Simplified version of kdb+-tick
Scripts in this folder will run a real time tickerplant demo.
The original example is now on github.
https://github.com/KxSystems/cookbook/tree/master/start/tick
and is discussed here.
http://code.kx.com/q/tutorials/startingq/tick/
--
This is an even more simplified version for demonstration.
It only runs on Linux. It uses rlwrap and screen.
- Structure
This is discussed in the tutorial, but a ticker-plant (ticker.q) publishes to its clients, or subscribers, these are (r.q and cx.q instances). It receives ticks from a feed (this demo uses feed.q.)
The ticker-plant interoperates with a special client, the real-time database, the r.q instance. This is not in this demo. The rdb is implemented using cx.q - the generic client with a number of methods - vwap, last and others. rdb is one of those.
- Installation and Running
** Installation: top level source directory - no build
You'll need the q interpreter in the ~/q directory. (Soft links work for me.)
** Running
run.sh calls run1.sh run1.sh uses screen to display all the services in one console, and runs the q interpreters under rlwrap in it.
$ run.sh
starts everything. And a screen session called ticker-1234 will be active. (The 1234 is a more or less random number.)
Attach to that screen. The clients should be updating. The most useful is 'last', it should display the last trades and quotes.
*** Testing
You can change what services run by specifying them like this.
You would start the ticker.q script in your q/kdb+ IDE and run other scripts.
$ run.sh feed last
Is a very simple configuration for testing the subscribe and publish processes.
If you want to debug the feed.q then
$ run.sh ticker last
- Changes
** ticker.q and feed.q - timespans
The basic time unit is now the timespan. The sym.q needn't be used to define schema for the clients in cx.q.
** feed.q - back-loading
The feed process now generates some sample ticks over the last hour and replays all those to the ticker plant, (using a special method on the feed, init[]).
It also adds a sequence number - guaranteed to be in time order.
After that, it begins to replay in real-time, using feed, assigned to synchronous method .z.ts, and the .u.upd method.
** feed.q - price generation
I had to pick apart the price generation logic and implement my own. feed0.q is a simple script for debugging.
These notes are for the batch[] function
generate a batch of prices qx index, qb/qa margins, qp price, qn position.
Makes use of alias :: to set globals.
d is a set of deltas.
n0 is a set of indicators for which deltas to use qx is a clever index randomize qx=/:til cnt returns indicators, 1 if the index is in qx. So for qx === 3 2 3 9 8 5 8 6 1 1 n0 === null, 8 9, 1, 0 2, null , 5 , 7, null, 4 6, 3 ie. no zeroes appear in qx, so the delta at position 0 is not selected. because 1 1 in qx at positions 8 and 9, then then deltas at positions 8 and 9 are chosen because 2 in qx at position 1, then the delta at pos 1 is chosen.
d n0 is shorthand for d[n0]
qp raze n - removes any nulls
t[] uses an 'n' but it isn't clear what it is.
There is a problem with this. p begins as the initial prices and is overwritten each time batch is called, but batch can introduce nulls. And, eventually, all the prices are assigned null and stay that way.
** ticker.q
This can support a variety of time-stamps. Currently, it zero-bases a timespan.
- Examples
** ticker, feed and last
There is a log-file of a run of
$ run.sh ticker last feed
There is an image of couple of updates.
- Usefulness
The terms of the licence for the use of the 32 bit free download are such that you can't use this implementation in commercial use.
** Developing clients
If you want to develop clients this will work well enough.
** Coding in q
q/kdb+ systems are tailored for their environment.
-
Much use is made of the host OS to do scheduling and file system management, so no need to learn large APIs.
-
There are no configuration files that adapt your process's application to the environment. So you don't usually start up in debug mode and load special functions. You write three of so variants of a function and use the text editor to define one.
-
Only do expensive application logic coding in the startup.
-
There is no debugger to speak of.
- Postamble
** New to q/kdb+?
I'd recommend the kx website and the excellent tutorials. This ticker plant isn't a very good test-bed for learning q/kdb+.
** Ticker plants in operation
The production kx q/kdb+ ticker plant is much more evolved. And its key parts are usually implemented in 'k' (the programming language that q is derived from.)
** This file's Emacs file variables
[ Local Variables: ] [ mode:text ] [ mode:outline-minor ] [ mode:auto-fill ] [ fill-column: 75 ] [ coding: iso-8859-1-unix ] [ comment-column:50 ] [ comment-start: "[ " ] [ comment-end:"]" ] [ End: ]