RPC Reader now retrying on DQ (ydb-platform#4035)

maximyurchuk · Apr 26, 2024 · f76e6ef · f76e6ef
1 parent d30eee9
commit f76e6ef
Show file tree

Hide file tree

Showing 2 changed files with 7 additions and 2 deletions.
diff --git a/ydb/library/yql/providers/dq/task_runner_actor/task_runner_actor.cpp b/ydb/library/yql/providers/dq/task_runner_actor/task_runner_actor.cpp
@@ -255,11 +255,12 @@ class TTaskRunnerActor
 
         bool fallback = false;
         bool retry = false;
+        bool rpcReaderFalledBack = false;
         for (TStringBuf line: StringSplitter(input).SplitByString("\n")) {
             if (line.Contains("mlockall failed")) {
                 // skip
             } else {
-                if (!fallback) {
+                if (!fallback || rpcReaderFalledBack) {
                     if (line.Contains("FindColumnInfo(): requirement memberType->GetKind() == TType::EKind::Data")) {
                     // YQL-14757: temporary workaround for part6/produce-reduce_lambda_list_table-default.txt
                         fallback = true;
@@ -272,6 +273,10 @@ class TTaskRunnerActor
                     } else if (line.Contains("YT RPC Reader exception:")) {
                         // RPC reader fallback to YT
                         fallback = true;
+                        rpcReaderFalledBack = true;
+                    } else if (line.Contains("Attachments stream write timed out") || line.Contains("No alive peers found")) {
+                        // RPC reader DQ retry
+                        retry = true;
                     } else if (line.Contains("Transaction") && line.Contains("aborted")) {
                         // YQL-15542
                         fallback = true;

diff --git a/ydb/library/yql/providers/yt/comp_nodes/dq/dq_yt_rpc_reader.cpp b/ydb/library/yql/providers/yt/comp_nodes/dq/dq_yt_rpc_reader.cpp
@@ -86,7 +86,7 @@ void TParallelFileInputState::Finish() {
 
 void TParallelFileInputState::CheckError() const {
     if (!InnerState_->Error.IsOK()) {
-        Cerr << "YT RPC Reader exception:\n";
+        Cerr << "YT RPC Reader exception:\n" << InnerState_->Error.GetMessage();
         InnerState_->Error.ThrowOnError();
     }
 }