Last modified: 2014-05-19 20:54:01 UTC
Created attachment 13404 [details] Filename and timestamp with empty oi_archive_name (Bawolff in bug 35637 comment #8) > Well the easiest to find would be everything select oi_name, oi_timestamp > from > oldimage where oi_archive_name = ''; this could be done by anyone with labs > > After that one can look in the thumbnail log. From what I've seen of it, its > full of line about thumbnail failed due to missing src path (this seems to be > the main cause of failing png thumbnails now that vips has removed the size > limit on that format) > > As an aside, It'd be nice if we graphed number of missing files somewhere in > ganglia. Ancedotally it seems like there are more of them then there used to > be. It would be good to get real stats on this very scary problem. $ grep -oE "[0-9]{14}$" oi_archive_name-empty.txt | grep -oE "^[0-9]{6}" | sort | uniq -c 3 200410 32 200411 16 200412 11 200501 9 200502 13 200503 11 200504 6 200505 18 200506 17 200507 7 200508 6 200509 10 200510 28 200511 12 200512 8 200601 20 200602 17 200603 27 200604 12 200605 11 200606 13 200607 16 200608 8 200609 9 200610 21 200611 10 200612 31 200701 13 200702 39 200703 5 200704 23 200705 16 200706 19 200707 18 200708 13 200709 15 200710 9 200711 16 200712 10 200801 9 200802 24 200803 14 200804 12 200805 3 200806 7 200807 4 200808 2 200809 1 200810 6 200811 4 200812 2 200901 1 200902 1 200903 1 200904 10 200905 44 200907 4 200910 2 200911 3 200912 5 201001 5 201002 2 201004 1 201005 7 201006 3 201007 2 201009 1 201012 1 201101 1 201104 2 201105 2 201106 1 201107 17 201108 253 201109 41 201110 73 201111 87 201112 53 201201 63 201202 102 201203 321 201204 366 201205 215 201206 833 201207 142 201208 2613 201209 636 201210 307 201211 351 201212 345 201301 383 201302 358 201303 495 201304 332 201305 379 201306 373 201307 521 201308 2374 201309
Most recent ones might be due to the upload wizard making it easier for users to hit race condition in file backend (don't have bug number off hand)
(In reply to comment #1) > Most recent ones might be due to the upload wizard making it easier for users > to hit race condition in file backend (don't have bug number off hand) Yes, that's my suspicion as well. Sometimes uploaded files go in the wrong place (multiple files under the same title) and when you fix it (by reuploading with Special:Upload) old versions sometimes disappear: bug 40304. There are peaks for every WikiLovesMonuments so it must be related to UploadWizard uploads, not – say – bot uploads.
(In reply to comment #1) > Most recent ones might be due to the upload wizard making it easier for users > to hit race condition in file backend (don't have bug number off hand) For reference bug 54750 was what i was thinking of.
