ANSCENTER/ANSCORE
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fix deadlock

Browse Source
This commit is contained in:
Tuan Nghia Nguyen
2026-04-24 12:54:16 +10:00
parent e2bf17289d
commit fd2394a85a
3 changed files with 138 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
core/ANSLicensingSystem/Utility.cpp
View File

@@ -11,17 +11,23 @@
// model folder — without that, thread A can finish extraction and begin opening
// train_last.onnx while thread B re-enters extraction and truncates the file,
// producing "system error number 13" (EACCES) on the first reader.
// Recursive so the same thread can re-acquire the lock through layered load
// calls — ANSALPR_OD::LoadEngine -> ANSONNXYOLO::LoadModelFromFolder both
// acquire the SAME folder lock on the SAME thread. A non-recursive
// timed_mutex deadlocks that nesting for 120 s then fails. Recursive keeps
// cross-thread serialization intact while allowing legitimate re-entry from
// the lock-holding thread.
static std::mutex g_zipPathMapMutex;
static std::map<std::string, std::shared_ptr<std::timed_mutex>> g_zipPathLocks;
static std::map<std::string, std::shared_ptr<std::recursive_timed_mutex>> g_zipPathLocks;
static std::shared_ptr<std::timed_mutex> GetZipPathLock(const std::string& path) {
static std::shared_ptr<std::recursive_timed_mutex> GetZipPathLock(const std::string& path) {
std::lock_guard<std::mutex> guard(g_zipPathMapMutex);
auto& ptr = g_zipPathLocks[path];
if (!ptr) ptr = std::make_shared<std::timed_mutex>();
if (!ptr) ptr = std::make_shared<std::recursive_timed_mutex>();
return ptr;
}
std::shared_ptr<std::timed_mutex> GetModelFolderLock(const std::string& folderPath) {
std::shared_ptr<std::recursive_timed_mutex> GetModelFolderLock(const std::string& folderPath) {
auto ptr = GetZipPathLock(folderPath);
ANS_DBG("ModelLock", "GetModelFolderLock: folder=%s mutex=%p",
folderPath.c_str(), (void*)ptr.get());
@@ -464,7 +470,7 @@ bool AddFolderContentsToZip(zip* archive, const char* folderPath, const char* zi
bool ZipFolderWithPassword(const char* folderPath, const char* zipFilePath, const char* password) {
auto pathLock = GetZipPathLock(std::string(zipFilePath));
std::lock_guard<std::timed_mutex> zipGuard(*pathLock);
std::lock_guard<std::recursive_timed_mutex> zipGuard(*pathLock);
zip* zipArchive;
zip_flags_t flags = ZIP_CREATE | ZIP_TRUNCATE;
@@ -850,7 +856,7 @@ std::string GetDateTimeString(const std::string& format) {
bool ExtractProtectedZipFile(const std::string& zipFileName, const std::string& password, const std::string& modelName, const std::string outputFolder)
{
auto pathLock = GetZipPathLock(outputFolder);
std::lock_guard<std::timed_mutex> zipGuard(*pathLock);
std::lock_guard<std::recursive_timed_mutex> zipGuard(*pathLock);
int error;
if (!FileExist(zipFileName))return false;

19
core/ANSLicensingSystem/Utility.h
View File

@@ -96,9 +96,12 @@ namespace fs = std::filesystem;
// creation on the same extracted folder so concurrent CreateANSODHandle calls
// cannot truncate/rewrite a model file while another thread is loading it.
// Keyed by folder path (not zip path) so both extractor and consumer agree.
// Returns std::timed_mutex so callers can bound their wait and avoid a hang
// if a peer thread deadlocks inside extraction or ORT session creation.
ANSLICENSE_API std::shared_ptr<std::timed_mutex> GetModelFolderLock(const std::string& folderPath);
// Returns std::recursive_timed_mutex so callers can bound their wait and
// recursion — layered load paths (e.g. ANSALPR_OD::LoadEngine ->
// ANSONNXYOLO::LoadModelFromFolder) legitimately re-enter on the same
// thread; a non-recursive timed_mutex self-deadlocks that nesting. Cross-
// thread serialization is unchanged.
ANSLICENSE_API std::shared_ptr<std::recursive_timed_mutex> GetModelFolderLock(const std::string& folderPath);
// ============================================================================
// ModelFolderLockGuard
@@ -111,8 +114,10 @@ namespace fs = std::filesystem;
// error number 13" (EACCES) on the reader.
//
// Backed by GetModelFolderLock() above which returns a process-wide
// std::timed_mutex keyed on the folder path. The extractor takes the same
// lock, so extract ↔ open is mutually exclusive.
// std::recursive_timed_mutex keyed on the folder path. The extractor takes
// the same lock, so extract ↔ open is mutually exclusive across threads,
// while same-thread re-entry (layered loaders) is permitted without
// deadlocking.
//
// Acquisition is bounded by `timeout` (default 120 s) so a deadlocked peer
// cannot hang the caller thread forever. On timeout, .acquired() is false and
@@ -165,7 +170,7 @@ namespace ANSCENTER {
return;
}
auto lock = GetModelFolderLock(folderPath);
_guard = std::unique_lock<std::timed_mutex>(*lock, std::defer_lock);
_guard = std::unique_lock<std::recursive_timed_mutex>(*lock, std::defer_lock);
ANS_DBG("EngineLoad",
"%s: waiting on folder lock (%llds): %s",
_caller, (long long)timeout.count(), folderPath.c_str());
@@ -196,7 +201,7 @@ namespace ANSCENTER {
private:
const char* _caller;
std::string _folder;
std::unique_lock<std::timed_mutex> _guard;
std::unique_lock<std::recursive_timed_mutex> _guard;
bool _ok = false;
};
} // namespace ANSCENTER