapids.python.memory.gpu.maxAllocFraction | The fraction of total GPU memory that limits the maximum size of the RMM pool for all the Python workers. It supposes to be less than (1 - $(spark.rapids.memory.gpu.maxAllocFraction)), since the executor will share the GPU with its owning Python workers. when setting to 0 it means no limit. | 0.0 | Runtime | 21.06.0 |
| spark.rapids.python.memory.gpu.pooling.enabled | Should RMM in Python workers act as a pooling allocator for GPU memory, or should it just pass through to CUDA memory allocation directly. When not specified, It will honor the value of config ‘spark.rapids.memory.gpu.pool’, but now only ‘DEFAULT’ and ‘NONE’ are supported. If ‘ASYNC’ or ‘ARENA’ is specified, it will fall back to ‘DEFAULT’. | None | Runtime | 21.06.0 |
| spark.rapids.shuffle.enabled | Enable or disable the RAPIDS Shuffle Manager implementation at runtime. On supported Spark versions, including Spark 4.0.0 and later, the RAPIDS Shuffle Manager is configured automatically unless spark.shuffle.manager is explicitly set. Note: Databricks runtimes may explicitly set spark.shuffle.manager, preventing automatic configuration from taking effect. To override it, explicitly set spark.shuffle.manager to the RAPIDS Shuffle Manager class. This automatic configuration is skipped on Dataproc runtimes that set spark.dataproc.engine, including Lightning Engine runtimes; on those runtimes, spark.shuffle.manager remains unset unless explicitly configured. On earlier Spark versions, the RAPIDS Shuffle Manager must already be configured. When set to false, the built-in Spark shuffle implementation will be used. | true | Runtime | 21.10.0 |
| spark.rapids.shuffle.mode | RAPIDS Shuffle Manager mode. “MULTITHREADED”: shuffle file writes and reads are parallelized using a thread pool. “UCX”: (requires UCX installation) uses accelerated transports for transferring shuffle blocks. “CACHE_ONLY”: use when running a single executor, for short-circuit cached shuffle (for testing purposes). | MULTITHREADED | Startup | 22.08.0 |
| spark.rapids.shuffle.multiThreaded.maxBytesInFlight | The size limit, in bytes, that the RAPIDS shuffle manager configured in “MULTITHREADED” mode will allow to be serialized or deserialized concurrently per task. This is also the maximum amount of memory that will be used per task. This should be set larger than Spark’s default maxBytesInFlight (48MB). The larger this setting is, the more compressed shuffle chunks are processed concurrently. In practice, care needs to be taken to not go over the amount of off-heap memory that Netty has available. See https://github.com/NVIDIA/cudf-spark/issues/9153. | 134217728 | Startup | 22.10.0 |
| spark.rapids.shuffle.multiThreaded.reader.threads | The number of threads to use for reading shuffle blocks per executor in the RAPIDS shuffle manager configured in “MULTITHREADED” mode. There are two special values: 0 = feature is disabled, falls back to Spark built-in shuffle reader; 1 = our implementation of Spark’s built-in shuffle reader with extra metrics. | 20 | Startup | 22.10.0 |
| spark.rapids.shuffle.multiThreaded.writer.threads | The number of threads to use for writing shuffle blocks per executor in the RAPIDS shuffle manager configured in “MULTITHREADED” mode. There are two special values: 0 = feature is disabled, falls back to Spark built-in shuffle writer; 1 = our implementation of Spark’s built-in shuffle writer with extra metrics. | 20 | Startup | 22.08.0 |
| spark.rapids.shuffle.multithreaded.skipMerge | When using MULTITHREADED shuffle mode, skip merging partial shuffle files and instead serve data directly from the MultithreadedShuffleBufferCatalog. This avoids I/O overhead from merging but requires: (1) External Shuffle Service (ESS) to be disabled, and (2) spark.rapids.memory.host.offHeapLimit.enabled=true (off-heap memory limits enabled) to prevent OOM from unbounded buffer growth. When set to false (default), partial files will be merged into a single shuffle file per map task as in standard Spark shuffle. Set to true when both requirements are met and shuffle data is not reused across SQL queries (e.g., avoid on Databricks with shuffle reuse enabled). | false | Startup | 26.06.0 |
| spark.rapids.shuffle.transport.earlyStart | Enable early connection establishment for RAPIDS Shuffle | true | Startup | 21.06.0 |
| spark.rapids.shuffle.transport.earlyStart.heartbeatInterval | Shuffle early start heartbeat interval (milliseconds). Executors will send a heartbeat RPC message to the driver at this interval | 5000 | Startup | 21.06.0 |
| spark.rapids.shuffle.transport.earlyStart.heartbeatTimeout | Shuffle early start heartbeat timeout (milliseconds). Executors that don’t heartbeat within this timeout will be considered stale. This timeout must be higher than the value for spark.rapids.shuffle.transport.earlyStart.heartbeatInterval | 10000 | Startup | 21.08.0 |
| spark.rapids.shuffle.transport.maxReceiveInflightBytes | Maximum aggregate amount of bytes that be fetched at any given time from peers during shuffle | 1073741824 | Startup | 21.06.0 |
| spark.rapids.shuffle.ucx.activeMessages.forceRndv | Set to true to force ‘rndv’ mode for all UCX Active Messages. This should only be required with UCX 1.10.x. UCX 1.11.x deployments should set to false. | false | Startup | 21.08.0 |
| spark.rapids.shuffle.ucx.managementServerHost | The host to be used to start the management server | null | Startup | 0.1.0 |
| spark.rapids.shuffle.ucx.useWakeup | When set to true, use UCX’s event-based progress (epoll) in order to wake up the progress thread when needed, instead of a hot loop. | true | Startup | 0.1.0 |
| spark.rapids.sql.acceleratedColumnarToRow.enabled | When set to true (default) the GPU columnar-to-row transition uses the GPU transpose kernel (AcceleratedColumnarToRowIterator) for wide fixed-width / STRING schemas. Setting it to false forces the slower per-row CPU iterator (ColumnarToRowIterator). Mainly useful for troubleshooting and performance comparisons; production workloads should leave this on. | true | Runtime | 26.08.0 |
| spark.rapids.sql.allowMultipleJars | Allow multiple rapids-4-spark, cudf-spark-jni, and cudf jars on the classpath. Spark will take the first one it finds, so the version may not be expected. Possible values are ALWAYS: allow all jars, SAME_REVISION: only allow jars with the same revision, NEVER: do not allow multiple jars at all. | SAME_REVISION | Startup | 24.02.0 |
| spark.rapids.sql.castDecimalToFloat.enabled | Casting from decimal to floating point types on the GPU returns results that have tiny difference compared to results returned from CPU. | true | Runtime | 21.10.0 |
| spark.rapids.sql.castFloatToDecimal.enabled | Casting from floating point types to decimal on the GPU returns results that have tiny difference compared to results returned from CPU. | true | Runtime | 21.06.0 |
| spark.rapids.sql.castFloatToIntegralTypes.enabled | Casting from floating point types to integral types on the GPU supports a slightly different range of values when using Spark 3.1.0 or later. Refer to the CAST documentation for more details. | true | Runtime | 21.06.0 |
| spark.rapids.sql.castFloatToString.enabled | Casting from floating point types to string on the GPU returns results that have a different precision than the default results of Spark. | true | Runtime | 21.06.0 |
| spark.rapids.sql.castStringToFloat.enabled | When set to true, enables casting from strings to float types (float, double) on the GPU. Currently hex values aren’t supported on the GPU. Also note that casting from string to float types on the GPU returns incorrect results when the string represents any number “1.7976931348623158E308” <= x < “1.7976931348623159E308” and “-1.7976931348623158E308” >= x > “-1.7976931348623159E308” in both these cases the GPU returns Double.MaxValue while CPU returns “+Infinity” and “-Infinity” respectively | true | Runtime | 21.06.0 |
| spark.rapids.sql.castStringToTimestamp.enabled | When set to false, this disables casting from string to timestamp on the GPU. | true | Runtime | 21.06.0 |
| spark.rapids.sql.coalescing.reader.numFilterParallel | This controls the number of files the coalescing reader will run in each thread when it filters blocks for reading. If this value is greater than zero the files will be filtered in a multithreaded manner where each thread filters the number of files set by this config. If this is set to zero the files are filtered serially. This uses the same thread pool as the multithreaded reader, see spark.rapids.sql.multiThreadedRead.numThreads. | 0 | Runtime | 22.10.0 |
| spark.rapids.sql.concurrentGpuTasks | Set the initial number of tasks that can execute concurrently per GPU. By default the number of tasks allowed on the GPU will adjust dynamically to try and provide optimal performance. This sets the starting point for each stage. If this is not set the amount of GPU memory will be used to come up with a starting estimate. | None | Runtime | 0.1.0 |
| spark.rapids.sql.concurrentGpuTasks.dynamic | Set to false if the system should not dynamically adjust the concurrent task amount, but keep it to be a static number | true | Runtime | 25.06.0 |
| spark.rapids.sql.concurrentWriterPartitionFlushSize | The flush size of the concurrent writer cache in bytes for each partition. If specified spark.sql.maxConcurrentOutputFileWriters, use concurrent writer to write data. Concurrent writer first caches data for each partition and begins to flush the data if it finds one partition with a size that is greater than or equal to this config. The default value is 0, which will try to select a size based off of file type specific configs. E.g.: It uses write.parquet.row-group-size-bytes config for Parquet type and orc.stripe.size config for Orc type. If the value is greater than 0, will use this positive value.Max value may get better performance but not always, because concurrent writer uses spillable cache and big value may cause more IO swaps. | 0 | Runtime | 22.10.0 |
| spark.rapids.sql.csv.read.decimal.enabled | CSV reading is not 100% compatible when reading decimals. | false | Runtime | 22.04.0 |
| spark.rapids.sql.csv.read.double.enabled | CSV reading is not 100% compatible when reading doubles. | true | Runtime | 21.06.0 |
| spark.rapids.sql.csv.read.float.enabled | CSV reading is not 100% compatible when reading floats. | true | Runtime | 21.06.0 |
| spark.rapids.sql.decimalOverflowGuarantees | FOR TESTING ONLY. DO NOT USE IN PRODUCTION. Please see the decimal section of the compatibility documents for more information on this config. | true | Runtime | 21.12.0 |
| spark.rapids.sql.delta.lowShuffleMerge.deletionVector.broadcast.threshold | Currently we need to broadcast deletion vector to all executors to perform low shuffle merge. When we detect the deletion vector broadcast size is larger than this value, we will fallback to normal shuffle merge. | 20971520 | Runtime | 24.08.1 |
| spark.rapids.sql.delta.lowShuffleMerge.enabled | Option to turn on the low shuffle merge for Delta Lake. Currently there are some limitations for this feature: 1. We support Delta Lake 2.4 and Databricks Runtime 17.3. 2. The file scan mode must be set to PERFILE. 3. The deletion vector size must be smaller than spark.rapids.sql.delta.lowShuffleMerge.deletionVector.broadcast.threshold | false | Runtime | 24.08.1 |
| spark.rapids.sql.detectDeltaCheckpointQueries | Queries against Delta Lake _delta_log checkpoint Parquet files are not efficient on the GPU. When this option is enabled, the plugin will attempt to detect these queries and fall back to the CPU. | true | Runtime | 22.12.0 |
| spark.rapids.sql.detectDeltaLogQueries | Queries against Delta Lake _delta_log JSON files are not efficient on the GPU. When this option is enabled, the plugin will attempt to detect these queries and fall back to the CPU. | true | Runtime | 22.08.0 |
| spark.rapids.sql.exec.opTimeTrackingRDD.enabled | Enable OpTimeTrackingRDD for all GPU operations. When true, OpTimeTrackingRDD wrappers will be created to track operation time. When false, can improve performance by avoiding overhead of operation time tracking. | true | Runtime | 25.10.0 |
| spark.rapids.sql.expression.cpuBridge.enabled | Enable CPU-GPU bridge expressions that allow CPU expression subtrees to run while keeping the overall plan on GPU. When enabled, expressions that have no GPU implementation will automatically be wrapped in bridge expressions instead of causing plan fallbacks. | true | Runtime | 26.06.0 |
| spark.rapids.sql.fast.sample | Option to turn on fast sample. If enable it is inconsistent with CPU sample because of GPU sample algorithm is inconsistent with CPU. | false | Runtime | 22.02.0 |
| spark.rapids.sql.format.avro.enabled | When set to true enables all avro input and output acceleration. (only input is currently supported anyways) | false | Runtime | 22.04.0 |
| spark.rapids.sql.format.avro.multiThreadedRead.maxNumFilesParallel | A limit on the maximum number of files per task processed in parallel on the CPU side before the file is sent to the GPU. This affects the amount of host memory used when reading the files in parallel. Used with MULTITHREADED reader, see spark.rapids.sql.format.avro.reader.type. | 2147483647 | Runtime | 22.06.0 |
| spark.rapids.sql.format.avro.multiThreadedRead.numThreads | The maximum number of threads, on one executor, to use for reading small Avro files in parallel. This can not be changed at runtime after the executor has started. Used with MULTITHREADED reader, see spark.rapids.sql.format.avro.reader.type. DEPRECATED: use spark.rapids.sql.multiThreadedRead.numThreads | None | Startup | 22.06.0 |
| spark.rapids.sql.format.avro.read.enabled | When set to true enables avro input acceleration | false | Runtime | 22.04.0 |
| spark.rapids.sql.format.avro.reader.type | Sets the Avro reader type. We support different types that are optimized for different environments. The original Spark style reader can be selected by setting this to PERFILE which individually reads and copies files to the GPU. Loading many small files individually has high overhead, and using either COALESCING or MULTITHREADED is recommended instead. The COALESCING reader is good when using a local file system where the executors are on the same nodes or close to the nodes the data is being read on. This reader coalesces all the files assigned to a task into a single host buffer before sending it down to the GPU. It copies blocks from a single file into a host buffer in separate threads in parallel, see spark.rapids.sql.multiThreadedRead.numThreads. MULTITHREADED is good for cloud environments where you are reading from a blobstore that is totally separate and likely has a higher I/O read cost. Many times the cloud environments also get better throughput when you have multiple readers in parallel. This reader uses multiple threads to read each file in parallel and each file is sent to the GPU separately. This allows the CPU to keep reading while GPU is also doing work. See spark.rapids.sql.multiThreadedRead.numThreads and spark.rapids.sql.format.avro.multiThreadedRead.maxNumFilesParallel to control the number of threads and amount of memory used. By default this is set to AUTO so we select the reader we think is best. This will either be the COALESCING or the MULTITHREADED based on whether we think the file is in the cloud. See spark.rapids.cloudSchemes. | AUTO | Runtime | 22.06.0 |
| spark.rapids.sql.format.csv.enabled | When set to false disables all csv input and output acceleration. (only input is currently supported anyways) | true | Runtime | 21.06.0 |
| spark.rapids.sql.format.csv.read.enabled | When set to false disables csv input acceleration | true | Runtime | 21.06.0 |
| spark.rapids.sql.format.delta.write.enabled | When set to false disables Delta Lake output acceleration. | true | Runtime | 22.12.0 |
| spark.rapids.sql.format.hive.text.enabled | When set to false disables Hive text table acceleration. Array/Struct/Map columns are unsupported for acceleration. | true | Runtime | 22.12.0 |
| spark.rapids.sql.format.hive.text.read.decimal.enabled | Hive text file reading is not 100% compatible when reading decimals. Hive has more limitations on what is valid compared to the GPU implementation in some corner cases. See https://github.com/NVIDIA/cudf-spark/issues/7246 | true | Runtime | 23.02.0 |
| spark.rapids.sql.format.hive.text.read.double.enabled | Hive text file reading is not 100% compatible when reading doubles. | true | Runtime | 23.02.0 |
| spark.rapids.sql.format.hive.text.read.enabled | When set to false disables Hive text table read acceleration. Array/Struct/Map columns are unsupported for read acceleration. | true | Runtime | 22.12.0 |
| spark.rapids.sql.format.hive.text.read.float.enabled | Hive text file reading is not 100% compatible when reading floats. | true | Runtime | 23.02.0 |
| spark.rapids.sql.format.hive.text.write.enabled | When set to false disables Hive text table write acceleration. Array/Struct/Map columns are unsupported for write acceleration. | true | Runtime | 23.02.0 |
| spark.rapids.sql.format.iceberg.deletionVector.crcCheck.enabled | When set to false, skips validation of the CRC-32 checksum of each Iceberg deletion vector. Disabling validation avoids an additional CPU pass over the deletion-vector data, but may allow corrupted deletion vectors to be read. | true | Runtime | Unreleased |
| spark.rapids.sql.format.iceberg.enabled | When set to false disables all Iceberg acceleration | true | Runtime | 22.08.0 |
| spark.rapids.sql.format.iceberg.read.enabled | When set to false disables Iceberg input acceleration | true | Runtime | 22.08.0 |
| spark.rapids.sql.format.iceberg.write.enabled | When set to false disables Iceberg write acceleration | true | Runtime | 25.10.0 |
| spark.rapids.sql.format.json.enabled | When set to true enables all json input and output acceleration. (only input is currently supported anyways) | true | Runtime | 22.02.0 |
| spark.rapids.sql.format.json.read.enabled | When set to true enables json input acceleration | true | Runtime | 22.02.0 |
| spark.rapids.sql.format.orc.enabled | When set to false disables all orc input and output acceleration | true | Runtime | 21.06.0 |
| spark.rapids.sql.format.orc.floatTypesToString.enable | When reading an ORC file, the source data schemas(schemas of ORC file) may differ from the target schemas (schemas of the reader), we need to handle the castings from source type to target type. Since float/double numbers in GPU have different precision with CPU, when casting float/double to string, the result of GPU is different from result of CPU spark. Its default value is true (this means the strings result will differ from result of CPU). If it’s set false explicitly and there exists casting from float/double to string in the job, then such behavior will cause an exception, and the job will fail. | true | Runtime | 22.10.0 |
| spark.rapids.sql.format.orc.multiThreadedRead.maxNumFilesParallel | A limit on the maximum number of files per task processed in parallel on the CPU side before the file is sent to the GPU. This affects the amount of host memory used when reading the files in parallel. Used with MULTITHREADED reader, see spark.rapids.sql.format.orc.reader.type. | 2147483647 | Runtime | 21.08.0 |
| spark.rapids.sql.format.orc.multiThreadedRead.numThreads | The maximum number of threads, on the executor, to use for reading small ORC files in parallel. This can not be changed at runtime after the executor has started. Used with MULTITHREADED reader, see spark.rapids.sql.format.orc.reader.type. DEPRECATED: use spark.rapids.sql.multiThreadedRead.numThreads | None | Startup | 21.08.0 |
| spark.rapids.sql.format.orc.read.enabled | When set to false disables orc input acceleration | true | Runtime | 21.06.0 |
| spark.rapids.sql.format.orc.reader.type | Sets the ORC reader type. We support different types that are optimized for different environments. The original Spark style reader can be selected by setting this to PERFILE which individually reads and copies files to the GPU. Loading many small files individually has high overhead, and using either COALESCING or MULTITHREADED is recommended instead. The COALESCING reader is good when using a local file system where the executors are on the same nodes or close to the nodes the data is being read on. This reader coalesces all the files assigned to a task into a single host buffer before sending it down to the GPU. It copies blocks from a single file into a host buffer in separate threads in parallel, see spark.rapids.sql.multiThreadedRead.numThreads. MULTITHREADED is good for cloud environments where you are reading from a blobstore that is totally separate and likely has a higher I/O read cost. Many times the cloud environments also get better throughput when you have multiple readers in parallel. This reader uses multiple threads to read each file in parallel and each file is sent to the GPU separately. This allows the CPU to keep reading while GPU is also doing work. See spark.rapids.sql.multiThreadedRead.numThreads and spark.rapids.sql.format.orc.multiThreadedRead.maxNumFilesParallel to control the number of threads and amount of memory used. By default this is set to AUTO so we select the reader we think is best. This will either be the COALESCING or the MULTITHREADED based on whether we think the file is in the cloud. See spark.rapids.cloudSchemes. | AUTO | Runtime | 21.08.0 |
| spark.rapids.sql.format.orc.write.dateType.enabled | When set to true, allows GPU ORC writes containing DATE columns, including nested DATE columns. The cuDF ORC writer does not record calendar metadata, so this is safe only when every DATE value written is on or after 1582-10-15. The plugin does not validate the values; the application must guarantee this range. By default, these writes fall back to the CPU writer. | false | Runtime | 26.10.0 |
| spark.rapids.sql.format.orc.write.enabled | When set to false disables orc output acceleration | true | Runtime | 21.06.0 |
| spark.rapids.sql.format.parquet.enabled | When set to false disables all parquet input and output acceleration | true | Runtime | 21.06.0 |
| spark.rapids.sql.format.parquet.multiThreadedRead.maxNumFilesParallel | A limit on the maximum number of files per task processed in parallel on the CPU side before the file is sent to the GPU. This affects the amount of host memory used when reading the files in parallel. Used with MULTITHREADED reader, see spark.rapids.sql.format.parquet.reader.type. | 2147483647 | Runtime | 21.06.0 |
| spark.rapids.sql.format.parquet.multiThreadedRead.numThreads | The maximum number of threads, on the executor, to use for reading small Parquet files in parallel. This can not be changed at runtime after the executor has started. Used with COALESCING and MULTITHREADED reader, see spark.rapids.sql.format.parquet.reader.type. DEPRECATED: use spark.rapids.sql.multiThreadedRead.numThreads | None | Startup | 21.06.0 |
| spark.rapids.sql.format.parquet.multithreaded.combine.sizeBytes | The target size in bytes to combine multiple small files together when using the MULTITHREADED parquet reader. With combine disabled, the MULTITHREADED reader reads the files in parallel and sends individual files down to the GPU, but that can be inefficient for small files. When combine is enabled, files that are ready within spark.rapids.sql.format.parquet.multithreaded.combine.waitTime together, up to this threshold size, are combined before sending down to GPU. This can be disabled by setting it to 0. Note that combine also will not go over the spark.rapids.sql.reader.batchSizeRows or spark.rapids.sql.reader.batchSizeBytes limits. DEPRECATED: use spark.rapids.sql.reader.multithreaded.combine.sizeBytes instead. | None | Runtime | 23.02.0 |
| spark.rapids.sql.format.parquet.multithreaded.combine.waitTime | When using the multithreaded parquet reader with combine mode, how long to wait, in milliseconds, for more files to finish if haven’t met the size threshold. Note that this will wait this amount of time from when the last file was available, so total wait time could be larger then this. DEPRECATED: use spark.rapids.sql.reader.multithreaded.combine.waitTime instead. | None | Runtime | 23.02.0 |
| spark.rapids.sql.format.parquet.multithreaded.read.keepOrder | When using the MULTITHREADED reader, if this is set to true we read the files in the same order Spark does, otherwise the order may not be the same. DEPRECATED: use spark.rapids.sql.reader.multithreaded.read.keepOrder instead. | None | Runtime | 23.02.0 |
| spark.rapids.sql.format.parquet.read.enabled | When set to false disables parquet input acceleration | true | Runtime | 21.06.0 |
| spark.rapids.sql.format.parquet.reader.footer.type | In some cases reading the footer of the file is very expensive. Typically this happens when there are a large number of columns and relatively few of them are being read on a large number of files. This provides the ability to use a different path to parse and filter the footer. AUTO is the default and decides which path to take using a heuristic. JAVA follows closely with what Apache Spark does. NATIVE will parse and filter the footer using C++. | AUTO | Runtime | 22.06.0 |
| spark.rapids.sql.format.parquet.reader.type | Sets the Parquet reader type. We support different types that are optimized for different environments. The original Spark style reader can be selected by setting this to PERFILE which individually reads and copies files to the GPU. Loading many small files individually has high overhead, and using either COALESCING or MULTITHREADED is recommended instead. The COALESCING reader is good when using a local file system where the executors are on the same nodes or close to the nodes the data is being read on. This reader coalesces all the files assigned to a task into a single host buffer before sending it down to the GPU. It copies blocks from a single file into a host buffer in separate threads in parallel, see spark.rapids.sql.multiThreadedRead.numThreads. MULTITHREADED is good for cloud environments where you are reading from a blobstore that is totally separate and likely has a higher I/O read cost. Many times the cloud environments also get better throughput when you have multiple readers in parallel. This reader uses multiple threads to read each file in parallel and each file is sent to the GPU separately. This allows the CPU to keep reading while GPU is also doing work. See spark.rapids.sql.multiThreadedRead.numThreads and spark.rapids.sql.format.parquet.multiThreadedRead.maxNumFilesParallel to control the number of threads and amount of memory used. By default this is set to AUTO so we select the reader we think is best. This will either be the COALESCING or the MULTITHREADED based on whether we think the file is in the cloud. See spark.rapids.cloudSchemes. | AUTO | Runtime | 21.06.0 |
| spark.rapids.sql.format.parquet.write.enabled | When set to false disables parquet output acceleration | true | Runtime | 21.06.0 |
| spark.rapids.sql.format.parquet.writer.int96.enabled | When set to false, disables accelerated parquet write if the spark.sql.parquet.outputTimestampType is set to INT96 | true | Runtime | 21.06.0 |
| spark.rapids.sql.formatNumberFloat.enabled | format_number with floating point types on the GPU returns results that have a different precision than the default results of Spark. | true | Runtime | 23.10.0 |
| spark.rapids.sql.hasExtendedYearValues | Spark 3.2.0+ extended parsing of years in dates and timestamps to support the full range of possible values. Prior to this it was limited to a positive 4 digit year. The Accelerator does not support the extended range yet. This config indicates if your data includes this extended range or not, or if you don’t care about getting the correct values on values with the extended range. | true | Runtime | 21.10.0 |
| spark.rapids.sql.hashOptimizeSort.enabled | Whether sorts should be inserted after some hashed operations to improve output ordering. This can improve output file sizes when saving to columnar formats. | false | Runtime | 0.1.0 |
| spark.rapids.sql.improvedFloatOps.enabled | For some floating point operations spark uses one way to compute the value and the underlying cudf implementation can use an improved algorithm. In some cases this can result in cudf producing an answer when spark overflows. | true | Runtime | 21.06.0 |
| spark.rapids.sql.incompatibleDateFormats.enabled | When parsing strings as dates and timestamps in functions like unix_timestamp, some formats are fully supported on the GPU and some are unsupported and will fall back to the CPU. Some formats behave differently on the GPU than the CPU. Spark on the CPU interprets date formats with unsupported trailing characters as nulls, while Spark on the GPU will parse the date with invalid trailing characters. More detail can be found at parsing strings as dates or timestamps. | false | Runtime | 21.06.0 |
| spark.rapids.sql.incompatibleOps.enabled | For operations that work, but are not 100% compatible with the Spark equivalent set if they should be enabled by default or disabled by default. | true | Runtime | 0.1.0 |
| spark.rapids.sql.join.cross.enabled | When set to true cross joins are enabled on the GPU | true | Runtime | 21.06.0 |
| spark.rapids.sql.join.existence.enabled | When set to true existence joins are enabled on the GPU | true | Runtime | 22.04.0 |
| spark.rapids.sql.join.fullOuter.enabled | When set to true full outer joins are enabled on the GPU | true | Runtime | 21.06.0 |
| spark.rapids.sql.join.hashTable.reuse | Enable reuse of hash tables across GPU hash-join probes. Currently this supports caching broadcast hash tables. With AUTO build-side selection a heuristic is used to determine whether to use the cached broadcast-side or rebuild with the stream-side. FIXED and SMALLEST retain their configured behavior. | false | Runtime | Unreleased |
| spark.rapids.sql.join.inner.enabled | When set to true inner joins are enabled on the GPU | true | Runtime | 21.06.0 |
| spark.rapids.sql.join.leftAnti.enabled | When set to true left anti joins are enabled on the GPU | true | Runtime | 21.06.0 |
| spark.rapids.sql.join.leftOuter.enabled | When set to true left outer joins are enabled on the GPU | true | Runtime | 21.06.0 |
| spark.rapids.sql.join.leftSemi.enabled | When set to true left semi joins are enabled on the GPU | true | Runtime | 21.06.0 |
| spark.rapids.sql.join.rightOuter.enabled | When set to true right outer joins are enabled on the GPU | true | Runtime | 21.06.0 |
| spark.rapids.sql.json.read.datetime.enabled | JSON reading is not 100% compatible when reading dates and timestamps. | false | Runtime | 24.12.0 |
| spark.rapids.sql.json.read.decimal.enabled | When reading a quoted string as a decimal Spark supports reading non-ascii unicode digits, and the cuDF plugin does not. | true | Runtime | 22.04.0 |
| spark.rapids.sql.json.read.double.enabled | JSON reading is not 100% compatible when reading doubles. | true | Runtime | 22.04.0 |
| spark.rapids.sql.json.read.float.enabled | JSON reading is not 100% compatible when reading floats. | true | Runtime | 22.04.0 |
| spark.rapids.sql.lore.dumpPath | The path to dump the LORE nodes’ input data. This must be set if spark.rapids.sql.lore.idsToDump has been set. The data of each LORE node will be dumped to a subfolder with name ‘loreId- | None | Runtime | 24.08.1 |
| spark.rapids.sql.lore.idsToDump | Specify the LORE ids of operators to dump. The format is a comma separated list of LORE ids. For example: “1[0]” will dump partition 0 of input of gpu operator with lore id 1. For more details, please refer to the LORE documentation. If this is not set, no data will be dumped. | None | Runtime | 24.08.1 |
| spark.rapids.sql.lore.nonStrictMode.enabled | Allow LoRE dumping to continue when a selected lore id fails. When enabled, failing lore ids are skipped with a warning, previously dumped data under the dump path is kept, and the rest of the query continues executing. | false | Runtime | 26.06.0 |
| spark.rapids.sql.lore.parquet.useOriginalSchemaNames | When enabled, LORE writes Parquet files using the original Spark schema names instead of auto-generated type-based names. This makes the dumped Parquet data easier to consume directly via Spark/other tools. | true | Runtime | 25.10.0 |
| spark.rapids.sql.mode | Set the mode for the cuDF plugin. The supported modes are explainOnly and executeOnGPU. This config can not be changed at runtime, you must restart the application for it to take affect. The default mode is executeOnGPU, which means the cuDF plugin converts Spark operations and executes them on the GPU when possible. The explainOnly mode allows running queries on the CPU and the cuDF plugin will evaluate the queries as if it was going to run on the GPU. The explanations of what would have run on the GPU and why are output in log messages. When using explainOnly mode, the default explain output is ALL, this can be changed by setting spark.rapids.sql.explain. See that config for more details. | executeongpu | Startup | 22.02.0 |
| spark.rapids.sql.optimizer.joinReorder.enabled | When enabled, joins may be reordered for improved query performance | true | Runtime | 23.04.0 |
| spark.rapids.sql.python.gpu.enabled | This is an experimental feature and is likely to change in the future. Enable (true) or disable (false) support for scheduling Python Pandas UDFs with GPU resources. When enabled, pandas UDFs are assumed to share the same GPU that the RAPIDs accelerator uses and will honor the python GPU configs | false | Runtime | 21.06.0 |
| spark.rapids.sql.reader.chunked | Enable a chunked reader where possible. A chunked reader allows reading highly compressed data that could not be read otherwise, but at the expense of more GPU memory, and in some cases more GPU computation. Currently this only supports ORC and Parquet formats. | true | Runtime | 22.12.0 |
| spark.rapids.sql.reader.chunked.limitMemoryUsage | Enable a soft limit on the internal memory usage of the chunked reader (if being used). Such limit is calculated as the multiplication of ‘spark.rapids.sql.batchSizeBytes’ and ‘spark.rapids.sql.reader.chunked.memoryUsageRatio’.For example, if batchSizeBytes is set to 1GB and memoryUsageRatio is 4, the chunked reader will try to keep its memory usage under 4GB. | None | Runtime | 24.06.0 |
| spark.rapids.sql.reader.chunked.subPage | Enable a chunked reader where possible for reading data that is smaller than the typical row group/page limit. Currently deprecated and replaced by ‘spark.rapids.sql.reader.chunked.limitMemoryUsage’. | None | Runtime | 24.02.0 |
| spark.rapids.sql.reader.multithreaded.combine.sizeBytes | The target size in bytes to combine multiple small files together when using the MULTITHREADED parquet or orc reader. With combine disabled, the MULTITHREADED reader reads the files in parallel and sends individual files down to the GPU, but that can be inefficient for small files. When combine is enabled, files that are ready within spark.rapids.sql.reader.multithreaded.combine.waitTime together, up to this threshold size, are combined before sending down to GPU. This can be disabled by setting it to 0. Note that combine also will not go over the spark.rapids.sql.reader.batchSizeRows or spark.rapids.sql.reader.batchSizeBytes limits. | 67108864 | Runtime | 23.06.0 |
| spark.rapids.sql.reader.multithreaded.combine.waitTime | When using the multithreaded parquet or orc reader with combine mode, how long to wait, in milliseconds, for more files to finish if haven’t met the size threshold. Note that this will wait this amount of time from when the last file was available, so total wait time could be larger then this. | 200 | Runtime | 23.06.0 |
| spark.rapids.sql.reader.multithreaded.read.keepOrder | When using the MULTITHREADED reader, if this is set to true we read the files in the same order Spark does, otherwise the order may not be the same. Now it is supported only for parquet and orc. | true | Runtime | 23.06.0 |
| spark.rapids.sql.reader.useReadEstimateFromSchema | Use a schema based estimate of GPU memory to limit a read batch. The estimate is a worst case guess that ignores compression, so it can stop a batch short of ‘spark.rapids.sql.reader.batchSizeBytes’. If unset, the estimate is used only when there is no chunked reader (see ‘spark.rapids.sql.reader.chunked’), since a chunked reader already bounds its own output to ‘spark.rapids.sql.batchSizeBytes’. Set to true to use it everywhere, or false to use it nowhere and let batches fill up to the row and byte limits. Text based readers such as CSV and JSON never use it, they measure the bytes they actually read. | None | Runtime | Unreleased |
| spark.rapids.sql.regexp.enabled | Specifies whether supported regular expressions will be evaluated on the GPU. Unsupported expressions will fall back to CPU. However, there are some known edge cases that will still execute on GPU and produce incorrect results and these are documented in the compatibility guide. Setting this config to false will make all regular expressions run on the CPU instead. | true | Runtime | 22.04.0 |
| spark.rapids.sql.replaceSortMergeJoin.enabled | Allow replacing sortMergeJoin with HashJoin | true | Runtime | 0.1.0 |
| spark.rapids.sql.rowBasedUDF.enabled | Deprecated. When set to true, enables the legacy row-based UDF path in a GPU operation by transferring only the data it needs between GPU and CPU inside a query operation, instead of falling this operation back to CPU. CPU bridge is the preferred replacement when it can safely bridge the expression, and this config might be removed in the future. | false | Runtime | 21.12.0 |
| spark.rapids.sql.stableSort.enabled | Enable or disable stable sorting. Apache Spark’s sorting is typically a stable sort, but sort stability cannot be guaranteed in distributed work loads because the order in which upstream data arrives to a task is not guaranteed. Sort stability then only matters when reading and sorting data from a file using a single task/partition. Because of limitations in the plugin when you enable stable sorting all of the data for a single task will be combined into a single batch before sorting. This currently disables spilling from GPU memory if the data size is too large. | false | Runtime | 21.06.0 |
| spark.rapids.sql.suppressPlanningFailure | Option to fallback an individual query to CPU if an unexpected condition prevents the query plan from being converted to a GPU-enabled one. Note this is different from a normal CPU fallback for a yet-to-be-supported Spark SQL feature. If this happens the error should be reported and investigated as a GitHub issue. | false | Runtime | 21.12.0 |
| spark.rapids.sql.variableFloatAgg.enabled | Spark assumes that all operations produce the exact same result each time. This is not true for some floating point aggregations, which can produce slightly different results on the GPU as the aggregation is done in parallel. This can enable those operations if you know the query is only computing it once. | true | Runtime | 0.1.0 |
| spark.rapids.sql.window.batched.bounded.row.max | Max value for bounded row window preceding/following extents permissible for the window to be evaluated in batched mode. This value affects both the preceding and following bounds, potentially doubling the window size permitted for batched execution | 100 | Runtime | 24.02.0 |
| spark.rapids.sql.window.collectList.enabled | The output size of collect list for a window operation is proportional to the window size squared. The current GPU implementation does not handle this well and is disabled by default. If you know that your window size is very small you can try to enable it. | false | Runtime | 24.02.0 |
| spark.rapids.sql.window.collectSet.enabled | The output size of collect set for a window operation can be proportional to the window size squared. The current GPU implementation does not handle this well and is disabled by default. If you know that your window size is very small you can try to enable it. | false | Runtime | 24.02.0 |
| spark.rapids.sql.window.range.byte.enabled | When the order-by column of a range based window is byte type and the range boundary calculated for a value has overflow, CPU and GPU will get the different results. When set to false disables the range window acceleration for the byte type order-by column | false | Runtime | 21.06.0 |
| spark.rapids.sql.window.range.decimal.enabled | When set to false, this disables the range window acceleration for the DECIMAL type order-by column | true | Runtime | 22.10.0 |
| spark.rapids.sql.window.range.double.enabled | When set to false, this disables the range window acceleration for the double type order-by column | true | Runtime | 23.08.1 |
| spark.rapids.sql.window.range.float.enabled | When set to false, this disables the range window acceleration for the FLOAT type order-by column | true | Runtime | 23.08.1 |
| spark.rapids.sql.window.range.int.enabled | When the order-by column of a range based window is int type and the range boundary calculated for a value has overflow, CPU and GPU will get the different results. When set to false disables the range window acceleration for the int type order-by column | true | Runtime | 21.06.0 |
| spark.rapids.sql.window.range.long.enabled | When the order-by column of a range based window is long type and the range boundary calculated for a value has overflow, CPU and GPU will get the different results. When set to false disables the range window acceleration for the long type order-by column | true | Runtime | 21.06.0 |
| spark.rapids.sql.window.range.short.enabled | When the order-by column of a range based window is short type and the range boundary calculated for a value has overflow, CPU and GPU will get the different results. When set to false disables the range window acceleration for the short type order-by column | false | Runtime | 21.06.0 |
Supported GPU Operators and Fine Tuning
The cuDF plugin can be configured to enable or disable specific GPU accelerated expressions. Enabled expressions are candidates for GPU execution. If the expression is configured as disabled, the accelerator plugin will not attempt replacement, and it will run on the CPU.
Please leverage the spark.rapids.sql.explain setting to get feedback from the plugin as to why parts of a query may not be executing on the GPU.
NOTE: Setting spark.rapids.sql.incompatibleOps.enabled=true will enable all the settings in the table below which are not enabled by default due to incompatibilities.
Expressions
| Name | SQL Function(s) | Description | Default Value | Notes | Since Version |
|---|---|---|---|---|---|
| spark.rapids.sql.expression.Abs | abs | Absolute value | true | None | 0.1.0 |
| spark.rapids.sql.expression.Acos | acos | Inverse cosine | true | None | 0.1.0 |
| spark.rapids.sql.expression.Acosh | acosh | Inverse hyperbolic cosine | true | None | 0.1.0 |
| spark.rapids.sql.expression.Add | + | Addition | true | None | 0.1.0 |
| spark.rapids.sql.expression.Alias | Gives a column a name | true | None | 0.1.0 | |
| spark.rapids.sql.expression.And | and | Logical AND | true | None | 0.1.0 |
| spark.rapids.sql.expression.AnsiCast | Convert a column of one type of data into another type | true | None | 21.06.0 | |
| spark.rapids.sql.expression.ArrayAggregate | aggregate | Aggregate elements in an array using an accumulator function and finishing transformation. Currently only lambdas of the form (acc, x) -> op(acc, g(x)) with an identity finish are executed on the GPU, where op is one of SUM/PRODUCT/MAX/MIN/ALL/ANY. If/CaseWhen branches are accepted as long as each branch is itself op-of-acc (or bare acc) with op consistent across branches; other shapes fall back to CPU. SUM/PRODUCT on float/double share the same parallel-reduction non-determinism as GpuSum and are gated by spark.rapids.sql.variableFloatAgg.enabled. SUM/PRODUCT on integer/decimal in ANSI mode fall back to CPU because cuDF segmented reduce wraps on overflow instead of raising. | true | None | 26.06.0 |
| spark.rapids.sql.expression.ArrayContains | array_contains | Returns a boolean if the array contains the passed in key | true | None | 21.06.0 |
| spark.rapids.sql.expression.ArrayDistinct | array_distinct | Removes duplicate values from the array | true | None | 25.06.0 |
| spark.rapids.sql.expression.ArrayExcept | array_except | Returns an array of the elements in array1 but not in array2, without duplicates | true | This is not 100% compatible with the Spark version because the GPU implementation treats -0.0 and 0.0 as equal, but the CPU implementation currently does not (see SPARK-39845). Also, Apache Spark 3.1.3 fixed issue SPARK-36741 where NaNs in these set like operators were not treated as being equal. We have chosen to break with compatibility for the older versions of Spark in this instance and handle NaNs the same as 3.1.3+ | 22.08.0 |
| spark.rapids.sql.expression.ArrayExists | exists | Return true if any element satisfies the predicate LambdaFunction | true | None | 22.04.0 |
| spark.rapids.sql.expression.ArrayFilter | filter | Filter an input array using a given predicate | true | None | 24.06.0 |
| spark.rapids.sql.expression.ArrayIntersect | array_intersect | Returns an array of the elements in the intersection of array1 and array2, without duplicates | true | This is not 100% compatible with the Spark version because the GPU implementation treats -0.0 and 0.0 as equal, but the CPU implementation currently does not (see SPARK-39845). Also, Apache Spark 3.1.3 fixed issue SPARK-36741 where NaNs in these set like operators were not treated as being equal. We have chosen to break with compatibility for the older versions of Spark in this instance and handle NaNs the same as 3.1.3+ | 22.08.0 |
| spark.rapids.sql.expression.ArrayJoin | array_join | Concatenates the elements of the given array using the delimiter and an optional string to replace nulls. If no value is set for nullReplacement, any null value is filtered. | true | None | 24.10.0 |
| spark.rapids.sql.expression.ArrayMax | array_max | Returns the maximum value in the array | true | None | 21.10.0 |
| spark.rapids.sql.expression.ArrayMin | array_min | Returns the minimum value in the array | true | None | 21.10.0 |
| spark.rapids.sql.expression.ArrayPosition | array_position | Returns the (1-based) index of the first matching element of the array as long, or 0 if no match is found. | true | None | 25.04.0 |
| spark.rapids.sql.expression.ArrayRemove | array_remove | Returns the array after removing all elements that equal to the input element (right) from the input array (left) | true | None | 22.12.0 |
| spark.rapids.sql.expression.ArrayRepeat | array_repeat | Returns the array containing the given input value (left) count (right) times | true | None | 22.06.0 |
| spark.rapids.sql.expression.ArraySort | array_sort | Sorts the input array in ascending order with nulls last according to the natural ordering of the elements (the default comparator). A custom comparator, or a nested element type (array or struct), falls back to the CPU | true | None | 26.08.0 |
| spark.rapids.sql.expression.ArrayTransform | transform | Transform elements in an array using the transform function. This is similar to a map in functional programming | true | None | 21.10.0 |
| spark.rapids.sql.expression.ArrayUnion | array_union | Returns an array of the elements in the union of array1 and array2, without duplicates. | true | This is not 100% compatible with the Spark version because the GPU implementation treats -0.0 and 0.0 as equal, but the CPU implementation currently does not (see SPARK-39845). Also, Apache Spark 3.1.3 fixed issue SPARK-36741 where NaNs in these set like operators were not treated as being equal. We have chosen to break with compatibility for the older versions of Spark in this instance and handle NaNs the same as 3.1.3+ | 22.08.0 |
| spark.rapids.sql.expression.ArraysOverlap | arrays_overlap | Returns true if a1 contains at least a non-null element present also in a2. If the arrays have no common element and they are both non-empty and either of them contains a null element null is returned, false otherwise. | true | This is not 100% compatible with the Spark version because the GPU implementation treats -0.0 and 0.0 as equal, but the CPU implementation currently does not (see SPARK-39845). Also, Apache Spark 3.1.3 fixed issue SPARK-36741 where NaNs in these set like operators were not treated as being equal. We have chosen to break with compatibility for the older versions of Spark in this instance and handle NaNs the same as 3.1.3+ | 22.08.0 |
| spark.rapids.sql.expression.ArraysZip | arrays_zip | Returns a merged array of structs in which the N-th struct contains all N-th values of input arrays. | true | None | 22.06.0 |
| spark.rapids.sql.expression.Ascii | ascii | The numeric value of the first character of string data. | false | This is disabled by default because it only supports strings starting with ASCII or Latin-1 characters after Spark 3.2.3, 3.3.1 and 3.4.0. Otherwise the results will not match the CPU. | 24.02.0 |
| spark.rapids.sql.expression.Asin | asin | Inverse sine | true | None | 0.1.0 |
| spark.rapids.sql.expression.Asinh | asinh | Inverse hyperbolic sine | true | None | 0.1.0 |
| spark.rapids.sql.expression.AtLeastNNonNulls | Checks if number of non null/Nan values is greater than a given value | true | None | 0.1.0 | |
| spark.rapids.sql.expression.Atan | atan | Inverse tangent | true | None | 0.1.0 |
| spark.rapids.sql.expression.Atanh | atanh | Inverse hyperbolic tangent | true | None | 0.1.0 |
| spark.rapids.sql.expression.AttributeReference | References an input column | true | None | 0.1.0 | |
| spark.rapids.sql.expression.BRound | bround | Round an expression to d decimal places using HALF_EVEN rounding mode | true | None | 21.06.0 |
| spark.rapids.sql.expression.Bin | bin | Returns the string representation of the long value expr represented in binary | true | None | 25.02.0 |
| spark.rapids.sql.expression.BitLength | bit_length | The bit length of string data | true | None | 22.04.0 |
| spark.rapids.sql.expression.BitwiseAnd | & | Returns the bitwise AND of the operands | true | None | 0.1.0 |
| spark.rapids.sql.expression.BitwiseCount | bit_count | Returns the number of bits that are set in the input as unsigned 64-bit integer | true | None | 25.06.0 |
| spark.rapids.sql.expression.BitwiseNot | ~ | Returns the bitwise NOT of the operands | true | None | 0.1.0 |
| spark.rapids.sql.expression.BitwiseOr | \| | Returns the bitwise OR of the operands | true | None | 0.1.0 |
| spark.rapids.sql.expression.BitwiseXor | ^ | Returns the bitwise XOR of the operands | true | None | 0.1.0 |
| spark.rapids.sql.expression.BloomFilterMightContain | Bloom filter query | true | None | 26.06.0 | |
| spark.rapids.sql.expression.BoundReference | Reference to a bound variable | true | None | 24.06.0 | |
| spark.rapids.sql.expression.CaseWhen | when | CASE WHEN expression | true | None | 0.1.0 |
| spark.rapids.sql.expression.Cast | bigint, binary, boolean, cast, date, decimal, double, float, int, smallint, string, timestamp, tinyint | Convert a column of one type of data into another type | true | None | 0.1.0 |
| spark.rapids.sql.expression.Cbrt | cbrt | Cube root | true | None | 0.1.0 |
| spark.rapids.sql.expression.Ceil | Ceiling of a number | true | None | 0.1.0 | |
| spark.rapids.sql.expression.CheckOverflow | CheckOverflow after arithmetic operations between DecimalType data | true | None | 21.06.0 | |
| spark.rapids.sql.expression.Coalesce | coalesce | Returns the first non-null argument if exists. Otherwise, null | true | None | 0.1.0 |
| spark.rapids.sql.expression.Concat | concat | List/String concatenate | true | None | 21.06.0 |
| spark.rapids.sql.expression.ConcatWs | concat_ws | Concatenates multiple input strings or array of strings into a single string using a given separator | true | None | 21.06.0 |
| spark.rapids.sql.expression.Contains | Contains | true | None | 0.1.0 | |
| spark.rapids.sql.expression.Conv | conv | Convert string representing a number from one base to another | true | None | 23.10.0 |
| spark.rapids.sql.expression.Cos | cos | Cosine | true | None | 0.1.0 |
| spark.rapids.sql.expression.Cosh | cosh | Hyperbolic cosine | true | None | 0.1.0 |
| spark.rapids.sql.expression.Cot | cot | Cotangent | true | None | 21.06.0 |
| spark.rapids.sql.expression.CreateArray | array | Returns an array with the given elements | true | None | 21.06.0 |
| spark.rapids.sql.expression.CreateMap | map | Create a map | true | None | 21.10.0 |
| spark.rapids.sql.expression.CreateNamedStruct | named_struct, struct | Creates a struct with the given field names and values | true | None | 21.06.0 |
| spark.rapids.sql.expression.CurrentRow$ | Special boundary for a window frame, indicating stopping at the current row | true | None | 21.06.0 | |
| spark.rapids.sql.expression.DateAdd | date_add | Returns the date that is num_days after start_date | true | None | 21.06.0 |
| spark.rapids.sql.expression.DateAddInterval | Adds interval to date | true | None | 21.06.0 | |
| spark.rapids.sql.expression.DateDiff | datediff | Returns the number of days from startDate to endDate | true | None | 0.1.0 |
| spark.rapids.sql.expression.DateFormatClass | date_format | Converts timestamp to a value of string in the format specified by the date format | true | None | 21.06.0 |
| spark.rapids.sql.expression.DateSub | date_sub | Returns the date that is num_days before start_date | true | None | 21.06.0 |
| spark.rapids.sql.expression.DayOfMonth | day, dayofmonth | Returns the day of the month from a date or timestamp | true | None | 0.1.0 |
| spark.rapids.sql.expression.DayOfWeek | dayofweek | Returns the day of the week (1 = Sunday…7=Saturday) | true | None | 21.06.0 |
| spark.rapids.sql.expression.DayOfYear | dayofyear | Returns the day of the year from a date or timestamp | true | None | 21.06.0 |
| spark.rapids.sql.expression.DenseRank | dense_rank | Window function that returns the dense rank value within the aggregation window | true | None | 21.08.0 |
| spark.rapids.sql.expression.Divide | / | Division | true | None | 0.1.0 |
| spark.rapids.sql.expression.DivideDTInterval | Day-time interval * operator | true | None | 26.06.0 | |
| spark.rapids.sql.expression.DivideYMInterval | Year-month interval / number | true | None | 26.06.0 | |
| spark.rapids.sql.expression.DynamicPruningExpression | Dynamic pruning expression marker | true | None | 23.10.0 | |
| spark.rapids.sql.expression.ElementAt | element_at | Returns element of array at given(1-based) index in value if column is array. Returns value for the given key in value if column is map. | true | None | 21.06.0 |
| spark.rapids.sql.expression.EndsWith | Ends with | true | None | 0.1.0 | |
| spark.rapids.sql.expression.EqualNullSafe | <=> | Check if the values are equal including nulls <=> | true | None | 21.06.0 |
| spark.rapids.sql.expression.EqualTo | ==, = | Check if the values are equal | true | None | 0.1.0 |
| spark.rapids.sql.expression.Exp | exp | Euler’s number e raised to a power | true | None | 0.1.0 |
| spark.rapids.sql.expression.Explode | explode_outer, explode | Given an input array produces a sequence of rows for each value in the array | true | None | 21.06.0 |
| spark.rapids.sql.expression.Expm1 | expm1 | Euler’s number e raised to a power minus 1 | true | None | 0.1.0 |
| spark.rapids.sql.expression.Flatten | flatten | Creates a single array from an array of arrays | true | None | 23.08.1 |
| spark.rapids.sql.expression.Floor | Floor of a number | true | None | 0.1.0 | |
| spark.rapids.sql.expression.FormatNumber | format_number | Formats the number x like ‘#,###,###.##’, rounded to d decimal places. | true | None | 23.10.0 |
| spark.rapids.sql.expression.FromUTCTimestamp | from_utc_timestamp | Render the input UTC timestamp in the input timezone | true | None | 22.10.0 |
| spark.rapids.sql.expression.FromUnixTime | from_unixtime | Get the string from a unix timestamp | true | None | 0.1.0 |
| spark.rapids.sql.expression.GetArrayItem | Gets the field at ordinal in the Array | true | None | 21.06.0 | |
| spark.rapids.sql.expression.GetArrayStructFields | Extracts the ordinal-th fields of all array elements for the data with the type of array of struct | true | None | 22.04.0 | |
| spark.rapids.sql.expression.GetJsonObject | get_json_object | Extracts a json object from path | true | None | 21.06.0 |
| spark.rapids.sql.expression.GetMapValue | Gets Value from a Map based on a key | true | None | 21.06.0 | |
| spark.rapids.sql.expression.GetStructField | Gets the named field of the struct | true | None | 21.06.0 | |
| spark.rapids.sql.expression.GetTimestamp | Gets timestamps from strings using given pattern. | true | None | 21.06.0 | |
| spark.rapids.sql.expression.GreaterThan | > | > operator | true | None | 0.1.0 |
| spark.rapids.sql.expression.GreaterThanOrEqual | >= | >= operator | true | None | 0.1.0 |
| spark.rapids.sql.expression.Greatest | greatest | Returns the greatest value of all parameters, skipping null values | true | None | 21.06.0 |
| spark.rapids.sql.expression.Hex | hex | Returns the hex string representation of a value | true | None | 26.06.0 |
| spark.rapids.sql.expression.HiveHash | hive hash operator | true | None | 24.08.1 | |
| spark.rapids.sql.expression.Hour | hour | Returns the hour component of the string/timestamp | true | None | 21.06.0 |
| spark.rapids.sql.expression.Hypot | hypot | Pythagorean addition (Hypotenuse) of real numbers | true | None | 22.04.0 |
| spark.rapids.sql.expression.If | if | IF expression | true | None | 0.1.0 |
| spark.rapids.sql.expression.In | in | IN operator | true | Non-literal list expressions must be deterministic and side-effect-free | 0.1.0 |
| spark.rapids.sql.expression.InSet | INSET operator | true | None | 0.1.0 | |
| spark.rapids.sql.expression.InitCap | initcap | Returns str with the first letter of each word in uppercase. All other letters are in lowercase | true | This is not 100% compatible with the Spark version because the Unicode version used by cuDF and the JVM may differ, resulting in some corner-case characters not changing case correctly. | 21.06.0 |
| spark.rapids.sql.expression.InputFileBlockLength | input_file_block_length | Returns the length of the block being read, or -1 if not available | true | None | 21.06.0 |
| spark.rapids.sql.expression.InputFileBlockStart | input_file_block_start | Returns the start offset of the block being read, or -1 if not available | true | None | 21.06.0 |
| spark.rapids.sql.expression.InputFileName | input_file_name | Returns the name of the file being read, or empty string if not available | true | None | 21.06.0 |
| spark.rapids.sql.expression.IntegralDivide | div | Division with a integer result | true | None | 0.1.0 |
| spark.rapids.sql.expression.IsNaN | isnan | Checks if a value is NaN | true | None | 0.1.0 |
| spark.rapids.sql.expression.IsNotNull | isnotnull | Checks if a value is not null | true | None | 0.1.0 |
| spark.rapids.sql.expression.IsNull | isnull | Checks if a value is null | true | None | 0.1.0 |
| spark.rapids.sql.expression.JsonToStructs | from_json | Returns a struct value with the given jsonStr and schema | true | None | 22.10.0 |
| spark.rapids.sql.expression.JsonTuple | json_tuple | Returns a tuple like the function get_json_object, but it takes multiple names. All the input parameters and output column types are string. | true | None | 23.02.0 |
| spark.rapids.sql.expression.KnownFloatingPointNormalized | Tag to prevent redundant normalization | true | None | 0.1.0 | |
| spark.rapids.sql.expression.KnownNotNull | Tag an expression as known to not be null | true | None | 21.08.0 | |
| spark.rapids.sql.expression.Lag | lag | Window function that returns N entries behind this one | true | None | 21.06.0 |
| spark.rapids.sql.expression.LambdaFunction | Holds a higher order SQL function | true | None | 21.10.0 | |
| spark.rapids.sql.expression.LastDay | last_day | Returns the last day of the month which the date belongs to | true | None | 21.06.0 |
| spark.rapids.sql.expression.Lead | lead | Window function that returns N entries ahead of this one | true | None | 21.06.0 |
| spark.rapids.sql.expression.Least | least | Returns the least value of all parameters, skipping null values | true | None | 21.06.0 |
| spark.rapids.sql.expression.Length | char_length, character_length, length | String character length or binary byte length | true | None | 21.06.0 |
| spark.rapids.sql.expression.LessThan | < | < operator | true | None | 0.1.0 |
| spark.rapids.sql.expression.LessThanOrEqual | <= | <= operator | true | None | 0.1.0 |
| spark.rapids.sql.expression.Like | like | Like | true | None | 21.06.0 |
| spark.rapids.sql.expression.Literal | Holds a static value from the query | true | None | 0.1.0 | |
| spark.rapids.sql.expression.Log | ln | Natural log | true | None | 0.1.0 |
| spark.rapids.sql.expression.Log10 | log10 | Log base 10 | true | None | 21.06.0 |
| spark.rapids.sql.expression.Log1p | log1p | Natural log 1 + expr | true | None | 21.06.0 |
| spark.rapids.sql.expression.Log2 | log2 | Log base 2 | true | None | 21.06.0 |
| spark.rapids.sql.expression.Logarithm | log | Log variable base | true | None | 21.06.0 |
| spark.rapids.sql.expression.Lower | lcase, lower | String lowercase operator | true | This is not 100% compatible with the Spark version because the Unicode version used by cuDF and the JVM may differ, resulting in some corner-case characters not changing case correctly. | 0.1.0 |
| spark.rapids.sql.expression.MakeDecimal | Create a Decimal from an unscaled long value for some aggregation optimizations | true | None | 21.06.0 | |
| spark.rapids.sql.expression.MapConcat | map_concat | Returns the union of all the given maps | true | None | 22.06.0 |
| spark.rapids.sql.expression.MapEntries | map_entries | Returns an unordered array of all entries in the given map | true | None | 21.10.0 |
| spark.rapids.sql.expression.MapFilter | map_filter | Filters entries in a map using the function | true | None | 22.06.0 |
| spark.rapids.sql.expression.MapFromArrays | map_from_arrays | Creates a new map from two arrays | true | None | 24.08.1 |
| spark.rapids.sql.expression.MapFromEntries | map_from_entries | Creates a map from an array of entries (structs of key-value pairs) | true | None | 25.12.0 |
| spark.rapids.sql.expression.MapKeys | map_keys | Returns an unordered array containing the keys of the map | true | None | 21.10.0 |
| spark.rapids.sql.expression.MapValues | map_values | Returns an unordered array containing the values of the map | true | None | 21.10.0 |
| spark.rapids.sql.expression.MapZipWith | map_zip_with | Filters entries in a map using the function | true | None | 25.10.0 |
| spark.rapids.sql.expression.Md5 | md5 | MD5 hash operator | true | None | 21.06.0 |
| spark.rapids.sql.expression.MicrosToTimestamp | timestamp_micros | Converts the number of microseconds from unix epoch to a timestamp | true | None | 23.08.1 |
| spark.rapids.sql.expression.MillisToTimestamp | timestamp_millis | Converts the number of milliseconds from unix epoch to a timestamp | true | None | 23.08.1 |
| spark.rapids.sql.expression.Minute | minute | Returns the minute component of the string/timestamp | true | None | 21.06.0 |
| spark.rapids.sql.expression.MonotonicallyIncreasingID | monotonically_increasing_id | Returns monotonically increasing 64-bit integers | true | None | 21.06.0 |
| spark.rapids.sql.expression.Month | month | Returns the month from a date or timestamp | true | None | 0.1.0 |
| spark.rapids.sql.expression.MonthsBetween | months_between | If timestamp1 is later than timestamp2, then the result is positive. If timestamp1 and timestamp2 are on the same day of month, or both are the last day of month, time of day will be ignored. Otherwise, the difference is calculated based on 31 days per month, and rounded to 8 digits unless roundOff=false. | true | None | 24.12.0 |
| spark.rapids.sql.expression.Multiply | * | Multiplication | true | None | 0.1.0 |
| spark.rapids.sql.expression.MultiplyDTInterval | Day-time interval * number | true | None | 26.06.0 | |
| spark.rapids.sql.expression.MultiplyYMInterval | Year-month interval * number | true | None | 26.06.0 | |
| spark.rapids.sql.expression.Murmur3Hash | hash | Murmur3 hash operator | true | None | 21.06.0 |
| spark.rapids.sql.expression.NTile | ntile | Window function that divides the rows in each partition into buckets | true | None | Unreleased |
| spark.rapids.sql.expression.NaNvl | nanvl | Evaluates to left iff left is not NaN, right otherwise | true | None | 0.1.0 |
| spark.rapids.sql.expression.NamedLambdaVariable | A parameter to a higher order SQL function | true | None | 21.10.0 | |
| spark.rapids.sql.expression.Not | !, not | Boolean not operator | true | None | 0.1.0 |
| spark.rapids.sql.expression.NthValue | nth_value | nth window operator | true | None | 22.08.0 |
| spark.rapids.sql.expression.OctetLength | octet_length | The byte length of string data | true | None | 22.04.0 |
| spark.rapids.sql.expression.Or | or | Logical OR | true | None | 0.1.0 |
| spark.rapids.sql.expression.ParseUrl | parse_url | Extracts a part from a URL | true | None | 24.02.0 |
| spark.rapids.sql.expression.PercentRank | percent_rank | Window function that returns the percent rank value within the aggregation window | true | None | 22.04.0 |
| spark.rapids.sql.expression.Pmod | pmod | Pmod | true | None | 21.06.0 |
| spark.rapids.sql.expression.PosExplode | posexplode_outer, posexplode | Given an input array produces a sequence of rows for each value in the array | true | None | 21.06.0 |
| spark.rapids.sql.expression.Pow | pow, power | lhs ^ rhs | true | None | 0.1.0 |
| spark.rapids.sql.expression.PreciseTimestampConversion | Expression used internally to convert the TimestampType to Long and back without losing precision, i.e. in microseconds. Used in time windowing | true | None | 21.10.0 | |
| spark.rapids.sql.expression.PromotePrecision | PromotePrecision before arithmetic operations between DecimalType data | true | None | 21.06.0 | |
| spark.rapids.sql.expression.PythonUDF | UDF run in an external python process. Does not actually run on the GPU, but the transfer of data to/from it can be accelerated | true | None | 21.06.0 | |
| spark.rapids.sql.expression.Quarter | quarter | Returns the quarter of the year for date, in the range 1 to 4 | true | None | 21.06.0 |
| spark.rapids.sql.expression.RLike | regexp_like, regexp, rlike | Regular expression version of Like | true | None | 21.12.0 |
| spark.rapids.sql.expression.RaiseError | raise_error | Throw an exception | true | None | 22.06.0 |
| spark.rapids.sql.expression.Rand | rand, random | Generate a random column with i.i.d. uniformly distributed values in [0, 1) | true | None | 0.1.0 |
| spark.rapids.sql.expression.Rank | rank | Window function that returns the rank value within the aggregation window | true | None | 21.08.0 |
| spark.rapids.sql.expression.RegExpExtract | regexp_extract | Extract a specific group identified by a regular expression | true | None | 22.02.0 |
| spark.rapids.sql.expression.RegExpExtractAll | regexp_extract_all | Extract all strings matching a regular expression corresponding to the regex group index | true | None | 22.08.0 |
| spark.rapids.sql.expression.RegExpReplace | regexp_replace | String replace using a regular expression pattern | true | None | 21.06.0 |
| spark.rapids.sql.expression.Remainder | %, mod | Remainder or modulo | true | None | 0.1.0 |
| spark.rapids.sql.expression.ReplicateRows | Given an input row replicates the row N times | true | None | 22.02.0 | |
| spark.rapids.sql.expression.Reverse | reverse | Returns a reversed string or an array with reverse order of elements | true | None | 23.02.0 |
| spark.rapids.sql.expression.Rint | rint | Rounds up a double value to the nearest double equal to an integer | true | None | 0.1.0 |
| spark.rapids.sql.expression.Round | round | Round an expression to d decimal places using HALF_UP rounding mode | true | None | 21.06.0 |
| spark.rapids.sql.expression.RoundCeil | Computes the ceiling of the given expression to d decimal places | true | None | 26.06.0 | |
| spark.rapids.sql.expression.RoundFloor | Computes the floor of the given expression to d decimal places | true | None | 26.06.0 | |
| spark.rapids.sql.expression.RowNumber | row_number | Window function that returns the index for the row within the aggregation window | true | None | 21.06.0 |
| spark.rapids.sql.expression.ScalaUDF | User Defined Function, the UDF can choose to implement a RAPIDS accelerated interface to get better performance. | true | None | 21.06.0 | |
| spark.rapids.sql.expression.Second | second | Returns the second component of the string/timestamp | true | None | 21.06.0 |
| spark.rapids.sql.expression.SecondsToTimestamp | timestamp_seconds | Converts the number of seconds from unix epoch to a timestamp | true | None | 23.08.1 |
| spark.rapids.sql.expression.Sequence | sequence | Sequence | true | None | 22.02.0 |
| spark.rapids.sql.expression.Sha1 | sha1, sha | Sha1 hash operator | true | None | 25.06.0 |
| spark.rapids.sql.expression.Sha2 | sha2 | Sha2 hash operator | true | None | 26.06.0 |
| spark.rapids.sql.expression.ShiftLeft | shiftleft | Bitwise shift left («) | true | None | 0.1.0 |
| spark.rapids.sql.expression.ShiftRight | shiftright | Bitwise shift right (») | true | None | 0.1.0 |
| spark.rapids.sql.expression.ShiftRightUnsigned | shiftrightunsigned | Bitwise unsigned shift right (»>) | true | None | 0.1.0 |
| spark.rapids.sql.expression.Signum | sign, signum | Returns -1.0, 0.0 or 1.0 as expr is negative, 0 or positive | true | None | 21.06.0 |
| spark.rapids.sql.expression.Sin | sin | Sine | true | None | 0.1.0 |
| spark.rapids.sql.expression.Sinh | sinh | Hyperbolic sine | true | None | 0.1.0 |
| spark.rapids.sql.expression.Size | cardinality, size | The size of an array or a map | true | None | 21.06.0 |
| spark.rapids.sql.expression.Slice | slice | Subsets array x starting from index start (array indices start at 1, or starting from the end if start is negative) with the specified length. | true | None | 25.04.0 |
| spark.rapids.sql.expression.SortArray | sort_array | Returns a sorted array with the input array and the ascending / descending order | true | None | 21.08.0 |
| spark.rapids.sql.expression.SortOrder | Sort order | true | None | 0.1.0 | |
| spark.rapids.sql.expression.SparkPartitionID | spark_partition_id | Returns the current partition id | true | None | 21.06.0 |
| spark.rapids.sql.expression.SpecifiedWindowFrame | Specification of the width of the group (or “frame”) of input rows around which a window function is evaluated | true | None | 21.06.0 | |
| spark.rapids.sql.expression.Sqrt | sqrt | Square root | true | None | 0.1.0 |
| spark.rapids.sql.expression.Stack | stack | Separates expr1, …, exprk into n rows. | true | None | 23.12.0 |
| spark.rapids.sql.expression.StartsWith | Starts with | true | None | 0.1.0 | |
| spark.rapids.sql.expression.StringDecode | Decodes binary data from a charset to a UTF-8 string | true | None | 26.06.0 | |
| spark.rapids.sql.expression.StringInstr | instr | Instr string operator | true | None | 22.12.0 |
| spark.rapids.sql.expression.StringLPad | Pad a string on the left | true | None | 21.06.0 | |
| spark.rapids.sql.expression.StringLocate | locate, position | Substring search operator | true | None | 0.1.0 |
| spark.rapids.sql.expression.StringRPad | Pad a string on the right | true | None | 21.06.0 | |
| spark.rapids.sql.expression.StringRepeat | repeat | StringRepeat operator that repeats the given strings with numbers of times given by repeatTimes | true | None | 21.10.0 |
| spark.rapids.sql.expression.StringReplace | replace | StringReplace operator | true | None | 21.06.0 |
| spark.rapids.sql.expression.StringSplit | split | Splits str around occurrences that match regex | true | None | 21.06.0 |
| spark.rapids.sql.expression.StringToMap | str_to_map | Creates a map after splitting the input string into pairs of key-value strings | true | None | 22.04.0 |
| spark.rapids.sql.expression.StringTranslate | translate | StringTranslate operator | true | This is not 100% compatible with the Spark version because the GPU implementation supports all unicode code points. In Spark versions < 3.2.0, translate() does not support unicode characters with code point >= U+10000 (See SPARK-34094) | 23.06.0 |
| spark.rapids.sql.expression.StringTrim | trim | StringTrim operator | true | None | 21.06.0 |
| spark.rapids.sql.expression.StringTrimLeft | ltrim | StringTrimLeft operator | true | None | 21.06.0 |
| spark.rapids.sql.expression.StringTrimRight | rtrim | StringTrimRight operator | true | None | 21.06.0 |
| spark.rapids.sql.expression.StructsToJson | to_json | Converts structs to JSON text format | false | This is disabled by default because it is currently in beta and undergoes continuous enhancements. Please consult the compatibility documentation to determine whether you can enable this configuration for your use case | 23.12.0 |
| spark.rapids.sql.expression.Substring | substr, substring | Substring operator | true | None | 0.1.0 |
| spark.rapids.sql.expression.SubstringIndex | substring_index | substring_index operator | true | None | 21.06.0 |
| spark.rapids.sql.expression.Subtract | - | Subtraction | true | None | 0.1.0 |
| spark.rapids.sql.expression.Tan | tan | Tangent | true | None | 0.1.0 |
| spark.rapids.sql.expression.Tanh | tanh | Hyperbolic tangent | true | None | 0.1.0 |
| spark.rapids.sql.expression.TimeAdd | Adds interval to timestamp | true | None | 21.06.0 | |
| spark.rapids.sql.expression.ToDegrees | degrees | Converts radians to degrees | true | None | 21.06.0 |
| spark.rapids.sql.expression.ToRadians | radians | Converts degrees to radians | true | None | 21.06.0 |
| spark.rapids.sql.expression.ToUTCTimestamp | to_utc_timestamp | Render the input timestamp in UTC | true | None | 24.02.0 |
| spark.rapids.sql.expression.ToUnixTimestamp | to_unix_timestamp | Returns the UNIX timestamp of the given time | true | None | 21.06.0 |
| spark.rapids.sql.expression.TransformKeys | transform_keys | Transform keys in a map using a transform function | true | None | 21.10.0 |
| spark.rapids.sql.expression.TransformValues | transform_values | Transform values in a map using a transform function | true | None | 21.10.0 |
| spark.rapids.sql.expression.TruncDate | trunc | Truncate the date to the unit specified by the given string format | true | None | 25.02.0 |
| spark.rapids.sql.expression.TruncTimestamp | date_trunc | Truncate the timestamp to the unit specified by the given string format | true | None | 25.02.0 |
| spark.rapids.sql.expression.UnaryMinus | negative | Negate a numeric value | true | None | 0.1.0 |
| spark.rapids.sql.expression.UnaryPositive | positive | A numeric value with a + in front of it | true | None | 0.1.0 |
| spark.rapids.sql.expression.UnboundedFollowing$ | Special boundary for a window frame, indicating all rows preceding the current row | true | None | 21.06.0 | |
| spark.rapids.sql.expression.UnboundedPreceding$ | Special boundary for a window frame, indicating all rows preceding the current row | true | None | 21.06.0 | |
| spark.rapids.sql.expression.UnixTimestamp | unix_timestamp | Returns the UNIX timestamp of current or specified time | true | None | 21.06.0 |
| spark.rapids.sql.expression.UnscaledValue | Convert a Decimal to an unscaled long value for some aggregation optimizations | true | None | 21.06.0 | |
| spark.rapids.sql.expression.Upper | ucase, upper | String uppercase operator | true | This is not 100% compatible with the Spark version because the Unicode version used by cuDF and the JVM may differ, resulting in some corner-case characters not changing case correctly. | 0.1.0 |
| spark.rapids.sql.expression.Uuid | uuid | Uuid | true | None | 25.10.0 |
| spark.rapids.sql.expression.WeekDay | weekday | Returns the day of the week (0 = Monday…6=Sunday) | true | None | 21.06.0 |
| spark.rapids.sql.expression.WindowExpression | Calculates a return value for every input row of a table based on a group (or “window”) of rows | true | None | 21.06.0 | |
| spark.rapids.sql.expression.WindowSpecDefinition | Specification of a window function, indicating the partitioning-expression, the row ordering, and the width of the window | true | None | 21.06.0 | |
| spark.rapids.sql.expression.XxHash64 | xxhash64 | xxhash64 hash operator | true | None | 23.08.1 |
| spark.rapids.sql.expression.Year | year | Returns the year from a date or timestamp | true | None | 0.1.0 |
| spark.rapids.sql.expression.AggregateExpression | Aggregate expression | true | None | 0.1.0 | |
| spark.rapids.sql.expression.ApproximatePercentile | approx_percentile, percentile_approx | Approximate percentile | true | This is not 100% compatible with the Spark version because the GPU implementation of approx_percentile is not bit-for-bit compatible with Apache Spark | 21.10.0 |
| spark.rapids.sql.expression.Average | avg, mean | Average aggregate operator | true | None | 0.1.0 |
| spark.rapids.sql.expression.BitAndAgg | bit_and | Returns the bitwise AND of all non-null input values | true | None | 25.06.0 |
| spark.rapids.sql.expression.BitOrAgg | bit_or | Returns the bitwise OR of all non-null input values | true | None | 25.06.0 |
| spark.rapids.sql.expression.BitXorAgg | bit_xor | Returns the bitwise XOR of all non-null input values | true | None | 25.06.0 |
| spark.rapids.sql.expression.BloomFilterAggregate | Bloom filter build | true | None | 26.06.0 | |
| spark.rapids.sql.expression.CollectList | array_agg, collect_list | Collect a list of non-unique elements, not supported in reduction | true | None | 21.06.0 |
| spark.rapids.sql.expression.CollectSet | collect_set | Collect a set of unique elements, not supported in reduction | true | None | 21.08.0 |
| spark.rapids.sql.expression.Count | count | Count aggregate operator | true | None | 0.1.0 |
| spark.rapids.sql.expression.First | first_value, first | first aggregate operator | true | None | 0.1.0 |
| spark.rapids.sql.expression.HyperLogLogPlusPlus | approx_count_distinct | Aggregation approximate count distinct | true | None | 25.04.0 |
| spark.rapids.sql.expression.Last | last_value, last | last aggregate operator | true | None | 0.1.0 |
| spark.rapids.sql.expression.Max | max | Max aggregate operator | true | None | 0.1.0 |
| spark.rapids.sql.expression.MaxBy | max_by | MaxBy aggregate operator. It may produce different results than CPU when multiple rows in a group have same minimum value in the ordering column and different associated values in the value column. | true | None | 24.10.0 |
| spark.rapids.sql.expression.Min | min | Min aggregate operator | true | None | 0.1.0 |
| spark.rapids.sql.expression.MinBy | min_by | MinBy aggregate operator. It may produce different results than CPU when multiple rows in a group have same minimum value in the ordering column and different associated values in the value column. | true | None | 24.10.0 |
| spark.rapids.sql.expression.Percentile | percentile | Aggregation computing exact percentile | true | None | 23.12.0 |
| spark.rapids.sql.expression.PivotFirst | PivotFirst operator | true | None | 21.06.0 | |
| spark.rapids.sql.expression.StddevPop | stddev_pop | Aggregation computing population standard deviation | true | None | 21.10.0 |
| spark.rapids.sql.expression.StddevSamp | std, stddev_samp, stddev | Aggregation computing sample standard deviation | true | None | 21.10.0 |
| spark.rapids.sql.expression.Sum | sum | Sum aggregate operator | true | None | 0.1.0 |
| spark.rapids.sql.expression.VariancePop | var_pop | Aggregation computing population variance | true | None | 21.10.0 |
| spark.rapids.sql.expression.VarianceSamp | var_samp, variance | Aggregation computing sample variance | true | None | 21.10.0 |
| spark.rapids.sql.expression.StaticInvoke | StaticInvoke | true | The supported types are not deterministic since it’s a dynamic expression | 25.10.0 | |
| spark.rapids.sql.expression.NormalizeNaNAndZero | Normalize NaN and zero | true | None | 0.1.0 | |
| spark.rapids.sql.expression.InSubqueryExec | Evaluates to true if values are in a subquery’s result set | true | None | 26.06.0 | |
| spark.rapids.sql.expression.ScalarSubquery | Subquery that will return only one row and one column | true | None | 21.06.0 | |
| spark.rapids.sql.expression.HiveGenericUDF | Hive Generic UDF, the UDF can choose to implement a RAPIDS accelerated interface to get better performance | true | None | 21.06.0 | |
| spark.rapids.sql.expression.HiveSimpleUDF | Hive UDF, the UDF can choose to implement a RAPIDS accelerated interface to get better performance | true | None | 21.06.0 |
Execution
Commands
| Name | Description | Default Value | Notes | Since Version |
|---|---|---|---|---|
| spark.rapids.sql.command.SaveIntoDataSourceCommand | Write to a data source | true | None | 23.02.0 |