Browse Source

YARN-8640. Restore previous state in container-executor after failure. Contributed by Jim Brennan

Jason Lowe 6 years ago
parent
commit
a896a011d7

+ 24 - 17
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager/src/main/native/container-executor/impl/container-executor.c

@@ -199,10 +199,12 @@ static int change_effective_user(uid_t user, gid_t group) {
  * cgroup_file: Path to cgroup file where pid needs to be written to.
  */
 static int write_pid_to_cgroup_as_root(const char* cgroup_file, pid_t pid) {
+  int rc = 0;
   uid_t user = geteuid();
   gid_t group = getegid();
   if (change_effective_user(0, 0) != 0) {
-    return -1;
+    rc = -1;
+    goto cleanup;
   }
 
   // open
@@ -210,7 +212,8 @@ static int write_pid_to_cgroup_as_root(const char* cgroup_file, pid_t pid) {
   if (cgroup_fd == -1) {
     fprintf(LOGFILE, "Can't open file %s as node manager - %s\n", cgroup_file,
            strerror(errno));
-    return -1;
+    rc = -1;
+    goto cleanup;
   }
 
   // write pid
@@ -221,15 +224,17 @@ static int write_pid_to_cgroup_as_root(const char* cgroup_file, pid_t pid) {
   if (written == -1) {
     fprintf(LOGFILE, "Failed to write pid to file %s - %s\n",
        cgroup_file, strerror(errno));
-    return -1;
+    rc = -1;
+    goto cleanup;
   }
 
+cleanup:
   // Revert back to the calling user.
   if (change_effective_user(user, group)) {
-    return -1;
+    rc = -1;
   }
 
-  return 0;
+  return rc;
 }
 #endif
 
@@ -238,15 +243,18 @@ static int write_pid_to_cgroup_as_root(const char* cgroup_file, pid_t pid) {
  * pid_file: Path to pid file where pid needs to be written to
  */
 static int write_pid_to_file_as_nm(const char* pid_file, pid_t pid) {
+  char *temp_pid_file = NULL;
+  int rc = 0;
   uid_t user = geteuid();
   gid_t group = getegid();
   if (change_effective_user(nm_uid, nm_gid) != 0) {
     fprintf(ERRORFILE, "Could not change to effective users %d, %d\n", nm_uid, nm_gid);
     fflush(ERRORFILE);
-    return -1;
+    rc = -1;
+    goto cleanup;
   }
 
-  char *temp_pid_file = concatenate("%s.tmp", "pid_file_path", 1, pid_file);
+  temp_pid_file = concatenate("%s.tmp", "pid_file_path", 1, pid_file);
   fprintf(LOGFILE, "Writing to tmp file %s\n", temp_pid_file);
   fflush(LOGFILE);
   // create with 700
@@ -255,8 +263,8 @@ static int write_pid_to_file_as_nm(const char* pid_file, pid_t pid) {
     fprintf(LOGFILE, "Can't open file %s as node manager - %s\n", temp_pid_file,
            strerror(errno));
     fflush(LOGFILE);
-    free(temp_pid_file);
-    return -1;
+    rc = -1;
+    goto cleanup;
   }
 
   // write pid to temp file
@@ -268,8 +276,8 @@ static int write_pid_to_file_as_nm(const char* pid_file, pid_t pid) {
     fprintf(LOGFILE, "Failed to write pid to file %s as node manager - %s\n",
        temp_pid_file, strerror(errno));
     fflush(LOGFILE);
-    free(temp_pid_file);
-    return -1;
+    rc = -1;
+    goto cleanup;
   }
 
   // rename temp file to actual pid file
@@ -279,18 +287,18 @@ static int write_pid_to_file_as_nm(const char* pid_file, pid_t pid) {
         temp_pid_file, pid_file, strerror(errno));
     fflush(LOGFILE);
     unlink(temp_pid_file);
-    free(temp_pid_file);
-    return -1;
+    rc = -1;
+    goto cleanup;
   }
 
+cleanup:
   // Revert back to the calling user.
   if (change_effective_user(user, group)) {
-	free(temp_pid_file);
-    return -1;
+    rc = -1;
   }
 
   free(temp_pid_file);
-  return 0;
+  return rc;
 }
 
 /**
@@ -1210,7 +1218,6 @@ char** tokenize_docker_command(const char *input, int *split_counter) {
   char *line = (char *)calloc(strlen(input) + 1, sizeof(char));
   char **linesplit = (char **) malloc(sizeof(char *));
   char *p = NULL;
-  int c = 0;
   *split_counter = 0;
   strncpy(line, input, strlen(input));