Use pthread_exit to terminate raft service. (#536)

* Use pthread_exit to terminate raft service. Allow the process to handle termination of raft service rather than force process wide `exit()`. This is compile time configurable during CMake configuration. Addresses #535 --------- Co-authored-by: Jung-Sang Ahn <[email protected]>
eBay · Sep 18, 2024 · 6741a45 · 6741a45
1 parent ea1385e
commit 6741a45
Show file tree

Hide file tree

Showing 6 changed files with 43 additions and 15 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -78,6 +78,7 @@ endif()
 
 
 # === Compiler flags ===
+option(USE_PTHREAD_EXIT "Call pthread_exit on server threads" OFF)
 if (NOT WIN32)
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wall")
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-pessimizing-move")
@@ -95,11 +96,19 @@ if (NOT WIN32)
     else()
         set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -pthread")
     endif ()
+    if (USE_PTHREAD_EXIT)
+        message(STATUS "Using ::pthread_exit for termination")
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DUSE_PTHREAD_EXIT")
+    endif()
 
 else ()
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /wd5045 /wd4571 /wd4774 /wd4820 /wd5039 /wd4626 /wd4625 /wd5026 /wd5027 /wd4623 /wd4996 /wd4530 /wd4267 /wd4244 /W3")
     message(STATUS "---- WIN32 ----")
     set(DISABLE_SSL 1)
+
+    if (USE_PTHREAD_EXIT)
+        message(FATAL_ERROR "Using ::pthread_exit not supported on Windows")
+    endif()
 endif ()
 
 # === Disable SSL ===

diff --git a/src/exit_handler.hxx b/src/exit_handler.hxx
@@ -0,0 +1,15 @@
+#pragma once
+
+// What should be called to shutdown when raft_server is in abnormal state
+#ifdef USE_PTHREAD_EXIT
+extern "C" {
+#include <pthread.h>
+}
+#define _sys_exit(status) ::pthread_exit(nullptr)
+
+#else
+
+#include <cstdlib>
+#define _sys_exit(status) ::exit((status))
+
+#endif
diff --git a/src/handle_append_entries.cxx b/src/handle_append_entries.cxx
@@ -25,6 +25,7 @@ limitations under the License.
 #include "cluster_config.hxx"
 #include "error_code.hxx"
 #include "event_awaiter.hxx"
+#include "exit_handler.hxx"
 #include "handle_custom_notification.hxx"
 #include "peer.hxx"
 #include "snapshot.hxx"
@@ -406,7 +407,7 @@ ptr<req_msg> raft_server::create_append_entries_req(ptr<peer>& pp) {
         p_er( "Peer's lastLogIndex is too large %" PRIu64 " v.s. %" PRIu64 ", ",
               last_log_idx, cur_nxt_idx );
         ctx_->state_mgr_->system_exit(raft_err::N8_peer_last_log_idx_too_large);
-        ::exit(-1);
+        _sys_exit(-1);
         return ptr<req_msg>();
         // LCOV_EXCL_STOP
     }

diff --git a/src/handle_commit.cxx b/src/handle_commit.cxx
@@ -23,6 +23,7 @@ limitations under the License.
 
 #include "cluster_config.hxx"
 #include "error_code.hxx"
+#include "exit_handler.hxx"
 #include "handle_client_request.hxx"
 #include "global_mgr.hxx"
 #include "peer.hxx"
@@ -150,7 +151,7 @@ void raft_server::commit_in_bg() {
               "exiting to protect the system",
               err.what() );
         ctx_->state_mgr_->system_exit(raft_err::N20_background_commit_err);
-        ::exit(-1);
+        _sys_exit(-1);
         // LCOV_EXCL_STOP
      }
     }
@@ -228,7 +229,7 @@ bool raft_server::commit_in_bg_exec(size_t timeout_ms) {
             // LCOV_EXCL_START
             p_ft( "failed to get log entry with idx %" PRIu64 "", index_to_commit );
             ctx_->state_mgr_->system_exit(raft_err::N19_bad_log_idx_for_term);
-            ::exit(-1);
+            _sys_exit(-1);
             // LCOV_EXCL_STOP
         }
 
@@ -239,7 +240,7 @@ bool raft_server::commit_in_bg_exec(size_t timeout_ms) {
             p_ft( "empty log at idx %" PRIu64 ", must be log corruption",
                   index_to_commit );
             ctx_->state_mgr_->system_exit(raft_err::N19_bad_log_idx_for_term);
-            ::exit(-1);
+            _sys_exit(-1);
             // LCOV_EXCL_STOP
         }
 
@@ -304,7 +305,7 @@ void raft_server::commit_app_log(ulong idx_to_commit,
         p_ft( "pre-commit index %" PRIu64 " is smaller than commit index %" PRIu64,
               pc_idx, sm_idx );
         ctx_->state_mgr_->system_exit(raft_err::N23_precommit_order_inversion);
-        ::exit(-1);
+        _sys_exit(-1);
     }
     ret_value = state_machine_->commit_ext
                 ( state_machine::ext_op_params( sm_idx, buf ) );
@@ -583,7 +584,7 @@ bool raft_server::snapshot_and_compact(ulong committed_idx, bool forced_creation
                      "cannot be found in current committed logs, "
                      "this is a system error, exiting");
                 ctx_->state_mgr_->system_exit(raft_err::N6_no_snapshot_found);
-                ::exit(-1);
+                _sys_exit(-1);
                 return false;
                 // LCOV_EXCL_STOP
             }
@@ -600,7 +601,7 @@ bool raft_server::snapshot_and_compact(ulong committed_idx, bool forced_creation
                  ", committed idx %" PRIu64,
                  conf->get_log_idx(), conf->get_prev_log_idx(), committed_idx);
             //ctx_->state_mgr_->system_exit(raft_err::N7_no_config_at_idx_one);
-            //::exit(-1);
+            //_sys_exit(-1);
             //return;
         }
 

diff --git a/src/handle_snapshot_sync.cxx b/src/handle_snapshot_sync.cxx
@@ -23,6 +23,7 @@ limitations under the License.
 #include "context.hxx"
 #include "error_code.hxx"
 #include "event_awaiter.hxx"
+#include "exit_handler.hxx"
 #include "peer.hxx"
 #include "snapshot.hxx"
 #include "snapshot_sync_ctx.hxx"
@@ -123,7 +124,7 @@ ptr<req_msg> raft_server::create_sync_snapshot_req(ptr<peer>& pp,
                      last_log_idx, snp->get_last_log_idx());
             }
             ctx_->state_mgr_->system_exit(raft_err::N16_snapshot_for_peer_not_found);
-            ::exit(-1);
+            _sys_exit(-1);
             return ptr<req_msg>();
             // LCOV_EXCL_STOP
         }
@@ -135,7 +136,7 @@ ptr<req_msg> raft_server::create_sync_snapshot_req(ptr<peer>& pp,
                  "machine implementation, stop the system to prevent "
                  "further errors");
             ctx_->state_mgr_->system_exit(raft_err::N17_empty_snapshot);
-            ::exit(-1);
+            _sys_exit(-1);
             return ptr<req_msg>();
             // LCOV_EXCL_STOP
         }
@@ -187,7 +188,7 @@ ptr<req_msg> raft_server::create_sync_snapshot_req(ptr<peer>& pp,
                   "bytes are expected, must be something wrong, exit.",
                   sz_rd, data->size() );
             ctx_->state_mgr_->system_exit(raft_err::N18_partial_snapshot_block);
-            ::exit(-1);
+            _sys_exit(-1);
             return ptr<req_msg>();
             // LCOV_EXCL_STOP
         }
@@ -251,7 +252,7 @@ ptr<resp_msg> raft_server::handle_install_snapshot_req(req_msg& req, std::unique
                   req.get_src() );
             ctx_->state_mgr_->system_exit
                 ( raft_err::N10_leader_receive_InstallSnapshotRequest );
-            ::exit(-1);
+            _sys_exit(-1);
             return ptr<resp_msg>();
             // LCOV_EXCL_STOP
 
@@ -543,7 +544,7 @@ bool raft_server::handle_snapshot_sync_req(snapshot_sync_req& req, std::unique_l
             // LCOV_EXCL_START
             p_er("bad server role for applying a snapshot, exit for debugging");
             ctx_->state_mgr_->system_exit(raft_err::N11_not_follower_for_snapshot);
-            ::exit(-1);
+            _sys_exit(-1);
             // LCOV_EXCL_STOP
         }
 
@@ -564,7 +565,7 @@ bool raft_server::handle_snapshot_sync_req(snapshot_sync_req& req, std::unique_l
                 p_er("failed to apply the snapshot after log compacted, "
                      "to ensure the safety, will shutdown the system");
                 ctx_->state_mgr_->system_exit(raft_err::N12_apply_snapshot_failed);
-                ::exit(-1);
+                _sys_exit(-1);
                 return false;
                 // LCOV_EXCL_STOP
             }
@@ -608,7 +609,7 @@ bool raft_server::handle_snapshot_sync_req(snapshot_sync_req& req, std::unique_l
     // LCOV_EXCL_START
     p_er("failed to handle snapshot installation due to system errors");
     ctx_->state_mgr_->system_exit(raft_err::N13_snapshot_install_failed);
-    ::exit(-1);
+    _sys_exit(-1);
     return false;
     // LCOV_EXCL_STOP
  }

diff --git a/src/raft_server.cxx b/src/raft_server.cxx
@@ -24,6 +24,7 @@ limitations under the License.
 #include "context.hxx"
 #include "error_code.hxx"
 #include "event_awaiter.hxx"
+#include "exit_handler.hxx"
 #include "global_mgr.hxx"
 #include "handle_client_request.hxx"
 #include "handle_custom_notification.hxx"
@@ -1628,7 +1629,7 @@ ulong raft_server::term_for_log(ulong log_idx) {
         }
         p_lv(log_lv, "log_store_->start_index() %" PRIu64, log_store_->start_index());
         //ctx_->state_mgr_->system_exit(raft_err::N19_bad_log_idx_for_term);
-        //::exit(-1);
+        //_sys_exit(-1);
         return 0L;
     }