Fix issue with pylibmc binary mode never being set.

[rogerhu]

No description provided.

[gusdan]

Good catch. But that is interesting Django doesn't support it in 1.7 as well.
https://github.com/django/django/blob/stable/1.7.x/django/core/cache/backends/memcached.py#L185
did you try that option? Any speed changing in case of using with ElasticCache?

[rogerhu]

I haven't benchmarked myself but there are several perf measurements that people have done in the past..

http://www.creapptives.com/post/46449295377/binary-protocol-benchmarks-for-memcached

[gusdan]

I was wondering why django comunity didn't set that flag. And i performed my own simple bacnhmarks and binary protocol works slowly for long strings (not numbers). I did simple test:

1. clone you repo
2. add it to virtual env
3. run django shell
4. perform 1000 reads for same key with value = '1'*1000

old version (no binary) takes 1.359 - 1.405s
new version (binary) takes 1.419 - 1.438s

i think it's common usecase for memcache: storing long string (html pages, pickle, jsons). and binary protocol does some otherhead here without pros of minimize transfering.

So i'm going to reject your pull request unless you have some objections.

[robbyt]

Just a note about this, I've seen issues with pickled objects silently becoming corrupt when forcing memcached binary mode. It was a deep bug, and suggest you stay away from memcached binary mode w/ django.

[gusdan]

Thank you for update.

On Tue, Oct 6, 2015 at 3:18 AM, Robert Terhaar notifications@github.com
wrote:

Just a note about this, I've seen issues with pickled objects silently
becoming corrupt when forcing memcached binary mode. It was a deep bug, and
suggest you stay away from memcached binary mode w/ django.

—
Reply to this email directly or view it on GitHub
#6 (comment)
.