diff --git a/be/src/common/config.cpp b/be/src/common/config.cpp
index 02ebefbc0f32d0..a4efb631664feb 100644
--- a/be/src/common/config.cpp
+++ b/be/src/common/config.cpp
@@ -950,8 +950,6 @@ DEFINE_Int32(doris_remote_scanner_thread_pool_thread_num, "48");
 // number of s3 scanner thread pool queue size
 DEFINE_Int32(doris_remote_scanner_thread_pool_queue_size, "102400");
 DEFINE_mInt64(block_cache_wait_timeout_ms, "1000");
-DEFINE_mInt64(cache_lock_long_tail_threshold, "1000");
-DEFINE_Int64(file_cache_recycle_keys_size, "1000000");
 
 // limit the queue of pending batches which will be sent by a single nodechannel
 DEFINE_mInt64(nodechannel_pending_queue_max_bytes, "67108864");
@@ -1043,6 +1041,9 @@ DEFINE_Bool(enable_ttl_cache_evict_using_lru, "true");
 DEFINE_mBool(enbale_dump_error_file, "true");
 // limit the max size of error log on disk
 DEFINE_mInt64(file_cache_error_log_limit_bytes, "209715200"); // 200MB
+DEFINE_mInt64(cache_lock_long_tail_threshold, "1000");
+DEFINE_Int64(file_cache_recycle_keys_size, "1000000");
+DEFINE_mBool(enable_file_cache_keep_base_compaction_output, "false");
 
 DEFINE_mInt32(index_cache_entry_stay_time_after_lookup_s, "1800");
 DEFINE_mInt32(inverted_index_cache_stale_sweep_time_sec, "600");
diff --git a/be/src/common/config.h b/be/src/common/config.h
index 56d5e8e648a86c..b840f87f725440 100644
--- a/be/src/common/config.h
+++ b/be/src/common/config.h
@@ -1004,8 +1004,6 @@ DECLARE_mInt64(nodechannel_pending_queue_max_bytes);
 // The batch size for sending data by brpc streaming client
 DECLARE_mInt64(brpc_streaming_client_batch_bytes);
 DECLARE_mInt64(block_cache_wait_timeout_ms);
-DECLARE_mInt64(cache_lock_long_tail_threshold);
-DECLARE_Int64(file_cache_recycle_keys_size);
 
 DECLARE_Bool(enable_brpc_builtin_services);
 
@@ -1084,6 +1082,15 @@ DECLARE_Bool(enable_ttl_cache_evict_using_lru);
 DECLARE_mBool(enbale_dump_error_file);
 // limit the max size of error log on disk
 DECLARE_mInt64(file_cache_error_log_limit_bytes);
+DECLARE_mInt64(cache_lock_long_tail_threshold);
+DECLARE_Int64(file_cache_recycle_keys_size);
+// Base compaction may retrieve and produce some less frequently accessed data,
+// potentially affecting the file cache hit rate.
+// This configuration determines whether to retain the output within the file cache.
+// Make your choice based on the following considerations:
+// If your file cache is ample enough to accommodate all the data in your database,
+// enable this option; otherwise, it is recommended to leave it disabled.
+DECLARE_mBool(enable_file_cache_keep_base_compaction_output);
 
 // inverted index searcher cache
 // cache entry stay time after lookup
diff --git a/be/src/olap/compaction.cpp b/be/src/olap/compaction.cpp
index e608033701ed04..d5337b419aae5a 100644
--- a/be/src/olap/compaction.cpp
+++ b/be/src/olap/compaction.cpp
@@ -1181,8 +1181,12 @@ Status CloudCompactionMixin::construct_output_rowset_writer(RowsetWriterContext&
         ctx.compaction_level = _engine.cumu_compaction_policy(compaction_policy)
                                        ->new_compaction_level(_input_rowsets);
     }
-
-    ctx.write_file_cache = compaction_type() == ReaderType::READER_CUMULATIVE_COMPACTION;
+    // We presume that the data involved in cumulative compaction is sufficiently 'hot'
+    // and should always be retained in the cache.
+    // TODO(gavin): Ensure that the retention of hot data is implemented with precision.
+    ctx.write_file_cache = (compaction_type() == ReaderType::READER_CUMULATIVE_COMPACTION) ||
+                           (config::enable_file_cache_keep_base_compaction_output &&
+                            compaction_type() == ReaderType::READER_BASE_COMPACTION);
     ctx.file_cache_ttl_sec = _tablet->ttl_seconds();
     _output_rs_writer = DORIS_TRY(_tablet->create_rowset_writer(ctx, _is_vertical));
     RETURN_IF_ERROR(_engine.meta_mgr().prepare_rowset(*_output_rs_writer->rowset_meta().get()));