be/src/udf/python/python_udf_runtime.cpp

Source
// Licensed to the Apache Software Foundation (ASF) under one
// or more contributor license agreements.  See the NOTICE file
// distributed with this work for additional information
// regarding copyright ownership.  The ASF licenses this file
// to you under the Apache License, Version 2.0 (the
// "License"); you may not use this file except in compliance
// with the License.  You may obtain a copy of the License at
//
//   http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing,
// software distributed under the License is distributed on an
// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
// KIND, either express or implied.  See the License for the
// specific language governing permissions and limitations
// under the License.

#include "udf/python/python_udf_runtime.h"

#include <butil/fd_utility.h>
#include <signal.h>
#include <string.h>
#include <sys/wait.h>
#include <unistd.h>

#include <algorithm>
#include <boost/process.hpp>
#include <cerrno>
#include <chrono>
#include <condition_variable>
#include <deque>
#include <mutex>
#include <thread>
#ifdef BE_TEST
#include <atomic>
#endif

#include "common/logging.h"
#include "runtime/thread_context.h"

namespace doris {

#ifdef BE_TEST
static constexpr std::chrono::milliseconds PROCESS_TERMINATE_TIMEOUT {100};
#else
static constexpr std::chrono::milliseconds PROCESS_TERMINATE_TIMEOUT {1000};
#endif
static constexpr std::chrono::milliseconds BACKGROUND_REAP_INTERVAL {1000};

#ifdef BE_TEST
static std::atomic<int> FORCED_CHILD_EXIT_TIMEOUTS {0};

static bool consume_forced_child_exit_timeout() {
    int remaining = FORCED_CHILD_EXIT_TIMEOUTS.load(std::memory_order_relaxed);
    while (remaining > 0) {
        if (FORCED_CHILD_EXIT_TIMEOUTS.compare_exchange_weak(remaining, remaining - 1,
                                                             std::memory_order_relaxed)) {
            return true;
        }
    }
    return false;
}
#endif

struct BackgroundChildReaper {
    std::mutex mutex;
    std::condition_variable cv;
    std::deque<pid_t> pids;
#ifdef BE_TEST
    std::deque<pid_t> reaped_pids;
#endif
    std::thread thread;
};

static BackgroundChildReaper& background_child_reaper() {
    static auto* reaper = new BackgroundChildReaper();
    return *reaper;
}

void PythonUDFProcess::enqueue_child_for_reap(pid_t pid) {
    if (pid <= 0) [[unlikely]] {
        return;
    }

    auto& reaper = background_child_reaper();
    {
        std::lock_guard<std::mutex> lock(reaper.mutex);
        if (std::find(reaper.pids.begin(), reaper.pids.end(), pid) != reaper.pids.end()) {
            return;
        }
        reaper.pids.push_back(pid);
        if (!reaper.thread.joinable()) {
            // This thread only owns pids that were already SIGKILLed but could not be reaped within
            // the bounded shutdown wait. Such processes may be stuck in uninterruptible I/O; if they
            // exit later and nobody calls waitpid(), they stay as zombies under BE. Keep reaping them
            // asynchronously so foreground shutdown remains bounded without dropping wait ownership.
            reaper.thread = std::thread([]() {
                SCOPED_INIT_THREAD_CONTEXT();
                std::deque<pid_t> pending_pids;
                while (true) {
                    auto& reaper_ref = background_child_reaper();
                    std::unique_lock<std::mutex> lock(reaper_ref.mutex);
                    if (pending_pids.empty()) {
                        reaper_ref.cv.wait(lock,
                                           [&reaper_ref]() { return !reaper_ref.pids.empty(); });
                    } else {
                        reaper_ref.cv.wait_for(lock, BACKGROUND_REAP_INTERVAL);
                    }
                    pending_pids.insert(pending_pids.end(), reaper_ref.pids.begin(),
                                        reaper_ref.pids.end());
                    reaper_ref.pids.clear();
                    std::deque<pid_t> pids;
                    pids.swap(pending_pids);
                    lock.unlock();

                    for (pid_t pending_pid : pids) {
                        int exit_status = 0;
                        auto wait_result = PythonUDFProcess::wait_child_exit(
                                pending_pid, std::chrono::milliseconds(0), &exit_status);
                        if (wait_result == ChildExitWaitResult::EXITED ||
                            wait_result == ChildExitWaitResult::ALREADY_REAPED) {
                            LOG(INFO) << "Background reaped Python process pid=" << pending_pid;
#ifdef BE_TEST
                            {
                                std::lock_guard<std::mutex> reaped_lock(reaper_ref.mutex);
                                reaper_ref.reaped_pids.push_back(pending_pid);
                            }
                            reaper_ref.cv.notify_all();
#endif
                        } else if (wait_result == ChildExitWaitResult::TIMEOUT) {
                            pending_pids.push_back(pending_pid);
                        } else {
                            LOG(WARNING) << "Background failed to reap Python process pid="
                                         << pending_pid;
                        }
                    }
                }
            });
        }
    }
    reaper.cv.notify_one();
}

#ifdef BE_TEST
bool PythonUDFProcess::wait_background_reaped_for_test(pid_t pid,
                                                       std::chrono::milliseconds timeout) {
    auto& reaper = background_child_reaper();
    std::unique_lock<std::mutex> lock(reaper.mutex);
    return reaper.cv.wait_for(lock, timeout, [&reaper, pid]() {
        return std::find(reaper.reaped_pids.begin(), reaper.reaped_pids.end(), pid) !=
               reaper.reaped_pids.end();
    });
}

void PythonUDFProcess::force_child_exit_timeouts_for_test(int count) {
    FORCED_CHILD_EXIT_TIMEOUTS.store(count, std::memory_order_relaxed);
}
#endif

PythonUDFProcess::ChildExitWaitResult PythonUDFProcess::wait_child_exit(
        pid_t pid, std::chrono::milliseconds timeout, int* exit_status) {
#ifdef BE_TEST
    if (consume_forced_child_exit_timeout()) {
        return ChildExitWaitResult::TIMEOUT;
    }
#endif
    const auto deadline = std::chrono::steady_clock::now() + timeout;
    while (true) {
        pid_t ret = waitpid(pid, exit_status, WNOHANG);
        if (ret == pid) {
            return ChildExitWaitResult::EXITED;
        }
        if (ret < 0) {
            if (errno == EINTR) {
                // retry if interrupted
                continue;
            }
            // Another owner may already have observed the child exit through boost::process.
            if (errno == ECHILD) {
                return ChildExitWaitResult::ALREADY_REAPED;
            }
            LOG(WARNING) << "Failed to wait Python process pid=" << pid << ": " << strerror(errno);
            return ChildExitWaitResult::ERROR;
        }
        if (std::chrono::steady_clock::now() >= deadline) {
            return ChildExitWaitResult::TIMEOUT;
        }
        std::this_thread::sleep_for(std::chrono::milliseconds(10));
    }
}

void PythonUDFProcess::remove_unix_socket() {
    if (_uri.empty() || _unix_socket_file_path.empty()) return;

    if (unlink(_unix_socket_file_path.c_str()) == 0) {
        LOG(INFO) << "Successfully removed unix socket: " << _unix_socket_file_path;
        return;
    }

    if (errno == ENOENT) {
        // File does not exist, this is fine, no need to warn
        LOG(INFO) << "Unix socket not found (already removed): " << _uri;
    } else {
        LOG(WARNING) << "Failed to remove unix socket " << _uri << ": " << std::strerror(errno)
                     << " (errno=" << errno << ")";
    }
}

void PythonUDFProcess::shutdown() {
    if (!_child.valid() || _is_shutdown) return;

    int exit_status = 0;
    bool exited = !_child.running();
    bool status_available = false;
    bool already_reaped = false;
    if (!exited) {
        ::kill(_child_pid, SIGTERM);
        auto wait_result = wait_child_exit(_child_pid, PROCESS_TERMINATE_TIMEOUT, &exit_status);
        exited = wait_result == ChildExitWaitResult::EXITED ||
                 wait_result == ChildExitWaitResult::ALREADY_REAPED;
        status_available = wait_result == ChildExitWaitResult::EXITED;
        already_reaped = wait_result == ChildExitWaitResult::ALREADY_REAPED;
    } else {
        auto wait_result = wait_child_exit(_child_pid, std::chrono::milliseconds(0), &exit_status);
        status_available = wait_result == ChildExitWaitResult::EXITED;
        already_reaped = wait_result == ChildExitWaitResult::ALREADY_REAPED;
    }

    if (!exited) {
        LOG(WARNING) << "Python process did not terminate gracefully, sending SIGKILL, pid="
                     << _child_pid;
        ::kill(_child_pid, SIGKILL);
        auto wait_result = wait_child_exit(_child_pid, PROCESS_TERMINATE_TIMEOUT, &exit_status);
        exited = wait_result == ChildExitWaitResult::EXITED ||
                 wait_result == ChildExitWaitResult::ALREADY_REAPED;
        status_available = wait_result == ChildExitWaitResult::EXITED;
        already_reaped = wait_result == ChildExitWaitResult::ALREADY_REAPED;
    }
    _child.detach();

    if (!exited) [[unlikely]] {
        LOG(WARNING) << "Python process did not exit after SIGKILL, enqueue background reap, pid="
                     << _child_pid;
        enqueue_child_for_reap(_child_pid);
    } else if (already_reaped) {
        LOG(INFO) << "Python process already reaped by another owner, pid=" << _child_pid;
    } else if (!status_available) {
        LOG(INFO) << "Python process exited but exit status is unavailable, pid=" << _child_pid;
    } else {
        if (WIFSIGNALED(exit_status)) {
            LOG(INFO) << "Python process was killed by signal " << WTERMSIG(exit_status);
        } else if (WIFEXITED(exit_status)) {
            LOG(INFO) << "Python process exited normally with code: " << WEXITSTATUS(exit_status);
        } else {
            LOG(INFO) << "Python process exited";
        }
    }

    _output_stream.close();
    remove_unix_socket();
    _is_shutdown = true;
}

std::string PythonUDFProcess::to_string() const {
    return fmt::format(
            "PythonUDFProcess(child_pid={}, uri={}, "
            "unix_socket_file_path={}, is_shutdown={})",
            _child_pid, _uri, _unix_socket_file_path, _is_shutdown);
}

} // namespace doris

Coverage Report

Created: 2026-06-09 14:23

Line	Count	Source
1		// Licensed to the Apache Software Foundation (ASF) under one
2		// or more contributor license agreements. See the NOTICE file
3		// distributed with this work for additional information
4		// regarding copyright ownership. The ASF licenses this file
5		// to you under the Apache License, Version 2.0 (the
6		// "License"); you may not use this file except in compliance
7		// with the License. You may obtain a copy of the License at
8		//
9		// http://www.apache.org/licenses/LICENSE-2.0
10		//
11		// Unless required by applicable law or agreed to in writing,
12		// software distributed under the License is distributed on an
13		// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
14		// KIND, either express or implied. See the License for the
15		// specific language governing permissions and limitations
16		// under the License.
17
18		#include "udf/python/python_udf_runtime.h"
19
20		#include <butil/fd_utility.h>
21		#include <signal.h>
22		#include <string.h>
23		#include <sys/wait.h>
24		#include <unistd.h>
25
26		#include <algorithm>
27		#include <boost/process.hpp>
28		#include <cerrno>
29		#include <chrono>
30		#include <condition_variable>
31		#include <deque>
32		#include <mutex>
33		#include <thread>
34		#ifdef BE_TEST
35		#include <atomic>
36		#endif
37
38		#include "common/logging.h"
39		#include "runtime/thread_context.h"
40
41		namespace doris {
42
43		#ifdef BE_TEST
44		static constexpr std::chrono::milliseconds PROCESS_TERMINATE_TIMEOUT {100};
45		#else
46		static constexpr std::chrono::milliseconds PROCESS_TERMINATE_TIMEOUT {1000};
47		#endif
48		static constexpr std::chrono::milliseconds BACKGROUND_REAP_INTERVAL {1000};
49
50		#ifdef BE_TEST
51		static std::atomic<int> FORCED_CHILD_EXIT_TIMEOUTS {0};
52
53		static bool consume_forced_child_exit_timeout() {
54		int remaining = FORCED_CHILD_EXIT_TIMEOUTS.load(std::memory_order_relaxed);
55		while (remaining > 0) {
56		if (FORCED_CHILD_EXIT_TIMEOUTS.compare_exchange_weak(remaining, remaining - 1,
57		std::memory_order_relaxed)) {
58		return true;
59		}
60		}
61		return false;
62		}
63		#endif
64
65		struct BackgroundChildReaper {
66		std::mutex mutex;
67		std::condition_variable cv;
68		std::deque<pid_t> pids;
69		#ifdef BE_TEST
70		std::deque<pid_t> reaped_pids;
71		#endif
72		std::thread thread;
73		};
74
75	0	static BackgroundChildReaper& background_child_reaper() {
76	0	static auto* reaper = new BackgroundChildReaper();
77	0	return *reaper;
78	0	}
79
80	0	void PythonUDFProcess::enqueue_child_for_reap(pid_t pid) {
81	0	if (pid <= 0) [[unlikely]] {
82	0	return;
83	0	}
84
85	0	auto& reaper = background_child_reaper();
86	0	{
87	0	std::lock_guard<std::mutex> lock(reaper.mutex);
88	0	if (std::find(reaper.pids.begin(), reaper.pids.end(), pid) != reaper.pids.end()) {
89	0	return;
90	0	}
91	0	reaper.pids.push_back(pid);
92	0	if (!reaper.thread.joinable()) {
93		// This thread only owns pids that were already SIGKILLed but could not be reaped within
94		// the bounded shutdown wait. Such processes may be stuck in uninterruptible I/O; if they
95		// exit later and nobody calls waitpid(), they stay as zombies under BE. Keep reaping them
96		// asynchronously so foreground shutdown remains bounded without dropping wait ownership.
97	0	reaper.thread = std::thread([]() {
98	0	SCOPED_INIT_THREAD_CONTEXT();
99	0	std::deque<pid_t> pending_pids;
100	0	while (true) {
101	0	auto& reaper_ref = background_child_reaper();
102	0	std::unique_lock<std::mutex> lock(reaper_ref.mutex);
103	0	if (pending_pids.empty()) {
104	0	reaper_ref.cv.wait(lock,
105	0	[&reaper_ref]() { return !reaper_ref.pids.empty(); });
106	0	} else {
107	0	reaper_ref.cv.wait_for(lock, BACKGROUND_REAP_INTERVAL);
108	0	}
109	0	pending_pids.insert(pending_pids.end(), reaper_ref.pids.begin(),
110	0	reaper_ref.pids.end());
111	0	reaper_ref.pids.clear();
112	0	std::deque<pid_t> pids;
113	0	pids.swap(pending_pids);
114	0	lock.unlock();
115
116	0	for (pid_t pending_pid : pids) {
117	0	int exit_status = 0;
118	0	auto wait_result = PythonUDFProcess::wait_child_exit(
119	0	pending_pid, std::chrono::milliseconds(0), &exit_status);
120	0	if (wait_result == ChildExitWaitResult::EXITED \|\|
121	0	wait_result == ChildExitWaitResult::ALREADY_REAPED) {
122	0	LOG(INFO) << "Background reaped Python process pid=" << pending_pid;
123		#ifdef BE_TEST
124		{
125		std::lock_guard<std::mutex> reaped_lock(reaper_ref.mutex);
126		reaper_ref.reaped_pids.push_back(pending_pid);
127		}
128		reaper_ref.cv.notify_all();
129		#endif
130	0	} else if (wait_result == ChildExitWaitResult::TIMEOUT) {
131	0	pending_pids.push_back(pending_pid);
132	0	} else {
133	0	LOG(WARNING) << "Background failed to reap Python process pid="
134	0	<< pending_pid;
135	0	}
136	0	}
137	0	}
138	0	});
139	0	}
140	0	}
141	0	reaper.cv.notify_one();
142	0	}
143
144		#ifdef BE_TEST
145		bool PythonUDFProcess::wait_background_reaped_for_test(pid_t pid,
146		std::chrono::milliseconds timeout) {
147		auto& reaper = background_child_reaper();
148		std::unique_lock<std::mutex> lock(reaper.mutex);
149		return reaper.cv.wait_for(lock, timeout, [&reaper, pid]() {
150		return std::find(reaper.reaped_pids.begin(), reaper.reaped_pids.end(), pid) !=
151		reaper.reaped_pids.end();
152		});
153		}
154
155		void PythonUDFProcess::force_child_exit_timeouts_for_test(int count) {
156		FORCED_CHILD_EXIT_TIMEOUTS.store(count, std::memory_order_relaxed);
157		}
158		#endif
159
160		PythonUDFProcess::ChildExitWaitResult PythonUDFProcess::wait_child_exit(
161	0	pid_t pid, std::chrono::milliseconds timeout, int* exit_status) {
162		#ifdef BE_TEST
163		if (consume_forced_child_exit_timeout()) {
164		return ChildExitWaitResult::TIMEOUT;
165		}
166		#endif
167	0	const auto deadline = std::chrono::steady_clock::now() + timeout;
168	0	while (true) {
169	0	pid_t ret = waitpid(pid, exit_status, WNOHANG);
170	0	if (ret == pid) {
171	0	return ChildExitWaitResult::EXITED;
172	0	}
173	0	if (ret < 0) {
174	0	if (errno == EINTR) {
175		// retry if interrupted
176	0	continue;
177	0	}
178		// Another owner may already have observed the child exit through boost::process.
179	0	if (errno == ECHILD) {
180	0	return ChildExitWaitResult::ALREADY_REAPED;
181	0	}
182	0	LOG(WARNING) << "Failed to wait Python process pid=" << pid << ": " << strerror(errno);
183	0	return ChildExitWaitResult::ERROR;
184	0	}
185	0	if (std::chrono::steady_clock::now() >= deadline) {
186	0	return ChildExitWaitResult::TIMEOUT;
187	0	}
188	0	std::this_thread::sleep_for(std::chrono::milliseconds(10));
189	0	}
190	0	}
191
192	0	void PythonUDFProcess::remove_unix_socket() {
193	0	if (_uri.empty() \|\| _unix_socket_file_path.empty()) return;
194
195	0	if (unlink(_unix_socket_file_path.c_str()) == 0) {
196	0	LOG(INFO) << "Successfully removed unix socket: " << _unix_socket_file_path;
197	0	return;
198	0	}
199
200	0	if (errno == ENOENT) {
201		// File does not exist, this is fine, no need to warn
202	0	LOG(INFO) << "Unix socket not found (already removed): " << _uri;
203	0	} else {
204	0	LOG(WARNING) << "Failed to remove unix socket " << _uri << ": " << std::strerror(errno)
205	0	<< " (errno=" << errno << ")";
206	0	}
207	0	}
208
209	0	void PythonUDFProcess::shutdown() {
210	0	if (!_child.valid() \|\| _is_shutdown) return;
211
212	0	int exit_status = 0;
213	0	bool exited = !_child.running();
214	0	bool status_available = false;
215	0	bool already_reaped = false;
216	0	if (!exited) {
217	0	::kill(_child_pid, SIGTERM);
218	0	auto wait_result = wait_child_exit(_child_pid, PROCESS_TERMINATE_TIMEOUT, &exit_status);
219	0	exited = wait_result == ChildExitWaitResult::EXITED \|\|
220	0	wait_result == ChildExitWaitResult::ALREADY_REAPED;
221	0	status_available = wait_result == ChildExitWaitResult::EXITED;
222	0	already_reaped = wait_result == ChildExitWaitResult::ALREADY_REAPED;
223	0	} else {
224	0	auto wait_result = wait_child_exit(_child_pid, std::chrono::milliseconds(0), &exit_status);
225	0	status_available = wait_result == ChildExitWaitResult::EXITED;
226	0	already_reaped = wait_result == ChildExitWaitResult::ALREADY_REAPED;
227	0	}
228
229	0	if (!exited) {
230	0	LOG(WARNING) << "Python process did not terminate gracefully, sending SIGKILL, pid="
231	0	<< _child_pid;
232	0	::kill(_child_pid, SIGKILL);
233	0	auto wait_result = wait_child_exit(_child_pid, PROCESS_TERMINATE_TIMEOUT, &exit_status);
234	0	exited = wait_result == ChildExitWaitResult::EXITED \|\|
235	0	wait_result == ChildExitWaitResult::ALREADY_REAPED;
236	0	status_available = wait_result == ChildExitWaitResult::EXITED;
237	0	already_reaped = wait_result == ChildExitWaitResult::ALREADY_REAPED;
238	0	}
239	0	_child.detach();
240
241	0	if (!exited) [[unlikely]] {
242	0	LOG(WARNING) << "Python process did not exit after SIGKILL, enqueue background reap, pid="
243	0	<< _child_pid;
244	0	enqueue_child_for_reap(_child_pid);
245	0	} else if (already_reaped) {
246	0	LOG(INFO) << "Python process already reaped by another owner, pid=" << _child_pid;
247	0	} else if (!status_available) {
248	0	LOG(INFO) << "Python process exited but exit status is unavailable, pid=" << _child_pid;
249	0	} else {
250	0	if (WIFSIGNALED(exit_status)) {
251	0	LOG(INFO) << "Python process was killed by signal " << WTERMSIG(exit_status);
252	0	} else if (WIFEXITED(exit_status)) {
253	0	LOG(INFO) << "Python process exited normally with code: " << WEXITSTATUS(exit_status);
254	0	} else {
255	0	LOG(INFO) << "Python process exited";
256	0	}
257	0	}
258
259	0	_output_stream.close();
260	0	remove_unix_socket();
261	0	_is_shutdown = true;
262	0	}
263
264	1.33k	std::string PythonUDFProcess::to_string() const {
265	1.33k	return fmt::format(
266	1.33k	"PythonUDFProcess(child_pid={}, uri={}, "
267	1.33k	"unix_socket_file_path={}, is_shutdown={})",
268	1.33k	_child_pid, _uri, _unix_socket_file_path, _is_shutdown);
269	1.33k	}
270
271		} // namespace doris