Bug#31341888 RECURRING TEMPORARY ERRORS STALL NDB METADATA SYNC

Post push fix: Temporary errors are bumped up to permanent ones only when ndb_metadata_sync is enabled. Recurring temporary errors are expected in the general use case for auto sync especially when the restore takes longer to complete. Objects being synchronized are more likely to be prematurely blacklisted in such scenarios. Change-Id: I6f8f4fb40f1059ebb936da0ee0ba16093cef6125

Author: arnabray21

storage/ndb/plugin/ndb_metadata_sync.cc

@@ -486,9 +486,25 @@ unsigned int Ndb_metadata_sync::get_blacklist_count() {
  return m_blacklist.size();
 }
 
+extern bool opt_ndb_metadata_sync;
+
 bool Ndb_metadata_sync::retry_limit_exceeded(const std::string &schema_name,
  const std::string &name,
  object_detected_type type) {
+ if (!opt_ndb_metadata_sync) {
+ /*
+ The ndb_metadata_sync variable hasn't been set. This is then the default
+ automatic sync mechanism where it's better to retry indefinitely under the
+ assumption that the temporary error will have disappeared by the time the
+ next discovery + sync attempt occurs.
+ */
+ return false;
+ }
+ /*
+ The ndb_metadata_sync variable has been set. Check if the retry limit (10)
+ has been hit in which case the object is added to the blacklist by the
+ caller.
+ */
  std::lock_guard<std::mutex> guard(m_retry_objects_mutex);
  for (Detected_object &object : m_retry_objects) {
  if (object.m_type == type && object.m_schema_name == schema_name &&

storage/ndb/plugin/ndb_metadata_sync.h

@@ -336,7 +336,8 @@ class Ndb_metadata_sync {
 
  /*
  @brief Checks if the number of times the synchronization of an object has
- been retried has exceeded the retry limit
+ been retried has exceeded the retry limit. This is applicable only
+ when ndb_metadata_sync is used
 
  @param schema_name Name of the schema
  @param name Name of the object