Redis as Storage Service for SP

[Cantor, Scott]

> Scott, sorry to hear that there are currently no plans to develop Redis
> support for Shibboleth, where apparently more people are looking for this
> solution.

I don't see any issue in Jira for it.

> In my opinion, Memcache is kind of dead, while Redis these days is a more
> suitable solution which is also developed with clustering and HA in mind.

I was opposed to supporting memcache to begin with because I think session persistence should be, well, persistent, but forgive me when I speculate that soon Redis will be "kind of dead" and the next fad comes along.

You're asking me to now take on the ongoing support of *another* technology I don't know (one which for all I know has no standard libraries or proper Windows support, which is/was the situation with memcache). And I still have to support memcache for the foreseeable future. I don't think you appreciate the impact of those kinds of decisions.

> We're using OwnCloud in combination with Apache, in which we're using
> Shibboleth for the single sign-on authentication. OwnCloud supports Redis as
> memory cache solution, but for the authentication sessions I'm afraid I'm still
> depending on the options which Shibboleth provides.

As Peter outlined, I'm suggesting you not do that so as to make the problem much simpler. The SP's session feature is necessary because of its generality, but in practice, it's not targeted at large deployments and clustered applications, where there are already going to be better options.

> I understand your point about; "the people asking for new code projects
> would be funding the project.". But how many time do you think it will take
> to implement Redis support?

Implementation is not the major issue, maintenance is. But to answer that, since I had nothing to do with the memcache code and would have to really bottom it out, it would likely take me on the order of 40 hours to deal with all the implications of such an addition.

> Unfortunately I'm not a developer, but I think the changes in the code will
> not be very large, because the way of saving keys in Memcached is very
> similar to Redis.

Changes in code I didn't write and have never used.

I would note that memcache doesn't support the two-part key layout used by the SP either, and only handles it very imperfectly, as I understand it. If Redis is the same, then it's also a bad fit.

If somebody shows up and wants to build and maintain a plugin, I have no objection to that and I would help them get started.

-- Scott