Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
stephenjudkins
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
24 ms
·
151.
▲
by
stephenjudkins
17y ago
We've been disappointed, in some way or another, with almost every one we've tried as well. Interestingly, one of our uses of Tokyo cabinet completely choked at somewhere between 60-70gb of data. At the same time, these articles can be ver
152.
▲
by
stephenjudkins
17y ago
I understand it refers to Global File System. Given our experience with GlusterFS, I imagine Github would have crashed and burned a long time ago if they were using it. Our use-case (checking if a file exist and grabbing images to occasion
153.
▲
by
stephenjudkins
17y ago
We've gotten consistently sub-100ms latencies from EC2 instances, relatively predictably.
154.
▲
by
stephenjudkins
17y ago
Yes, Engineyard offered by far the best impression of competence and having sysadmins who would actually work well with us. We've heard great things from everyone we've heard of who uses them. I've also heard great things about the Porsche
155.
▲
by
stephenjudkins
17y ago
There were several hosting services that were a good deal cheaper. It costs $576/mo to run an extra large EC2 instance fulltime, while purchasing a server with similar specs only costs $2k or so. The data center across town could give us
156.
▲
by
stephenjudkins
17y ago
When evaluating hosting providers, the quote Engineyard gave us was tremendously expensive. They offered a great deal of value-added services, but we found that it would be cheaper to go with the second-most expensive option (Amazon EC2) wh
157.
▲
by
stephenjudkins
17y ago
When a Memcached instance goes under, all Diggs stored on that instance in the last N days disappear. Not the end of the world for this application, but very undesirable for most things. You could potentially "fall back" to MySQL for thi
158.
▲
by
stephenjudkins
17y ago
Ahh, looks like it's being rolled in sometime soon: https://rails.lighthouseapp.com/projects/8994-ruby-on-rails/...
159.
▲
by
stephenjudkins
17y ago
I would assume Twitter is using conventional ERB templates since their front end is written using Rails. The ERB templates used by rails don't escape HTML by default--you have to write <%= h user_provided_value %> instead of <%=
160.
▲
by
stephenjudkins
17y ago
The study is linked to in the second paragraph. It appeared in what seems to be a reputable, peer-reviewed journal and controls for socioeconomic status. I see this all the time online: a study is published, and a chorus of commenters ech
161.
▲
by
stephenjudkins
17y ago
It seems he may have some good points, but he definitely tried pretty hard to make some of the examples ugly. The way currying is generally done in Scala is much cleaner-looking than this, as the first several google results for "scala cur
162.
▲
by
stephenjudkins
18y ago
We looked at Sphinx but ultimately decided to use Solr/Lucene instead. There were a couple reasons: * We're not doing fulltext search, but searching by a set of tags that we didn't want tokenized/stemmed. Doing this with Lucene was easy,
163.
▲
by
stephenjudkins
18y ago
Why not just keep the crontab file in your repo? A rake task that does `cp #{project_dir}/crontab /etc/crontab` would be sufficient to meet the same needs.
164.
▲
by
stephenjudkins
18y ago
Why not just use cron for this? Yes, the syntax for this does look a lot cleaner, but cron really isn't that hard to use. Why add another layer of complexity (which may break and be difficult to debug) when you could simply add lines to a
165.
▲
by
stephenjudkins
18y ago
Keep in mind that the problems we're having aren't with XMPP, per se, just with ejabberd. Ejabberd worked flawlessly for several months until it started exhibiting some very strange behavior and performance characteristics. We upgraded to