KIKIMR-19215: process error in case many retries TEvScanError

1 files changed, 7 insertions, 1 deletions

diff --git a/ydb/core/kqp/compute_actor/kqp_scan_fetcher_actor.cpp b/ydb/core/kqp/compute_actor/kqp_scan_fetcher_actor.cpp
index 07eb0cf8875..d92a50f04ce 100644
--- a/ydb/core/kqp/compute_actor/kqp_scan_fetcher_actor.cpp
+++ b/ydb/core/kqp/compute_actor/kqp_scan_fetcher_actor.cpp
@@ -193,7 +193,6 @@ void TKqpScanFetcherActor::HandleExecute(TEvKqpCompute::TEvScanError::TPtr& ev)
 
     YQL_ENSURE(state->Generation == msg.GetGeneration());
 
-
     if (state->State == EShardState::Starting) {
         if (FindSchemeErrorInIssues(status, issues)) {
             return EnqueueResolveShard(state);
@@ -208,6 +207,13 @@ void TKqpScanFetcherActor::HandleExecute(TEvKqpCompute::TEvScanError::TPtr& ev)
         InFlightShards.ClearAckState(state);
         state->ResetRetry();
         ++TotalRetries;
+        if (state->TotalRetries >= MAX_SHARD_RETRIES || TotalRetries >= MAX_TOTAL_SHARD_RETRIES) {
+            CA_LOG_E("TKqpScanFetcherActor: broken tablet for this request " << state->TabletId
+                << ", retries limit exceeded (" << state->TotalRetries << "/" << TotalRetries << ")");
+            SendGlobalFail(NDqProto::COMPUTE_STATE_FAILURE, YdbStatusToDqStatus(status), issues);
+            return PassAway();
+        }
+
         return StartReadShard(state);
     }
 }