Thread Previous • Date Previous • Date Next • Thread Next |
On 06/04/2012 07:15 PM, Nguyen, Liem Manh wrote:
I don't think the SSL overhead is that great. If there is a large hue and cry from the community, I'll provide it as an option, but I am thinking that there is little benefit in backwards compatibility, and a large amount of code that then needs to be maintained.Looks good, Adam... I have a couple comments/questions: 1) We probably want to maintain backward-compatibility with the old token auth. So, PKI can be turned on or off, maybe via a middleware filter that sits in front of the old token auth (instead of a cache for example which does not make sense for PKI anymore). While PKI comes with benefits, not everyone may want to deal with certificates, etc... 2-way SSL is configurable for this very same reason.
This is different from 2 Way SSL, which is implemented in the Python Layer, and is likely a performance bottleneck. You don't want to do Crypto in Python.
I'm not sure that the cache should go. I can see benefits to using it: it obviates the need for spawning a process for additional signed tokens checks.
2) I don't know if you have looked at pyopenssl yet? Maybe a better alternative than spawning processes...
I'm pretty sure that any call into a native library is going to be more of an issue than spawning a process, unless that library has been specifically implemented with non-blocking code in mind, and explicitly avoids taking the GIL. However, Since openssl doesn't seem to support CMS signing (the PKCS7 module is pretty much read only information) I think the issue is moot.
3) Is PKI v3 or v2? I assume v3, but just want to double-check. /v2.0 should be /v3.0 if that's the case.
V2, as there is no change to the API, just the implementation.
Thanks, Liem ________________________________________ From: openstack-bounces+liem_m_nguyen=hp.com@xxxxxxxxxxxxxxxxxxx [openstack-bounces+liem_m_nguyen=hp.com@xxxxxxxxxxxxxxxxxxx] on behalf of Adam Young [ayoung@xxxxxxxxxx] Sent: Friday, June 01, 2012 2:56 PM To: openstack Subject: [Openstack] Signed Tokens The signed tokens work has been updated. I think this is the final architecture. https://github.com/admiyo/keystone/commits/signed-tokens-5 Not all of the unit tests run. Some of the Memcache tests are suspect, and I wonder if we even need memcache support for tokens in the middle ware. I think we don't. Also, the Diablo tokens are not supported. I think we can safely deprecate them for Folsom. _______________________________________________ Mailing list: https://launchpad.net/~openstack Post to : openstack@xxxxxxxxxxxxxxxxxxx Unsubscribe : https://launchpad.net/~openstack More help : https://help.launchpad.net/ListHelp
Thread Previous • Date Previous • Date Next • Thread Next |