I want to use kdb to maintain my order book, I need to keep this order book always sorted.
At this now, I always resort (xasc,xdesc) my table after I change my table(insert/update/delete),
but I found this way is very slow when my data group to 10k rows.
Maybe this can help: You can have a raw table and a sorted view on it. The view will sort it on demand iff the raw table changes.
András
Maybe you can save a local state for the current updates so that you don’t have to sort a table with other data.
msg.l2update:{
/* handle level2 update messages */
x:“SSZ*”$x; //cast dictionary to relevant types
s:.Q.id x`product_id; //extract sym, remove bad chars
c:“SFF”$/:x`changes; //extract and cast changes
{.[`.gdax.askst`.gdax.bidst y[0]=`buy;(x;y 1);:;y 2]}[s]'[c]; //update state dict(s)
sort.state[s]; //sort state dicts
rec.book[x`time;s]; //record current state of book
}
sort.state:{[s]
/* sort state dictionaries & drop empty levels */
@[;s;{(where 0=x)_x}]'[.gdax.bidst.gdax.askst]; //drop all zeros
@[`.gdax.askst;s;{stdepth sublist asc[key x]#x}]; //sort asks ascending
@[`.gdax.bidst;s;{stdepth sublist desc[key x]#x}]; //sort bids descending
}
depth:5 //depth to maintain in book table
stdepth:100*depth //depth to maintain in state dicts
exchange: “Gdax”
bidst:(u#enlist)!enlist(float$())!float$() //bid state dict
askst:(u#enlist)!enlist(float$())!float$() //ask state dict
lb:(u#enlist)!enlist(bidsbsizesasksasizes!()) //last book state
/* Redefine publish function to pass to TP for real FH */
publish:upsert //define publish function to upsert for example FH
rec.book:{[t;s]
/* determine if book record needs published & publish if so */
bk: bidsbsizes!depth sublist’(key;value)@:bidst[s]; //get current bid book up to depth
bk,:asksasizes!depth sublist’(key;value)@:askst[s]; //get current ask book up to depth
if[not bk~lb[s]; //compare to last book
publish[`book;@[bk;`ex`sym`time;:;(`Gdax;s;“p”$t)]]; //publish record if changed
lb[s]:bk; //record state of last book
];
}
Probably because it is time-series DB?
On Fri, Jul 17, 2020 at 10:08 AM Carfield Yim wrote:>> Probably because it is time-series DB?I think the question is, “Why doesn’t Q support mutable trees, so wecan have things like always sorted dictionaries, etc?”