premature abort of upload if some shares were already present and some servers fail

[zooko]

There is logic in src/allmydata/immutable/encode.py@20090107034822-92b7f-4b8bbe1a2c1bf374154bc1cd32ff8040e998ca00#L474 to abort an upload if there aren't enough remaining shareholders. This fails to take into account that some shares might already have been present on servers before this upload began. So for example if M=10, and there were 5 shares present on servers when you started, and shares_of_happiness is 7, then you'll be talking to 5 landlords, and if one of them disconnects you should not then give up on the whole upload, per this line:

        if len(self.landlords) < self.shares_of_happiness:
            msg = "lost too many shareholders during upload: %s" % why

To test this, the test needs to create a file with 10 shares, delete 5 of them, start an upload, and then arrange it that during the upload 3 of the servers fail -- the upload should proceed normally (since share_of_happiness is 7). Also test what happens when 4 of them fail -- the upload should stop immediately.

[zooko]

Eh, this doesn't have to be fixed for tahoe-1.3.0.

[zooko]

I'm going to add cc: tahoe-dev to this ticket, and then I'm going to post the original contents of this ticket to tahoe-dev along with a link to this ticket.

[zooko]

Kevan's patch from #778 ("shares of happiness" is the wrong measure; "servers of happiness" is better) might fix this.

[davidsarah]

Related to #610 and #873.

[zooko]

#778 did indeed fix this ticket.