[wip][performance] Use pread when possible to avoid read amplificatio…

…n caused by buffering

core/ratarmountcore/SQLiteIndexedTar.py

@@ -1163,7 +1163,17 @@ def read(self, fileInfo: FileInfo, size: int, offset: int) -> bytes:
                 return file.read(size)
 
         # For non-sparse files, we can simply seek to the offset and read from it.
-        self.tarFileObject.seek(tarFileInfo.offset + offset, os.SEEK_SET)
+        fileno = None
+        if hasattr(os, 'pread'):
+            if hasattr(self.tarFileObject, 'fileno'):
+                try:
+                    fileno = self.tarFileObject.fileno()
+                except Exception:
+                    pass
+
+        if isinstance(fileno, int) and fileno >= 0:
+            return os.pread(fileno, size, tarFileInfo.offset + offset)
+        self.tarFileObject.seek(tarFileInfo.offset + offset, io.SEEK_SET)
         return self.tarFileObject.read(size)
 
     @overrides(MountSource)

core/ratarmountcore/StenciledFile.py

@@ -151,11 +151,28 @@ def read(self, size: int = -1) -> bytes:
             offsetInsideStencil = self.offset - self.cumsizes[i]
             assert offsetInsideStencil >= 0
             assert offsetInsideStencil < self.sizes[i]
-            self.fileObjects[i].seek(self.offsets[i] + offsetInsideStencil, io.SEEK_SET)
+            offset = self.offsets[i] + offsetInsideStencil
 
             # Read as much as requested or as much as the current contiguous region / stencil still contains
             readableSize = min(size, self.sizes[i] - (self.offset - self.cumsizes[i]))
-            tmp = self.fileObjects[i].read(readableSize)
+
+            fileObject = self.fileObjects[i]
+            fileno = None
+            if hasattr(os, 'pread'):
+                if hasattr(fileObject, 'fileno'):
+                    try:
+                        fileno = fileObject.fileno()
+                    except Exception:
+                        pass
+
+#            print("[StenciledFile.read] fileno:", fileno, "readableSize:", readableSize)
+
+            if isinstance(fileno, int) and fileno >= 0:
+                tmp = os.pread(fileno, readableSize, offset)
+            else:
+                fileObject.seek(offset, io.SEEK_SET)
+                tmp = fileObject.read(readableSize)
+
             self.offset += len(tmp)
             result += tmp
 
@@ -294,11 +311,26 @@ def read(self, size: int = -1) -> bytes:
             offsetInsideStencil = self.offset - self.cumsizes[i]
             assert offsetInsideStencil >= 0
             assert offsetInsideStencil < self.sizes[i]
-            fileObject.seek(offsetInsideStencil, io.SEEK_SET)
 
             # Read as much as requested or as much as the current contiguous region / stencil still contains
             readableSize = min(size, self.sizes[i] - (self.offset - self.cumsizes[i]))
-            tmp = fileObject.read(readableSize)
+
+            fileno = None
+            print("TRY PREAD:")
+            if hasattr(os, 'pread'):
+                if hasattr(fileObject, 'fileno'):
+                    try:
+                        fileno = fileObject.fileno()
+                    except Exception:
+                        pass
+
+            print("FILENO:", fileno)
+            if isinstance(fileno, int) and fileno >= 0:
+                tmp = os.pread(fileObject.fileno(), readableSize, offsetInsideStencil)
+            else:
+                fileObject.seek(offsetInsideStencil, io.SEEK_SET)
+                tmp = fileObject.read(readableSize)
+
             self.offset += len(tmp)
             result += tmp