Merge 'cnch_dev_s3_unique_index_handler' into 'cnch-dev'

fix(clickhousech@m-5283670005): fix s3 handler for unique index read

See merge request: !25432

src/Storages/IndexFile/Env.h

@@ -116,6 +116,10 @@ public:
     //
     // Safe for concurrent use by multiple threads.
     virtual Status Read(uint64_t offset, size_t n, Slice * result, char * scratch, bool * from_local) const = 0;
+
+    // After the iter call ends, we need to explicitly release the buffer, otherwise there may be too many open files due to remote handles.
+    // Safe for concurrent use by multiple threads.
+    virtual void releaseRemoteFD() const { }
 };
 
 // A file abstraction for sequential writing.  The implementation

src/Storages/IndexFile/IndexFileReader.cpp

@@ -47,7 +47,11 @@ Status IndexFileReader::Open(const RemoteFileInfo & remote_file)
     std::unique_ptr<RandomAccessFile> file;
     Status s = rep->options.env->NewRandomAccessRemoteFileWithCache(remote_file, rep->options.remote_file_cache, &file);
     if (s.ok())
+    {
         s = Table::Open(rep->options, std::move(file), remote_file.size, &rep->table_reader);
+        /// directly release remote FD, as SSTable read from footer, meta, etc, then data block
+        rep->table_reader->releaseRemoteFD();
+    }
     return s;
 }
 

src/Storages/IndexFile/PosixEnv.cpp

@@ -141,6 +141,13 @@ namespace
             return s;
         }
 
+        virtual void releaseRemoteFD() const override
+        {
+            std::lock_guard lock(remote_fd_mutex);
+            if (buffer)
+                buffer = nullptr;
+        }
+
     private:
         RemoteFileInfo file;
         RemoteFileCachePtr cache;

src/Storages/IndexFile/Table.cpp

@@ -275,6 +275,11 @@ Iterator * Table::NewIterator(const ReadOptions & options) const
         rep_->options.comparator, rep_->index_block->NewIterator(rep_->options.comparator), rep_->filter, &Table::BlockReader, const_cast<Table *>(this), options);
 }
 
+void Table::releaseRemoteFD() const
+{
+    rep_->file->releaseRemoteFD();
+}
+
 Status Table::Get(const ReadOptions & options, const Slice & k, std::string * value)
 {
     Status notfound = Status::NotFound(Slice());

src/Storages/IndexFile/Table.h

@@ -68,6 +68,10 @@ public:
     // be close to the file length.
     // uint64_t ApproximateOffsetOf(const Slice & key) const;
 
+    // After the iter call ends, we need to explicitly release the buffer, otherwise there may be too many open files due to remote handles.
+    // Safe for concurrent use by multiple threads.
+    void releaseRemoteFD() const;
+
 private:
     struct Rep;
 

src/Storages/IndexFile/TwoLevelIterator.cpp

@@ -93,7 +93,11 @@ namespace
     {
     }
 
-    TwoLevelIterator::~TwoLevelIterator() = default;
+    TwoLevelIterator::~TwoLevelIterator()
+    {
+        Table * table = reinterpret_cast<Table *>(arg_);
+        table->releaseRemoteFD();
+    }
 
     void TwoLevelIterator::Seek(const Slice & target)
     {
@@ -157,7 +161,7 @@ namespace
             // move to the next block
             index_iter_.Next();
 
-            // Filter data using min max index. 
+            // Filter data using min max index.
             // If target is greater than index.key(max), then find target in next data block
             if (index_iter_.Valid() && Compare(index_iter_.key(), target) < 0)
                 continue; // skip the next block