![](https://secure.gravatar.com/avatar/68ff00c1752bf8aede68a05f77ebd28b.jpg?s=120&d=mm&r=g)
Hi Dan, On 09/24/2015 12:35 AM, Dan Ragle wrote:
On 9/22/2015 5:12 PM, Alexander Barkov wrote:
On 09/23/2015 01:00 AM, Alexander Barkov wrote:
Hi Dan,
On 09/20/2015 12:53 AM, Dan Ragle wrote:
We're converting from MySQL 5.1 to MariaDB 5.5.44 and I'm hoping someone can help clarify for me what's going on in the following scenario. In brief, we're getting milliseconds displayed in a particular date_sub calculation (and possibly other selects with date arithmetic, trying to figure out how to track them down is part of the goal of this post) where we aren't expecting them.
MariaDB [testdb]> show variables like '%version%'; +-------------------------+---------------------+ | Variable_name | Value | +-------------------------+---------------------+ | innodb_version | 5.5.43-MariaDB-37.2 | | protocol_version | 10 | | slave_type_conversions | | | version | 5.5.44-MariaDB-log | | version_comment | MariaDB Server | | version_compile_machine | x86_64 | | version_compile_os | Linux | +-------------------------+---------------------+ 7 rows in set (0.00 sec)
MariaDB [testdb]> show create table test \G *************************** 1. row *************************** Table: test Create Table: CREATE TABLE `test` ( `datetime` datetime NOT NULL ) ENGINE=MyISAM DEFAULT CHARSET=latin1 1 row in set (0.00 sec)
MariaDB [testdb]> select datetime from test; Field 1: `datetime` Catalog: `def` Database: `velocitynorep` Table: `test` Org_table: `test` Type: DATETIME Collation: binary (63) Length: 19 Max_length: 19 Decimals: 0 Flags: NOT_NULL BINARY NO_DEFAULT_VALUE
+---------------------+ | datetime | +---------------------+ | 2015-09-18 09:23:21 | +---------------------+ 1 row in set (0.00 sec)
MariaDB [testdb]> select date_sub(datetime,interval 1 second) from test; Field 1: `date_sub(datetime,interval 1 second)` Catalog: `def` Database: `` Table: `` Org_table: `` Type: DATETIME Collation: binary (63) Length: 19 Max_length: 19 Decimals: 0 Flags: BINARY
+--------------------------------------+ | date_sub(datetime,interval 1 second) | +--------------------------------------+ | 2015-09-18 09:23:20 | +--------------------------------------+ 1 row in set (0.00 sec)
MariaDB [testdb]> select substr(datetime,1,13) from test; Field 1: `substr(datetime,1,13)` Catalog: `def` Database: `` Table: `` Org_table: `` Type: VAR_STRING Collation: utf8_general_ci (33) Length: 39 Max_length: 13 Decimals: 31 Flags: NOT_NULL
+-----------------------+ | substr(datetime,1,13) | +-----------------------+ | 2015-09-18 09 | +-----------------------+ 1 row in set (0.00 sec)
MariaDB [testdb]> select concat(substr(datetime,1,13),':00:00') -> from test; Field 1: `concat(substr(datetime,1,13),':00:00')` Catalog: `def` Database: `` Table: `` Org_table: `` Type: VAR_STRING Collation: utf8_general_ci (33) Length: 57 Max_length: 19 Decimals: 31 Flags: NOT_NULL
+----------------------------------------+ | concat(substr(datetime,1,13),':00:00') | +----------------------------------------+ | 2015-09-18 09:00:00 | +----------------------------------------+ 1 row in set (0.00 sec)
MariaDB [testdb]> select concat('2015-09-18 09',':00:00'); Field 1: `concat('2015-09-18 09',':00:00')` Catalog: `def` Database: `` Table: `` Org_table: `` Type: VAR_STRING Collation: utf8_general_ci (33) Length: 57 Max_length: 19 Decimals: 31 Flags: NOT_NULL
+----------------------------------+ | concat('2015-09-18 09',':00:00') | +----------------------------------+ | 2015-09-18 09:00:00 | +----------------------------------+ 1 row in set (0.00 sec)
MariaDB [testdb]> select date_sub(concat('2015-09-18 09',':00:00'), -> interval 1 second); Field 1: `date_sub(concat('2015-09-18 09',':00:00'), interval 1 second)` Catalog: `def` Database: `` Table: `` Org_table: `` Type: STRING Collation: binary (63) Length: 19 Max_length: 19 Decimals: 0 Flags: BINARY
+---------------------------------------------------------------+ | date_sub(concat('2015-09-18 09',':00:00'), interval 1 second) | +---------------------------------------------------------------+ | 2015-09-18 08:59:59 | +---------------------------------------------------------------+ 1 row in set (0.00 sec)
MariaDB [testdb]> select -> date_sub(concat(substr(datetime,1,13),':00:00'), -> interval 1 second) from test; Field 1: `date_sub(concat(substr(datetime,1,13),':00:00'), interval 1 second)` Catalog: `def` Database: `` Table: `` Org_table: `` Type: STRING Collation: binary (63) Length: 26 Max_length: 26 Decimals: 6 Flags: BINARY
+---------------------------------------------------------------------+ | date_sub(concat(substr(datetime,1,13),':00:00'), interval 1 second) | +---------------------------------------------------------------------+ | 2015-09-18 08:59:59.000000 | +---------------------------------------------------------------------+ 1 row in set (0.00 sec)
The last one is the one that has me stumped.
1.) Why does it include milliseconds in the output? According to this page:
https://mariadb.com/kb/en/mariadb/microseconds-in-mariadb/
where no precision is specified the result should not display milliseconds, which is what happens in all the other examples above except this one.
2.) The statement on a 5.1 MySQL server (5.1.73-log, also 64 bit Linux) produces this:
mysql> select date_sub(concat(substr(datetime,1,13),':00:00'), -> interval 1 second) from test; Field 1: `date_sub(concat(substr(datetime,1,13),':00:00'), interval 1 second)` Catalog: `def` Database: `` Table: `` Org_table: `` Type: STRING Collation: binary (63) Length: 29 Max_length: 19 Decimals: 31 Flags: BINARY
+---------------------------------------------------------------------+ | date_sub(concat(substr(datetime,1,13),':00:00'), interval 1 second) | +---------------------------------------------------------------------+ | 2015-09-18 08:59:59 | +---------------------------------------------------------------------+ 1 row in set (0.00 sec)
Is there some global variable I can set to revert to the 5.1 behavior? I realize I can CAST the result to a datetime to make sure I don't get the fractional seconds, but as I mentioned we're converting our system over to 5.5 and there's boatloads of code that I would have to scan through to try and find these differences.
3.) If there is no way to revert to the 5.1 behavior, can someone help me to understand the exact reason why the milliseconds get displayed on the final select in MariaDB 5.5? I'll need to look for any similar types of results and other than the exact reduction above I'm not sure what statements I would need to identify and test.
Thanks for any assistance!
DATE_ADD() and DATE_SUB() are hybrid type functions. They can return the following data types: - DATETIME - DATE - TIME - VARCHAR
depending on its argument data types.
In case when the first argument is not of a temporal type, the result data type is VARCHAR, the result fractional precision is detected from the argument, and the exact output format is detected per row! It can return DATE-alike and DATETIME-alike strings in the same result set. This query demonstrates this hybrid nature of DATE_ADD/DATE_SUB in case of a VARCHAR result:
SELECT DATE_ADD(a, INTERVAL 1 DAY) FROM (SELECT '2001-01-01' AS a UNION SELECT '2001-01-01 10:20:30') t1;
+-----------------------------+ | DATE_ADD(a, INTERVAL 1 DAY) | +-----------------------------+ | 2001-01-02 | | 2001-01-02 10:20:30.000000 | +-----------------------------+
Now returning to your examples:
In case when the first argument is a VARCHAR constant, like in this query:
select date_sub(concat('2015-09-18 09',':00:00'), interval 1 second);
the result data type is VARCHAR, and its fractional precision is calculated before the actual execution of the query: - the first argument is evaluated to '2015-09-18 09:00:00' - the constant is parsed and the number of fractional digits is detected to be 0 - the query is executed, returning no fractional digits
This is why the query returns '2015-09-18 08:59:59'
In case when the first argument is a VARCHAR non-constant, like in this query:
select date_sub(concat(substr(datetime,1,13),':00:00'), interval 1 second) from test;
it cannot know in advance how the input string data will look like, so it sets the maximum possible precision, which is 6.
This is why you get this result:
'2015-09-18 08:59:59.000000'
Notice, the result type in both cases is VARCHAR!
A small correction:
- VARCHAR in MariaDB-10.0 and higher
- VARBINARY in earlier versions
As a workaround, I can suggest this query:
SELECT DATE_SUB(TRUNCATE(datetime,-4), INTERVAL 1 SECOND) FROM test;
It does effectively the same thing with substr followed by concat, but it returns '2015-09-18 08:59:59' with no fractional digits.
The two queries work differently because at the prepare stage of a query execution:
- The number of fractional digits in the string which is a result of CONCAT(SUBSTR(...)) is NOT known in advance. So it reserves the maximum possible fractional precision of 6.
- The number of fractional digits in the numeric value which is a result of TRUNCATE(...,-4) is perfectly known in advance to be 0 for any possible input, so DATE_SUB() returns no fractional digits.
Notice, the result of:
SELECT DATE_SUB(TRUNCATE(datetime,-4), INTERVAL 1 SECOND) FROM test;
is still VARCHAR, because the first argument is not of a temporal data type: it's a number!
So perhaps you might want to do even this:
SELECT DATE_SUB(CAST(TRUNCATE(datetime,-4) AS DATETIME), INTERVAL 1 SECOND) FROM test;
It returns '2015-09-18 08:59:59', and the result type is DATETIME.
Hope this helps.
Yes, the discussion is helpful, thank you.
So to sum up (and probably oversimplify, but enough so that I know generally what I should be looking for) any function that potentially returns a datetime and relies on a varchar-non constant -- i.e., something the value of which isn't known until the query is actually executed -- will, when returning a datetime value, present the result with 6 digits of precision.
There are 3 functions that can decide per row: 1. DATE_ADD() and its synonyms and variations, like DATE_SUB(), ADDDATE(), etc. 2. ADDTIME() Example:
MariaDB [test]> SELECT ADDTIME(a,1) FROM (SELECT '10:20:30' AS a UNION SELECT '2001-01-01 10:20:30' UNION SELECT '10:20:30.123456') t1; +---------------------+ | ADDTIME(a,1) | +---------------------+ | 10:20:31 | | 2001-01-01 10:20:31 | | 10:20:31.123456 | +---------------------+ 3 rows in set (0.00 sec)
ADDTIME() works as follows: - If first arg is a MYSQL_TYPE_DATETIME or MYSQL_TYPE_TIMESTAMP result is MYSQL_TYPE_DATETIME - If first arg is a MYSQL_TYPE_TIME result is MYSQL_TYPE_TIME - Otherwise the result is MYSQL_TYPE_STRING, and it decides per row 3. str_to_date(). All other functions that can return a temporal data type choose the exact data type and temporal precision before the query execution (not per row). This includes COALESCE(), CASE, NULLIF(), IF(), etc., they also choose before the query execution.
So presumably something like this:
select date_sub(concat(substr(now(),1,13),':00:00'), -> interval 1 second) from test;
returns no fractional seconds because now() is executed and the result known before the query itself is actually launched.
Right, this is exactly what happens. I just traced it in the debugger to make sure.
And in my original case, even though datetime is defined in my test table as a datetime with no (zero) precision, because the concatenation doesn't happen until after the query is generated (on a row-by-row basis) the precision of any returned datetime has to be set to the max possible value.
Yes? No?
Correct. It cannot go inside CONCAT(SUBSTR()) and see that: - the original data source is actually DATETIME(0) - the concatenated part also cannot introduce fractional digits DATE_SUB() just sees that the argument is a VARCHAR non-constant, so it switches to the "per-row" mode. To summarize: - DATE_SUB() and ADDTIME() guarantee a predictable output data type and a predictable output fractional precision if the input is of a temporal data type - they can't guarantee the exact output format and fractional precision case of a non-temporal input This is where CAST() is very helpful, e.g.: DATE_SUB(CAST(TRUNCATE(...) AS DATETIME)...) DATE_SUB(CAST(CONCAT(SUBSTR(...) AS DATETIME)...)
Dan Ragle
_______________________________________________ Mailing list: https://launchpad.net/~maria-discuss Post to : maria-discuss@lists.launchpad.net Unsubscribe : https://launchpad.net/~maria-discuss More help : https://help.launchpad.net/ListHelp
_______________________________________________ Mailing list: https://launchpad.net/~maria-discuss Post to : maria-discuss@lists.launchpad.net Unsubscribe : https://launchpad.net/~maria-discuss More help : https://help.launchpad.net/ListHelp