Worker init clarification; Inform on error thread notification failure

eduardocl1212 · May 1, 2017 · 280e2a9 · 280e2a9
1 parent e7f453b
commit 280e2a9
Show file tree

Hide file tree

Showing 2 changed files with 7 additions and 3 deletions.
diff --git a/torch/lib/THD/master_worker/common/CommandChannel.cpp b/torch/lib/THD/master_worker/common/CommandChannel.cpp
@@ -11,6 +11,7 @@
 #include <string>
 #include <stdexcept>
 #include <utility>
+#include <iostream>
 
 namespace thd {
 namespace {
@@ -55,7 +56,9 @@ MasterCommandChannel::MasterCommandChannel()
 
 MasterCommandChannel::~MasterCommandChannel() {
   if (_error_thread.joinable()) {
-    ::write(_error_pipe, "exit", 4);
+    if (::write(_error_pipe, "exit", 4) != 4) {
+      std::cerr << "Failed to notify error thread" << std::endl;
+    }
     _error_thread.join();
 
     ::close(_error_pipe);

diff --git a/torch/lib/THD/master_worker/master/Master.cpp b/torch/lib/THD/master_worker/master/Master.cpp
@@ -20,8 +20,9 @@ bool THDMasterWorkerInit(THDChannelType channel_type) {
 
   if (dataChannel->getRank() > 0) {
     /*
-     * Worker initialization. If initialization succeeds it goes into
-     * infinite loop in which waits for commands from master.
+     * Worker initialization. It goes into infinite loop in which waits
+     * for commands from master. Returning from `THDWorkerMain` indicates
+     * a failure so it will `return false`.
      */
     THDWorkerMain();
     return false;