Download the PHP package iliaal/statgrab without Composer

statgrab

A native PHP extension wrapping libstatgrab, the cross-platform system-statistics library. Originally released to PECL in 2006 against libstatgrab 0.6; the 2.0.0 line is a full modernization for the PHP 8.0+ era against libstatgrab 0.92+.

Supports PHP 8.0 through 8.5 on glibc Linux, musl, macOS, and *BSD.

⚖️ Why native

Without an extension, reading system stats from PHP means one of these:

• Shell out to top, vmstat, df, or ps and parse the output. Each call forks a process, which adds up if you poll every few seconds, and the output format drifts between OS releases.
• Parse /proc by hand. That's Linux-only, and each file (/proc/meminfo, /proc/loadavg, /proc/diskstats, /proc/net/dev) has its own format and edge cases.
• Query a stats daemon or monitoring framework. That adds a network hop and a daemon to deploy, which is overkill if all you need is a CPU number for a health endpoint.

statgrab calls libstatgrab in-process. libstatgrab handles the per-OS path (Linux /proc, FreeBSD kvm, macOS host_* APIs) and is packaged for Debian/Ubuntu, Homebrew, and FreeBSD. The 2006 PECL binding never shipped a PHP 8 build, and its quirks (stringified counters, swapped page-stat keys, a flat name_list for users) made it awkward even when you could compile it.

✨ Key features

🚀 Quick start

PIE (recommended on PHP 8.x)

PIE is the PHP Foundation's PECL successor. It installs from Packagist and builds against the active php-config. Install libstatgrab first (package names are under From source), then:

Then add extension=statgrab to your php.ini.

PECL

statgrab is also on PECL:

From source

Then add extension=statgrab to your php.ini.

config.m4 finds libstatgrab through pkg-config first, then falls back to a path probe (/usr and /usr/local). Pass --with-statgrab=<prefix> to point at a custom install.

Bundled libstatgrab (statically linked, leak-fixed)

The repo carries a vendored copy of libstatgrab 0.92.1 under vendor/libstatgrab/ with local patches (see vendor/libstatgrab/LOCAL_PATCHES.md) for a process-exit leak, a stack buffer overflow in the diff getters, uninitialised process session IDs, and a heap buffer overflow when parsing Linux process command lines, which upstream has not released yet, plus a local fix for two undefined shifts. A Git checkout does not include the generated libstatgrab configure script, so generate it with autoreconf first. This requires Autoconf, Automake, and Libtool in addition to the usual C compiler, Make, and PHP development tools (on Debian/Ubuntu, install autoconf automake libtool php-dev build-essential pkg-config). To use it:

The resulting .so has no libstatgrab.so runtime dependency. PHP's make clean also deletes the bundled libstatgrab.a; after it, run make -C vendor/libstatgrab before make, or the build stops.

The vendored libstatgrab tree stays LGPL 2.1+ (see LICENSE.libstatgrab); the extension code stays PHP-3.01 (see LICENSE). Dynamic-link or static-link, neither license infects the other.

API

Procedural

The 2006 function names are preserved.

CPU sampling and timing

With bundled libstatgrab 0.92.1 on the supported Unix platforms, CPU results carry the legacy previous_run key with these meanings (in whole seconds). The corresponding Statgrab methods use the same behavior:

• sg_cpu_totals() / cpuStats() and sg_cpu_percent_usage(CPU_PERCENT_ENTIRE) / cpu(CPU_PERCENT_ENTIRE): the Unix timestamp when the current cumulative sample was collected.
• sg_cpu_diff() / cpuDiff() and sg_cpu_percent_usage(CPU_PERCENT_NEW_DIFF) / cpu(CPU_PERCENT_NEW_DIFF): collect a new CPU sample and report elapsed wall-clock seconds since the previous CPU sample. If no baseline exists yet, the first call returns cumulative data and its collection timestamp instead.
• sg_cpu_percent_usage(CPU_PERCENT_LAST_DIFF) / cpu(CPU_PERCENT_LAST_DIFF): reuse the most recently computed diff and its elapsed seconds without collecting another sample. If no diff exists, collect one using the same first-call behavior as CPU_PERCENT_NEW_DIFF.

The CPU_PERCENT_* names above are Statgrab class constants. CPU sampling state is shared by both APIs within the calling thread: cumulative CPU calls also replace its baseline. Whole-second wall-clock differences may be zero for rapid samples and can be affected by system clock adjustments.

To measure a fresh interval, call sg_cpu_totals(), wait for the desired interval, then call sg_cpu_percent_usage(Statgrab::CPU_PERCENT_NEW_DIFF). Avoid other CPU sampling calls between them. sg_snapshot() is a no-op in bundled libstatgrab on Unix; it does not seed or reset these counters. System-linked builds use the installed libstatgrab's backend behavior.

Rapid consecutive CPU diff samples can contain no elapsed CPU ticks. In that case CPU percentages are 0.0 rather than NaN, so samples remain JSON-encodable; wait between samples to measure meaningful utilization.

sg_diskio_stats() / sg_diskio_stats_diff(), sg_network_stats() / sg_network_stats_diff(), and sg_network_iface_stats() key rows by device/interface name; a duplicate name (multipath/LVM can repeat one) falls back to a numeric key instead of overwriting the earlier row.

proc_title from sg_process_stats() / processes() holds at most 4096 bytes of each process's command line. Any local user controls their own argv, so the cap keeps a few huge command lines from exhausting memory_limit. libstatgrab still reads each full command line into the C heap while it collects the list, so that memory is used transiently outside memory_limit. When the cut lands inside a UTF-8 sequence, the title ends before that sequence, so a truncated title can be up to 3 bytes shorter.

The valid-filesystems filter is process-wide in libstatgrab, not per-request. In NTS builds it persists across requests in long-lived workers such as FPM. Under a threaded (ZTS) SAPI, sg_set_valid_filesystems() in one request also changes the sg_fs_stats() filter for every other thread. The extension serializes sg_fs_stats(), sg_valid_filesystems(), and sg_set_valid_filesystems() (and their Statgrab methods) across threads so concurrent calls are memory-safe. A hung mount that blocks sg_fs_stats() therefore stalls all of these calls in every thread.

Object-oriented

Class constants:

• Statgrab::DUPLEX_FULL | DUPLEX_HALF | DUPLEX_UNKNOWN
• Statgrab::SORT_NAME | PID | UID | GID | SIZE | RES | CPU | TIME
• Statgrab::STATE_RUNNING | SLEEPING | STOPPED | ZOMBIE | UNKNOWN
• Statgrab::CPU_PERCENT_ENTIRE | LAST_DIFF | NEW_DIFF: $source modes for cpu() / sg_cpu_percent_usage()
• Statgrab::HOST_STATE_UNKNOWN | PHYSICAL | VIRTUAL_MACHINE | PARAVIRTUAL_MACHINE | HARDWARE_VIRTUALIZED: interprets the host_state field on host() / sg_general_stats()
• Statgrab::FS_UNKNOWN | REGULAR | SPECIAL | LOOPBACK | REMOTE | LOCAL | ALLTYPES: bitmask values for the device_type field on filesystems() / sg_fs_stats()
• Statgrab::ERROR_NONE | INVALID_ARGUMENT | OPEN | OPENDIR | PERMISSION | UNSUPPORTED: common libstatgrab error codes to match on

Errors

Library-side errors emit E_WARNING with the libstatgrab error string and code, and the function returns false. The OO surface follows the same convention. Argument-count violations on no-arg functions throw ArgumentCountError.

sg_set_valid_filesystems() and Statgrab::setValidFilesystems() throw instead of warning on bad input: ValueError on an empty array or an entry containing NUL bytes, TypeError on a non-string entry. A libstatgrab runtime failure from the setter (or any other stat call) takes the usual warning and false path. Call sg_error_details() / Statgrab::errorDetails() after a false return for the code, message, errno, and arg.

Notable 2.0 BC breaks

• sg_user_stats() returns per-user records, not a flat array of usernames. The underlying name_list field was removed from libstatgrab 0.91+. Migrate callers to read login_name from each record.
• Numeric counters (memory totals, fs sizes, jiffies) are returned as int instead of stringified numbers. The 2006 release stringified via snprintf("%lld") because 32-bit PHP couldn't hold them; modern 64-bit zend_long does. On 32-bit PHP builds these fields still truncate above 2^31, so use a 64-bit build.
• sg_page_stats() / sg_page_stats_diff() were swapped in 2006 and are now correct.
• sg_process_stats() fields gid and egid are now distinct from uid and euid (2006 had a copy-paste bug returning uid/euid for both).

See CHANGELOG.md for the full list.

🔗 Native PHP extensions

Companion native PHP extensions:

• php_excel: native Excel I/O via LibXL. 7-10× faster than PhpSpreadsheet, full XLS/XLSX with formulas, formatting, and styling.
• mdparser: native CommonMark + GFM markdown parser via md4c. 15-30× faster than pure-PHP libraries.
• php_clickhouse: native ClickHouse client speaking the wire protocol directly. Picks up where SeasClick left off.
• pdo_duckdb: PDO driver for DuckDB, analytical SQL in your PHP stack.
• fastjson: drop-in faster ext/json, backed by yyjson. 6× encode, 2.7× decode, 5× validate.
• phpser: decoder-optimized binary serializer for cache workloads. Faster than igbinary on packed numerics and DTO batches.
• fast_uuid: high-throughput UUID generation (v1/v4/v7), batched CSPRNG and SIMD hex formatter, ramsey-compatible API.
• fastchart: native chart-rendering extension. 38 chart types behind one fluent OO API, SVG-canonical with PNG/JPG/WebP and optional PDF output.
• phonetic: native phonetic name matching (Double Metaphone, Beider-Morse, Daitch-Mokotoff, NYSIIS, Match Rating), the encoders PHP core lacks.

License

PHP License 3.01.

Follow @iliaa on X • Blog

If this saved you from parsing /proc by hand, ⭐ star it!