The basic problem is that the page content dumps are ordered by revision number within each page, which makes good sense for dumps users but means that the addition of a single revision to a page will shift all of the remaining data ,resulting in different compressed blocks. That's going to be true regardless of the compression type. In the not too distant future we might switch over to multi-stream output files for all page content, fixing the page id range per stream for bz2 files. This might let a user check the current list of page ids against the previous one and only get the streams with the pages they want, in the brave new Hadoop-backed object store of my dreams. 7z files are another matter altogether and I don't see how we can do better there without rethinking them altogether. Can you describe which dump files you are keeping and why having them in sequence is useful? Maybe we can find a workaround that will let you get what you need without keeping a bunch of older files. Ariel On Tue, Jul 28, 2020 at 8:48 AM Count Count <countvoncount123456(a)gmail.com> wrote:

Are you looking for the most current version of each page? Do you want articles or also talk pages, user pages and the rest? In any case, There are a couple of projects that might be of interest to you. One is the so-called adds/changes dumps, available here: https://dumps.wikimedia.org/other/incr/ The other istwork being done on producing HTML dumps; you may follow the progress of that on Phabricator: https://phabricator.wikimedia.org/T254275 Note that parsing wikitext to generate HTML is quite intensive; you might look at the Kwix project for more information about how they do it: https://github.com/openzim/mwoffliner Ariel On Wed, Jul 29, 2020 at 1:48 PM griffin tucker <gtucker4.une(a)hotmail.com> wrote:

I want all revisions of pages, and also user pages and talk pages. Kiwix looks like an interesting project, however I want all revisions of pages, not just the latest. I can think about adding Images/vectors/other attachments later. For now, I just want to download a few sequential dumps and watch the data savings. If I have to use mwoffliner then I’ll use that. I’ve read through most of the phabricator link, however I don’t understand everything in it. I set the blocksize to 4k instead of the default 128k on a Debian virtualbox using zfs, however as soon as I switched on dedup, it slowed to a crawl, even with 8gb of dedicated RAM. As long as it’s not writing data, it’s not so slow, so for my purposes this isn’t a problem. Also, this is just a temporary setup to get familiar with zfs – eventually I’ll be using aws, pending budget requirements. I’m still not sure how I’ll approach indexing them and using revision id’s, but I’ll figure that out as I go along this adventure. From: Xmldatadumps-l &lt;xmldatadumps-l-bounces(a)lists.wikimedia.org&gt; On Behalf Of Richard Farmbrough Sent: Thursday, 30 July 2020 2:27 AM To: xmldatadumps-l(a)lists.wikimedia.org Subject: Re: [Xmldatadumps-l] Has anyone had success with data deduplication? If you extract the records from the XML into individual files, then the ZFS deduplication should take effect. I'm not sure this is a great way to handle the data, because you will potentially waste a lot of disk space on "slack". (This is actually a good use case for slackless files, since they won't grow. But this would break the block alignment property.) You will also have to decide what to call each file. If you use page names you will have duplicates, if you use revision IDs then you have effectively done the deduplication yourself - and you will need to create an index of some form. A nice alternative would be to parse each file and remove revisionID's you have already seen. The first file you process will not shrink, subsequent files will shrink a lot. You will however be missing many revisions if you only use pages-current dumps. On Wed, 29 Jul 2020 at 16:33, Ariel Glenn WMF <ariel@wikimedia.org<mailto:ariel@wikimedia.org>> wrote: Are you looking for the most current version of each page? Do you want articles or also talk pages, user pages and the rest? In any case, There are a couple of projects that might be of interest to you. One is the so-called adds/changes dumps, available here: https://dumps.wikimedia.org/other/incr/ The other istwork being done on producing HTML dumps; you may follow the progress of that on Phabricator: https://phabricator.wikimedia.org/T254275 Note that parsing wikitext to generate HTML is quite intensive; you might look at the Kwix project for more information about how they do it: https://github.com/openzim/mwoffliner Ariel On Wed, Jul 29, 2020 at 1:48 PM griffin tucker <gtucker4.une@hotmail.com<mailto:gtucker4.une@hotmail.com>> wrote: I figured I would decompress the .bz2 and .gz files and that subsequent downloads of dumps would only store the changes, disregarding the compressed .bz2 , .gz , and .7z files. My purposes are just experimenting/learning (I’m a first year comp-sci student) and I really like the idea of downloading multiple dumps and it not taking up much more space. My plan was to download a few dumps of enwikinews as a test, and then go for enwikipedia when it’s tested successfully. I’ve just been doing this locally, however I was planning on using cloud virtual machines like aws, and then moving them to glacier for long-term storage (copies of the massive volumes). I’ve tried following the guides for using mediawiki to reproduce the dumps, but it runs into errors after only a few thousand pages. I was going to reproduce each dump and then scrape locally for .html files and store those. Images would be a bonus. Then, every month I want to run a script that would do all of this automatically, storing to a dedup volume. That’s my plan, anyway. From: Ariel Glenn WMF <ariel@wikimedia.org<mailto:ariel@wikimedia.org>> Sent: Wednesday, 29 July 2020 4:49 PM To: Count Count <countvoncount123456@gmail.com<mailto:countvoncount123456@gmail.com>> Cc: griffin tucker <gtucker4.une@hotmail.com<mailto:gtucker4.une@hotmail.com>>; xmldatadumps-l@lists.wikimedia.org<mailto:xmldatadumps-l@lists.wikimedia.org> Subject: Re: [Xmldatadumps-l] Has anyone had success with data deduplication? The basic problem is that the page content dumps are ordered by revision number within each page, which makes good sense for dumps users but means that the addition of a single revision to a page will shift all of the remaining data ,resulting in different compressed blocks. That's going to be true regardless of the compression type. In the not too distant future we might switch over to multi-stream output files for all page content, fixing the page id range per stream for bz2 files. This might let a user check the current list of page ids against the previous one and only get the streams with the pages they want, in the brave new Hadoop-backed object store of my dreams. 7z files are another matter altogether and I don't see how we can do better there without rethinking them altogether. Can you describe which dump files you are keeping and why having them in sequence is useful? Maybe we can find a workaround that will let you get what you need without keeping a bunch of older files. Ariel On Tue, Jul 28, 2020 at 8:48 AM Count Count <countvoncount123456@gmail.com<mailto:countvoncount123456@gmail.com>> wrote: Hi! The underlying filesystem (ZFS) uses block-level deduplication, so unique chunks of 128KiB (default value) are only stored once. The 128KB chunks making up dumps are mostly unique since there is no alignment so deduplication will not help as far as I can see. Best regards, Count Count On Tue, Jul 28, 2020 at 3:51 AM griffin tucker <gtucker4.une@hotmail.com<mailto:gtucker4.une@hotmail.com>> wrote: I’ve tried using freenas/truenas with a data deduplication volume to store multiple sequential dumps, however it doesn’t seem to save much space at all – I was hoping someone could point me in the right direction so that I can download multiple dumps and not have it take up so much room (uncompressed). Has anyone tried anything similar and had success with data deduplication? Is there a guide? _______________________________________________ Xmldatadumps-l mailing list Xmldatadumps-l@lists.wikimedia.org<mailto:Xmldatadumps-l@lists.wikimedia.org> https://lists.wikimedia.org/mailman/listinfo/xmldatadumps-l _______________________________________________ Xmldatadumps-l mailing list Xmldatadumps-l@lists.wikimedia.org<mailto:Xmldatadumps-l@lists.wikimedia.org> https://lists.wikimedia.org/mailman/listinfo/xmldatadumps-l _______________________________________________ Xmldatadumps-l mailing list Xmldatadumps-l@lists.wikimedia.org<mailto:Xmldatadumps-l@lists.wikimedia.org> https://lists.wikimedia.org/mailman/listinfo/xmldatadumps-l -- Landline (UK) 01780 757 250 Mobile (UK) 0798 1995 792