on 09-18-2015 9:39 AM
Dear all,
I have created a stored procedure in HANA which will select the parameter "Property" into an array, iterate over this array and call a second stored procedure ("pattern_Identification") with the parameter "Property" as input parameter. As I have around 20 different "Property" parameters the procedure "pattern_Identification" is called several times sequential and makes the whole process slowly. That's why I would like to parallelize the process.
Please let me know how I can use HANA capability to call this stored procedure parallelly ?
Regards,
Matthias
PROCEDURE "MY_SYSTEM"."My_Project.procedures::run_Pattern_Identification" (
IN personID ALPHANUM(10)
)
LANGUAGE SQLSCRIPT
SQL SECURITY INVOKER
DEFAULT SCHEMA TEST_SYSTEM
AS
BEGIN
DECLARE property_Array VARCHAR(50) ARRAY;
DECLARE a_Length INT;
DECLARE a_Index INT;
DECLARE property VARCHAR(50);
property_Tab = SELECT DISTINCT "PROPERTY" FROM "ENVIRONMENTAL_DATA";
property_Array := ARRAY_AGG(:property_Tab.PROPERTY);
a_Length := CARDINALITY(:property_Array);
FOR a_Index IN 2 .. a_Length DO
property := :property_Array[:a_Index];
CALL "test_Project.procedures::pattern_Identification"(:property); -- Call could be parallelly
END FOR;
END;
Hi Matthias,
You can follow below steps to run these procedures parallelize.
1> Combine the data of Array into a string by storing Property names as comma separated. Use below codes
DECLARE v_property VARCHAR(300);
FOR a_index IN 1..a_Length DO
IF a_index<a_Length THEN
v_property:= :v_property||:property_array[:a_index]||','
ELSE
v_property:= :v_property||:property_array[:a_index]
ENDIF;
END FOR;
2>Now change the input parameter of inner procedure as IN <paramter_name >VARCHAR(300)
and call the procedure making v_property as input paremeter.
CALL "test_Project.procedures::pattern_Identification"(:v_property);
3>Inside this pattern_identification procedure split this string into the array and use the data of the array in the logic so procedures will run just like parallel process.
Let me know in case any issue.
Regards,
Jyoti
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
Hi Jyoti,
Thank you for your help.
Sorry but I don't fully understand Step 3.
When I split the string into the array I have again a array with all Properties inside
and I must iterate over it and run my logic.
Where is the difference between my current solution?
May I just didn't understood something correctly.
Regards,
Matthias
Hi Matthias,
No need to iterate it gain inside the pattern_identification procedure after making string into array. If you are using the values of property inside a SELECT Statement than just put below codes.
DECLARE v_propert_array VARCHAR(100) ARRAY;
DECLARE v_count_arry INTEGER;
DECLARE v_index INTEGER;
WHILE(LENGTH(:v_property)>0) DO
v_index:=v_index+1;
IF (SUBSTR_BEFORE(:v_property, ',') ='') THEN
v_propert_array[:v_index] := :v_property;
ELSE
v_propert_array[:v_index] := SUBSTR_BEFORE(:v_property, ',');
END IF ;
v_property := SUBSTR_AFTER(:v_property,',');
END WHILE;
v_count_arry := CARDINALITY(:v_propert_array);
v_sel_properties := UNNEST (:v_property_array) AS (SELECTED_PROP);
Now you can use statemet ("SELECT SELECTED_PROPS FROM :v_sel_properties") in any SELECT Statement for any type of operation without using any iteration.
If you still face any issue then please paste that piece of code where you are putting your logic.
Regards,
Jyoti
Hi Jyoti
Thank you for your example. I still have problem to adopt it on my logic, that's why I will post the logic code also here.
My main problem is avoiding the use of cursor.
I have a table with dates of discomfort. And want to select from my environmental data for all properties the measured values from the days before a discomfort.
Regards
Matthias
PROCEDURE "MY_SYSTEM"."My_Project.procedures::pattern_Identification" (
IN property VARCHAR(50)
)
LANGUAGE SQLSCRIPT
SQL SECURITY INVOKER
DEFAULT SCHEMA TEST_SYSTEM
AS
BEGIN
DECLARE eventID INT;
DECLARE sequenceID INT :=1;
DECLARE prev_Day DATE;
DECLARE discomfort_Day DATE;
DECLARE item FLOAT;
DECLARE timeframe_length INT = 5;
DECLARE CURSOR c_cursor1 FOR SELECT "DATE" FROM "DISCOMFORT_T" ORDER BY "DATE";
FOR cur_Row as c_cursor1 DO
SELECT cur_Row."DATE" INTO discomfort_Day FROM DUMMY;
FOR eventID IN 1 .. :timeframe_length DO
prev_Day := ADD_DAYS (:discomfort_Day , :eventID * (-1));
SELECT "VALUE" INTO value FROM "ENVIRONMENTAL_DATA" WHERE "PROPERTY" = :property AND "DATE" = :prev_Day ;
INSERT INTO "PATTERN_T" VALUES (:sequenceID, :eventID, :value, :property);
END FOR;
sequenceID := sequenceID +1;
END FOR;
END;
Hi Matthias,
I know I am a bit late in this discussion, but it stuck to the back of my mind.
The main question is of course: why the procedure calls at all?
Looking at what happens in the procedures I figured it's this:
- find all different properties that had been collected so far
- for each of the properties
- find all days with discomfort PLUS the 5 days before that day
- for every day
- read the date and the value of the current property (if available on this date)
- store the result in a table
That's definitively doable in a single SQL statement:
select row_number() over(partition by dates.date order by dates.off desc ) sequence_id
, dates.date
, dates.off as "eventID"
, day_before
, prop.property
, prop."VALUE"
FROM
(select dc."DATE", date_range.off, add_days(dc."DATE", date_range.off) as day_before
from discomfort_t dc
cross join (select 0 as off from dummy union all
select -1 as off from dummy union all
select -2 as off from dummy union all
select -3 as off from dummy union all
select -4 as off from dummy union all
select -5 as off from dummy) date_range) dates
inner join
(select "DATE", property, "VALUE" from environmental_data) prop
on dates.day_before = prop."DATE"
order by dates.date asc, "eventID" desc
The dates "union"-sub-select generates the date offsets you are looking for.
This gets joined with the dates of days of discomfort.
We now have a list of dates of discomfort and the five days before that.
This then gets joined with all the measures properties and values.
Finally the row_number() produces the "sequence_id" starting new with every date of day of discomfort.
I don't have any test data of reasonable size but I don't see why this shouldn't be quicker then the two procedure approach...
Cheers,
Lars
So, now
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
User | Count |
---|---|
88 | |
23 | |
11 | |
9 | |
8 | |
5 | |
5 | |
5 | |
5 | |
4 |
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.