Select less and less records over time? - mysql

I'd like to write a query or stored procedure to retrieve less and less records over time from a relational database.
Think of this like populating the Google Finance stock chart: The past few days will have all ticks fit the day, and the further you go back, less and less ticks are displayed on each date. All ticks will show for today, 50% of ticks will show for one week ago, 30% for one month ago, and 10% for one year ago. Think of this like a gradient.
Is it possible to achieve this with one query? Or perhaps it would be necessary to use multiple queries? What might this look like?
Note that record ids are non-contiguous (there are gaps), but each record has a timestamp for determining order.
Also note that I am using MySQL.
Here is the structure of my table:
quotes
id
security_id
last_price
bid_price
ask_price
date
timestamp
trade_volume
cumulative_volume
average_volume
created_at

Sounds like you are looking for a constant set of records that represent the time-span. You can do so by defining a control date set.
Here's a sample query (doesn't account for weekends and holidays but that can be added):
POPULATE:
CREATE TABLE #quotes
(
id int identity(1,1)
,security_id VARCHAR(50)
,last_price FLOAT
,bid_price FLOAT
,ask_price FLOAT
,[date] DATETIME
,[timestamp] DATETIME
,trade_volume FLOAT
,cumulative_volume FLOAT
,average_volume FLOAT
,created_at DATETIME
)
DECLARE #i int
set #i = 100000
WHILE #i > 0
BEGIN
INSERT INTO #quotes (
security_id
,last_price
,bid_price
,ask_price
,[date]
,[timestamp]
,trade_volume
,cumulative_volume
,average_volume
,created_at
)
values( 'IBM US'
, 100.00 + RAND()
, 100.00 + RAND()
, 100.00 + RAND()
, DATEADD(MINUTE, -1* #i, GETDATE())
, DATEADD(MINUTE, -1* #i, GETDATE())
, 10000000.00 + RAND()*1000000.00
, 10000000.00 + RAND()*1000000.00
, 10000000.00 + RAND()*1000000.00
,getdate())
set #i= #i-1
END
You can change around the time span, but the following will give you around 1000 records that represent the set from start to finish.
DECLARE #StartDate DATETIME,
#EndDate DATETIME,
#j FLOAT,
#step FLOAT
set #StartDate = GETDATE()-20
SET #EndDAte = GETDATE()
set #j = 0.0
CREATE TABLE #TimeTable
(
IntervalDate DATETIME
)
--say you always want 1000 measures
--use the datediff value to define the step size:
select #step = DATEDIFF(MINUTE, #StartDate, #EndDate)/1000.0
WHILE #j < DATEDIFF(MINUTE, #StartDate, #EndDate)
BEGIN
INSERT #TimeTable (IntervalDate) VALUES (DATEADD(minute, #j, #StartDate))
SET #j = #j+#step
print #j
END
select security_id
,last_price
,bid_price
,ask_price
,[date]
,[timestamp]
,trade_volume
,cumulative_volume
,average_volume
,created_at
from #Quotes q
join #TimeTable t on dateadd(mi, datediff(mi, 0, q.date), 0) = dateadd(mi, datediff(mi, 0, t.IntervalDate), 0)

Related

Decrease the execution time of the query

I write a query the result of the query is fine but issue with the time that is taking to complete the execution. For increase the performance i create index as per my query so this decrease from 5 minutes to 0.44 seconds. But i want to make this more faster. The issue i release with my function that i comment then its execution time is 0.00 but the function i am using is important for me. Is any option to decrease my execution time of my query .
**My function*
ALTER FUNCTION [dbo].[fnCount]
(
#StartDate DATE,#EndDate DATE,#DropDate DATE
)
RETURNS INT
AS
BEGIN
DECLARE #WeekStartDate DATE
DECLARE #WeekEndDate DATE
DECLARE #Count AS INT
SET #Count=0
SET #WeekStartDate=CONVERT(DATE,DATEADD(dd, -(DATEPART(dw, #DropDate)-1), #DropDate))
SET #WeekEndDate=CONVERT(DATE,DATEADD(dd, 7-(DATEPART(dw, #DropDate)), #DropDate))
IF #StartDate <= #WeekStartDate
SET #StartDate=#WeekStartDate
IF #EndDate >=#WeekEndDate
SET #EndDate=#WeekEndDate
SELECT #Count=COUNT(distinct DROPDATE)
FROM orde_
WHERE orde_.CANCELLED = 0
AND DROPDATE >= #StartDate AND DROPDATE <= #EndDate
AND (DISPID IS NULL OR DISPID NOT IN ( '291', '327') )
RETURN #Count
END
Query
SELECT READYDATE,TOTAL,DRV1PAY,ORDERNUM,
CASE WHEN dbo.fnCount('20170901','20171031',DROPDATE) >=4
THEN (CONVERT(VARCHAR(4),LEFT(DROPDATE,4)) + '' + CONVERT(VARCHAR(4),DATEPART( wk, DROPDATE)))
ELSE NULL
END AS WEEKOfYear1,
FROM orde_
WHERE CANCELLED = 0 AND DROPDATE >='20170901' AND DROPDATE <='20171031'
AND (DISPID IS NULL OR DISPID NOT IN ( '291', '327') )
ORDER BY DISPID
Thanks for your help and comments

convert int to date to add a dayn and convert to int back

I have int value with YYYYMM. I want to:
1. convert it into datetime
2. add one day DATEADD(Day, +1, #date)
3. convert it back into int
What's the easiest way to do this?
Here is a nice exercise and I hope it works out for you...
declare #date date
declare #newDate date
set #date = convert(date, '20100101')
set #newdate = DateAdd(dd, 1, #Date)
select #date
select #newdate
select convert(int, convert(varchar, #newdate, 112)) -- this is your final conversion back to int
If your initial int is, say, 201310 (October 2013), then what I think you want is this:
select convert(datetime, rtrim(201310 * 100 + 1))
The function RTRIM is a trick to convert int to string type.
The result is this:
2013-10-01 00:00:00.000
If you don't want to use RTRIM, the command below will get you the same result:
select convert(datetime, convert(char, 201310 * 100 + 1))

SQL Server - calculate elapsed time between two datetime stamps in HH:MM:SS format

I have a SQL Server table that has a "Time" column. The table is a log table the houses status messages and timestamps for each message. The log table is inserted into via a batch file. There is an ID column that groups rows together. Each time the batch file runs it initializes the ID and writes records. What I need to do is get the elapsed time from the first record in an ID set to the last record of the same ID set. I started toying with select Max(Time) - Min(Time) from logTable where id = but couldn't figure out how to format it correctly. I need it in HH:MM:SS.
SQL Server doesn't support the SQL standard interval data type. Your best bet is to calculate the difference in seconds, and use a function to format the result. The native function CONVERT() might appear to work fine as long as your interval is less than 24 hours. But CONVERT() isn't a good solution for this.
create table test (
id integer not null,
ts datetime not null
);
insert into test values (1, '2012-01-01 08:00');
insert into test values (1, '2012-01-01 09:00');
insert into test values (1, '2012-01-01 08:30');
insert into test values (2, '2012-01-01 08:30');
insert into test values (2, '2012-01-01 10:30');
insert into test values (2, '2012-01-01 09:00');
insert into test values (3, '2012-01-01 09:00');
insert into test values (3, '2012-01-02 12:00');
Values were chosen in such a way that for
id = 1, elapsed time is 1 hour
id = 2, elapsed time is 2 hours, and
id = 3, elapsed time is 3 hours.
This SELECT statement includes one column that calculates seconds, and one that uses CONVERT() with subtraction.
select t.id,
min(ts) start_time,
max(ts) end_time,
datediff(second, min(ts),max(ts)) elapsed_sec,
convert(varchar, max(ts) - min(ts), 108) do_not_use
from test t
group by t.id;
ID START_TIME END_TIME ELAPSED_SEC DO_NOT_USE
1 January, 01 2012 08:00:00 January, 01 2012 09:00:00 3600 01:00:00
2 January, 01 2012 08:30:00 January, 01 2012 10:30:00 7200 02:00:00
3 January, 01 2012 09:00:00 January, 02 2012 12:00:00 97200 03:00:00
Note the misleading "03:00:00" for the 27-hour difference on id number 3.
Function to format elapsed time in SQL Server
UPDATED:
Correctly calculate a timespan in SQL Server, even if more than 24 hours:
-- Setup test data
declare #minDate datetime = '2012-12-12 20:16:47.160'
declare #maxDate datetime = '2012-12-13 15:10:12.050'
-- Get timespan in hh:mi:ss
select cast(
(cast(cast(#maxDate as float) - cast(#minDate as float) as int) * 24) /* hours over 24 */
+ datepart(hh, #maxDate - #minDate) /* hours */
as varchar(10))
+ ':' + right('0' + cast(datepart(mi, #maxDate - #minDate) as varchar(2)), 2) /* minutes */
+ ':' + right('0' + cast(datepart(ss, #maxDate - #minDate) as varchar(2)), 2) /* seconds */
-- Returns 18:53:24
Edge cases that show inaccuracy are especially welcome!
DECLARE #EndTime AS DATETIME, #StartTime AS DATETIME
SELECT #StartTime = '2013-03-08 08:00:00', #EndTime = '2013-03-08 08:30:00'
SELECT CAST(#EndTime - #StartTime AS TIME)
Result: 00:30:00.0000000
Format result as you see fit.
The best and simple way:
Convert(varchar, {EndTime} - {StartTime}, 108)
Just like Anri noted.
Use the DATEDIFF to return value in milliseconds, seconds, minutes, hours, ...
DATEDIFF(interval, date1, date2)
interval REQUIRED - The time/date part to return. Can be one of the following values:
year, yyyy, yy = Year
quarter, qq, q = Quarter
month, mm, m = month
dayofyear = Day of the year
day, dy, y = Day
week, ww, wk = Week
weekday, dw, w = Weekday
hour, hh = hour
minute, mi, n = Minute
second, ss, s = Second
millisecond, ms = Millisecond
date1, date2 REQUIRED - The two dates to calculate the difference between
select convert(varchar, Max(Time) - Min(Time) , 108) from logTable where id=...
See if this helps. I can set variables for Elapsed Days, Hours, Minutes, Seconds.
You can format this to your liking or include in a user defined function.
Note: Don't use DateDiff(hh,#Date1,#Date2). It is not reliable! It rounds in unpredictable ways
Given two dates...
(Sample Dates: two days, three hours, 10 minutes, 30 seconds difference)
declare #Date1 datetime = '2013-03-08 08:00:00'
declare #Date2 datetime = '2013-03-10 11:10:30'
declare #Days decimal
declare #Hours decimal
declare #Minutes decimal
declare #Seconds decimal
select #Days = DATEDIFF(ss,#Date1,#Date2)/60/60/24 --Days
declare #RemainderDate as datetime = #Date2 - #Days
select #Hours = datediff(ss, #Date1, #RemainderDate)/60/60 --Hours
set #RemainderDate = #RemainderDate - (#Hours/24.0)
select #Minutes = datediff(ss, #Date1, #RemainderDate)/60 --Minutes
set #RemainderDate = #RemainderDate - (#Minutes/24.0/60)
select #Seconds = DATEDIFF(SS, #Date1, #RemainderDate)
select #Days as ElapsedDays, #Hours as ElapsedHours, #Minutes as ElapsedMinutes, #Seconds as ElapsedSeconds
Hope this helps you in getting the exact time between two time stamps
Create PROC TimeDurationbetween2times(#iTime as time,#oTime as time)
As
Begin
DECLARE #Dh int, #Dm int, #Ds int ,#Im int, #Om int, #Is int,#Os int
SET #Im=DATEPART(MI,#iTime)
SET #Om=DATEPART(MI,#oTime)
SET #Is=DATEPART(SS,#iTime)
SET #Os=DATEPART(SS,#oTime)
SET #Dh=DATEDIFF(hh,#iTime,#oTime)
SET #Dm = DATEDIFF(mi,#iTime,#oTime)
SET #Ds = DATEDIFF(ss,#iTime,#oTime)
DECLARE #HH as int, #MI as int, #SS as int
if(#Im>#Om)
begin
SET #Dh=#Dh-1
end
if(#Is>#Os)
begin
SET #Dm=#Dm-1
end
SET #HH = #Dh
SET #MI = #Dm-(60*#HH)
SET #SS = #Ds-(60*#Dm)
DECLARE #hrsWkd as varchar(8)
SET #hrsWkd = cast(#HH as char(2))+':'+cast(#MI as char(2))+':'+cast(#SS as char(2))
select #hrsWkd as TimeDuration
End

Rounding datetime to quarter minutes

I have a function to round a datetime to the nearest quarter hour.
But is there a method to round down to the nearest quarter instead?
Example.
08:14:00 becomes 08:00:00
08:03:00 becomes 08:00:00
08:29:00 becomes 08:15:00
08:55:00 becomes 08:45:00
This is what I have now to round to the nearest quarter.
(
#dt datetime
)
returns datetime
as
begin
declare #result datetime
declare #mm int
set #mm=datepart(minute,#dt)
set #result = dateadd(minute,-#mm + (round(#mm/cast(15 as float),0)*15) , #dt )
return #result
Using SQL Server:
select cast(
FLOOR( cast( GetDate() as float)*(24*4)) / (24*4)
as smalldatetime) AS "datetime_quarter"
The strategy is:
Convert the date to a float number, *24 to get number of hours, *4 to get number of quarters
Round down with FLOOR
Convert back to number of days by /(24*4)
Convert number of days to a datetime. smalldatetime is used to avoid float rounding issues.
This can easily be adjusted to use ROUNDor CEILING instead; or to use other hour multiples instead of 4 (quarters).
This is a procedure to get the current quarter upper and lower bound.
declare QUARTER_FLOOR datetime;
declare QUARTER_CEIL datetime;
declare CURRENT_DATETIME datetime;
declare CURRENT_MINUTE int;
// get current datetime without second and millisecond
SET CURRENT_DATETIME = dateformat(getdate(), 'YYYY-MM-DD hh:NN:00.000');
set CURRENT_MINUTE = datepart(minute,CURRENT_DATETIME); // get current minute
// GET CURRENT QUARTER FLOOR
set QUARTER_FLOOR = dateadd(minute, -CURRENT_MINUTE + (round(CURRENT_MINUTE/cast(15 as float),0)*15) , CURRENT_DATETIME );
// GET CURRENT QUARTER FLOOR
set QUARTER_CEIL = dateadd(minute, ((round(CURRENT_MINUTE/cast(15 as float),0)+1)*15) - CURRENT_MINUTE , CURRENT_DATETIME );
// RETURN
select QUARTER_FLOOR, QUARTER_CEIL;
This method relies on integer division when we divide the month by 4. In principle it should be portable, subject of course to modifying the syntax of dateadd() / date_add() and month()/year().
SELECT DATEADD(
MONTH,
3 * (
MONTH( my_date ) / 4
),
CONCAT(
YEAR( my_date ),
'0101' )
) AS my_date_rounded_to_quarter

Create a date range in mysql

Best way to create on the fly, date ranges, for use with report.
So I can avoid empty rows on my report if there's no activity for a given day.
Mostly to avoid this issue: What is the most straightforward way to pad empty dates in sql results (on either mysql or perl end)?
My advice is: don't make your life harder, make it easier. Just create a table with one row for each calendar day, having as many rows as you think you reasonably need to last. In datawarehousing, this is the common solution, and it is so widely implemented this way that a dwh that doesn't have it, has a code smell.
Many people used to dealing with more traditional oltp/data entry apps feel a natural revulsion against this idea, because the feel the can generate the data anyway, and therefore it shouldn't be stored. But if you do create a table like that, you can adorn it with many useful attributes, such as whether it's a holdiday or a weekend, and you can store many common date representations (iso, european, us format etc) inside it, which can save you a ton of time when creating reports (since you don't have to bother figuring out how the date formatting works in each reporting tool you come by. Or you can go a step further and update your date table everyday to mark flags for the current day, current week, current month, current year, etc - all kinds of useful tools that make it much, much easier to build reports that need to work against some date range.
MySQL sample code as per request in comment:
delimiter //
DROP PROCEDURE IF EXISTS p_load_dim_date
//
CREATE PROCEDURE p_load_dim_date (
p_from_date DATE
, p_to_date DATE
)
BEGIN
DECLARE v_date DATE DEFAULT p_from_date;
DECLARE v_month tinyint;
CREATE TABLE IF NOT EXISTS dim_date (
date_key int primary key
, date_value date
, date_iso char(10)
, year smallint
, quarter tinyint
, quarter_name char(2)
, month tinyint
, month_name varchar(10)
, month_abbreviation varchar(10)
, week char(2)
, day_of_month tinyint
, day_of_year smallint
, day_of_week smallint
, day_name varchar(10)
, day_abbreviation varchar(10)
, is_weekend tinyint
, is_weekday tinyint
, is_today tinyint
, is_yesterday tinyint
, is_this_week tinyint
, is_last_week tinyint
, is_this_month tinyint
, is_last_month tinyint
, is_this_year tinyint
, is_last_year tinyint
);
WHILE v_date < p_to_date DO
SET v_month := month(v_date);
INSERT INTO dim_date(
date_key
, date_value
, date_iso
, year
, quarter
, quarter_name
, month
, month_name
, month_abbreviation
, week
, day_of_month
, day_of_year
, day_of_week
, day_name
, day_abbreviation
, is_weekend
, is_weekday
) VALUES (
v_date + 0
, v_date
, DATE_FORMAT(v_date, '%y-%c-%d')
, year(v_date)
, ((v_month - 1) DIV 3) + 1
, CONCAT('Q', ((v_month - 1) DIV 3) + 1)
, v_month
, DATE_FORMAT(v_date, '%M')
, DATE_FORMAT(v_date, '%b')
, DATE_FORMAT(v_date, '%u')
, DATE_FORMAT(v_date, '%d')
, DATE_FORMAT(v_date, '%j')
, DATE_FORMAT(v_date, '%w') + 1
, DATE_FORMAT(v_date, '%W')
, DATE_FORMAT(v_date, '%a')
, IF(DATE_FORMAT(v_date, '%w') IN (0,6), 1, 0)
, IF(DATE_FORMAT(v_date, '%w') IN (0,6), 0, 1)
);
SET v_date := v_date + INTERVAL 1 DAY;
END WHILE;
CALL p_update_dim_date();
END;
//
DROP PROCEDURE IF EXISTS p_update_dim_date;
//
CREATE PROCEDURE p_update_dim_date()
UPDATE dim_date
SET is_today = IF(date_value = current_date, 1, 0)
, is_yesterday = IF(date_value = current_date - INTERVAL 1 DAY, 1, 0)
, is_this_week = IF(year = year(current_date) AND week = DATE_FORMAT(current_date, '%u'), 1, 0)
, is_last_week = IF(year = year(current_date - INTERVAL 7 DAY) AND week = DATE_FORMAT(current_date - INTERVAL 7 DAY, '%u'), 1, 0)
, is_this_month = IF(year = year(current_date) AND month = month(current_date), 1, 0)
, is_last_month = IF(year = year(current_date - INTERVAL 1 MONTH) AND month = month(current_date - INTERVAL 1 MONTH), 1, 0)
, is_this_year = IF(year = year(current_date), 1, 0)
, is_last_year = IF(year = year(current_date - INTERVAL 1 YEAR), 1, 0)
WHERE is_today
OR is_yesterday
OR is_this_week
OR is_last_week
OR is_this_month
OR is_last_month
OR is_this_year
OR is_last_year
OR IF(date_value = current_date, 1, 0)
OR IF(date_value = current_date - INTERVAL 1 DAY, 1, 0)
OR IF(year = year(current_date) AND week = DATE_FORMAT(current_date, '%u'), 1, 0)
OR IF(year = year(current_date - INTERVAL 7 DAY) AND week = DATE_FORMAT(current_date - INTERVAL 7 DAY, '%u'), 1, 0)
OR IF(year = year(current_date) AND month = month(current_date), 1, 0)
OR IF(year = year(current_date - INTERVAL 1 MONTH) AND month = month(current_date - INTERVAL 1 MONTH), 1, 0)
OR IF(year = year(current_date), 1, 0)
OR IF(year = year(current_date - INTERVAL 1 YEAR), 1, 0)
;
//
delimiter ;
Using p_load_dim_date you uinitially load the dim_date table with say 25 years of data. And daily, prefereabluy round midnight, you run p_update_dim_date. Then you can use the flag fields is_today, is_yesterday, is_this_week, is_last_week and so on to select common ranges. Of course, you should amend this code to suit your particular needs but this is the idea. So no generaging ranges on the fly, you just preload for a long enough period of time ahead. For the time of day, a similar design can be set up - you should be able to manage that yourself going by this code.
For even fancier date dimensions that take care of holidays, and localized names for month and days, you can take a look at:
http://rpbouman.blogspot.com/2007/04/kettle-tip-using-java-locales-for-date.html
and
http://rpbouman.blogspot.com/2010/01/easter-eggs-for-mysql-and-kettle.html
I've recently done some research to find and evaluate possible options. http://www.freeportmetrics.com/devblog/2012/11/02/how-to-quickly-add-date-dimension-to-pentaho-mondrian-olap-cube/.
You can use:
kettle
degenerated dimensions
lucidb build-in function
up-coming Mondrian built-in function
your own custom script to generate SQL
mysql script mentioned earlier
Please check the blog post for more details. It also contains improved version of Roland's sql script that will automatically calculate date range for given column and join it with date dimension.
There is no straightforward way to do that in MySQL. Your best bet is to generate a daterange array in your server-side language of choice, and then pull data from the database and merge the resulting array with your daterange array using the date as a key.
Which server side language are you using?
Edit:
Basically what you would do is (pseudocode):
// Create an array with all dates for a given range
dates = makeRange(startDate, endDate);
getData = mysqlQuery('SELECT date, x, y, z FROM a WHERE a AND b AND c');
while (r = fetchRowArray(getData)) {
dates[ date(r['date']) ] = Array ( x, y, z);
}
You end up with an array of dates you can loop through, with the dates that have or don't have activity data associated to them.
Can easily be modified to group / filter data by hours.
Try using a loop in a MySQL stored routine to create date ranges:
declare iterDate date;
set iterDate = startDate;
DROP TABLE IF EXISTS MyDates;
create temporary table MyDates (
theDate date
);
label1: LOOP
insert into MyDates(theDate) values (iterDate);
SET iterDate = DATE_ADD(iterDate, INTERVAL 1 DAY);
IF iterDate <= endDate THEN
ITERATE label1;
END IF;
LEAVE label1;
END LOOP label1;
select * from MyDates;
DROP TABLE IF EXISTS MyDates;
startDate and endDate constitute the endpoints of the range and are supplied as parameters to the routine.
I realise this is an old post but, to keep Stack Overflow a bit up-to-date, I feel the urge to respond.
With the new SEQUENCE engine in MariaDB, this is possible within a SELECT statement without any stored routine or temporary table:
SELECT
DATE_ADD(
CAST('2022-06-01' AS DATE),
INTERVAL `s1`.`seq` DAY
) AS `dates`
FROM `seq_0_to_364` AS `s1`;
Any interval will work as long as it is within the limits of BIGINT(20) UNSIGNED, as this is the limit of the SEQUENCE engine.