FTP upload silently fails when nodes < happiness

[slush]

When I upload file using FTP (SFTP not affected), the upload successfully passes these phases:

*) Upload locally to tmp dir

*) Upload to helper (if used)

*) Converted to CHK_encoding (when using helper)

And then the FTP transfer silently fails, pretending that upload successfully finished. When I refresh the directory listing, the file isn't there.

Expected behaviour (from my view of common user): FTP upload should finish after all phases of tahoe upload.

Used software: Node - Windows7 with patched Twisted. Helper - Debian without patch.

I tried scenario many times with different files and happiness and both with or without helper.

Note: Using SFTP in same scenario (nodes < happiness) leads to RemoteException? in SFTP client which indicates that something went wrong.

How to reproduce:

a) Wait until count of nodes is less than current happiness in config.

b) or set happiness in config to something large.

Reproducibility: always

[zooko]

I think we should treat this as a v1.7 issue as far as documenting that it can happen and instructing people on how to avoid it or to diagnose that this is the cause of their problem (but not treat it as a v1.7 issue as far as changing any code to improve it).

So, ideally we would update some docs, probably in docs/frontends/FTP-and-SFTP.txt, explaining that if you have insufficient servers for servers-of-happiness then uploads will silently fail. (This will be along with docs changes per #1024 and #1082 which will hopefully make users more aware of the servers-of-happiness issues in general.) The FTP-and-SFTP docs should point users to other docs such as running.html / using.html which explain how to set up multiple servers, how to see how many servers you have set up, how to configure servers-of-happiness, and how to see what your current servers-of-happiness configuration is.

Then, having written this doc, we'll move this ticket from the "1.7" Milestone to the "soon" Milestone and ship Tahoe-LAFS v1.7 final!

Then we'll see if we can somehow change our code to signal failure reliably to the user of an FTP or SFTP frontend so that they can detect and diagnose failures.

[davidsarah]

Documented in 29a9059c94eef955.

[warner]

not making it into 1.9