fix listener can't receive snapshot

vesoft-inc · May 10, 2022 · 6881810 · 6881810
1 parent 1fd89af
commit 6881810
Show file tree

Hide file tree

Showing 3 changed files with 17 additions and 15 deletions.
diff --git a/src/kvstore/Listener.cpp b/src/kvstore/Listener.cpp
@@ -147,6 +147,9 @@ void Listener::doApply() {
   if (isStopped()) {
     return;
   }
+  if (needToCleanupSnapshot()) {
+    cleanupSnapshot();
+  }
   // todo(doodle): only put is handled, all remove is ignored for now
   folly::via(executor_.get(), [this] {
     SCOPE_EXIT {

diff --git a/src/kvstore/raftex/RaftPart.h b/src/kvstore/raftex/RaftPart.h
@@ -557,6 +557,17 @@ class RaftPart : public std::enable_shared_from_this<RaftPart> {
    */
   void removePeer(const HostAddr& peer);
 
+  /**
+   * @brief Return whether need to clean snapshot when a node has not received the snapshot for a
+   * period of time
+   */
+  bool needToCleanupSnapshot();
+
+  /**
+   * @brief Convert to follower when snapshot has been outdated
+   */
+  void cleanupSnapshot();
+
  private:
   // A list of <idx, resp>
   // idx  -- the index of the peer
@@ -625,17 +636,6 @@ class RaftPart : public std::enable_shared_from_this<RaftPart> {
    */
   bool needToStartElection();
 
-  /**
-   * @brief Return whether need to clean snapshot when a node has not received the snapshot for a
-   * period of time
-   */
-  bool needToCleanupSnapshot();
-
-  /**
-   * @brief Convert to follower when snapshot has been outdated
-   */
-  void cleanupSnapshot();
-
   /**
    * @brief The method sends out AskForVote request. Return true if I have been granted majority
    * votes on proposedTerm, no matter isPreVote or not

diff --git a/src/kvstore/raftex/SnapshotManager.cpp b/src/kvstore/raftex/SnapshotManager.cpp
@@ -78,11 +78,10 @@ folly::Future<StatusOr<std::pair<LogID, TermID>>> SnapshotManager::sendSnapshot(
                 }
                 return true;
               } else {
-                VLOG(2) << part->idStr_ << "Sending snapshot failed, we don't retry anymore! "
-                        << "The error code is "
+                VLOG(2) << part->idStr_ << "Sending snapshot failed, the error code is "
                         << apache::thrift::util::enumNameSafe(resp.get_error_code());
-                p.setValue(Status::Error("Send snapshot failed!"));
-                return false;
+                sleep(1);
+                continue;
               }
             } catch (const std::exception& e) {
               VLOG(3) << part->idStr_ << "Send snapshot failed, exception " << e.what()