preload chunks for hardlink slaves w/o preloaded master, fixes #4350

also split the hardlink extraction test into 2 tests.

(cherry picked from commit f33f318d81)

src/borg/archive.py

@@ -192,7 +192,7 @@ class DownloadPipeline:
         self.repository = repository
         self.key = key
 
-    def unpack_many(self, ids, filter=None, preload=False):
+    def unpack_many(self, ids, filter=None, partial_extract=False, preload=False, hardlink_masters=None):
         """
         Return iterator of items.
 
@@ -209,12 +209,40 @@ class DownloadPipeline:
             for item in items:
                 if 'chunks' in item:
                     item.chunks = [ChunkListEntry(*e) for e in item.chunks]
+
+            def preload(chunks):
+                self.repository.preload([c.id for c in chunks])
+
             if filter:
                 items = [item for item in items if filter(item)]
+
             if preload:
+                if filter and partial_extract:
+                    # if we do only a partial extraction, it gets a bit
+                    # complicated with computing the preload items: if a hardlink master item is not
+                    # selected (== not extracted), we will still need to preload its chunks if a
+                    # corresponding hardlink slave is selected (== is extracted).
+                    # due to a side effect of the filter() call, we now have hardlink_masters dict populated.
+                    masters_preloaded = set()
+                    for item in items:
+                        if 'chunks' in item:  # regular file, maybe a hardlink master
+                            preload(item.chunks)
+                            # if this is a hardlink master, remember that we already preloaded it:
+                            if 'source' not in item and hardlinkable(item.mode) and item.get('hardlink_master', True):
+                                masters_preloaded.add(item.path)
+                        elif 'source' in item and hardlinkable(item.mode):  # hardlink slave
+                            source = item.source
+                            if source not in masters_preloaded:
+                                # we only need to preload *once* (for the 1st selected slave)
+                                chunks, _ = hardlink_masters[source]
+                                preload(chunks)
+                                masters_preloaded.add(source)
+                else:
+                    # easy: we do not have a filter, thus all items are selected, thus we need to preload all chunks.
                     for item in items:
                         if 'chunks' in item:
-                        self.repository.preload([c.id for c in item.chunks])
+                            preload(item.chunks)
+
             for item in items:
                 yield item
 
@@ -433,8 +461,10 @@ Utilization of max. archive size: {csize_max:.0%}
             return False
         return filter(item) if filter else True
 
-    def iter_items(self, filter=None, preload=False):
-        for item in self.pipeline.unpack_many(self.metadata.items, preload=preload,
+    def iter_items(self, filter=None, partial_extract=False, preload=False, hardlink_masters=None):
+        assert not (filter and partial_extract and preload) or hardlink_masters is not None
+        for item in self.pipeline.unpack_many(self.metadata.items, partial_extract=partial_extract,
+                                              preload=preload, hardlink_masters=hardlink_masters,
                                               filter=lambda item: self.item_filter(item, filter)):
             yield item
 

src/borg/archiver.py

@@ -755,7 +755,8 @@ class Archiver:
         else:
             pi = None
 
-        for item in archive.iter_items(filter, preload=True):
+        for item in archive.iter_items(filter, partial_extract=partial_extract,
+                                       preload=True, hardlink_masters=hardlink_masters):
             orig_path = item.path
             if strip_components:
                 item.path = os.sep.join(orig_path.split(os.sep)[strip_components:])
@@ -997,7 +998,7 @@ class Archiver:
                 return None, stream
             return tarinfo, stream
 
-        for item in archive.iter_items(filter, preload=True):
+        for item in archive.iter_items(filter, preload=True, hardlink_masters=hardlink_masters):
             orig_path = item.path
             if strip_components:
                 item.path = os.sep.join(orig_path.split(os.sep)[strip_components:])

src/borg/testsuite/archiver.py

@@ -823,7 +823,18 @@ class ArchiverTestCase(ArchiverTestCaseBase):
             assert open('input/dir1/subdir/hardlink', 'rb').read() == b'123456'
 
     @requires_hardlinks
-    def test_extract_hardlinks(self):
+    def test_extract_hardlinks1(self):
+        self._extract_hardlinks_setup()
+        with changedir('output'):
+            self.cmd('extract', self.repository_location + '::test')
+            assert os.stat('input/source').st_nlink == 4
+            assert os.stat('input/abba').st_nlink == 4
+            assert os.stat('input/dir1/hardlink').st_nlink == 4
+            assert os.stat('input/dir1/subdir/hardlink').st_nlink == 4
+            assert open('input/dir1/subdir/hardlink', 'rb').read() == b'123456'
+
+    @requires_hardlinks
+    def test_extract_hardlinks2(self):
         self._extract_hardlinks_setup()
         with changedir('output'):
             self.cmd('extract', self.repository_location + '::test', '--strip-components', '2')
@@ -839,13 +850,6 @@ class ArchiverTestCase(ArchiverTestCaseBase):
             assert open('input/dir1/subdir/hardlink', 'rb').read() == b'123456'
             assert os.stat('input/dir1/aaaa').st_nlink == 2
             assert os.stat('input/dir1/source2').st_nlink == 2
-        with changedir('output'):
-            self.cmd('extract', self.repository_location + '::test')
-            assert os.stat('input/source').st_nlink == 4
-            assert os.stat('input/abba').st_nlink == 4
-            assert os.stat('input/dir1/hardlink').st_nlink == 4
-            assert os.stat('input/dir1/subdir/hardlink').st_nlink == 4
-            assert open('input/dir1/subdir/hardlink', 'rb').read() == b'123456'
 
     def test_extract_include_exclude(self):
         self.cmd('init', '--encryption=repokey', self.repository_location)