30002033718

Committed 23 Jul 2026 11:08AM UTC coverage: 52.742% (+0.08%) from 52.66%

Build # 30002033718

Build Type

Pull #99

github

Committed by

web-flow

Commit Message

Merge df0b9588d into 06bc84ec9

Pull Request Pull Request #99: Support CM time_metric (NS/MS/SEC/US) across reader and viz

Coverage Stats

41941 of 102829 branches covered (40.79%)

Branch coverage included in aggregate %.

2221 of 2839 new or added lines in 58 files covered. (78.23%)

137 existing lines in 13 files now uncovered.

37042 of 46924 relevant lines covered (78.94%)

59523.07 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

59.54

/src/dftracer/utils/core/runtime.cpp

#include <dftracer/utils/core/common/error.h>
#include <dftracer/utils/core/common/logging.h>
#include <dftracer/utils/core/common/platform_compat.h>
#include <dftracer/utils/core/env.h>
#include <dftracer/utils/core/runtime.h>

#include <algorithm>
#include <cstdlib>
#include <stdexcept>
#include <string>
#include <string_view>
#include <thread>

namespace dftracer::utils {

namespace {
// Resolve the worker-thread count. DFTRACER_UTILS_THREADS overrides everything
// (a debugging lever: force a fixed count, e.g. 1 for a single-threaded async
// loop). Otherwise 0 means hardware_concurrency.
std::size_t resolve_threads(std::size_t requested) {
    if (auto env = Env::get<std::string_view>("DFTRACER_UTILS_THREADS");
        env.has_value()) {
        const long long n =
            std::strtoll(std::string(*env).c_str(), nullptr, 10);
        if (n > 0) return static_cast<std::size_t>(n);
    }
    return requested == 0 ? dftracer_utils_hardware_concurrency() : requested;
}

// Resolve the I/O-pool size. DFTRACER_UTILS_IO_THREADS overrides everything (a
// lever to bound the epoll/kqueue thread pool independently of the compute
// pool, e.g. so the coordinator does not open a full-machine-sized pool).
// Otherwise 0 means hardware_concurrency, matching the executor default.
std::size_t resolve_io_threads(std::size_t requested) {
    if (auto env = Env::get<std::string_view>("DFTRACER_UTILS_IO_THREADS");
        env.has_value()) {
        const long long n =
            std::strtoll(std::string(*env).c_str(), nullptr, 10);
        if (n > 0) return static_cast<std::size_t>(n);
    }
    return requested == 0 ? dftracer_utils_hardware_concurrency() : requested;
}
}  // namespace

Runtime::Runtime(std::size_t threads) : threads_(resolve_threads(threads)) {
    ExecutorConfig config;
    config.num_threads = threads_;
    config.io_pool_size = resolve_io_threads(config.io_pool_size);
    executor_ = std::make_unique<Executor>(config);
    executor_->start();

    watchdog_ = std::make_unique<Watchdog>();
    watchdog_->set_executor(executor_.get());
}

Runtime::Runtime(const ExecutorConfig& config, bool enable_watchdog)
    : threads_(resolve_threads(config.num_threads)) {
    ExecutorConfig cfg = config;
    cfg.num_threads = threads_;
    cfg.io_pool_size = resolve_io_threads(config.io_pool_size);
    executor_ = std::make_unique<Executor>(cfg);
    executor_->start();

    if (enable_watchdog) {
        watchdog_ = std::make_unique<Watchdog>();
        watchdog_->set_executor(executor_.get());
    }
}

Runtime::Runtime(const ExecutorConfig& config,
                 std::unique_ptr<Watchdog> watchdog)
    : threads_(resolve_threads(config.num_threads)) {
    ExecutorConfig cfg = config;
    cfg.num_threads = threads_;
    cfg.io_pool_size = resolve_io_threads(config.io_pool_size);
    executor_ = std::make_unique<Executor>(cfg);
    executor_->start();

    watchdog_ = std::move(watchdog);
    if (watchdog_) {
        watchdog_->set_executor(executor_.get());
    }
}

Runtime::~Runtime() { shutdown(); }

TaskHandle Runtime::submit(coro::CoroTask<void> task, std::string name) {
    if (shutdown_called_.load(std::memory_order_acquire)) {
        throw DFTUtilsException(ErrorCode::PIPELINE, "Runtime is shut down");
    }
    if (name.empty()) {
        name = "task-" + std::to_string(task_name_counter_++);
    }

    auto promise = std::make_shared<std::promise<void>>();
    auto future = promise->get_future().share();
    auto tid = std::make_shared<std::atomic<TaskIndex>>(-1);

    auto wrapper =
        [](coro::CoroTask<void> t, std::shared_ptr<std::promise<void>> p,
           Executor* exec,
           std::shared_ptr<std::atomic<TaskIndex>> task_id) -> coro::Coro {
        try {
            co_await std::move(t);
            t = coro::CoroTask<void>{
                std::coroutine_handle<coro::CoroTask<void>::promise_type>{}};
            exec->mark_coro_completed(task_id->load(std::memory_order_acquire));
        } catch (...) {
            t = coro::CoroTask<void>{
                std::coroutine_handle<coro::CoroTask<void>::promise_type>{}};
            exec->mark_coro_completed(task_id->load(std::memory_order_acquire));
            p->set_exception(std::current_exception());
            co_return;
        }
        p->set_value();
    };

    // Set the executor on the task's promise so awaitables (e.g. channels)
    // that capture `get_root_promise()->get_executor()` can schedule
    // resumption. Without this, awaiters end up with executor=nullptr because
    // the wrapping `coro::Coro` doesn't extend PromiseBase and the
    // root-promise chain stops at the user's CoroTask.
    if (task.handle()) {
        task.handle().promise().set_executor(executor_.get());
    }
    auto coro = wrapper(std::move(task), promise, executor_.get(), tid);
    TaskIndex id = executor_->enqueue_tracked(std::move(coro), name, tid);

    {
        std::lock_guard<std::mutex> lock(futures_mutex_);
        cleanup_completed_futures();
        outstanding_futures_.push_back(future);
    }

    return TaskHandle{future, id, std::move(name)};
}

void Runtime::wait_all() {
    std::vector<std::shared_future<void>> futures;
    {
        std::lock_guard<std::mutex> lock(futures_mutex_);
        futures = std::move(outstanding_futures_);
        outstanding_futures_.clear();
    }
    for (auto& f : futures) {
        f.wait();
    }
}

void Runtime::cleanup_completed_futures() {
    outstanding_futures_.erase(
        std::remove_if(outstanding_futures_.begin(), outstanding_futures_.end(),
                       [](const std::shared_future<void>& f) {
                           return f.wait_for(std::chrono::seconds(0)) ==
                                  std::future_status::ready;
                       }),
        outstanding_futures_.end());
}

ExecutorProgress Runtime::get_progress() const {
    return executor_->get_progress();
}

bool Runtime::is_responsive() const { return executor_->is_responsive(); }

void Runtime::set_global_timeout(std::chrono::milliseconds timeout) {
    if (!watchdog_) {
        throw DFTUtilsException(
            ErrorCode::PIPELINE,
            "Cannot set timeout: Runtime created without watchdog");
    }
    watchdog_->set_global_timeout(timeout);
}

void Runtime::set_default_task_timeout(std::chrono::milliseconds timeout) {
    if (!watchdog_) {
        throw DFTUtilsException(
            ErrorCode::PIPELINE,
            "Cannot set timeout: Runtime created without watchdog");
    }
    watchdog_->set_default_task_timeout(timeout);
}

void Runtime::shutdown() {
    bool expected = false;
    if (!shutdown_called_.compare_exchange_strong(expected, true)) return;
    if (watchdog_) watchdog_->stop();
    if (executor_) executor_->shutdown();
}

std::size_t Runtime::threads() const { return threads_; }

std::size_t Runtime::io_threads() const {
    return executor_ ? executor_->get_io_pool_size() : 0;
}

}  // namespace dftracer::utils

1	#include <dftracer/utils/core/common/error.h>
2	#include <dftracer/utils/core/common/logging.h>
3	#include <dftracer/utils/core/common/platform_compat.h>
4	#include <dftracer/utils/core/env.h>
5	#include <dftracer/utils/core/runtime.h>
6
7	#include <algorithm>
8	#include <cstdlib>
9	#include <stdexcept>
10	#include <string>
11	#include <string_view>
12	#include <thread>
13
14	namespace dftracer::utils {
15
16	namespace {
17	// Resolve the worker-thread count. DFTRACER_UTILS_THREADS overrides everything
18	// (a debugging lever: force a fixed count, e.g. 1 for a single-threaded async
19	// loop). Otherwise 0 means hardware_concurrency.
20	std::size_t resolve_threads(std::size_t requested) {	942✔
21	if (auto env = Env::get<std::string_view>("DFTRACER_UTILS_THREADS");	1,884!
22	env.has_value()) {	942✔
23	const long long n =
24	std::strtoll(std::string(*env).c_str(), nullptr, 10);	×
25	if (n > 0) return static_cast<std::size_t>(n);	×
26	}
27	return requested == 0 ? dftracer_utils_hardware_concurrency() : requested;	942✔
28	}	471✔
29
30	// Resolve the I/O-pool size. DFTRACER_UTILS_IO_THREADS overrides everything (a
31	// lever to bound the epoll/kqueue thread pool independently of the compute
32	// pool, e.g. so the coordinator does not open a full-machine-sized pool).
33	// Otherwise 0 means hardware_concurrency, matching the executor default.
34	std::size_t resolve_io_threads(std::size_t requested) {	942✔
35	if (auto env = Env::get<std::string_view>("DFTRACER_UTILS_IO_THREADS");	1,884!
36	env.has_value()) {	942✔
37	const long long n =
NEW 38	std::strtoll(std::string(*env).c_str(), nullptr, 10);	×
NEW 39	if (n > 0) return static_cast<std::size_t>(n);	×
40	}
41	return requested == 0 ? dftracer_utils_hardware_concurrency() : requested;	942✔
42	}	471✔
43	} // namespace
44
45	Runtime::Runtime(std::size_t threads) : threads_(resolve_threads(threads)) {	438!
46	ExecutorConfig config;	292!
47	config.num_threads = threads_;	292✔
48	config.io_pool_size = resolve_io_threads(config.io_pool_size);	292!
49	executor_ = std::make_unique<Executor>(config);	292!
50	executor_->start();	292!
51
52	watchdog_ = std::make_unique<Watchdog>();	292!
53	watchdog_->set_executor(executor_.get());	292!
54	}	438✔
55
56	Runtime::Runtime(const ExecutorConfig& config, bool enable_watchdog)	234✔
57	: threads_(resolve_threads(config.num_threads)) {	234!
58	ExecutorConfig cfg = config;	156✔
59	cfg.num_threads = threads_;	156✔
60	cfg.io_pool_size = resolve_io_threads(config.io_pool_size);	156!
61	executor_ = std::make_unique<Executor>(cfg);	156!
62	executor_->start();	156!
63
64	if (enable_watchdog) {	156!
65	watchdog_ = std::make_unique<Watchdog>();	156!
66	watchdog_->set_executor(executor_.get());	156!
67	}	78✔
68	}	234✔
69
70	Runtime::Runtime(const ExecutorConfig& config,	741✔
71	std::unique_ptr<Watchdog> watchdog)	247✔
72	: threads_(resolve_threads(config.num_threads)) {	741!
73	ExecutorConfig cfg = config;	494✔
74	cfg.num_threads = threads_;	494✔
75	cfg.io_pool_size = resolve_io_threads(config.io_pool_size);	494!
76	executor_ = std::make_unique<Executor>(cfg);	494!
77	executor_->start();	494!
78
79	watchdog_ = std::move(watchdog);	494✔
80	if (watchdog_) {	494✔
81	watchdog_->set_executor(executor_.get());	162!
82	}	81✔
83	}	741✔
84
85	Runtime::~Runtime() { shutdown(); }	1,413!
86
87	TaskHandle Runtime::submit(coro::CoroTask<void> task, std::string name) {	1,066✔
88	if (shutdown_called_.load(std::memory_order_acquire)) {	1,066✔
89	throw DFTUtilsException(ErrorCode::PIPELINE, "Runtime is shut down");	4!
90	}
91	if (name.empty()) {	1,062✔
92	name = "task-" + std::to_string(task_name_counter_++);	6!
93	}	3✔
94
95	auto promise = std::make_shared<std::promise<void>>();	1,062!
96	auto future = promise->get_future().share();	1,062!
97	auto tid = std::make_shared<std::atomic<TaskIndex>>(-1);	1,062!
98
99	auto wrapper =	531✔
100	[](coro::CoroTask<void> t, std::shared_ptr<std::promise<void>> p,	4,242!
101	Executor* exec,
102	std::shared_ptr<std::atomic<TaskIndex>> task_id) -> coro::Coro {	531!
103	try {
104	co_await std::move(t);	2,649!
105	t = coro::CoroTask<void>{	1,056!
106	std::coroutine_handle<coro::CoroTask<void>::promise_type>{}};	528✔
107	exec->mark_coro_completed(task_id->load(std::memory_order_acquire));	528!
108	} catch (...) {	531✔
109	t = coro::CoroTask<void>{	6!
110	std::coroutine_handle<coro::CoroTask<void>::promise_type>{}};	3✔
111	exec->mark_coro_completed(task_id->load(std::memory_order_acquire));	3!
112	p->set_exception(std::current_exception());	3!
113	co_return;	3✔
114	}	3!
115	p->set_value();	528!
116	};	2,121!
117
118	// Set the executor on the task's promise so awaitables (e.g. channels)
119	// that capture `get_root_promise()->get_executor()` can schedule
120	// resumption. Without this, awaiters end up with executor=nullptr because
121	// the wrapping `coro::Coro` doesn't extend PromiseBase and the
122	// root-promise chain stops at the user's CoroTask.
123	if (task.handle()) {	1,062✔
124	task.handle().promise().set_executor(executor_.get());	1,062!
125	}	531✔
126	auto coro = wrapper(std::move(task), promise, executor_.get(), tid);	1,593!
127	TaskIndex id = executor_->enqueue_tracked(std::move(coro), name, tid);	1,062!
128
129	{
130	std::lock_guard<std::mutex> lock(futures_mutex_);	1,062!
131	cleanup_completed_futures();	1,062!
132	outstanding_futures_.push_back(future);	1,062!
133	}	1,062✔
134
135	return TaskHandle{future, id, std::move(name)};	1,593!
136	}	1,064✔
137
138	void Runtime::wait_all() {	200✔
139	std::vector<std::shared_future<void>> futures;	200✔
140	{
141	std::lock_guard<std::mutex> lock(futures_mutex_);	200!
142	futures = std::move(outstanding_futures_);	200✔
143	outstanding_futures_.clear();	200✔
144	}	200✔
145	for (auto& f : futures) {	274✔
146	f.wait();	74!
147	}
148	}	200✔
149
150	void Runtime::cleanup_completed_futures() {	1,234✔
151	outstanding_futures_.erase(	2,468!
152	std::remove_if(outstanding_futures_.begin(), outstanding_futures_.end(),	1,234!
153	[](const std::shared_future<void>& f) {	952✔
154	return f.wait_for(std::chrono::seconds(0)) ==	952!
155	std::future_status::ready;	456✔
156	}),
157	outstanding_futures_.end());	1,234✔
158	}	1,234✔
159
160	ExecutorProgress Runtime::get_progress() const {	38✔
161	return executor_->get_progress();	38✔
162	}
163
164	bool Runtime::is_responsive() const { return executor_->is_responsive(); }	4✔
165
166	void Runtime::set_global_timeout(std::chrono::milliseconds timeout) {	×
167	if (!watchdog_) {	×
168	throw DFTUtilsException(	×
169	ErrorCode::PIPELINE,
170	"Cannot set timeout: Runtime created without watchdog");	×
171	}
172	watchdog_->set_global_timeout(timeout);	×
173	}	×
174
175	void Runtime::set_default_task_timeout(std::chrono::milliseconds timeout) {	×
176	if (!watchdog_) {	×
177	throw DFTUtilsException(	×
178	ErrorCode::PIPELINE,
179	"Cannot set timeout: Runtime created without watchdog");	×
180	}
181	watchdog_->set_default_task_timeout(timeout);	×
182	}	×
183
184	void Runtime::shutdown() {	1,788✔
185	bool expected = false;	1,788✔
186	if (!shutdown_called_.compare_exchange_strong(expected, true)) return;	1,788✔
187	if (watchdog_) watchdog_->stop();	942!
188	if (executor_) executor_->shutdown();	942!
189	}	894✔
190
191	std::size_t Runtime::threads() const { return threads_; }	334✔
192
193	std::size_t Runtime::io_threads() const {	×
194	return executor_ ? executor_->get_io_pool_size() : 0;	×
195	}
196
197	} // namespace dftracer::utils

llnl / dftracer-utils / 30002033718

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous