search.twitter.com is unreachable from internet connections routed through Sprint/Nextel, as per the unfortunate news
Update: this has been corrected
search.twitter.com is unreachable from internet connections routed through Sprint/Nextel, as per the unfortunate news
Update: this has been corrected
In the past 2 hours, we’ve experienced 2 periods of network issues as we installed new networking hardware.
These periods lasted approximately 3 minutes each. We do not expect any further issues.
Monday and Tuesday of this week we have had connection problems to twitter.com between 0730 and 1400 PDT. Some users were seeing timeouts in their browsers on a regular basis through these periods.
We discovered the source of problem this evening and have fixed it.
For the technically inclined: It had to do with a limit set in our load balancer. We crossed a threshold of traffic that caused it to use up the connection pool available to it. The problem was, it was failing without throwing the usual error, due to a bug. We now know about that bug (as does our hosting partner), so it shouldn’t happen again.
We’re currently diagnosing some network latency that is resulting in occasional timeouts and site slowness for some folks.
The bug reported on Friday about folks being able to see some protected updates by viewing XML source was fixed on Friday. If you protect your updates, only people you approve will receive them.
Twitter search is minutes behind realtime. We are working to resolve this.
Update: this has been resolved
We have temporarily turned off user search because of a problem affecting overall site performance. It will be re-enabled once we’ve solved the underlying problem.
Direct Messages are not currently being delivered via SMS. You can still see any direct messages sent to you on the web or via email (if you elect to receive DMs that way).
We hope to resolve this problem quickly.
API clients were seeing a higher than normal number of 500 errors this morning. We’re tracking down the problem and believe it is now resolved. More investigation is ongoing…
All email from Twitter is currently being blocked by users with email accounts on AT&T, SBC or BellSouth. If your email account is with one of these providers, you will not receive mail from us and, therefore, will not be able to recover your password.
We are working with these ISPs to resolve the problem. In the meantime, if 1) your email account is with AT&T, SBC or BellSouth and 2) you need assistance in resetting your password, please write to twitter.recovery@gmail.com and we will do our best to help.
In October 2006, just three months after Twitter launched publicly, we added IM support—i.e., the ability to get and send tweets via XMPP/Jabber/Google Talk. I was a big fan of this feature, because this interface, which millions of people were already familiar with, seemed a perfect fit for Twitter’s real-time nature.
In December of 2006, we extended that support to AIM, enabling a much bigger number of users to interface with Twitter via the same system they talk to their friends on all day.
While off to an early start, since then, our IM feature has been, well…spotty. We first killed AIM support after struggling for months to make it reliable (which was a side-project to trying to keep the service as a whole reliable). And our Jabber support has been up and down until about four-and-a-half months ago when it’s just been…down.
We’ve been unclear about its status and what to expect; I want to clarify that now.
First: I know a lot of people love this feature. I’ve heard personally from many folks who say it’s critical for their enjoyment of Twitter. So it kills me that we haven’t been able to deliver on this consistently. And the bad news is, we don’t have a quick fix.
A New Way of Doing Things
If you’ve been using Twitter for a while, you’ve probably noticed that things have gotten much more steady on the Twitter ship in the last few months. There’s a lot of reasons for this. Besides some key hires and herculean efforts from top-notch engineers, there’s been a conscious decision to try and not make promises we can’t keep. That is, we want to be solid on our current offerings before adding to the load and getting in a precarious situation again.
The fact is, our infrastructure for IM was never ready for prime time. Like a lot of things, it was built rapidly when we were small and had to be completely re-thought in order to support a much bigger user base. Since we were (famously) having trouble keeping up with that demand in the core service, IM, which was used by a minority of users, always got the shaft.
Startups are all about launching things that are not ready for prime time—because startups don’t live in prime time. You can always rush to make something ready if people like it. Except when you can’t. The process of going from being able to throw things at the wall and see if they stick, to really thinking carefully about how something’s going to scale and effect the system as a whole, is an inevitable evolution that companies go through (if they’re successful). Sometimes they do it gracefully, sometimes less so.
To summarize, we want to bring IM back. We intend to bring IM back. But we’ve officially moved it from our Things That are Broken list to our Things We Want to Build list. Based on our analysis, the cost-to-benefit for IM for the most users is not as high as some other things—so it will be a while before we tackle it. Like any budget (in this case, the budget of our limited engineering time), tough calls need to be made—especially in these times. And while we don’t expect everyone to agree with this decision, we at least want to be straightforward with you.
Sorry we haven’t been very communicative on this before. Saying it was “coming soon” was more of a case our our wishful thinking (that we would get to it real soon now) than wanting to mislead.
Thank you for your understanding and patience.
-Evan Williams
p.s. - Like other features that Twitter doesn’t have natively, IM support could be built by a third party using our API. In fact, it has been: Check out excla.im. It may or may not be ready for prime-time either, but it’s very promising. (Hi, @harper!). (There may be others.)
We will have a 30 minute downtime tonight at 10p Pacific to perform system maintenance.
We’re working on an issue with search not serving its formatting correctly, resulting in some funky looking results.
Update: This has been fixed.
In the past hour, some users will not have received all messages they should have via SMS. This includes the delivery of direct messages. All updates and DMs will be viewable on the web or over the API. This problem should be resolved in the next hour or so.
Yesterday we were throwing 500 errors on API requests to /users/show. This has now been fixed.