Put some comments in places we get asked about a lot

Code Issues Proposed changes

Change-Id: If932b82f4e225cfdda21f47e4215c6f4c59dae4a

This commit is contained in:

gholt

2013年05月23日 19:49:34 +00:00

parent b4b35c2561

commit 1c8d188872

2 changed files with 38 additions and 0 deletions

Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL

Download Patch File Download Diff File Expand all files Collapse all files

24
swift/common/memcached.py

View File

@@ -14,6 +14,30 @@

# limitations under the License.

"""

Why our own memcache client?

By Michael Barton

python-memcached doesn't use consistent hashing, so adding or

removing a memcache server from the pool invalidates a huge

percentage of cached items.

If you keep a pool of python-memcached client objects, each client

object has its own connection to every memcached server, only one of

which is ever in use. So you wind up with n * m open sockets and

almost all of them idle. This client effectively has a pool for each

server, so the number of backend connections is hopefully greatly

reduced.

python-memcache uses pickle to store things, and there was already a

huge stink about Swift using pickles in memcache

(http://osvdb.org/show/osvdb/86581). That seemed sort of unfair,

since nova and keystone and everyone else use pickles for memcache

too, but it's hidden behind a "standard" library. But changing would

be a security regression at this point.

Also, pylibmc wouldn't work for us because it needs to use python

sockets in order to play nice with eventlet.

Lucid comes with memcached: v1.4.2. Protocol documentation for that

version is at:

14
swift/common/swob.py

View File

@@ -19,6 +19,20 @@ Implementation of WSGI Request and Response objects.

This library has a very similar API to Webob. It wraps WSGI request

environments and response values into objects that are more friendly to

interact with.

Why Swob and not just use WebOb?

By Michael Barton

We used webob for years. The main problem was that the interface

wasn't stable. For a while, each of our several test suites required

a slightly different version of webob to run, and none of them worked

with the then-current version. It was a huge headache, so we just

scrapped it.

This is kind of a ton of code, but it's also been a huge relief to

not have to scramble to add a bunch of code branches all over the

place to keep Swift working every time webob decides some interface

needs to change.

"""

from collections import defaultdict