How is it possible that a question was asked "-1s ago"?
[screenshot no longer available]
Update:
The servers seem to be even 6 (or more?) seconds out of sync:
[screenshot no longer available]
-
Clearly it is ;)– ChrisF ModJan 22, 2010 at 12:44
-
I got one of those yesterday, "-2s" in my case.– nb69307Jan 22, 2010 at 12:48
-
2I guess Bart K. types really fast.– Graeme PerrowJan 22, 2010 at 12:48
-
8So was this question "no longer relevant" before it was even posted? ;)– TroggyJan 22, 2010 at 12:48
-
11Uh oh. Bart's figured out how Jon answers questions before they're asked. I suspect this will only be solved through a cage fight.– PollyannaJan 22, 2010 at 13:17
-
@Pollyanna: There are better solutions, remember? (I mean, you did start this post ... ) meta.stackexchange.com/questions/35864/full-contact-moderating/…– John RudyJan 22, 2010 at 14:32
-
3The price you pay for having a fast web site ...– Pop CatalinJan 22, 2010 at 16:04
-
With "Status-completed", was the bug fixed before or after the question was asked?– Andrew GrimmJan 28, 2010 at 6:54
-
1@Andrew: it's hard to measure.– perbertFeb 5, 2010 at 2:17
-
6This post was [closed] as Threat to Self-Awareness by Skynet♦ tomorrow– JaydlesAug 30, 2010 at 22:15
-
problem is back BIG time July 17th 2011 at 3:30am CEST... see meta.stackexchange.com/questions/98816/back-to-the-future– Jonas HeidelbergJul 17, 2011 at 2:34
Add a comment
|
4 Answers
All the servers sync via NTP with pool.ntp.org -- there was an issue previously that is documented on Server Fault.
Most recently, the database servers didn't have access to the internet any more, so their NTP calls to pool.ntp.org were failing.
We set up an internal NTP source for them.
edit: new problem related to time service failing to start. Crossing my fingers..
answered Jan 22, 2010 at 17:45
Whenever I see one of this bugs I'm reminded of Isaac Asimov's The Endochronic Properties of Resublimated Thiotimoline (Can't seem to find it online, sadly).
In Asimov's writing, thiotimoline is notable for the fact that when it is mixed with water, the chemical actually begins to break down before it contacts the water. This is explained by the fact that in the thiotimoline molecule, there is at least one carbon atom such that, while two of the carbon's four chemical bonds lie in normal space and time, one of the bonds projects into the future and another into the past. Thiotimoline is derived from the bark of the (fictitious) shrub Rosacea Karlsbadensis rufo, and the thiotimoline molecule includes at least fourteen hydroxy groups, two amino groups, and one sulfonic acid group, and possibly one nitro compound group as well. The nature of the hydrocarbon nucleus is unknown, although it seems in part to be an aromatic hydrocarbon.
I guess the time is recorded in the database, but the comparison done on the web server.
answered Jan 22, 2010 at 13:03
-
-
1Frankly I'm surprised they haven't done this already. Have two servers NTP to a stable time source elsewhere, and the others NTP to the two servers. They should all be in sync within tens of milliseconds. I can see how this might be exploited to get an answer in a few seconds before someone else... Jan 22, 2010 at 15:22
Rob is probably right. Turns out the built-in NTP client for windows servers is only expected to keep servers within several seconds of each other, not within a second or sub second.
http://blogs.technet.com/askds/archive/2007/10/23/high-accuracy-w32time-requirements.aspx
If they are simply using the built in time sync mechanisms, then you could have posts appearing 10 seconds in the future, and those users would beat others to the punch in the 'newest' listings, even if they posted several seconds after.
answered Jan 22, 2010 at 15:55
-
2So, in other words it's a feature, and not a bug? I wish somebody would teach Microsoft the difference. Jan 22, 2010 at 16:27