Merge pull request #1263 from enkore/issue/1262

Repository: cleanup on ENOSPC

borg/repository.py

@@ -84,6 +84,12 @@ def __enter__(self):
 
     def __exit__(self, exc_type, exc_val, exc_tb):
         if exc_type is not None:
+            no_space_left_on_device = exc_type is OSError and exc_val.errno == errno.ENOSPC
+            # The ENOSPC could have originated somewhere else besides the Repository. The cleanup is always safe, unless
+            # EIO or FS corruption ensues, which is why we specifically check for ENOSPC.
+            if self._active_txn and no_space_left_on_device:
+                logger.warning('No space left on device, cleaning up partial transaction to free space.')
+                self.io.cleanup(self.io.get_segments_transaction_id())
             self.rollback()
         self.close()
 
@@ -543,7 +549,7 @@ def get_latest_segment(self):
         return None
 
     def get_segments_transaction_id(self):
-        """Verify that the transaction id is consistent with the index transaction id
+        """Return last committed segment
         """
         for segment, filename in self.segment_iterator(reverse=True):
             if self.is_committed_segment(segment):

docs/internals.rst

@@ -280,6 +280,7 @@ emptied to 25%, its size is shrinked. So operations on it have a variable
 complexity between constant and linear with low factor, and memory overhead
 varies between 33% and 300%.
 
+.. _cache-memory-usage:
 
 Indexes / Caches memory usage
 -----------------------------

docs/quickstart.rst

@@ -11,11 +11,15 @@ The next section continues by showing how backups can be automated.
 Important note about free space
 -------------------------------
 
-Before you start creating backups, please make sure that there is **always**
+Before you start creating backups, please make sure that there is *always*
 a good amount of free space on the filesystem that has your backup repository
-(and also on ~/.cache). It is hard to tell how much, maybe 1-5%.
+(and also on ~/.cache). A few GB should suffice for most hard-drive sized
+repositories. See also :ref:`cache-memory-usage`.
 
-If you run out of disk space, it can be hard or impossible to free space,
+If |project_name| runs out of disk space, it tries to free as much space as it
+can while aborting the current operation safely, which allows to free more space
+by deleting/pruning archives. This mechanism is not bullet-proof though.
+If you *really* run out of disk space, it can be hard or impossible to free space,
 because |project_name| needs free space to operate - even to delete backup
 archives. There is a ``--save-space`` option for some commands, but even with
 that |project_name| will need free space to operate.