Allow ncclDebugLog to print messages longer than 1024 bytes

src/debug.cc

@@ -156,6 +156,8 @@ void ncclDebugLog(ncclDebugLogLevel level, unsigned long flags, const char *file
     va_start(vargs, fmt);
     (void) vsnprintf(ncclLastError, sizeof(ncclLastError), fmt, vargs);
     va_end(vargs);
+    // Make sure ncclLastError is always null-terminated
+    ncclLastError[sizeof(ncclLastError)-1] = '\0';
     pthread_mutex_unlock(&ncclDebugLock);
   }
   if (ncclDebugLevel < level || ((flags & ncclDebugMask) == 0)) return;
@@ -185,17 +187,22 @@ void ncclDebugLog(ncclDebugLogLevel level, unsigned long flags, const char *file
     len = snprintf(buffer, sizeof(buffer), "%s:%d:%d [%d] %f %s:%d NCCL TRACE ",
                    hostname, pid, tid, cudaDev, timestamp, filefunc, line);
   }
-
+  if (len > sizeof(buffer)) {
+    len = sizeof(buffer);
+  }
   if (len) {
+    len += snprintf(buffer+len, sizeof(buffer)-len, "%s\n", fmt);
+    // vsnprintf may return len > sizeof(buffer) in the case of a truncated output.
+    // Rewind len so that we can replace the final \0 by \n
+    if (len > sizeof(buffer)-1) {
+      len = sizeof(buffer)-1;
+    }
+    buffer[len-1] = '\n';
+    buffer[len] = '\0';
     va_list vargs;
     va_start(vargs, fmt);
-    len += vsnprintf(buffer+len, sizeof(buffer)-len, fmt, vargs);
+    vfprintf(ncclDebugFile, buffer, vargs);
     va_end(vargs);
-    // vsnprintf may return len > sizeof(buffer) in the case of a truncated output.
-    // Rewind len so that we can replace the final \0 by \n
-    if (len > sizeof(buffer)) len = sizeof(buffer)-1;
-    buffer[len++] = '\n';
-    fwrite(buffer, 1, len, ncclDebugFile);
   }
 }