|
I mean obviously there are many other projects that are less wasteful uses of my time but then again here I am posting in yospos
|
# ? Nov 24, 2014 20:41 |
|
|
# ? May 20, 2024 04:40 |
|
Dessert Rose posted:fair, but spotlight/Alfred only search my local machine, and they are pull only; I can't see my web server data on my desktop without some serious work, and then I still have to do yet more work to get it on my iPhone or my work machine. alfred workflows can run literally any script you want or just pipe anything you want to syslog and have tail -f going
|
# ? Nov 24, 2014 20:44 |
|
power botton posted:alfred workflows can run literally any script you want that's cool actually, I should look at what Alfred is capable of still though all the code would be running on one of my machines locally and I want to be able to connect from any device and see essentially the same data
|
# ? Nov 24, 2014 21:08 |
|
tail -fing syslog was sort of the inspiration here but not quite "rich" enough if that makes sense
|
# ? Nov 24, 2014 21:10 |
|
nope. nothing about this makes any drat sense.
|
# ? Nov 24, 2014 21:31 |
|
power botton posted:nope. nothing about this makes any drat sense. welp
|
# ? Nov 24, 2014 21:48 |
|
make custom filesystems for each thing and spotlight to search them
|
# ? Nov 24, 2014 21:55 |
|
it sounds like you want a webpage honestly. have a bunch of scripts that periodically poll whatever service you are interested in and dump it as json in postgres and then write a simple rails app that queries that
|
# ? Nov 24, 2014 22:14 |
|
power botton posted:nope. nothing about anything makes any drat sense.
|
# ? Nov 24, 2014 22:20 |
|
Snapchat A Titty posted:make custom filesystems for each thing and spotlight to search them plan 9 was a bad idea
|
# ? Nov 24, 2014 22:58 |
|
the talent deficit posted:it sounds like you want a webpage honestly. have a bunch of scripts that periodically poll whatever service you are interested in and dump it as json in postgres and then write a simple rails app that queries that great now im using postgres as an unstructured data store for transient data and writing a loving rails app thats definitely what i want to do i dont necessarily want any of this data to exist past its origination, like the log reader agent will just go "here is the latest line from the nginx error log on the butts server" and then anyone interested in that stream will get it and if some chain of agents processing that stream results in it going into a database, cool, if it ends up on a screen somewhere, great but to me the main feature is that i can modify that signal chain on the fly, write new agents to run wherever, from anywhere, without having to copy files around, and it should Just Work, all the time (so the agents have to end up on a box in my cloud somewhere) that's why alfred [probably] doesn't work for me, it does all the processing on a local box, and i dont want to have to work out syncing this poo poo between my desktop and laptop a web page could work as a "screen" in the half assed uncommunicable design i have in my head but i definitely want to do more than just roll up some logs and tail -f them ill freely admit that 99% of this will be overengineered crap but it'll be fun overengineered crap
|
# ? Nov 24, 2014 23:09 |
|
publish the log as an rss feed
|
# ? Nov 24, 2014 23:10 |
|
Malcolm XML posted:lets say hashing takes 5us. 1k instances, goes up and down between 600-1100 as demand goes up and down so: 5ms. Network latency is 10s-100s of ms, who gives a gently caress doesn't that assume that your system only has to do one lookup to service the request? if you have to call out to a few other distributed systems in turn, things could get noticeable. (20ms overhead would be noticeable in some applications, like ad bidding or typeahead.) 5us does seem like a generous budget for hashing, though!
|
# ? Nov 24, 2014 23:48 |
|
Bloody posted:publish the log as an rss feed yeah, reinventing RSS/Atom is a good place to start. we use thrift/scribe/not-apple's-swift underneath FB ticker, but the latter two are mostly to get scale
|
# ? Nov 24, 2014 23:54 |
|
Subjunctive posted:doesn't that assume that your system only has to do one lookup to service the request? if you have to call out to a few other distributed systems in turn, things could get noticeable. (20ms overhead would be noticeable in some applications, like ad bidding or typeahead.) rendevous hashing is good when you have a fantastic hash + small amount of servers, and you're servicing requests as they come i am exploring using rendevous hashing instead of vnodes but i'll tell you how well it works
|
# ? Nov 25, 2014 01:50 |
|
tef posted:rendevous hashing is good when you have a fantastic hash + small amount of servers, and you're servicing requests as they come are you implementing this on top of riak? want a hand?
|
# ? Nov 25, 2014 05:39 |
|
Subjunctive posted:doesn't that assume that your system only has to do one lookup to service the request? if you have to call out to a few other distributed systems in turn, things could get noticeable. (20ms overhead would be noticeable in some applications, like ad bidding or typeahead.) remember the hash calc can be parallelized, embarrassingly. maybe if u have a cool hash function u could do some precalculation. 5us is kind of a lot as well i think it's about 10-50x overkill also yes 5ms is basically pointless for low latency stuff ( < 100ms say) yeah if u have to make more than one or two lookups its gonna be bad solution: dont do that
|
# ? Nov 25, 2014 05:47 |
|
tef posted:rendevous hashing is good when you have a fantastic hash + small amount of servers, and you're servicing requests as they come tell me more about your fantastic hash
|
# ? Nov 25, 2014 16:47 |
|
SHA-420
|
# ? Nov 25, 2014 21:25 |
|
Soricidus posted:SHA-420 SHA420_WITH_TCC_219
|
# ? Nov 25, 2014 21:25 |
|
steveklabnik's "The Guide" posted:Next, "Hello, world!" is a string. Strings are a surprisingly complicated topic in a systems programming language, and Oh, Klabbs
|
# ? Nov 26, 2014 05:43 |
|
sarehu posted:Oh, Klabbs strings are an opaque blob with optional encoding metadata that can only be handled by the appropriate library (icu). period
|
# ? Nov 27, 2014 01:21 |
|
since all you need is a semi-even distribution, shouldn't CRC32 be fine? then you can calculate a new hash(message, server) every cycle
|
# ? Nov 28, 2014 01:30 |
|
lock-free data structures papers and talks are baffling. from a note to slide 30: "I have a nagging feeling that we actually need four barriers", "avoids a possibly theoretical problem", "often implies correctness", "it is probably not needed" does anyone actually write lock-free data structures, as opposed to lock-free data structure papers
|
# ? Nov 28, 2014 15:19 |
|
hackbunny posted:lock-free data structures papers and talks are baffling. from a note to slide 30: "I have a nagging feeling that we actually need four barriers", "avoids a possibly theoretical problem", "often implies correctness", "it is probably not needed" yeah and what they usually mean is memory barriers
|
# ? Nov 28, 2014 15:37 |
|
it's painful to read, there are so many pitfalls and portability issues. volatile objects may or may not have barriers. CPU barriers may or may not count as compiler barriers. correctness is "probable". I want to write lock-free code but even the experts goof up so often, compilers don't give a poo poo and actively work against you, and libraries are a little pathetic
|
# ? Nov 28, 2014 15:44 |
|
writing lock-free code is for people who think doing exception safety in C++ was for casuals and make strict-aliasing jokes at dinner parties. on basically every team I've ever managed, once performance work gets to a certain point someone will demand to jump down the rabbit hole in search of lock-free bliss. I've learned that there's nothing to be done but let them do it and pull them up when the tugs on the rope become frantic. (exception: deploying read-copy-update, if the stars align and you can meet the scheduler requirements without detonating the architecture of your code.)
|
# ? Nov 28, 2014 16:29 |
|
Give up and use locks. Or a single thread.
|
# ? Nov 28, 2014 19:59 |
|
Zombywuf posted:Give up and use locks. use python and get both for free
|
# ? Nov 28, 2014 20:03 |
|
hackbunny posted:does anyone actually write lock-free data structures, as opposed to lock-free data structure papers do single-producer single-consumer queues count?
|
# ? Nov 28, 2014 20:13 |
|
hackbunny posted:it's painful to read, there are so many pitfalls and portability issues. volatile objects may or may not have barriers. CPU barriers may or may not count as compiler barriers. correctness is "probable". I want to write lock-free code but even the experts goof up so often, compilers don't give a poo poo and actively work against you, and libraries are a little pathetic I like to look at new open source releases for supar spiffy library(tm) and they always gently caress up barriers or side effects on atomic ops, it's a funny train wreck. Almost every single implementation is wrong, even GCC and GMP are fruity.
|
# ? Nov 28, 2014 22:19 |
|
Soricidus posted:use python and get both for free locks
|
# ? Nov 29, 2014 01:09 |
|
Zombywuf posted:locks the gil is so powerful that you don't need any other locks
|
# ? Nov 29, 2014 01:17 |
|
the talent deficit posted:strings are an opaque blob with optional encoding metadata that can only be handled by the appropriate library (icu). period You also need support for different encodings of different substrings, in order to support string concatenation, because pairs of character sets exist that do not have a common superset.
|
# ? Nov 29, 2014 01:48 |
|
sarehu posted:You also need support for different encodings of different substrings, in order to support string concatenation, because pairs of character sets exist that do not have a common superset. or you can choose to not support that model and let the program manage string sequences itself since that really never happens.
|
# ? Nov 29, 2014 02:37 |
|
so i need to make a dead simple api. it has only one endpoint, and the result is going to be a json object with a name and a date and a time. it will accomplish this by querying some database table full of datetimes and names and returning the next closest one in the future. I would say .NET but I don't want to pay for azure for it. What is the most way to accomplish this with a p-lang? I was thinking Rails but idk.
|
# ? Nov 29, 2014 02:39 |
|
Careful Drums posted:so i need to make a dead simple api. it has only one endpoint, and the result is going to be a json object with a name and a date and a time. it will accomplish this by querying some database table full of datetimes and names and returning the next closest one in the future. kill ruby then yourself
|
# ? Nov 29, 2014 02:40 |
|
php it is
|
# ? Nov 29, 2014 02:45 |
|
/index_bak.php?sql=SELECT%20*%20FROM%20items%20ORDER%20BY%20time%20DESC%20LIMIT%201 hth op
|
# ? Nov 29, 2014 03:13 |
|
|
# ? May 20, 2024 04:40 |
|
when did "api" start meaning "rest api"
|
# ? Nov 29, 2014 03:16 |