pg_test_timing

Run the default timing test (3 seconds)

$ pg_test_timing

$ pg_test_timing [[-d|--duration]] [seconds]

pg_test_timing [option...]

-d delay_usec, --delay=delay_usec
    Specify a delay in microseconds between timing calls. This helps measure overhead when calls are not back-to-back. Default is 0.

-r repetitions, --repetitions=repetitions
    Set the number of timing repetitions to perform. A higher number provides more accurate averages. Default is 1,000,000.

-t type, --type=type
    Choose the timing method to test. Possible types on Linux often include gettimeofday and clock_gettime. If omitted, all available timing methods are tested.

-V, --version
    Display the PostgreSQL version information, then exit.

-?, --help
    Show help about pg_test_timing command line arguments, then exit.

The pg_test_timing utility is a diagnostic tool provided with PostgreSQL to measure the overhead of various timing functions available on the operating system. Accurate and efficient timing is critical for PostgreSQL's internal operations, including transaction timestamping, performance monitoring, and write-ahead log management. This command helps determine which timing methods (like gettimeofday or clock_gettime on Linux) offer the lowest overhead and highest resolution on a particular system. It runs a specified number of repetitions, optionally with a delay between calls, and reports the average time taken per timing call and its standard deviation, usually in nanoseconds. This information can be vital for database administrators to understand potential performance implications related to the system's timing capabilities.

Results from pg_test_timing are highly dependent on the specific hardware, operating system, and current system load. They reflect the overhead of timing function calls themselves, not the overall performance of the PostgreSQL server. These measurements should be taken on the actual system where PostgreSQL will run for meaningful results. Background processes can introduce noise into the measurements.

Accurate and low-overhead timing is crucial for PostgreSQL's transactional integrity and performance. High-resolution timestamps are used for transaction commit times, log entries, and various internal scheduling. If timing calls are inefficient, it can become a bottleneck, especially under heavy workloads. This utility helps identify systems where timing overhead might be a concern, allowing administrators to assess potential impacts on database performance.

Modern operating systems provide several ways to get the current time. gettimeofday is a traditional Unix system call, offering microsecond resolution. clock_gettime (with CLOCK_REALTIME or CLOCK_MONOTONIC) is part of POSIX.1-2001 and can offer higher resolution and better performance depending on the system's implementation. PostgreSQL selects the best available method based on system capabilities, and this tool helps verify their performance characteristics on a given environment.

postgres(1), pg_ctl(1), initdb(1), psql(1)