Environment Variables¶

The following environment variables can be used to affect the behavior of Arrow C++ at runtime. Many of these variables are inspected only once per process (for example, when the Arrow C++ DLL is loaded), so you cannot assume that changing their value later will have an effect.

ARROW_DEBUG_MEMORY_POOL¶

Enable rudimentary memory checks to guard against buffer overflows. The value of this environment variable selects the behavior when a buffer overflow is detected:

abort exits the processus with a non-zero return value;
trap issues a platform-specific debugger breakpoint / trap instruction;
warn prints a warning on stderr and continues execution;

If this variable is not set, or has empty an value, memory checks are disabled.

Note

While this functionality can be useful and has little overhead, it is not a replacement for more sophisticated memory checking utilities such as Valgrind or Address Sanitizer.

ARROW_DEFAULT_MEMORY_POOL¶: Override the backend to be used for the default memory pool. Possible values are among jemalloc, mimalloc and system, depending on which backends were enabled when building Arrow C++.

ARROW_IO_THREADS¶: Override the default number of threads for the global IO thread pool. The value of this environment variable should be a positive integer.

ARROW_LIBHDFS_DIR¶: The directory containing the C HDFS library (hdfs.dll on Windows, libhdfs.dylib on macOS, libhdfs.so on other platforms). Alternatively, one can set HADOOP_HOME.

ARROW_TRACING_BACKEND¶

The backend where to export OpenTelemetry-based execution traces. Possible values are:

ostream: emit textual log messages to stdout;
otlp_http: emit OTLP JSON encoded traces to a HTTP server (by default, the endpoint URL is “http://localhost:4318/v1/traces”);
arrow_otlp_stdout: emit JSON traces to stdout;
arrow_otlp_stderr: emit JSON traces to stderr.

If this variable is not set, no traces are exported.

This environment variable has no effect if Arrow C++ was not built with tracing enabled.

ARROW_USER_SIMD_LEVEL¶

The SIMD optimization level to select. By default, Arrow C++ detects the capabilities of the current CPU at runtime and chooses the best execution paths based on that information. One can override the detection by setting this environment variable to a well-defined value. Supported values are:

NONE disables any runtime-selected SIMD optimization;
SSE4_2 enables any SSE2-based optimizations until SSE4.2 (included);
AVX enables any AVX-based optimizations and earlier;
AVX2 enables any AVX2-based optimizations and earlier;
AVX512 enables any AVX512-based optimizations and earlier.

This environment variable only has an effect on x86 platforms. Other platforms currently do not implement any form of runtime dispatch.

Note

In addition to runtime dispatch, the compile-time SIMD level can be set using the ARROW_SIMD_LEVEL CMake configuration variable. Unlike runtime dispatch, compile-time SIMD optimizations cannot be changed at runtime (for example, if you compile Arrow C++ with AVX512 enabled, the resulting binary will only run on AVX512-enabled CPUs). Setting ARROW_USER_SIMD_LEVEL=NONE prevents the execution of explicit SIMD optimization code, but it does not rule out the execution of compiler generated SIMD instructions. E.g., on x86_64 platform, Arrow is built with ARROW_SIMD_LEVEL=SSE4_2 by default. Compiler may generate SSE4.2 instructions from any C/C++ source code. On legacy x86_64 platforms do not support SSE4.2, Arrow binary may fail with SIGILL (Illegal Instruction). User must rebuild Arrow and PyArrow from scratch by setting cmake option ARROW_SIMD_LEVEL=NONE.

GANDIVA_CACHE_SIZE¶: The number of entries to keep in the Gandiva JIT compilation cache. The cache is in-memory and does not persist accross processes.

HADOOP_HOME¶: The path to the Hadoop installation.

JAVA_HOME¶: Set the path to the Java Runtime Environment installation. This may be required for HDFS support if Java is installed in a non-standard location.

OMP_NUM_THREADS¶: The number of worker threads in the global (process-wide) CPU thread pool. If this environment variable is not defined, the available hardware concurrency is determined using a platform-specific routine.

OMP_THREAD_LIMIT¶

An upper bound for the number of worker threads in the global (process-wide) CPU thread pool.

For example, if the current machine has 4 hardware threads and OMP_THREAD_LIMIT is 8, the global CPU thread pool will have 4 worker threads. But if OMP_THREAD_LIMIT is 2, the global CPU thread pool will have 2 worker threads.

OpenTelemetry

Examples