Reset PG after failures

eBay · Dec 25, 2024 · ab2184a · ab2184a
1 parent a64147b
commit ab2184a
Show file tree

Hide file tree

Showing 3 changed files with 11 additions and 4 deletions.
diff --git a/src/lib/homestore_backend/hs_homeobject.hpp b/src/lib/homestore_backend/hs_homeobject.hpp
@@ -587,6 +587,7 @@ class HSHomeObject : public HomeObjectImpl {
 
     std::shared_ptr< BlobIndexTable > recover_index_table(homestore::superblk< homestore::index_table_sb >&& sb);
     std::optional< pg_id_t > get_pg_id_with_group_id(homestore::group_id_t group_id) const;
+    bool is_pg_present(pg_id_t pg_id);
 
 private:
     std::shared_ptr< BlobIndexTable > create_index_table();

diff --git a/src/lib/homestore_backend/hs_pg_manager.cpp b/src/lib/homestore_backend/hs_pg_manager.cpp
@@ -280,6 +280,12 @@ std::optional< pg_id_t > HSHomeObject::get_pg_id_with_group_id(homestore::group_
     }
 }
 
+bool HSHomeObject::is_pg_present(pg_id_t pg_id) {
+    auto lg = std::scoped_lock(_pg_lock);
+    auto iter = _pg_map.find(pg_id);
+    return iter != _pg_map.end();
+}
+
 void HSHomeObject::pg_destroy(pg_id_t pg_id) {
     mark_pg_destroyed(pg_id);
     destroy_shards(pg_id);

diff --git a/src/lib/homestore_backend/replication_state_machine.cpp b/src/lib/homestore_backend/replication_state_machine.cpp
@@ -363,12 +363,12 @@ void ReplicationStateMachine::write_snapshot_obj(std::shared_ptr< homestore::sna
 
         auto pg_data = GetSizePrefixedResyncPGMetaData(data_buf);
 
-        // Check if the snapshot context is same as the current snapshot context.
-        // If not, drop the previous context and re-init a new one
-        if (m_snp_rcv_handler->get_context_lsn() != context->get_lsn()) {
+        //Check if pg exists, if yes, clean the stale pg resources, may be due to previous snapshot failure. Let's resync on a pristine base
+        if (home_object_->is_pg_present(pg_data->pg_id())) {
+            LOGI("pg already exists, clean pg resources before snapshot, pg_id:{} {}", pg_data->pg_id(), log_suffix);
+            home_object_->pg_destroy(pg_data->pg_id());
             LOGI("reset context from lsn:{} to lsn:{}", m_snp_rcv_handler->get_context_lsn(), context->get_lsn());
             m_snp_rcv_handler->reset_context(context->get_lsn(), pg_data->pg_id());
-            // TODO: Reset all data of current PG - let's resync on a pristine base
         }
 
         auto ret = m_snp_rcv_handler->process_pg_snapshot_data(*pg_data);