From 6924637e64735b1c2cb8f8803a8d8cf494080fc7 Mon Sep 17 00:00:00 2001
From: stdpain <34912776+stdpain@users.noreply.github.com>
Date: Wed, 26 May 2021 10:02:39 +0800
Subject: [PATCH] [BUG] fix compression bug while compaction (#5893)

Because the maximum length of LZ4 compression is 2^32, it can cause some memory problems
---
 be/src/olap/fs/file_block_manager.cpp     |  2 +-
 be/src/olap/rowset/segment_v2/page_io.cpp | 30 ++++++++++++-----------
 be/src/util/block_compression.cpp         | 12 ++++++++-
 3 files changed, 28 insertions(+), 16 deletions(-)

diff --git a/be/src/olap/fs/file_block_manager.cpp b/be/src/olap/fs/file_block_manager.cpp
index 616734244e2a93..2f49136a014dc0 100644
--- a/be/src/olap/fs/file_block_manager.cpp
+++ b/be/src/olap/fs/file_block_manager.cpp
@@ -169,7 +169,7 @@ Status FileWritableBlock::appendv(const Slice* data, size_t data_cnt) {
 
     // Calculate the amount of data written
     size_t bytes_written = accumulate(data, data + data_cnt, static_cast<size_t>(0),
-                                      [&](int sum, const Slice& curr) { return sum + curr.size; });
+                                      [](size_t sum, const Slice& curr) { return sum + curr.size; });
     _bytes_appended += bytes_written;
     return Status::OK();
 }
diff --git a/be/src/olap/rowset/segment_v2/page_io.cpp b/be/src/olap/rowset/segment_v2/page_io.cpp
index 16037daa944120..8f455b1b186100 100644
--- a/be/src/olap/rowset/segment_v2/page_io.cpp
+++ b/be/src/olap/rowset/segment_v2/page_io.cpp
@@ -41,20 +41,22 @@ Status PageIO::compress_page_body(const BlockCompressionCodec* codec, double min
     size_t uncompressed_size = Slice::compute_total_size(body);
     if (codec != nullptr && uncompressed_size > 0) {
         size_t max_compressed_size = codec->max_compressed_len(uncompressed_size);
-        faststring buf;
-        buf.resize(max_compressed_size);
-        Slice compressed_slice(buf);
-        RETURN_IF_ERROR(codec->compress(body, &compressed_slice));
-        buf.resize(compressed_slice.get_size());
-
-        double space_saving = 1.0 - static_cast<double>(buf.size()) / uncompressed_size;
-        // return compressed body only when it saves more than min_space_saving
-        if (space_saving > 0 && space_saving >= min_space_saving) {
-            // shrink the buf to fit the len size to avoid taking
-            // up the memory of the size MAX_COMPRESSED_SIZE
-            buf.shrink_to_fit();
-            *compressed_body = buf.build();
-            return Status::OK();
+        if (max_compressed_size) {
+            faststring buf;
+            buf.resize(max_compressed_size);
+            Slice compressed_slice(buf);
+            RETURN_IF_ERROR(codec->compress(body, &compressed_slice));
+            buf.resize(compressed_slice.get_size());
+    
+            double space_saving = 1.0 - static_cast<double>(buf.size()) / uncompressed_size;
+            // return compressed body only when it saves more than min_space_saving
+            if (space_saving > 0 && space_saving >= min_space_saving) {
+                // shrink the buf to fit the len size to avoid taking
+                // up the memory of the size MAX_COMPRESSED_SIZE
+                buf.shrink_to_fit();
+                *compressed_body = buf.build();
+                return Status::OK();
+            }
         }
     }
     // otherwise, do not compress
diff --git a/be/src/util/block_compression.cpp b/be/src/util/block_compression.cpp
index 109848d630b47b..290ffec66c3fc2 100644
--- a/be/src/util/block_compression.cpp
+++ b/be/src/util/block_compression.cpp
@@ -26,6 +26,8 @@
 #include "gutil/strings/substitute.h"
 #include "util/faststring.h"
 
+#include <limits>
+
 namespace doris {
 
 using strings::Substitute;
@@ -71,7 +73,12 @@ class Lz4BlockCompression : public BlockCompressionCodec {
         return Status::OK();
     }
 
-    size_t max_compressed_len(size_t len) const override { return LZ4_compressBound(len); }
+    size_t max_compressed_len(size_t len) const override { 
+        if (len > std::numeric_limits<int32_t>::max()) {
+            return 0;
+        }
+        return LZ4_compressBound(len); 
+    }
 };
 
 // Used for LZ4 frame format, decompress speed is two times faster than LZ4.
@@ -120,6 +127,9 @@ class Lz4fBlockCompression : public BlockCompressionCodec {
     }
 
     size_t max_compressed_len(size_t len) const override {
+        if (len > std::numeric_limits<int32_t>::max()) {
+            return 0;
+        }
         return std::max(LZ4F_compressBound(len, &_s_preferences),
                         LZ4F_compressFrameBound(len, &_s_preferences));
     }