upload of file into dir doesn't appear on Recent Uploads and Downloads

[zooko]

When I upload a file into a directory, such as this one: ​http://pubgrid.tahoe-lafs.org/uri/URI:DIR2-RO:ixqhc4kdbjxc7o65xjnveoewym:5x6lwoxghrd5rxhwunzavft2qygfkt27oj3fbxlq4c6p45z5uneq/ , which the timestamps show got linked into this directory at 2010-06-12_03:45:30.369968, the upload of the immutable file doesn't appear on the Recent Uploads and Downloads page:

21:45:38 11-Jun-2010	retrieve	up4mutkj52kcm2l7c7nvg5j2ua	No	314B	100.0%	Finished
21:45:38 11-Jun-2010	mapupdate MODE_READ	up4mutkj52kcm2l7c7nvg5j2ua	No	-NA-	100.0%	Finished
21:45:31 11-Jun-2010	retrieve	up4mutkj52kcm2l7c7nvg5j2ua	No	314B	100.0%	Finished
21:45:31 11-Jun-2010	mapupdate MODE_READ	up4mutkj52kcm2l7c7nvg5j2ua	No	-NA-	100.0%	Finished
21:45:30 11-Jun-2010	publish	up4mutkj52kcm2l7c7nvg5j2ua	No	314B	100.0%	Finished
21:45:30 11-Jun-2010	retrieve	up4mutkj52kcm2l7c7nvg5j2ua	No	624B	100.0%	Finished
21:45:30 11-Jun-2010	mapupdate MODE_WRITE	up4mutkj52kcm2l7c7nvg5j2ua	No	-NA-	100.0%	Finished
21:45:30 11-Jun-2010	publish	up4mutkj52kcm2l7c7nvg5j2ua	No	624B	100.0%	Finished
21:45:30 11-Jun-2010	retrieve	up4mutkj52kcm2l7c7nvg5j2ua	No	310B	100.0%	Finished
21:45:30 11-Jun-2010	mapupdate MODE_WRITE	up4mutkj52kcm2l7c7nvg5j2ua	No	-NA-	100.0%	Finished
21:45:30 11-Jun-2010	retrieve	up4mutkj52kcm2l7c7nvg5j2ua	No	310B	100.0%	Finished
21:45:29 11-Jun-2010	mapupdate MODE_READ	up4mutkj52kcm2l7c7nvg5j2ua	No	-NA-	100.0%	Finished
21:45:24 11-Jun-2010	retrieve	up4mutkj52kcm2l7c7nvg5j2ua	No	310B	100.0%	Finished
21:45:24 11-Jun-2010	mapupdate MODE_READ	up4mutkj52kcm2l7c7nvg5j2ua	No	-NA-	100.0%	Finished
21:45:24 11-Jun-2010	publish	up4mutkj52kcm2l7c7nvg5j2ua	No	310B	100.0%	Finished
21:45:23 11-Jun-2010	retrieve	up4mutkj52kcm2l7c7nvg5j2ua	No	620B	100.0%	Finished
21:45:23 11-Jun-2010	mapupdate MODE_WRITE	up4mutkj52kcm2l7c7nvg5j2ua	No	-NA-	100.0%	Finished
21:45:14 11-Jun-2010	retrieve	up4mutkj52kcm2l7c7nvg5j2ua	No	620B	100.0%	Finished
21:45:13 11-Jun-2010	mapupdate MODE_READ	up4mutkj52kcm2l7c7nvg5j2ua	No	-NA-	100.0%	Finished
21:32:56 11-Jun-2010	retrieve	up4mutkj52kcm2l7c7nvg5j2ua	No	620B	100.0%	Finished

Argh! These timestamps are in different timezones and neither of them have explicit timezone indicators. Grumble. Also they are of different and non-standard formatting. I just created #1077 (consistent timestamp format and timezone).

Anyway, making some educated guesses about what these timestamps mean, the upload that I completed at around 2010-06-12 03:45:30Z should have appeared in the list as "11-Jun-2010 21:45:30".

It seems like immutable file uploads into a directory never appear on Recent Uploads/Downloads, but immutable file downloads and unlinked immutable file uploads do.

[zooko]

This shouldn't be hard to fix and it would be a big help with evaluating and improving Tahoe-LAFS performance if we could tell how fast uploads were.

[davidsarah]

Out of time for 1.8.

[warner]

I did some digging. The problem is that we aren't being consistent in passing the "History" object to all cases of file upload. (to get into the "Recent Uploads and Downloads" RUaD page, the Uploader must be given access to the History object, so it can call History.add_upload).

Client.upload(uploadable) does the right thing. The webapi has several different code paths that result in an immutable upload:

• PUT /uri : hits PUTUnlinkedCHK (unlinked.py#L11) which calls client.upload, ok
• PUT /uri/DIRCAP/childname: hits filenode.py#L37 which calls self.parentnode.add_file(), which does *not* handle history
• POST /uri/DIRCAP/newchildname?t=upload and POST /uri/DIRCAP/oldchildname?t=upload&replace=true : both hit replace_me_with_a_formpost (filenode.py#L86), uses self.parentnode.add_file(), *no* history

The root problem is that DirectoryNode.add_file(), while convenient, doesn't have access to the History object. The best fix is probably to delete this method, and rewrite web/filenode.py to make two separate calls: client.upload and self.parentnode.set_uri.

OTOH, add_file() is used in an awful lot of unit tests, because it's so darned convenient. It'd be easier to change DirectoryNode.add_file to use client.upload, since the client knows about History, but DirectoryNode doesn't currently hold a reference to the client (just the nodemaker and the uploader). Maybe we should move client.upload over to the nodemaker?

[davidsarah]

The SFTP and FTP frontends also use add_file.

[davidsarah]

Replying to davidsarah:

The SFTP and FTP frontends also use add_file.

As does the drop-upload frontend. All of these have access to the Client, so one possibility would be to add a client= argument to add_file. But since all uploads should be included in the history, I would think (without having looked in detail at the code) that the uploader should have (that is, be instantiated with) a reference to the History.

[Brian Warner <warner@…>]

In fd676a5846fce5da:

(The changeset message doesn't reference this ticket)

[warner]

excellent idea! Just landed it. I don't know why I didn't think of that before.. it might be that I've been hoping to get rid of the Uploader service (as we did with the old Downloader service), and so I didn't want to add anything extra to it. But that's a project for another day, and adding History to the Uploader is the perfect solution for this. Thanks!