Change PosixWritableFile Truncate to reseek to new end of file (#14088)

Summary:
Change PosixWritableFile's Truncate to the new end offset. This ensures that future appends are written with no holes or overwrites. RocksDB doesn't guarantee this in the FileSystem contract, and its left up to the specific implementation.

Pull Request resolved: https://github.com/facebook/rocksdb/pull/14088

Reviewed By: cbi42

Differential Revision: D85786398

Pulled By: anand1976

fbshipit-source-id: 3520d9d6336362f5128a17bbf396297d821a5da3

env/io_posix.cc

@@ -1406,6 +1406,7 @@ IOStatus PosixWritableFile::Truncate(uint64_t size, const IOOptions& /*opts*/,
                 filename_, errno);
   } else {
     filesize_ = size;
+    lseek(fd_, filesize_, SEEK_SET);
   }
   return s;
 }

env/io_posix_test.cc

@@ -4,6 +4,7 @@
 // (found in the LICENSE.Apache file in the root directory).
 
 #include "test_util/testharness.h"
+#include "util/random.h"
 
 #ifdef ROCKSDB_LIB_IO_POSIX
 #include "env/io_posix.h"
@@ -131,6 +132,48 @@ TEST_F(LogicalBlockSizeCacheTest, Ref) {
 }
 #endif
 
+class PosixWritableFileTest : public testing::Test {};
+
+TEST_F(PosixWritableFileTest, SeekAfterTruncate) {
+  std::shared_ptr<FileSystem> fs = FileSystem::Default();
+  std::string path =
+      test::PerThreadDBPath("PosixWritableFileTest_SeekAfterTruncate");
+  Random rnd(300);
+  std::unique_ptr<FSWritableFile> wfile;
+
+  ASSERT_OK(fs->NewWritableFile(path, FileOptions(), &wfile, nullptr));
+  ASSERT_OK(wfile->Append(rnd.RandomString(16384), IOOptions(), nullptr));
+  ASSERT_OK(wfile->Truncate(4096, IOOptions(), nullptr));
+  ASSERT_OK(wfile->Append(rnd.RandomString(4096), IOOptions(), nullptr));
+  ASSERT_OK(wfile->Close(IOOptions(), nullptr));
+  wfile.reset();
+
+  uint64_t size = 0;
+  ASSERT_OK(fs->GetFileSize(path, IOOptions(), &size, nullptr));
+  ASSERT_EQ(size, 8192);
+  ASSERT_OK(fs->DeleteFile(path, IOOptions(), nullptr));
+}
+
+TEST_F(PosixWritableFileTest, SeekAfterExtend) {
+  std::shared_ptr<FileSystem> fs = FileSystem::Default();
+  std::string path =
+      test::PerThreadDBPath("PosixWritableFileTest_SeekAfterTruncate");
+  Random rnd(300);
+  std::unique_ptr<FSWritableFile> wfile;
+
+  ASSERT_OK(fs->NewWritableFile(path, FileOptions(), &wfile, nullptr));
+  ASSERT_OK(wfile->Append(rnd.RandomString(4096), IOOptions(), nullptr));
+  ASSERT_OK(wfile->Truncate(8192, IOOptions(), nullptr));
+  ASSERT_OK(wfile->Append(rnd.RandomString(8192), IOOptions(), nullptr));
+  ASSERT_OK(wfile->Close(IOOptions(), nullptr));
+  wfile.reset();
+
+  uint64_t size = 0;
+  ASSERT_OK(fs->GetFileSize(path, IOOptions(), &size, nullptr));
+  ASSERT_EQ(size, 16384);
+  ASSERT_OK(fs->DeleteFile(path, IOOptions(), nullptr));
+}
+
 }  // namespace ROCKSDB_NAMESPACE
 #endif
 

include/rocksdb/file_system.h

@@ -1166,8 +1166,10 @@ class FSWritableFile {
 
   // Truncate is necessary to trim the file to the correct size
   // before closing. It is not always possible to keep track of the file
-  // size due to whole pages writes. The behavior is undefined if called
-  // with other writes to follow.
+  // size due to whole pages writes. If called with other writes to follow,
+  // the behavior is file system specific. Posix will reseek to the new EOF.
+  // Other file systems may behave differently. Its the caller's
+  // responsibility to check the file system contract.
   virtual IOStatus Truncate(uint64_t /*size*/, const IOOptions& /*options*/,
                             IODebugContext* /*dbg*/) {
     return IOStatus::OK();

unreleased_history/behavior_changes/posix_writable_file_truncate.md

@@ -0,0 +1 @@
+PosixWritableFile now repositions the seek pointer to the new end of file after a call to Truncate.