Bugfix: remove log in child process in case of deadlock in logger aft…

…er fork. Signed-off-by: Li Junlin <[email protected]>
PKUHPC · Nov 12, 2024 · dd082ef · dd082ef
1 parent a649dff
commit dd082ef
Show file tree

Hide file tree

Showing 3 changed files with 16 additions and 16 deletions.
diff --git a/src/Craned/CgroupManager.cpp b/src/Craned/CgroupManager.cpp
@@ -450,10 +450,10 @@ bool CgroupManager::ReleaseCgroup(uint32_t task_id, uid_t uid) {
     task_id_set_ptr->erase(task_id);
     if (task_id_set_ptr->empty()) {
       // In case of destroy a locked lock
-      task_id_set_empty=true;
+      task_id_set_empty = true;
     }
   }
-  if (task_id_set_empty)uid_task_ids_map_ptr->erase(uid);
+  if (task_id_set_empty) uid_task_ids_map_ptr->erase(uid);
 
   return true;
 }

diff --git a/src/Craned/CtldClient.cpp b/src/Craned/CtldClient.cpp
@@ -24,7 +24,7 @@ CtldClient::~CtldClient() {
   m_thread_stop_ = true;
 
   CRANE_TRACE("CtldClient is ending. Waiting for the thread to finish.");
-  if(m_async_send_thread_.joinable())m_async_send_thread_.join();
+  if (m_async_send_thread_.joinable()) m_async_send_thread_.join();
 }
 
 void CtldClient::InitChannelAndStub(const std::string& server_address) {

diff --git a/src/Craned/TaskManager.cpp b/src/Craned/TaskManager.cpp
@@ -714,8 +714,8 @@ CraneErr TaskManager::SpawnProcessInInstance_(TaskInstance* instance,
     const std::string& cwd = instance->task.cwd();
     rc = chdir(cwd.c_str());
     if (rc == -1) {
-      CRANE_ERROR("[Child Process] Error: chdir to {}. {}", cwd.c_str(),
-                  strerror(errno));
+      // CRANE_ERROR("[Child Process] Error: chdir to {}. {}", cwd.c_str(),
+      //             strerror(errno));
       std::abort();
     }
 
@@ -736,13 +736,13 @@ CraneErr TaskManager::SpawnProcessInInstance_(TaskInstance* instance,
 
     ok = ParseDelimitedFromZeroCopyStream(&msg, &istream, nullptr);
     if (!ok || !msg.ok()) {
-      if (!ok) {
-        int err = istream.GetErrno();
-        CRANE_ERROR("Failed to read socket from parent: {}", strerror(err));
-      }
+      // if (!ok) {
+      // int err = istream.GetErrno();
+      // CRANE_ERROR("Failed to read socket from parent: {}", strerror(err));
+      // }
 
-      if (!msg.ok())
-        CRANE_ERROR("Parent process ask not to start the subprocess.");
+      // if (!msg.ok())
+      // CRANE_ERROR("Parent process ask not to start the subprocess.");
 
       std::abort();
     }
@@ -758,8 +758,8 @@ CraneErr TaskManager::SpawnProcessInInstance_(TaskInstance* instance,
       stdout_fd =
           open(stdout_file_path.c_str(), O_RDWR | O_CREAT | O_TRUNC, 0644);
       if (stdout_fd == -1) {
-        CRANE_ERROR("[Child Process] Error: open {}. {}", stdout_file_path,
-                    strerror(errno));
+        // CRANE_ERROR("[Child Process] Error: open {}. {}", stdout_file_path,
+        // strerror(errno));
         std::abort();
       }
       dup2(stdout_fd, 1);
@@ -770,8 +770,8 @@ CraneErr TaskManager::SpawnProcessInInstance_(TaskInstance* instance,
         stderr_fd =
             open(stderr_file_path.c_str(), O_RDWR | O_CREAT | O_TRUNC, 0644);
         if (stderr_fd == -1) {
-          CRANE_ERROR("[Child Process] Error: open {}. {}", stderr_file_path,
-                      strerror(errno));
+          // CRANE_ERROR("[Child Process] Error: open {}. {}", stderr_file_path,
+          //             strerror(errno));
           std::abort();
         }
         dup2(stderr_fd, 2);  // stderr -> error file
@@ -795,7 +795,7 @@ CraneErr TaskManager::SpawnProcessInInstance_(TaskInstance* instance,
     ok = SerializeDelimitedToZeroCopyStream(child_process_ready, &ostream);
     ok &= ostream.Flush();
     if (!ok) {
-      CRANE_ERROR("[Child Process] Error: Failed to flush.");
+      // CRANE_ERROR("[Child Process] Error: Failed to flush.");
       std::abort();
     }