YQ-3013: add retry for CURLE_GOT_NOTHING for FQ (#3565)

5 files changed, 65 insertions, 27 deletions

diff --git a/ydb/core/fq/libs/init/init.cpp b/ydb/core/fq/libs/init/init.cpp
index 87fe099ae6..908a0f233b 100644
--- a/ydb/core/fq/libs/init/init.cpp
+++ b/ydb/core/fq/libs/init/init.cpp
@@ -195,7 +195,7 @@ void Init(
 
     if (protoConfig.GetPrivateApi().GetEnabled()) {
         const auto& s3readConfig = protoConfig.GetReadActorsFactoryConfig().GetS3ReadActorFactoryConfig();
-        auto s3HttpRetryPolicy = NYql::GetHTTPDefaultRetryPolicy(TDuration::Max());
+        auto s3HttpRetryPolicy = NYql::GetHTTPDefaultRetryPolicy(NYql::THttpRetryPolicyOptions{.MaxTime = TDuration::Max(), .RetriedCurlCodes = NYql::FqRetriedCurlCodes()});
         NYql::NDq::TS3ReadActorFactoryConfig readActorFactoryCfg;
         if (const ui64 rowsInBatch = s3readConfig.GetRowsInBatch()) {
             readActorFactoryCfg.RowsInBatch = rowsInBatch;
diff --git a/ydb/core/kqp/compute_actor/kqp_compute_actor.cpp b/ydb/core/kqp/compute_actor/kqp_compute_actor.cpp
index d8b9e14d4e..9ec068e045 100644
--- a/ydb/core/kqp/compute_actor/kqp_compute_actor.cpp
+++ b/ydb/core/kqp/compute_actor/kqp_compute_actor.cpp
@@ -77,8 +77,9 @@ NYql::NDq::IDqAsyncIoFactory::TPtr CreateKqpAsyncIoFactory(
     RegisterSequencerActorFactory(*factory, counters);
 
     if (federatedQuerySetup) {
-        RegisterS3ReadActorFactory(*factory, federatedQuerySetup->CredentialsFactory, federatedQuerySetup->HttpGateway);
-        RegisterS3WriteActorFactory(*factory,  federatedQuerySetup->CredentialsFactory, federatedQuerySetup->HttpGateway);
+        auto s3HttpRetryPolicy = NYql::GetHTTPDefaultRetryPolicy(NYql::THttpRetryPolicyOptions{.RetriedCurlCodes = NYql::FqRetriedCurlCodes()});
+        RegisterS3ReadActorFactory(*factory, federatedQuerySetup->CredentialsFactory, federatedQuerySetup->HttpGateway, s3HttpRetryPolicy);
+        RegisterS3WriteActorFactory(*factory,  federatedQuerySetup->CredentialsFactory, federatedQuerySetup->HttpGateway, s3HttpRetryPolicy);
 
         if (federatedQuerySetup->ConnectorClient) {
             RegisterGenericProviderFactories(*factory, federatedQuerySetup->CredentialsFactory, federatedQuerySetup->ConnectorClient);
diff --git a/ydb/library/yql/providers/common/http_gateway/yql_http_default_retry_policy.cpp b/ydb/library/yql/providers/common/http_gateway/yql_http_default_retry_policy.cpp
index eb94f33826..994ed88d5a 100644
--- a/ydb/library/yql/providers/common/http_gateway/yql_http_default_retry_policy.cpp
+++ b/ydb/library/yql/providers/common/http_gateway/yql_http_default_retry_policy.cpp
@@ -2,31 +2,50 @@
 
 namespace NYql {
 
-IHTTPGateway::TRetryPolicy::TPtr GetHTTPDefaultRetryPolicy(TDuration maxTime, size_t maxRetries) {
+std::unordered_set<CURLcode> YqlRetriedCurlCodes() {
+    return {
+        CURLE_COULDNT_CONNECT,
+        CURLE_WEIRD_SERVER_REPLY,
+        CURLE_WRITE_ERROR,
+        CURLE_READ_ERROR,
+        CURLE_OPERATION_TIMEDOUT,
+        CURLE_SSL_CONNECT_ERROR,
+        CURLE_BAD_DOWNLOAD_RESUME,
+        CURLE_SEND_ERROR,
+        CURLE_RECV_ERROR,
+        CURLE_NO_CONNECTION_AVAILABLE
+    };
+}
+
+std::unordered_set<CURLcode> FqRetriedCurlCodes() {
+    return {
+        CURLE_COULDNT_CONNECT,
+        CURLE_WEIRD_SERVER_REPLY,
+        CURLE_WRITE_ERROR,
+        CURLE_READ_ERROR,
+        CURLE_OPERATION_TIMEDOUT,
+        CURLE_SSL_CONNECT_ERROR,
+        CURLE_BAD_DOWNLOAD_RESUME,
+        CURLE_SEND_ERROR,
+        CURLE_RECV_ERROR,
+        CURLE_NO_CONNECTION_AVAILABLE,
+        CURLE_GOT_NOTHING
+    };
+}
+
+IHTTPGateway::TRetryPolicy::TPtr GetHTTPDefaultRetryPolicy(THttpRetryPolicyOptions&& options) {
+    auto maxTime = options.MaxTime;
+    auto maxRetries = options.MaxRetries;
     if (!maxTime) {
         maxTime = TDuration::Minutes(5);
     }
-    return IHTTPGateway::TRetryPolicy::GetExponentialBackoffPolicy([](CURLcode curlCode, long httpCode) {
-
-        switch (curlCode) {
-            case CURLE_OK:
-                // look to http code
-                break;
-            case CURLE_COULDNT_CONNECT:
-            case CURLE_WEIRD_SERVER_REPLY:
-            case CURLE_WRITE_ERROR:
-            case CURLE_READ_ERROR:
-            case CURLE_OPERATION_TIMEDOUT:
-            case CURLE_SSL_CONNECT_ERROR:
-            case CURLE_BAD_DOWNLOAD_RESUME:
-            case CURLE_SEND_ERROR:
-            case CURLE_RECV_ERROR:
-            case CURLE_NO_CONNECTION_AVAILABLE:
-                // retry small number of known errors
-                return ERetryErrorClass::ShortRetry;
-            default:
-                // do not retry others
-                return ERetryErrorClass::NoRetry;
+    return IHTTPGateway::TRetryPolicy::GetExponentialBackoffPolicy([options = std::move(options)](CURLcode curlCode, long httpCode) {
+        if (curlCode == CURLE_OK) {
+            // pass
+        } else if (options.RetriedCurlCodes.contains(curlCode)) {
+            return ERetryErrorClass::ShortRetry;
+        } else {
+            return ERetryErrorClass::NoRetry;
         }
 
         switch (httpCode) {
@@ -52,4 +71,8 @@ IHTTPGateway::TRetryPolicy::TPtr GetHTTPDefaultRetryPolicy(TDuration maxTime, si
     maxTime); // maxTime
 }
 
+IHTTPGateway::TRetryPolicy::TPtr GetHTTPDefaultRetryPolicy(TDuration maxTime, size_t maxRetries) {
+    return GetHTTPDefaultRetryPolicy(THttpRetryPolicyOptions{.MaxTime = maxTime, .MaxRetries = maxRetries});
+}
+
 }
diff --git a/ydb/library/yql/providers/common/http_gateway/yql_http_default_retry_policy.h b/ydb/library/yql/providers/common/http_gateway/yql_http_default_retry_policy.h
index 74169c215a..6c7d323932 100644
--- a/ydb/library/yql/providers/common/http_gateway/yql_http_default_retry_policy.h
+++ b/ydb/library/yql/providers/common/http_gateway/yql_http_default_retry_policy.h
@@ -2,8 +2,22 @@
 
 #include "yql_http_gateway.h"
 
+#include <curl/curl.h>
+#include <unordered_set>
+
 namespace NYql {
 
-IHTTPGateway::TRetryPolicy::TPtr GetHTTPDefaultRetryPolicy(TDuration maxTime = TDuration::Zero(), size_t maxRetries = std::numeric_limits<size_t>::max()); // Zero means default maxTime
+std::unordered_set<CURLcode> YqlRetriedCurlCodes();
+std::unordered_set<CURLcode> FqRetriedCurlCodes();
+
+struct THttpRetryPolicyOptions {
+    TDuration MaxTime = TDuration::Zero(); // Zero means default maxTime
+    size_t MaxRetries = std::numeric_limits<size_t>::max();
+    std::unordered_set<CURLcode> RetriedCurlCodes = YqlRetriedCurlCodes();
+};
+
+IHTTPGateway::TRetryPolicy::TPtr GetHTTPDefaultRetryPolicy(THttpRetryPolicyOptions&& options = {});
+
+IHTTPGateway::TRetryPolicy::TPtr GetHTTPDefaultRetryPolicy(TDuration maxTime, size_t maxRetries = std::numeric_limits<size_t>::max()); // Zero means default maxTime
 
 }
diff --git a/ydb/library/yql/providers/s3/actors/yql_s3_applicator_actor.cpp b/ydb/library/yql/providers/s3/actors/yql_s3_applicator_actor.cpp
index 46f997504e..5b8f95449f 100644
--- a/ydb/library/yql/providers/s3/actors/yql_s3_applicator_actor.cpp
+++ b/ydb/library/yql/providers/s3/actors/yql_s3_applicator_actor.cpp
@@ -231,7 +231,7 @@ public:
     , CredentialsFactory(credentialsFactory)
     , ExternalEffect(externalEffect)
     , ActorSystem(NActors::TActivationContext::ActorSystem())
-    , RetryPolicy(NYql::GetHTTPDefaultRetryPolicy(TDuration::Zero(), 3))
+    , RetryPolicy(NYql::GetHTTPDefaultRetryPolicy(NYql::THttpRetryPolicyOptions{.MaxRetries = 3, .RetriedCurlCodes = NYql::FqRetriedCurlCodes()}))
     , RetryCount(GLOBAL_RETRY_LIMIT) {
         // ^^^ 3 retries in HTTP GW per operation
         // up to 100 retries at app level for all operations ^^^