From 8da2a1fcd1b12988b7da2780b735dad316f82ee0 Mon Sep 17 00:00:00 2001
From: Mikhail Montsev <svartmetal@nebius.com>
Date: Tue, 24 Sep 2024 16:30:36 +0100
Subject: [PATCH] blockstore & filestore SDK retriable error codes should be in
 sync with cloud/storage/core/libs/common/error.cpp (#2119)

---
 .../blockstore/public/sdk/go/client/error.go  | 28 ++++++++++++++++--
 .../public/sdk/python/client/error.py         | 29 +++++++++++++++----
 cloud/filestore/public/sdk/go/client/error.go | 12 ++++++--
 .../public/sdk/python/client/error.py         | 14 +++++----
 4 files changed, 68 insertions(+), 15 deletions(-)

diff --git a/cloud/blockstore/public/sdk/go/client/error.go b/cloud/blockstore/public/sdk/go/client/error.go
index fc7b5d6889..e366718142 100644
--- a/cloud/blockstore/public/sdk/go/client/error.go
+++ b/cloud/blockstore/public/sdk/go/client/error.go
@@ -44,8 +44,14 @@ func (e *ClientError) IsRetriable() bool {
 	}
 
 	switch e.Facility() {
-	case FACILITY_GRPC, FACILITY_SYSTEM:
-		// system/network errors should be retriable
+	case FACILITY_GRPC:
+		if e.Code == E_GRPC_UNIMPLEMENTED {
+			return false
+		}
+		// network errors should be retriable
+		return true
+	case FACILITY_SYSTEM:
+		// system errors should be retriable
 		return true
 	case FACILITY_KIKIMR:
 		switch e.Status() {
@@ -59,6 +65,24 @@ func (e *ClientError) IsRetriable() bool {
 			20: // NKikimrProto::NOT_YET
 			return true
 		}
+	case FACILITY_SCHEMESHARD:
+		switch e.Status() {
+		case
+			13, // NKikimrScheme::StatusNotAvailable
+			8:  // NKikimrScheme::StatusMultipleModifications
+			return true
+		}
+	case FACILITY_TXPROXY:
+		switch e.Status() {
+		case
+			16, // NKikimr::NTxProxy::TResultStatus::ProxyNotReady
+			20, // NKikimr::NTxProxy::TResultStatus::ProxyShardNotAvailable
+			21, // NKikimr::NTxProxy::TResultStatus::ProxyShardTryLater
+			22, // NKikimr::NTxProxy::TResultStatus::ProxyShardOverloaded
+			51, // NKikimr::NTxProxy::TResultStatus::ExecTimeout
+			55: // NKikimr::NTxProxy::TResultStatus::ExecResultUnavailable
+			return true
+		}
 	}
 
 	// any other errors should not be retried automatically
diff --git a/cloud/blockstore/public/sdk/python/client/error.py b/cloud/blockstore/public/sdk/python/client/error.py
index 51aa256d03..0a3629298a 100644
--- a/cloud/blockstore/public/sdk/python/client/error.py
+++ b/cloud/blockstore/public/sdk/python/client/error.py
@@ -65,11 +65,14 @@ def is_retriable(self) -> bool:
 
         facility = self.facility
 
-        if facility in [
-            EFacility.FACILITY_GRPC.value,
-            EFacility.FACILITY_SYSTEM.value,
-        ]:
-            # system/network errors should be retriable
+        if facility == EFacility.FACILITY_GRPC.value:
+            if self.code == EResult.E_GRPC_UNIMPLEMENTED.value:
+                return False
+            # network errors should be retriable
+            return True
+
+        if facility == EFacility.FACILITY_SYSTEM.value:
+            # system errors should be retriable
             return True
 
         if facility == EFacility.FACILITY_KIKIMR.value and self.status in [
@@ -83,6 +86,22 @@ def is_retriable(self) -> bool:
         ]:
             return True
 
+        if facility == EFacility.FACILITY_SCHEMESHARD.value and self.status in [
+            13,  # NKikimrScheme::StatusNotAvailable
+            8,   # NKikimrScheme::StatusMultipleModifications
+        ]:
+            return True
+
+        if facility == EFacility.FACILITY_TXPROXY.value and self.status in [
+            16,  # NKikimr::NTxProxy::TResultStatus::ProxyNotReady
+            20,  # NKikimr::NTxProxy::TResultStatus::ProxyShardNotAvailable
+            21,  # NKikimr::NTxProxy::TResultStatus::ProxyShardTryLater
+            22,  # NKikimr::NTxProxy::TResultStatus::ProxyShardOverloaded
+            51,  # NKikimr::NTxProxy::TResultStatus::ExecTimeout:
+            55,  # NKikimr::NTxProxy::TResultStatus::ExecResultUnavailable:
+        ]:
+            return True
+
         # any other errors should not be retried automatically
         return False
 
diff --git a/cloud/filestore/public/sdk/go/client/error.go b/cloud/filestore/public/sdk/go/client/error.go
index 27936c3415..437895e0c5 100644
--- a/cloud/filestore/public/sdk/go/client/error.go
+++ b/cloud/filestore/public/sdk/go/client/error.go
@@ -38,14 +38,20 @@ func (e *ClientError) Error() string {
 
 func (e *ClientError) IsRetriable() bool {
 	switch e.Code {
-	case E_REJECTED, E_TIMEOUT:
+	case E_REJECTED, E_TIMEOUT, E_FS_OUT_OF_SPACE:
 		// special error code for retries
 		return true
 	}
 
 	switch e.Facility() {
-	case FACILITY_GRPC, FACILITY_SYSTEM:
-		// system/network errors should be retriable
+	case FACILITY_GRPC:
+		if e.Code == E_GRPC_UNIMPLEMENTED {
+			return false
+		}
+		// network errors should be retriable
+		return true
+	case FACILITY_SYSTEM:
+		// system errors should be retriable
 		return true
 	case FACILITY_KIKIMR:
 		switch e.Status() {
diff --git a/cloud/filestore/public/sdk/python/client/error.py b/cloud/filestore/public/sdk/python/client/error.py
index 821143c7e4..90bf50267b 100644
--- a/cloud/filestore/public/sdk/python/client/error.py
+++ b/cloud/filestore/public/sdk/python/client/error.py
@@ -57,16 +57,20 @@ def is_retriable(self):
         if self.code in [
             EResult.E_REJECTED.value,
             EResult.E_TIMEOUT.value,
+            EResult.E_FS_OUT_OF_SPACE.value,
         ]:
             return True
 
         facility = self.facility
 
-        if facility in [
-            EFacility.FACILITY_GRPC.value,
-            EFacility.FACILITY_SYSTEM.value,
-        ]:
-            # system/network errors should be retriable
+        if facility == EFacility.FACILITY_GRPC.value:
+            if self.code == EResult.E_GRPC_UNIMPLEMENTED.value:
+                return False
+            # network errors should be retriable
+            return True
+
+        if facility == EFacility.FACILITY_SYSTEM.value:
+            # system errors should be retriable
             return True
 
         if facility == EFacility.FACILITY_KIKIMR.value and self.status in [