openstack/swift
Code Issues Proposed changes

Drop cache after fsync

Browse Source 
Drop the cache (fadvise) after calling fsync to avoid redundant work
in the kernel visiting pages that are dirty. By dropping cache after
fsync(), the pages will be clean so can be easily removed from the
cache.

Change-Id: I3013d2ac1e27bbe43e58a81638d04805f5e2ae6e
Signed-off-by: Peter Portante <peter.portante@redhat.com>
This commit is contained in:
Peter Portante 2013-04-18 14:31:58 -04:00
parent a186c53b24
commit dec517e349
1 changed files with 12 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
swift/obj/server.py
View File

@ -315,11 +315,21 @@ class DiskFile(object):
assert self.tmppath is not None
metadata['name'] = self.name
timestamp = normalize_timestamp(metadata['X-Timestamp'])
# Write the metadata before calling fsync() so that both data and
# metadata are flushed to disk.
write_metadata(fd, metadata)
if 'Content-Length' in metadata:
self.drop_cache(fd, 0, int(metadata['Content-Length']))
# We call fsync() before calling drop_cache() to lower the amount of
# redundant work the drop cache code will perform on the pages (now
# that after fsync the pages will be all clean).
tpool.execute(fsync, fd)
if 'Content-Length' in metadata:
# From the Department of the Redundancy Department, make sure we
# call drop_cache() after fsync() to avoid redundant work (pages
# all clean).
self.drop_cache(fd, 0, int(metadata['Content-Length']))
invalidate_hash(os.path.dirname(self.datadir))
# After the rename completes, this object will be available for other
# requests to reference.
renamer(self.tmppath,
os.path.join(self.datadir, timestamp + extension))
self.metadata = metadata