Hello,
Thank you to all who have taken the time to answer.
As more people have asked, here are some details about the project. We want to build a feature in our smartphone app that allows users to read wikipedia articles and we want to make the articles and their images available offline to them, which is why we have to first download this content from wikipedia and wikimedia. We have installed a wikipedia mirror locally and extracted the desired article texts through the API. For the images, we first thought about getting the image dump tarballs. However, the articles (and consequently the images) are spread over more language domains, so this approach would have been both inefficient and too much space consuming.
I'll look into the rsync approach.
Once again, many thanks for all your suggestions. Mihai
-----Original Message----- From: wikitech-l-bounces@lists.wikimedia.org [mailto:wikitech-l-bounces@lists.wikimedia.org] On Behalf Of Jeremy Baron Sent: 23 September 2013 17:12 To: Wikimedia developers; Wikipedia Xmldatadumps-l Subject: Re: [Wikitech-l] Bulk download
On Sep 23, 2013 9:25 AM, "Mihai Chintoanu" mihai.chintoanu@skobbler.com wrote:
I have a list of about 1.8 million images which I have to download from
commons.wikimedia.org. Is there any simple way to do this which doesn't involve an individual HTTP hit for each image?
You mean full size originals, not thumbs scaled to a certain size, right?
You should rsync from a mirror[0] (rsync allows specifying a list of files to copy) and then fill in the missing images from upload.wikimedia.org ; for upload.wikimedia.org I'd say you should throttle yourself to 1 cache miss per second (you can check headers on a response to see if was a hit or miss and then back off when you get a miss) and you shouldn't use more than one or two simultaneous HTTP connections. In any case, make sure you have an accurate UA string with contact info (email address) so ops can contact you if there's an issue.
At the moment there's only one mirror and it's ~6-12 months out of date so there may be a substantial amount to fill in. And of course you should be getting checksums from somewhere (the API?) and verifying them. If your images are all missing from the mirror than it should take around 40 days at 0.5 img/sec but I guess you probably could do it in less than 10 days if you have a fast enough pipe. (depends on if you get a lit of misses or hits)
See also [1] but not all of that applies because upload.wikimedia.org isn't MediaWiki. so e.g. no maxlag param.
-Jeremy
[0] https://meta.wikimedia.org/wiki/Mirroring_Wikimedia_project_XML_dumps#Media [1] https://www.mediawiki.org/wiki/API:Etiquette _______________________________________________ Wikitech-l mailing list Wikitech-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l