Slashdot Mirror
Power Outage Takes Wikimedia Down
Baricom writes "Just a few weeks after a major power outage took out well-known blogging service LiveJournal for several hours, almost all of Wikimedia Foundation's services are offline due to a tripped circuit breaker at a different colo. Among other services, Wikimedia runs the well-known Wikipedia open encyclopedia. Coincidentally, the foundation is in the middle of a fundraising drive to pay for new servers. They have established an off-site backup of the fundraising page here until power returns."
They'll turn the lights off.
Coincidentally, the foundation is in the middle of a fundraising drive to pay for new servers.
"You see, guys? This is what could happen if we ever ran out of money. Now cough up some dough!"
The coolest voice ever.
What happened?
At about 14:15 PST some circuit breakers were tripped in the colocation facility where our servers are housed. Although the facility has a well-stocked generator, this took out power to places inside the facility, including the switch that connects us to the network and all our servers.
What's wrong?
After some minutes, the switch and most of our machines had rebooted. Some of our servers required additional work to get up, and a few may still be sitting there dead but can be worked around.
The sticky point is the database servers, where all the important stuff is. Although we use MySQL's transactional InnoDB tables, they can still sometimes be left in an unrecoverable state. Attempting to bring up the master database and one of the slaves immediately after the downtime showed corruption in parts of the database. We're currently running full backups of the raw data on two other database slave servers prior to attempting recovery on them (recovery alters the data).
If these machines also can't be recovered, we may have to restore from backup and replay log files which could take a while.
After returning from the power outage, the servers have just been slash-fried.
Although we use MySQL's transactional InnoDB tables, they can still sometimes be left in an unrecoverable state
Ya know, I just don't understand why so many projects with such high visibility and requirements for reliability use a toy database like MySQL.
Someone PLEASE tell me why. Because right now the only thing I can think is that people just don't know how to pronounce "Postgres".
This is not a troll or a flame at all but between this and the livejournal servers, it sure sounds like hell if your mysql servers ever go down unexpected.
Is mysql the only dbase like this or does postgres get corrupted as well during unplanned downtime? If I recall from using MSSQL servers , we never had a problem like this. We would simply reboot the servers and not worry about tables being left in unrecoverable states. Please correct me if I am wrong though.
Is there any way around this or will this always be a problem with mysql?
I found this useful information about power outages:o utage
http://www.wikipedia.org/search?/power_
A power outage has taken down wikipedia! as a community we must carry the torch!
Paul Grosfield - the quicker picker upper.
Even when the servers go back on, they'll be slashdotted.
As that economic genius, Eric Cartman taught us:
1) Get something other people love
2) Don't let them use it
3) Profit!
It doesn't hurt if you are running a fund drive at the same time, either.
To make laws that man cannot, and will not obey, serves to bring all law into contempt.
--E.C. Stanton
Meanwhile, the devs are working fairly furiously to get it back up
;)
Don't worry, we'll take care of your backup servers in the meantime.
The coolest voice ever.
So far one of our database servers has completed a successful recovery (we're working through them all). On a gigabit link it takes something between 90 minutes and 4 hours to rsync from one to another. As soon as we have two database servers working, we'll be restoring service in read only mode. Likely to be that 90 minutes to 4 hours from now as worst case.
I'll post followups to this post later, as we're closer to being fully recovered.
it's as though 300,000 people cried out and were suddently silenced ...
and then somebody diffed the change and made them speak again
This outage, as well as our beloved slashdotting, is yet another argument for URIs, rather than just URLs. URLs are like IP#s; they're absolute pointers to specific object locations, in terms of the storage/retrieval interface of a single instance. URIs are virtual, like domain names. They are distributed in DNS, a Netwide database, updated for current lookup values for actual retrieval. URLs need the same kind of layer. Of course, some other characteristics of these objects must be reflected in the URI model that are not appropriate to IP#/domain names, like multiple identical copies, or perhaps versions.
Just cacheing copies, either actively with a redirection URL, or passively in caching backbone webservers, isn't cutting it. Caching values is always better suited to solving performance problems, creating its own concurrency and identy problems. Not to mention the publication limits of "opt-in" caches, like Coral or Google, which are an afterthought (and usually unknown) to the published object itself. Google has a huge, high-performance URL lookup system. It's taken quite a bit of value from the Internet, and all the content creators it rides on to derive all its value. It give back quite a bit, with its simple, fast, effective interface. Google is perfectly positioned to make its name truly synonymous with an Internet revolution (not just a pinnacle of search evolution) by implementing URIs. If Google let objects get looked up by a URI code as simple as say, [A-Za-z0-9]+, it could get halfway to its namesake in objects with just 28 "digits"; just 7 digits would cover each object instance in its database right now, dozens of times over. If Google opened up such a URI protocol to anyone on the Web running such a "DIS" server, just like DNS, they could offload much of the work, avoid accusations of trying to "own the Internet", and improve their own service immeasurably, not least by making broken links in their database a quaint old curiosity. Will they rock our world, or will another big player, like Archive.org do it, before Microsoft, desperate to distinguish MSN Search, ruins it for everyone with some kind of proprietary hack that favors MS objects?
--
make install -not war
IIRC, that's the Fire Code. The breaker needs to be able to unconditionally kill all power inside the facility. Thus -- it kills the power post-UPS.
Sometimes it costs more to do things wrong, in the long term, than to do them right.
500GB of disk, 5TB of transfer, $5.95/mo
The database servers have dual redundant supplies and the colo tells us that TWO circuit breakers tripped. Fun. Not. Do try to avoid having the same happen to you - losing both circuits isn't fun.
You can look things up on answers.com.. They mirror wikimedia, as well as other dictionaries/encyclopedias.
Yes. It's in our plans regardless of what happens with Google.
- Distributed caches - now majority of hits are served by caches, and some of them are offsite. It was a pilot project for a while and now we're trying to design and build scalable infrastructure for that. But still, lots of edits are served uncached.
- Distributed file systems - are there any? NFS is single-server system, MS has something, PVFS has no redundancy, GoogleFS is closed and not released, Coda, AFS, all of those just don't work. Right now we're trying to develop MogileFS (the perl-based app-level file storage by LiveJournal) store and sure there are other ideas.
- Distributed database - there are no proper large database multimaster opensource solutions. MySQL with replication and transactional data store is used. In this event it would be great to have second datacenter nearby with additional DB replicas and gigabit interconnection, but that costs money. And app-level bidirectional replication is in plans for both MySQL and PostgreSQL. And SAN deployment is too costly.
And yes, MediaWiki code has PostgreSQL support, but migrating from one database to another without proper tests, benchmarks and insurance isn't very mature.Yes. I wrote that cached page and it's now a bit out of date. IF, and it's not certain, local fire regulations permit the use of UPS systems in the racks we're going to be installing them. Decided on that after LiveJournal's unfortunate experience. But don't yet have them.
As soon as I saw "Power corrupts. Power failure corrupts absolutely" I thought, the damn commies finally did it! But no, not hacked by commies...just by a renegade circuit breaker.
www.kiwilyrics.com - a wiki for lyrics
Our database masters do have dual power supplies. The circuit breakers were tripped on both sides.
Chu vi parolas Vikipedion?
The colocation facility has diesel generators to protect against the outside power going out. Thanks to the miracle of circuit breakers, power circuits inside the facility shut off (including both circuits feeding our dual-power supply machines).
Chu vi parolas Vikipedion?
A completely designed, 100% empty database.
A COMPLETE log of all the SQL statements that were applied to it IN the order they were used. This is obtained by the application logging the SQL statements to the SQL log file AFTER the SQL statement is succesfully executed.
When a data base failure occurs, stop everything, 'replay' the backed up SQL logfile (thats on a separate backup system) on a copy of the empty DB there. TADA! you are back in business back to the point of failure!
Read the Wikipedia page. That's exactly what they've done, but because the MySQL database got corrupted, instead of just falling back a few minutes, they may have to go right back to a full backup and replay the log since then, which takes a lot more time than replaying a few transactions.
The solution is to switch to a database that actually implements ACID (the second letter stands for "Consistency" and the last letter stands for "Durability" which is what failed here).
When all you have is a hammer, every problem starts to look like a thumb.
He thinks I'm a god now!
perhaps I just inadvertently reached clear
I find it an interesting coincidence the power outage happened so soon after that the Xenu article was featured.
Gee, you just had to mention the X-word! Now this thread won't load for most Scientologists because the keyword filters they were forced to install by their Church will see "Xenu" and block the site. After all the mere sight of the word could cause "pneumonia and death" if you haven't paid the Church of Scientology for the proper preparation.
Wikipedia's Xenu article has an interesting history if you look, as I did the other night when it was featured. Scientologists vandalize it regularly. You're supposed to pay them a half million (or some absurd sum of money) to find out about Xenu. After you find out, you're too embarrassed to admit to anybody that you paid a half million to learn that your problems are caused by bad science fiction, when you could have bought a house in Silicon Valley instead. So they obviously don't want a Wikipedia article giving away their half-million-dollar "trade secret" for free.
One trick I saw was to use HTML entities to spell out insults at the top of the article- like "only an idiot would believe this" or something. In the editor window, the entities weren't rendered and each letter appeared as a hex code.
A more effective attack took a different approach. The vandal in this case changed "Scientologists" to "Muslims", "Scientology" to "Islam", and inserted a boring-sounding sentence at the end of the first paragraph claiming that "Xenu" is another name that Muslims use for "Allah". It completely discouraged you from reading further. If you didn't know better you wouldn't find out how "Allah" distributed the thetans around volcanoes on various planets and blew them up with hydrogen bombs, and how their blown-up spirits cause problems in your personal life today.
This is OT, but what the hell, why not whack a beehive? Additional information on Xenu:
Operation Clambake (Hubbard maintained that humans are descended from clams)
The Xenu leaflet (all about Xenu- this information can save you lots of $$$$$)
The road to Xenu (authored by a woman who got suckered)
The Google cache of Wikipedia's Xenu article is also a must read.
I'm wondering if I'll get a lot of freaks, downmoderations, and hostile AC replies after I post this. After all, that's the kind of thing that Hubbard called "fair game". If it sinks below default visibility I'll repost it again with my karma bonus, so you theta-clear-wannabes out there can save your points for someone else.
:::eyes my UPS::::
::::ponders for a momment::::
:::eyes the serial cable that gracefully shuts down said computer in the event of a power failure::::
:::ponders some more::::
:::eyes the spare UPS sitting in the corner that used to be connected to a database server::::
Hmm, I think i'm almost onto something here, but i just can't seem to nail it down...
-Chris
--an unbreakable toy is useful for breaking other toys--
The link in the article is broken, here's the proper one:
http://wikimedia.org/fundraising/
Beware: In C++, your friends can see your privates!
If only I were a mod. Informative, and just plain funny if you ask me. I've read about that entire thing going back and forth and its kinda odd. On the one hand I think that Wikipedia should be limited to who can change it. But on the other its really neat and diverse to let everybody at it.
Oh well.. Slightly OT
Posted on the mailing list wikipedia-l 32 minutes ago:
From: Brion Vibber
Reply-To: wikipedia-l@wikimedia.org
To: Wikipedia-l, Wikimedia Foundation Mailing List, Wikimedia developers
Date: Tue, 22 Feb 2005 04:47:56 -0800
Subject: Re: [Wikipedia-l] Wiki Problems?
Brion Vibber wrote:
> There was some sort of power failure at the colocation facility. We're
> in the process of rebooting and recovering machines.
The power failure was due to circuit breakers being tripped within the colocation facility; some of our servers have redundant power supplies but *both* circuits failed, causing all our machines and the network switch to unceremoniously shut down.
Whether a problem in MySQL, with our server configurations, or with the hardware (or some combination thereof), most of our database servers managed to glitch the data on disk when they went down. (Yes, we use InnoDB tables. This ain't good enough, apparently.)
The good news: one server maintained a good copy, which we've been copying to the others to get things back on track. We're now serving all wikis read-only.
The bad news: that copy was a bit over a day behind synchronization (it was stopped to run maintenance jobs), so in addition to slogging around 170gb of data to each DB server we have to apply the last day's update logs before we can restore read/write service.
I don't know when exactly we'll have everything editable again, but it should be within 12 hours.