Friday, 14 February 2020

Working with the Recursive SQLScript limit in HANA

The SQLScript Reference Manual for HANA contains a description for using resursive logic within procedures and functions. Recursion in procedures has a limitation that sets the maximum call depth to 32. I ran across this as I was rewriting some database-related JAVA code as an SQLScript procedure.

While this call-depth check would normally be a showstopper, it looks like with some thought and knowledge of SQLScript capabilities, the limit can be easily overcome.

I assume this limit applies to procedures and not functions because it is related to the fact that procedures can execute inserts, updates, and deletes to tables, whereas functions cannot. The origin of the limit “32” is likewise not known. It does turn out that the SAP ASE database system likewise has the same type of recursive capability and limitation even though it is not documented at all, so I would not view the HANA limitation negatively.

The Business Application


The code under review involves inventory cost methods for equity stock security purchases and sales. When we sell a security, FIFO, LIFO, and moving average accounting methods are applied to the 2nd part of the accounting journal entry, which is the cost part rather than the revenue part. For cost, we calculate the reduction in inventory and the cost of securities sold. The recursive goal is to figure out which securities to use for costing. Although we could use an SQL loop to select the oldest securities (in the case of FIFO), recursion is also viable. When implementing this, we have many cases where positions for more than 32 small security purchases need to be unwound, in order match one large security sale.

The Recursion Limit Solution


The solution to the recursive call-depth limit is surprising simple.

First, the design of each recursive iteration must be transactionally complete, so that the input of the next iteration does not depend on the result of the previous iteration. This disqualifies things like the factorial example in the HANA documentation. Instead, we must have a design where each iteration commits results to the database so that the next iteration can continue from a known point. If SQL functions had a call-depth limit (which they do not), they too would have been disqualified since they cannot persist results in the form of updates, inserts, and deletes to tables.

Second, we need a way to continue the recursive iteration if it stops. Fortunately, SQLScript has EXIT and CONTINUE exception handlers. With this, we can trap SQL error code 1340 and continue the recursive iteration.

Here is a sample data set, code, and report for a recursive FIFO security costing method which does not abort on a ERR_SQLSCRIPT_NESTED_CALL_TOO_DEEP error:

Here is the Data Set


-- Chronological inventory of purchases (-) and sales (+)
create table inventory (
id         int          not null,
settleDate date         not null,
qty        int          not null,
cost       decimal(4,2)     null,
position   int          not null);

-- Hierarchy for fifo match
create table f_fifo (
parent_id  int              null,
node_id    int          not null,
ord        int          not null,
settleDate date         not null,
qty        int          not null,
cost       decimal(4,2) not null);

create sequence ord start with 1 increment by 1;

insert into inventory values(1004286312,to_date('10/01/2019','MM/DD/YYYY'),-100,10.00,-100);
insert into inventory values(1004286313,to_date('10/03/2019','MM/DD/YYYY'),-100,11.00,-100);
insert into inventory values(1004286314,to_date('10/05/2019','MM/DD/YYYY'), 150,null, 150);
insert into inventory values(1004286315,to_date('10/06/2019','MM/DD/YYYY'),-100,10.80,-100);
insert into inventory values(1004286316,to_date('10/07/2019','MM/DD/YYYY'),  50,null, 50);

Here is the Code


create or replace procedure sp_fifo (in sId int)
as
begin
  declare sPosition, diff, pPosition, pId int;
  declare sSettleDate date;
  declare pCost decimal(10,2);

  -- The oldest purchase (fifo)
  select id, position, cost
  into pId, pPosition, pCost default null, 0, null
  from inventory where id =
    (select min(id)
    from inventory where position < 0);

  -- The sale
  select position, settleDate
  into sPosition, sSettleDate default 0, null
  from inventory where id = :sId;

  if :pPosition < 0 and :sPositon > 0 then
    diff = case
      when :sPosition + :pPosition < 0 then 0
      else :sPosition + :pPosition
    end;

    update inventory set position = position + (:sPosition - :diff) where id = :pId;
    update inventory set position = :diff where id = :sId;
    insert into f_fifo values (:pId,:sId,ord.nextval,:sSettleDate, :sPosition - :diff,:pCost);

    call sp_fifo (:sId);
  end if;
end;

create or replace procedure sp_unwind (in sId int)
as
begin
  using sqlscript_print as prtlib;
  using sqlscript_string as strlib;
  declare msg nvarchar(5000) =
    strlib:format('recursion limit {} exceeded. continuing.',32);

  -- Restart fifo if recursion limit reached.
  declare continue handler for sql_error_code 1340
  begin
    prtlib:print_line(:msg);
    call sp_fifo(:sId);
  end;

  call sp_fifo(:sId);
end;

Here is how to Run it


call sp_unwind (1004286314); -- unwind the first sale
call sp_unwind (1004286316); -- unwind the second sale

And here is the Report


Here is the inventory input table, showing three purchases and two sales:

id          settleDate  ps        qty   cost
----------  ----------  --------  ----  -----
1004286312  2019-10-01  purchase  -100  10.00
1004286313  2019-10-03  purchase  -100  11.00
1004286314  2019-10-05  sale       150
1004286315  2019-10-06  purchase  -100  10.80
1004286316  2019-10-07  sale        50

Here is the f_fifo table result after applying FIFO recursion. I am using a Hierarchy data structure so that I can net the notionals (qty*cost) to compute a balance. For a parent purchase node which is a (-) quantity, sales are applied to it with a (+) quantity as a child node. Child nodes are added as the sale proceeds recursively. If the purchase has been completely unwound, then the balance is 0.0. The hierarchy_descendants_aggregate() function is used to net parent purchases with matching child sales:

-- Load all purchases for matching purposes.
-- Sales were already loaded by sp_unwind().
insert into f_fifo
select null,id,ord.nextval,settleDate,qty,cost
from inventory where qty < 0
and not exists (select * from f_fifo where qty < 0);

ps        parent_id   node_id     settleDate  qty   cost
--------  ----------  ----------  ----------  ----  -----
purchase  null        1004286312  2019-10-01  -100  10.00
purchase  null        1004286313  2019-10-03  -100  11.00
sale      1004286312  1004286314  2019-10-05   100  10.00
sale      1004286313  1004286314  2019-10-05    50  11.00
purchase  null        1004286315  2019-10-06  -100  10.80
sale      1004286313  1004286316  2019-10-07    50  11.00

And here is the final report showing notionals and balances. First we create the hierarchy data structure as a view over the f_fifo table which holds the results.

-- Create hierarchy view
create view h_fifo as
select *
from hierarchy (
  source f_fifo
  sibling order by ord);

Then we return the reporting results:

select
  case when hierarchy_aggregate_type = 0 then 'entry' else 'balance' end "type",
  case when hierarchy_level = 1 then 'purchase'
    when hierarchy_level != 1 and hierarchy_aggregate_type = 0 then 'sale'
    else null end "ps",
  node_id "id",
  settleDate "settleDate",
  qty "qty",
  cost "cost",
  qty*cost "notional",
  balance "balance"
from hierarchy_descendants_aggregate (
  source h_fifo
  measures (
    sum(qty*cost) as balance
  ) with total null
) order by hierarchy_aggregate_type,settleDate,ord;

type    ps        id           settleDate  qty   cost   notional  balance
-----   --------  -----------  ----------  ----  -----  --------  --------
entry   purchase   1004286312  2019-10-01  -100  10.00  -1000.00      0.00
entry   purchase   1004286313  2019-10-03  -100  11.00  -1100.00      0.00
entry   sale       1004286314  2019-10-05   100  10.00   1000.00   1000.00
entry   sale       1004286314  2019-10-05    50  11.00    550.00    550.00
entry   purchase   1004286315  2019-10-06  -100  10.80  -1080.00  -1080.00
entry   sale       1004286316  2019-10-07    50  11.00    550.00    550.00
balance                                                           -1080.00

With this, we show that a combination of HANA features for recursion and hierarchies can be used together in just a few lines of code to efficiently solve what used to be a considerable investment in an external JAVA application.

No comments:

Post a Comment