kogir

28 ms·

271.
Miscomputation: Learning to live with errors(tomasp.net)
54 points by kogir 11y ago|0 comments
272.
by kogir 11y ago
The article is not strictly correct. A more detailed explanation can be found here: https://msdn.microsoft.com/en-us/magazine/jj991977.aspx Exceptions actually bubble up to the active SynchronizationContext, and w
273.
by kogir 11y ago
SQL Server Web Edition is not the same as Standard or Enterprise. The reason you use SQL Server instead of PostgreSQL is for Database Encryption, Analysis Services, AlwaysOn Availability Groups, etc. Web Edition won't even do mirroring
274.
by kogir 11y ago
Back in 2005 when I started Loopt I went all in with Windows Server, .NET, and SQL Server. I already knew the tooling[1]. No regrets. It never held us back - product issues did. Now I'm using it again, for the same reasons. I already k
275.
by kogir 11y ago
I've never used a curved 4k monitor (or even seen one), but I use a 32" ASUS PQ321Q at its native resolution for coding. I love it and can't go back. It's right at the edge of what I can discern - any higher PPI and thin
276.
by kogir 11y ago
Oh, that makes sense. Older servers couldn't handle enough clients for connection related limits to matter.
277.
by kogir 11y ago
Crazy that IOCPs were in NT 3.5 in 1994[1], but it took FreeBSD until 2000 to add kqueue[2], and Linux until 2002 to add epoll[3][4]. I guess it just must not have mattered that much? Is it only recently machines became primarily IO bound?
278.
by kogir 12y ago
I imagine the thinking was similar to that behind FreeBSD's accf_http kernel module? https://www.freebsd.org/cgi/man.cgi?accf_http HTTP.sys obviously does a lot more, but especially back in the Windows 2000 days,
279.
by kogir 12y ago
It really depends what you're deploying. Presumably, you're using .NET or SQL Server, or you'd not be using Windows. In my experience making everything run in IIS (with activation, if needed) and using msdeploy to deploy to I
280.
by kogir 12y ago
Or get a 5-10 year support contract for things critical to your company's function?
281.
by kogir 12y ago
HN runs on a single, not so beefy server: FreeBSD 10.1 Nginx 1.7.10 Racket 6.1.1 with some HN and FreeBSD specific patches. 2x 3GHz Intel Xeon-IvyBridge (E5-2690-V2-DecaCore) 8x16GB Kingston 16GB DDR3 2Rx4 SuperMicro X9DRI-LN
282.
by kogir 12y ago
We offer a lot of data (in JSON and XML) via our Firebase powered API: https://github.com/HackerNews/API
283.
by kogir 12y ago
Sorry, we went down under load and figured it was better to prioritize current discussion. Working on resolving the problem.
284.
by kogir 12y ago
We're adding the new "descendants" property to all items. Thus they've changed.
285.
by kogir 12y ago
Thanks! We were trying to track down an error message we were seeing, and you found it.
286.
by kogir 12y ago
Racket is the runtime powering HN. Arc targets it.
287.
by kogir 12y ago
And now I'm doubling down. HN is once again running an entirely new HTTP pipeline. Please let us know if you run into any issues, especially related to cookies and logging in/out.
288.
by kogir 12y ago
And I missed it in code review and testing. Sorry!
289.
by kogir 12y ago
This is sad but not super surprising. Historically, if you had money and wanted a reasonable UI and cleaner integrations, you bought PGP (now from Symantec). GPG was always for people unwilling to pay. For the record I donated. I'm jus
290.
by kogir 12y ago
I'd love to see the results from visiting the Qualys SSL Client Test: https://www.ssllabs.com/ssltest/viewMyClient.html I imagine the issue is that we only support TLS 1.0+ and not SSLv2 or SSLv3. https:/&#x
291.
by kogir 12y ago
Honestly, all credit is due to CloudFlare. They get this stuff right and you need to worry much less about it. Not that the HN server tests poorly, but I can't check since they started disallowing IPs.
292.
by kogir 12y ago
In the age of DNS providers with APIs, you really should have at least two. At YC we've just moved everything to Route53 and will be adding more - likely CloudFlare and Google Cloud DNS. All of it will be managed by configuration files
293.
by kogir 12y ago
Actually, if the goal of a bounty program is to get reports instead of wild exploits, the only metric of success is getting the reports. In the case that someone would have reported it for reasons other than the bounty, the bounty is not on
294.
by kogir 12y ago
Imagine if Twitter limited you to exactly 160 characters when you were composing tweets, instead of indicating you were over. It would be maddening.
295.
by kogir 12y ago
You have to start with the original data, which is obviously de-anonymized. Full data -> [gender, time, origin neighborhood, destination neighborhood] leaves you with a pretty anonymous dataset, and is all that would be required for this
296.
by kogir 12y ago
I've had the chance to try the latest three versions and use it every time I cook now. Before I had no idea how wildly the temperature on my stove would vary, and would basically burn or undercook everything. I'd had luck using te
297.
by kogir 12y ago
100 was picked because it's a large round number. We need vote history to show vote arrows correctly, and for performance reasons we try to avoid loading entire vote files.
298.
by kogir 12y ago
Thanks for pointing that out. You're correct (and I was wrong) - it's not the default, but we format our salts to request it. The docs are here: https://www.freebsd.org/cgi/man.cgi?query=crypt&apropos=0&am
299.
by kogir 12y ago
Well, yes. That was the bug that was fixed.
300.
by kogir 12y ago
I'm not sure how to make it more clear than I did, but this data was not intentionally shared with a third party. Had we known it would happen, we'd obviously have prevented it. The only data we knowingly send to Firebase is alrea

More ›