Wikipedia:Wikipedia Signpost/2012-07-30/Technology report

Technology report

Talking performance with CT Woo and Green Semantic MediaWiki with Nischay Nahata

Talking performance with CT Woo

CT Woo relaxing on the second day of the 2011 Berlin Hackathon

In the light of recent questions over the long-term reliability of Wikimedia wikis, the Signpost caught up with CT Woo, the Wikimedia Foundation's director of technical operations.

Hey CT. Many users have reported timeouts and other performance problems over the last few months. Does the Foundation view these as separate incidents or as representative of a larger trend?
There are several reasons. For example, we are in the midst of changing file systems from NFS to an object storage system (OpenStack Swift). Since it is a very new product, we did discover a performance issue occuring during some image deletions. We have investigated, tracked it down and I am happy to report it is no longer an issue. Also recently, we hit a Linux kernel bug where systems started rebooting themselves after about 211 days of uptime. As a result, we had to patch all the affected servers. In addition, a number of development teams (especially Platform and Localisation) have changed their build-test-deploy process over the last few months and are now rolling out more frequent (albeit smaller) deployments. I do like to add that 2011/2012 has been a relatively good year for our site uptime metrics, better than 2010/2011. For readers of Wikipedia, the uptime was 99.97%. For editors, the uptime was 99.86%.
Does the Foundation feel that it has the resources at its disposal to make these kind of problems a thing of the past?
Resources are always a constraint. Whenever we encounter or discover a critical issue, we will all circle in to fix the problem. We usually gather the domain experts when we hit a hard problem and they could be from the Foundation or from the community. For example, the Varnish Software folks are helping us now to fix some issues when using Varnish for multimedia streaming purposes.
Is there not a tension between the operations team on the one hand and development teams on the other that could cause more issues in the future?
On the contrary, the teams work together very well. Yes, we do have differences in opinions occasionally but they are all healthy discussions. Most of the time, the operations team aren't the ones who perform the deployment but they are on standby. However, should we find performance issues with the deployment, and depending on the severity, we do revert the changes, using perform profiling to help identify bottlenecks.
CT, thank you.

Google Summer of Code: Green Semantic MediaWiki

The logo of Semantic MediaWiki, a collection of extensions for MediaWiki and the target of Nischay Nahata's Google Summer of Code work

In the second of our series looking at this year's eight ongoing Google Summer of Code projects, the Signpost caught up with developer Nischay Nahata. Nischay is working on performance improvements to Semantic MediaWiki (SMW), a collection of extensions not in use on any Wikimedia Projects, but nevertheless boasting a significant list of adopters. SMW is also regarded as an influential player when it comes to deciding the course of MediaWiki's potential adoption of so-called "structured data" forms, which have recently come to prominence with the establishment of the Wikidata project. While SMW and Wikidata are distinct projects, there is an active exchange of ideas (and developers) between them. Nischay explained to the Signpost what he has been trying to accomplish, and what its broader impact might be:

Nischay regularly updates a blog following his latest progress.

In brief

Signpost poll
Reader poll

Not all fixes may have gone live to WMF sites at the time of writing; some may not be scheduled to go live for several weeks.

  • Gerrit discussions continue: As reported in previous editions of the "Technology report", the start of code review tool Gerrit's own review period has sparked a series of discussions about its utility. In one, the possibility of making Gerrit compatible with popular Git repository management site GitHub was addressed (wikitech-l mailing list); in another, the possibility of changing the Gerrit visuals to something without "puke green/yellow colour schemes" (also wikitech-l). One thought that has come to prominence focusses on the soon-to-be-released Gerrit 2.5, which allows reusers such as Wikimedia to add their own plugins, an advance which will no doubt please those supporting an "improved Gerrit" outcome to the review. Improvements to Git statistical tracking were also in the news this week.
  • Lua to hit first WMF wikis in August: According to a recent update by Director of Platform Engineering, Lua scripts could be in operation on a Wikimedia wikis as soon as next month. Deployment will start on a test wiki, before moving to MediaWiki.org, when the possibilities afforded by MediaWiki's first serious attempt at providing a template programming language will begin to come under serious scrutiny. Talks regarding Lua (see previous Signpost coverage for context) were well received at both Berlin and Washington D.C.; any deployments are likely to attract significant developer attention.
  • Meet the Analytics teams: In a post on the Wikimedia blog, the WMF Analytics team introduced their work, focussed on an update to the Wikimedia Report Card and Kraken, a new "data services platform" aimed at providing a huge array of statistics generated from dozens of datasets. The blog post also stressed the Foundation's commitment to privacy under the heading "counting not tracking". WMF wikis have traditionally been praised for high privacy ratings, albeit at the potential expense of data collection (for example, see previous Signpost coverage).
  • Geolocation, geolocation, geolocation: The possibility of upgrading MediaWiki's geolocation abilities was raised this week (wikitech-l mailing list). Geolocation powers geonotices, messages delivered via the watchlist to editors from a specific area, usually advertising meetups and other real world events. The privacy implications of utilising data other than just publicly available IP addresses will no doubt also need to be considered.
  • One bot approved: 1 BRFA was recently approved for use on the English Wikipedia:
    • Legobot's 13th BRfA, creating a list of incorrectly moved pages for WP:AFC;
At the time of writing, 16 BRFAs are active. As usual, community input is encouraged.