Add process collector
This commit is contained in:
@@ -114,6 +114,7 @@ set(SOURCES
|
||||
src/arena_allocator.cpp
|
||||
src/format.cpp
|
||||
src/metric.cpp
|
||||
src/process_collector.cpp
|
||||
${CMAKE_BINARY_DIR}/json_tokens.cpp)
|
||||
|
||||
add_executable(weaseldb ${SOURCES})
|
||||
|
||||
@@ -1,10 +1,10 @@
|
||||
#include "config.hpp"
|
||||
#include "connection.hpp"
|
||||
#include "connection_handler.hpp"
|
||||
#include "http_handler.hpp"
|
||||
#include "metric.hpp"
|
||||
#include "perfetto_categories.hpp"
|
||||
#include "process_collector.hpp"
|
||||
#include "server.hpp"
|
||||
#include <atomic>
|
||||
#include <csignal>
|
||||
#include <cstring>
|
||||
#include <fcntl.h>
|
||||
@@ -176,6 +176,9 @@ int main(int argc, char *argv[]) {
|
||||
perfetto::TrackEvent::Register();
|
||||
#endif
|
||||
|
||||
// Register the process collector for default metrics.
|
||||
metric::register_collector(std::make_shared<ProcessCollector>());
|
||||
|
||||
std::string config_file = "config.toml";
|
||||
|
||||
// Parse command line arguments
|
||||
|
||||
@@ -12,6 +12,7 @@
|
||||
#include <cstring>
|
||||
#include <functional>
|
||||
#include <map>
|
||||
#include <memory>
|
||||
#include <mutex>
|
||||
#include <set>
|
||||
#include <string>
|
||||
@@ -451,6 +452,12 @@ struct Metric {
|
||||
return *threadArenas;
|
||||
}
|
||||
|
||||
static auto &get_collectors() {
|
||||
using CollectorRegistry = std::vector<std::shared_ptr<Collector>>;
|
||||
static CollectorRegistry *collectors = new CollectorRegistry();
|
||||
return *collectors;
|
||||
}
|
||||
|
||||
// Thread cleanup for per-family thread-local storage
|
||||
struct ThreadInit {
|
||||
ArenaAllocator arena;
|
||||
@@ -1697,6 +1704,11 @@ std::span<std::string_view> render(ArenaAllocator &arena) {
|
||||
// prevents races during static member initialization at program startup
|
||||
std::unique_lock<std::mutex> _{Metric::mutex};
|
||||
|
||||
// Call all registered collectors to update their metrics
|
||||
for (const auto &collector : Metric::get_collectors()) {
|
||||
collector->collect();
|
||||
}
|
||||
|
||||
// Phase 1: Compile - generate static text and instructions
|
||||
// Safe: cached_plan access/initialization protected by mutex above
|
||||
if (!Metric::cached_plan || Metric::cached_plan->registration_version !=
|
||||
@@ -1800,4 +1812,10 @@ void reset_metrics_for_testing() {
|
||||
// when threads exit naturally
|
||||
}
|
||||
|
||||
void register_collector(std::shared_ptr<Collector> collector) {
|
||||
std::unique_lock<std::mutex> _{Metric::mutex};
|
||||
++Metric::registration_version;
|
||||
Metric::get_collectors().push_back(std::move(collector));
|
||||
}
|
||||
|
||||
} // namespace metric
|
||||
|
||||
@@ -45,6 +45,7 @@
|
||||
|
||||
#include <functional>
|
||||
#include <initializer_list>
|
||||
#include <memory>
|
||||
#include <span>
|
||||
#include <type_traits>
|
||||
#include <vector>
|
||||
@@ -231,6 +232,36 @@ bool is_valid_label_value(std::string_view value);
|
||||
// when no metric objects are in use and no concurrent render() calls.
|
||||
void reset_metrics_for_testing();
|
||||
|
||||
/**
|
||||
* @brief Interface for a custom collector that can be registered with the
|
||||
* metrics system.
|
||||
*
|
||||
* This is used for complex metric gathering, such as reading from /proc, where
|
||||
* multiple metrics need to be updated from a single data source.
|
||||
*/
|
||||
struct Collector {
|
||||
/**
|
||||
* @brief Virtual destructor.
|
||||
*/
|
||||
virtual ~Collector() = default;
|
||||
|
||||
/**
|
||||
* @brief Called by the metrics system to update the metrics this collector is
|
||||
* responsible for.
|
||||
*/
|
||||
virtual void collect() = 0;
|
||||
};
|
||||
|
||||
/**
|
||||
* @brief Register a collector with the metrics system.
|
||||
*
|
||||
* The system will hold a shared_ptr to the collector and call its collect()
|
||||
* method during each metric rendering.
|
||||
*
|
||||
* @param collector A shared_ptr to the collector to be registered.
|
||||
*/
|
||||
void register_collector(std::shared_ptr<Collector> collector);
|
||||
|
||||
// Note: Histograms do not support callbacks due to their multi-value nature
|
||||
// (buckets + sum + count). Use static histogram metrics only.
|
||||
|
||||
|
||||
154
src/process_collector.cpp
Normal file
154
src/process_collector.cpp
Normal file
@@ -0,0 +1,154 @@
|
||||
#include "process_collector.hpp"
|
||||
|
||||
#include <cstdio>
|
||||
#include <cstring>
|
||||
#include <dirent.h>
|
||||
#include <sys/resource.h>
|
||||
#include <unistd.h>
|
||||
#include <vector>
|
||||
|
||||
namespace {
|
||||
// Helper function to read the system boot time from /proc/stat.
|
||||
// Returns boot time in seconds since epoch, or 0 on error.
|
||||
double get_boot_time() {
|
||||
FILE *fp = std::fopen("/proc/stat", "r");
|
||||
if (!fp) {
|
||||
return 0;
|
||||
}
|
||||
|
||||
char line[256];
|
||||
double boot_time = 0;
|
||||
while (std::fgets(line, sizeof(line), fp)) {
|
||||
if (std::strncmp(line, "btime ", 6) == 0) {
|
||||
if (std::sscanf(line + 6, "%lf", &boot_time) != 1) {
|
||||
boot_time = 0;
|
||||
}
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
std::fclose(fp);
|
||||
return boot_time;
|
||||
}
|
||||
} // namespace
|
||||
|
||||
ProcessCollector::ProcessCollector()
|
||||
: cpu_seconds_total_(metric::create_counter(
|
||||
"process_cpu_seconds_total",
|
||||
"Total user and system CPU time spent in seconds")
|
||||
.create({})),
|
||||
resident_memory_bytes_(
|
||||
metric::create_gauge("process_resident_memory_bytes",
|
||||
"Resident memory size in bytes")
|
||||
.create({})),
|
||||
virtual_memory_bytes_(metric::create_gauge("process_virtual_memory_bytes",
|
||||
"Virtual memory size in bytes")
|
||||
.create({})),
|
||||
open_fds_(metric::create_gauge("process_open_fds",
|
||||
"Number of open file descriptors")
|
||||
.create({})),
|
||||
max_fds_(metric::create_gauge("process_max_fds",
|
||||
"Maximum number of open file descriptors")
|
||||
.create({})),
|
||||
start_time_seconds_(
|
||||
metric::create_gauge(
|
||||
"process_start_time_seconds",
|
||||
"Start time of the process since unix epoch in seconds")
|
||||
.create({})) {
|
||||
// Set the constant max_fds metric.
|
||||
struct rlimit rlim;
|
||||
if (getrlimit(RLIMIT_NOFILE, &rlim) == 0) {
|
||||
max_fds_.set(rlim.rlim_cur);
|
||||
}
|
||||
|
||||
// Perform an initial collection to populate the other metrics and set the
|
||||
// initial counter values.
|
||||
collect();
|
||||
}
|
||||
|
||||
void ProcessCollector::collect() {
|
||||
// --- CPU Time, Memory, and Start Time from /proc/self/stat ---
|
||||
FILE *fp = std::fopen("/proc/self/stat", "r");
|
||||
if (!fp) {
|
||||
return;
|
||||
}
|
||||
|
||||
char buf[2048];
|
||||
if (std::fgets(buf, sizeof(buf), fp) == nullptr) {
|
||||
std::fclose(fp);
|
||||
return;
|
||||
}
|
||||
std::fclose(fp);
|
||||
|
||||
// Find the end of the command name, which is in parentheses
|
||||
const char *stats_start = std::strrchr(buf, ')');
|
||||
if (!stats_start) {
|
||||
return;
|
||||
}
|
||||
stats_start += 2; // Skip the ')' and the space
|
||||
|
||||
// Tokenize the rest of the string
|
||||
std::vector<const char *> stats;
|
||||
char *p = const_cast<char *>(stats_start);
|
||||
while (*p) {
|
||||
stats.push_back(p);
|
||||
while (*p && *p != ' ') {
|
||||
p++;
|
||||
}
|
||||
if (*p) {
|
||||
*p = '\0';
|
||||
p++;
|
||||
}
|
||||
}
|
||||
|
||||
// We need at least 22 fields for starttime, 24 for rss
|
||||
if (stats.size() < 24) {
|
||||
return;
|
||||
}
|
||||
|
||||
long clk_tck = sysconf(_SC_CLK_TCK);
|
||||
|
||||
// --- CPU Time ---
|
||||
unsigned long long utime_ticks = std::strtoull(stats[11], nullptr, 10);
|
||||
unsigned long long stime_ticks = std::strtoull(stats[12], nullptr, 10);
|
||||
unsigned long long current_total_ticks = utime_ticks + stime_ticks;
|
||||
|
||||
if (last_total_ticks_ > 0) { // If we have a previous value
|
||||
if (current_total_ticks > last_total_ticks_) {
|
||||
double delta_seconds =
|
||||
(double)(current_total_ticks - last_total_ticks_) / clk_tck;
|
||||
cpu_seconds_total_.inc(delta_seconds);
|
||||
}
|
||||
} else { // First run, initialize the counter
|
||||
cpu_seconds_total_.inc((double)current_total_ticks / clk_tck);
|
||||
}
|
||||
last_total_ticks_ = current_total_ticks;
|
||||
|
||||
// --- Memory ---
|
||||
unsigned long long vsize = std::strtoull(stats[20], nullptr, 10);
|
||||
long rss_pages = std::strtol(stats[21], nullptr, 10);
|
||||
virtual_memory_bytes_.set(vsize);
|
||||
resident_memory_bytes_.set(rss_pages * sysconf(_SC_PAGESIZE));
|
||||
|
||||
// --- Start Time (only needs to be set once) ---
|
||||
if (!start_time_set_) {
|
||||
long long start_time_ticks = std::strtoll(stats[19], nullptr, 10);
|
||||
double boot_time = get_boot_time();
|
||||
if (boot_time > 0) {
|
||||
start_time_seconds_.set(boot_time + (double)start_time_ticks / clk_tck);
|
||||
start_time_set_ = true;
|
||||
}
|
||||
}
|
||||
|
||||
// --- File Descriptors ---
|
||||
int fd_count = 0;
|
||||
DIR *dp = opendir("/proc/self/fd");
|
||||
if (dp) {
|
||||
while (readdir(dp) != nullptr) {
|
||||
fd_count++;
|
||||
}
|
||||
closedir(dp);
|
||||
// Subtract 3 for '.', '..', and the opendir handle itself
|
||||
open_fds_.set(fd_count > 3 ? fd_count - 3 : 0);
|
||||
}
|
||||
}
|
||||
34
src/process_collector.hpp
Normal file
34
src/process_collector.hpp
Normal file
@@ -0,0 +1,34 @@
|
||||
#pragma once
|
||||
|
||||
#include "metric.hpp"
|
||||
|
||||
/**
|
||||
* @brief A metric collector for standard process-level statistics.
|
||||
*
|
||||
* Gathers metrics like CPU usage, memory, and file descriptors by reading
|
||||
* files from the /proc filesystem.
|
||||
*/
|
||||
struct ProcessCollector : public metric::Collector {
|
||||
/**
|
||||
* @brief Constructs the collector and initializes the process metrics.
|
||||
*/
|
||||
ProcessCollector();
|
||||
|
||||
/**
|
||||
* @brief Called by the metrics system to update the process metrics.
|
||||
*/
|
||||
void collect() override;
|
||||
|
||||
private:
|
||||
// Metrics for process statistics
|
||||
metric::Counter cpu_seconds_total_;
|
||||
metric::Gauge resident_memory_bytes_;
|
||||
metric::Gauge virtual_memory_bytes_;
|
||||
metric::Gauge open_fds_;
|
||||
metric::Gauge max_fds_;
|
||||
metric::Gauge start_time_seconds_;
|
||||
|
||||
// Last observed values for calculating counter increments
|
||||
unsigned long long last_total_ticks_ = 0;
|
||||
bool start_time_set_ = false;
|
||||
};
|
||||
Reference in New Issue
Block a user