introducer remembers multiple announcements of the same server

[warner]

The new introducer-status web page reveals that the Introducer stores multiple announcements of the same server, when that server is upgraded and bounced while the Introducer remains running. This occurs because the announcements are indexed by the full announcement tuple, which includes the version of the service: so the first run of the service (at version1) is a distinct tuple from the second run of the service (at version2).

This doesn't cause a problem at the client end, because the client indexes everything by nodeid. But it would be nice to fix, to keep these items from piling up in the introducer, and to make the introducer status page more useful.

[warner]

This can also mess up the display of client nicknames, since the nickname is not necessarily grabbed from the latest stub_client announcement.

Also, now that clients are announcing stub_clients, the introducer will accumulate even more announcements.

[zooko]

This is step 2 of #68 (implement distributed introduction, remove Introducer as a single point of failure).

[zooko]

I mean #68, comment 10

[warner]

We have some munin graphs of count-of-storage-servers that are seeing artifacts because of this, so I'd like to see it finished sooner rather than later.

[warner]

fixed by 55dfb697a448dbc7. It may be a little while before we update the prodnet introducer to reflect these changes, however.