From 3d4b8814b75c6a32927558b589e0e119cd84f7a6 Mon Sep 17 00:00:00 2001
From: Yingchun Lai <laiyingchun@apache.org>
Date: Mon, 24 Apr 2023 21:02:12 +0800
Subject: [PATCH] refactor: fs manager

---
 src/replica/replica.cpp              | 7 +++++--
 src/replica/replica_failover.cpp     | 6 +++++-
 src/replica/replica_learn.cpp        | 8 ++++++--
 src/replica/replication_app_base.cpp | 6 ++++--
 4 files changed, 20 insertions(+), 7 deletions(-)

diff --git a/src/replica/replica.cpp b/src/replica/replica.cpp
index d2f9ccefbb..e23d389ed1 100644
--- a/src/replica/replica.cpp
+++ b/src/replica/replica.cpp
@@ -304,11 +304,14 @@ void replica::on_client_read(dsn::message_ex *request, bool ignore_throttling)
     auto storage_error = _app->on_request(request);
     if (dsn_unlikely(storage_error != ERR_OK)) {
         switch (storage_error) {
-        // TODO(yingchun): Now only kCorruption is dealt, consider to deal with more storage
-        //  engine errors.
+        // TODO(yingchun): Now only kCorruption and kIOError is dealt, consider to deal with
+        //  more storage engine errors.
         case rocksdb::Status::kCorruption:
             handle_local_failure(ERR_RDB_CORRUPTION);
             break;
+        case rocksdb::Status::kIOError:
+            handle_local_failure(ERR_DISK_IO_ERROR);
+            break;
         default:
             LOG_ERROR_PREFIX("client read encountered an unhandled error: {}", storage_error);
         }
diff --git a/src/replica/replica_failover.cpp b/src/replica/replica_failover.cpp
index 94edaab837..c465c221d7 100644
--- a/src/replica/replica_failover.cpp
+++ b/src/replica/replica_failover.cpp
@@ -33,8 +33,10 @@
  *     xxxx-xx-xx, author, fix bug about xxx
  */
 
+#include <atomic>
 #include <string>
 
+#include "common/fs_manager.h"
 #include "common/replication_common.h"
 #include "common/replication_enums.h"
 #include "dsn.layer2_types.h"
@@ -54,7 +56,9 @@ void replica::handle_local_failure(error_code error)
 {
     LOG_INFO_PREFIX("handle local failure error {}, status = {}", error, enum_to_string(status()));
 
-    if (error == ERR_RDB_CORRUPTION) {
+    if (error == ERR_DISK_IO_ERROR) {
+        _dir_node->status = disk_status::IO_ERROR;
+    } else if (error == ERR_RDB_CORRUPTION) {
         _data_corrupted = true;
     }
 
diff --git a/src/replica/replica_learn.cpp b/src/replica/replica_learn.cpp
index f65f775190..75fc9c5894 100644
--- a/src/replica/replica_learn.cpp
+++ b/src/replica/replica_learn.cpp
@@ -1238,8 +1238,12 @@ void replica::handle_learning_error(error_code err, bool is_local_error)
         err,
         is_local_error ? "local_error" : "remote error");
 
-    if (is_local_error && err == ERR_RDB_CORRUPTION) {
-        _data_corrupted = true;
+    if (is_local_error) {
+        if (err == ERR_DISK_IO_ERROR) {
+            _dir_node->status = disk_status::IO_ERROR;
+        } else if (err == ERR_RDB_CORRUPTION) {
+            _data_corrupted = true;
+        }
     }
 
     _stub->_counter_replicas_learning_recent_learn_fail_count->increment();
diff --git a/src/replica/replication_app_base.cpp b/src/replica/replication_app_base.cpp
index 5159fd757d..db92bdf17e 100644
--- a/src/replica/replication_app_base.cpp
+++ b/src/replica/replication_app_base.cpp
@@ -457,10 +457,12 @@ error_code replication_app_base::apply_mutation(const mutation *mu)
         //      an error.
         if (!has_ingestion_request) {
             switch (storage_error) {
-            // TODO(yingchun): Now only kCorruption is dealt, consider to deal with more storage
-            //  engine errors.
+            // TODO(yingchun): Now only kCorruption and kIOError are dealt, consider to deal with
+            //  more storage engine errors.
             case rocksdb::Status::kCorruption:
                 return ERR_RDB_CORRUPTION;
+            case rocksdb::Status::kIOError:
+                return ERR_DISK_IO_ERROR;
             default:
                 return ERR_LOCAL_APP_FAILURE;
             }