Ticket #573: A server selection framework.

[linshu]

The 3rd party server selection function can be passed in as
[client][server_selection_hook] in the tahoe.cfg file.
The format is python module syntax, like "module_a.submodule_b.function_c".
A default hook function "sort_servers" is implemented in allmydata.storage_client
to catch no 3rd party hook functoin configured case.

---

This change is 

[daira]

Please fix the test failures on this branch, then we'll review it. (If you need any help, don't hesitate to ask on Freenode #tahoe-lafs or the tahoe-dev list.)

[linshu]

Fixed the build. Please review it. Thanks.

[linshu]

Daira,

I fixed the build. Please review it. :-)

Thanks a lot!
-Shu

On Fri, Feb 6, 2015 at 5:02 PM, Daira Hopwood notifications@github.com
wrote:

Please fix the test failures on this branch, then we'll review it. (If you
need any help, don't hesitate to ask on Freenode #tahoe-lafs or the
tahoe-dev list.)

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

[linshu]

Hi Daira,

I have fixed all you suggested. Please review it.

Thanks.

[zooko]

Thank you very much for your contribution, Shu Lin! It is too bad that Daira is so busy right now with other work. I hope she can review your patch, too. I wonder if anyone else out there could review this patch?

[daira]

v1.10.1 is feature-frozen, but I will definitely review it for v1.11 after the v1.10.1 release.

[meejah]

This seems like a complicated API to export to users -- what they're supposed to do with key seems unclear (e.g. what if it's not specified? can it be ignored if it is specified?). I think it would be better if the "sort_servers" function takes only a single argument (the servers list). Or, maybe the second argument should be peer_selection_index?

Then for this default implementation, you could move the permuted inner-method from get_servers_for_psi into this function and thus still call sorted in the same fashion (with key=permuted).

[daira]

We discussed this in a Tahoe Nuts and Bolts meeting, and decided that this should be passed the peer_selection_index. If we don't pass that, then the hook does not have enough information to make changes to how the _permuted algorithm below works.

[exarkun]

Thanks to everyone for their efforts so far. This looks like a good start. The review feedback hasn't been addressed yet and the feature overall needs test coverage and probably additional documentation.

Please feel free to re-open this PR if there's still interest in contributing here.