In DB2, I need to do a SELECT FROM UPDATE, to put an update + select in a single transaction.
But I need to make sure to update only one record per transaction.
Familiar with the LIMIT clause from MySQL's UPDATE option
places a limit on the number of rows that can be updated
I looked for something similar in DB2's UPDATE reference but without success.
How can something similar be achieved in DB2?
Edit: In my scenario, I have to deliver 1000 coupon codes upon request. I just need to select (any)one that has not been given yet.
The question uses some ambiguous terminology that makes it unclear what needs to be accomplished. Fortunately, DB2 offers robust support for a variety of SQL patterns.
To limit the number of rows that are modified by an UPDATE:
UPDATE
( SELECT t.column1 FROM someschema.sometable t WHERE ... FETCH FIRST ROW ONLY
)
SET column1 = 'newvalue';
The UPDATE statement never sees the base table, just the expression that filters it, so you can control which rows are updated.
To INSERT a limited number of new rows:
INSERT INTO mktg.offeredcoupons( cust_id, coupon_id, offered_on, expires_on )
SELECT c.cust_id, 1234, CURRENT TIMESTAMP, CURRENT TIMESTAMP + 30 DAYS
FROM mktg.customers c
LEFT OUTER JOIN mktg.offered_coupons o
ON o.cust_id = c.cust_id
WHERE ....
AND o.cust_id IS NULL
FETCH FIRST 1000 ROWS ONLY;
This is how DB2 supports SELECT from an UPDATE, INSERT, or DELETE statement:
SELECT column1 FROM NEW TABLE (
UPDATE ( SELECT column1 FROM someschema.sometable
WHERE ... FETCH FIRST ROW ONLY
)
SET column1 = 'newvalue'
) AS x;
The SELECT will return data from only the modified rows.
You have two options. As noted by A Horse With No Name, you can use the primary key of the table to ensure that one row is updated at a time.
The alternative, if you're using a programming language and have control over cursors, is to use a cursor with the 'FOR UPDATE' option (though that may be probably optional; IIRC, cursors are 'FOR UPDATE' by default when the underlying SELECT means it can be), and then use an UPDATE statement with the WHERE CURRENT OF <cursor-name> in the UPDATE statement. This will update the one row currently addressed by the cursor. The details of the syntax vary with the language you're using, but the raw SQL looks like:
DECLARE CURSOR cursor_name FOR
SELECT *
FROM SomeTable
WHERE PKCol1 = ? AND PKCol2 = ?
FOR UPDATE;
UPDATE SomeTable
SET ...
WHERE CURRENT OF cursor_name;
If you can't write DECLARE in your host language, you have to do manual bashing to find the equivalent mechanism.
Related
Is there simple way to select updated rows?
I'm trying to store timestamp each time I am read row to be able to delete data that was not readed for a long time.
First I tried execute SELECT query first and even found little bit slow but simple solution like
UPDATE foo AS t, (SELECT id FROM foo WHERE statement=1)q
SET t.time=NOW() WHERE t.id=q.id
but I still want to find a normal way to do this.
I also think that updating time first and then just select updated rows should be much easier, but I didn't find anything even for this
For a single-row UPDATE in MySQL you could:
UPDATE foo
SET time = NOW()
WHERE statement = 1
AND #var := id
#var := id is always TRUE, but it writes the value of id to the variable #var before the update. Then you can:
SELECT #var;
In PostgreSQL you could use the RETURNING clause.
Oracle also has a RETURNING clause.
SQL-Server has an OUTPUT clause.
But MySQL doesn't have anything like that.
Declare the time column as follows:
CREATE TABLE foo (
...
time TIMESTAMP ON UPDATE CURRENT_TIMESTAMP(),
...)
Then whenever a row is updated, the column will be updated automatically.
UPDATE:
I don't think there's a way to update automatically during SELECT, so you have to do it in two steps:
UPDATE foo
SET time = NOW()
WHERE <conditions>;
SELECT <columns>
FROM foo
WHERE <conditions>;
As long as doesn't include the time column I think this should work. For maximum safety you'll need to use a transaction to prevent other queries from interfering.
#Erwin Brandstetter: Not difficult to extend the strategy of using user variables with CONCAT_WS() to get back multiple IDs. Sorry, still can't add comments...
As suggested here you can extract the modified primary keys to update their timestamp column afterwards.
SET #uids := null;
UPDATE footable
SET foo = 'bar'
WHERE fooid > 5
AND ( SELECT #uids := CONCAT_WS(',', fooid, #uids) );
SELECT #uids;
from https://gist.github.com/PieterScheffers/189cad9510d304118c33135965e9cddb
So you should replace the final SELECT #uids; with an update statement by splitting the resulting #uids value (it will be a varchar containing all the modified ids divided by ,).
I have a query written very poorly in SQL Server 2008
UPDATE PatientChartImages
SET PatientChartImages.IsLockDown = #IsLockdown
WHERE PatientChartImages.IsLockDown = #IsNotLockdown
AND PatientChartId IN (
SELECT PatientCharts.PatientChartId
FROM PatientCharts
WHERE ( PatientCharts.ChartStatusID = #ChartCompletedStatusID
OR PatientCharts.ChartStatusID = #ChartOnBaseStatusID
)
AND PatientCharts.IsLockDown = #IsNotLockdown
AND PatientCharts.CompletedOn IS NOT NULL
AND DATEDIFF(MINUTE, PatientCharts.CompletedOn, GETUTCDATE()) >= ( SELECT
tf.LockUpInterval
FROM
#tblFacCOnf tf
WHERE
tf.facilityId = PatientCharts.FacilityId
) )
This query locks the main table and results in TimeOut. IF i create a CTE first of all the updateable records and then update the main table by joining to the CTE. Will it help ??
First thing i advice you to do is to substitute IN condition with EXISTS. Second is to move all this conditional logic into CTE. Third is to substitute sub-select with #tblFacCOnf with join.
Last advice depends on your business logic and is not so important in my opinion
So at the end you will get something as
WITH search_cte as (
SELECT PatientCharts.PatientChartId
FROM PatientCharts
JOIN #tblFacCOnf tf on tf.facilityId = PatientCharts.FacilityId
WHERE ( PatientCharts.ChartStatusID = #ChartCompletedStatusID
OR PatientCharts.ChartStatusID = #ChartOnBaseStatusID
)
AND PatientCharts.IsLockDown = #IsNotLockdown
AND PatientCharts.CompletedOn IS NOT NULL
AND DATEDIFF(MINUTE, PatientCharts.CompletedOn, GETUTCDATE()) >= tf.LockUpInterval
) --cte end
UPDATE PatientChartImages
SET PatientChartImages.IsLockDown = #IsLockdown
WHERE PatientChartImages.IsLockDown = #IsNotLockdown
AND EXISTS (select 1 from PatientChartImages where PatientChartImages.PatientChartId = search_cte.PatientChartId)
One additional thing I might suggest if the other suggestions don't get you enough speed is not to use a table variable. Temp Tables are often faster for large data sets and can be indexed if need be.
The update lock is being held the time it takes to compute the CTE and the time for the update. The CTE time is probably causing the time out.
To reduce the lock time to the minimum required to update the target table. I suggest you create a temp table with two columns. Col1 is the primary key or cluster key of the target table and Col2 is the value you want in the target table. Wrap the temp table creation and fill the table with values according to your business logic within a transaction. Update the target table using a join to the temp table and the value from the temp table in a seperate transaction. After update drop the temp table.
I think you should create an SQL script (or stored procedure, if you will use it from a higher level) where you store the results of your selection into a cursor (you'll only have to find the PatientCartId's of the rows to be updated) and then you should use it in your update, so, the answer is yes.
It's easy to test this, you should put these commands into a transaction, rollback the transaction and before the rollback you should perform a selection to test your results. Good luck.
Using MySQL 5.1.49, I'm trying to implement a tagging system
the problem I have is with a table with two columns: id(autoincrement), tag(unique varchar) (InnoDB)
When using query, INSERT IGNORE INTO tablename SET tag="whatever", the auto increment id value increases even if the insert was ignored.
Normally this wouldn't be a problem, but I expect a lot of possible attempts to insert duplicates for this particular table which means that my next value for id field of a new row will be jumping way too much.
For example I'll end up with a table with say 3 rows but bad id's
1 | test
8 | testtext
678 | testtextt
Also, if I don't do INSERT IGNORE and just do regular INSERT INTO and handle the error, the auto increment field still increases so the next true insert is still a wrong auto increment.
Is there a way to stop auto increment if there's an INSERT duplicate row attempt?
As I understand for MySQL 4.1, this value wouldn't increment, but last thing I want to do is end up either doing a lot of SELECT statements in advance to check if the tags exist, or worse yet, downgrade my MySQL version.
You could modify your INSERT to be something like this:
INSERT INTO tablename (tag)
SELECT $tag
FROM tablename
WHERE NOT EXISTS(
SELECT tag
FROM tablename
WHERE tag = $tag
)
LIMIT 1
Where $tag is the tag (properly quoted or as a placeholder of course) that you want to add if it isn't already there. This approach won't even trigger an INSERT (and the subsequent autoincrement wastage) if the tag is already there. You could probably come up with nicer SQL than that but the above should do the trick.
If your table is properly indexed then the extra SELECT for the existence check will be fast and the database is going to have to perform that check anyway.
This approach won't work for the first tag though. You could seed your tag table with a tag that you think will always end up being used or you could do a separate check for an empty table.
I just found this gem...
http://www.timrosenblatt.com/blog/2008/03/21/insert-where-not-exists/
INSERT INTO [table name] SELECT '[value1]', '[value2]' FROM DUAL
WHERE NOT EXISTS(
SELECT [column1] FROM [same table name]
WHERE [column1]='[value1]'
AND [column2]='[value2]' LIMIT 1
)
If affectedRows = 1 then it inserted; otherwise if affectedRows = 0 there was a duplicate.
The MySQL documentation for v 5.5 says:
"If you use INSERT IGNORE and the row is ignored, the AUTO_INCREMENT counter
is **not** incremented and LAST_INSERT_ID() returns 0,
which reflects that no row was inserted."
Ref: http://dev.mysql.com/doc/refman/5.5/en/information-functions.html#function_last-insert-id
Since version 5.1 InnoDB has configurable Auto-Increment Locking. See also http://dev.mysql.com/doc/refman/5.1/en/innodb-auto-increment-handling.html#innodb-auto-inc...
Workaround: use option innodb_autoinc_lock_mode=0 (traditional).
I found mu is too short's answer helpful, but limiting because it doesn't do inserts on an empty table. I found a simple modification did the trick:
INSERT INTO tablename (tag)
SELECT $tag
FROM (select 1) as a #this line is different from the other answer
WHERE NOT EXISTS(
SELECT tag
FROM tablename
WHERE tag = $tag
)
LIMIT 1
Replacing the table in the from clause with a "fake" table (select 1) as a allowed that part to return a record which allowed the insert to take place. I'm running mysql 5.5.37. Thanks mu for getting me most of the way there ....
The accepted answer was useful, however I ran into a problem while using it that basically if your table had no entries it would not work as the select was using the given table, so instead I came up with the following, which will insert even if the table is blank, it also only needs you to insert the table in 2 places and the inserting variables in 1 place, less to get wrong.
INSERT INTO database_name.table_name (a,b,c,d)
SELECT
i.*
FROM
(SELECT
$a AS a,
$b AS b,
$c AS c,
$d AS d
/*variables (properly escaped) to insert*/
) i
LEFT JOIN
database_name.table_name o ON i.a = o.a AND i.b = o.b /*condition to not insert for*/
WHERE
o.a IS NULL
LIMIT 1 /*Not needed as can only ever be one, just being sure*/
Hope you find it useful
You can always add ON DUPLICATE KEY UPDATE Read here (not exactly, but solves your problem it seems).
From the comments, by #ravi
Whether the increment occurs or not depends on the
innodb_autoinc_lock_mode setting. If set to a non-zero value, the
auto-inc counter will increment even if the ON DUPLICATE KEY fires
I had the same problem but didn't want to use innodb_autoinc_lock_mode = 0 since it felt like I was killing a fly with a howitzer.
To resolve this problem I ended up using a temporary table.
create temporary table mytable_temp like mytable;
Then I inserted the values with:
insert into mytable_temp values (null,'valA'),(null,'valB'),(null,'valC');
After that you simply do another insert but use "not in" to ignore duplicates.
insert into mytable (myRow) select mytable_temp.myRow from mytable_temp
where mytable_temp.myRow not in (select myRow from mytable);
I haven't tested this for performance, but it does the job and is easy to read. Granted this was only important because I was working with data that was constantly being updated so I couldn't ignore the gaps.
modified the answer from mu is too short, (simply remove one line)
as i am newbie and i cannot make comment below his answer. Just post it here
the query below works for the first tag
INSERT INTO tablename (tag)
SELECT $tag
WHERE NOT EXISTS(
SELECT tag
FROM tablename
WHERE tag = $tag
)
I just put an extra statement after the insert/update query:
ALTER TABLE table_name AUTO_INCREMENT = 1
And then he automatically picks up the highest prim key id plus 1.
How can I edit the latest row in the database. I only know it's the last one. I don't know its id.
I don't know which language you are working with, in PHP's mySQL functions you can use
mysql_insert_id()
there are similar function in every other mySQL client library I know of.
Also, there is a native mySQL function!
LAST_INSERT_ID() (with no argument)
returns the first automatically
generated value that was set for an
AUTO_INCREMENT column by the most
recently executed INSERT statement to
affect such a column. For example,
after inserting a row that generates
an AUTO_INCREMENT value, you can get
the value like this:
mysql> SELECT LAST_INSERT_ID();
-> 195
Of course, a primary key with AUTO_INCREMENT is required for these functions to work.
For a table with an auto_increment id field:
UPDATE tbl SET col1 = 'val1' WHERE id = MAX(id);
If it's a row that has been inserted in your script (the same script from which you want to update it) and there is an auto_increment column on your table, you can get that auto_increment value, using functions such as those, for PHP :
mysql_insert_id
mysqli_insert_id
PDO::lastInsertId
There should be an equivalent for probably any language you can possibly be using for your application.
If your are trying to do an update from another script than the one in which you did the insert, and still have an auto_increment column, the best way will probably be to update the row that has the biggest value for that column :
update your_table
set your_column = ...
where id = max(id)
Or, in two steps (not sure it'll work in one) :
select max(id) as id from your_table
update your_table set your_column = ... where id = [what you got with thr first query]
You can also use UPDATE table SET ... WHERE id=LAST_INSERT_ID() (supposing the last insert was on the table you want to query).
I would not use TWO steps to find the last insert ID simply because a new record could be added in the mean time.
Depending on your version, you should be able to call $handle->last_id(); or $handle->{mysql_insertid};
Chris
I want to do all these update in one statement.
update table set ts=ts_1 where id=1
update table set ts=ts_2 where id=2
...
update table set ts=ts_n where id=n
Is it?
Use this:
UPDATE `table` SET `ts`=CONCAT('ts_', `id`);
Yes you can but that would require a table (if only virtual/temporary), where you's store the id + ts value pairs, and then run an UPDATE with the FROM syntax.
Assuming tmpList is a table with an id and a ts_value column, filled with the pairs of id value, ts value you wish to apply.
UPDATE table, tmpList
SET table.ts = tmpList.ts_value
WHERE table.id = tmpList.id
-- AND table.id IN (1, 2, 3, .. n)
-- above "AND" is only needed if somehow you wish to limit it, i.e
-- if tmpTbl has more idsthan you wish to update
A possibly table-less (but similar) approach would involve a CASE statement, as in:
UPDATE table
SET ts = CASE id
WHEN 1 THEN 'ts_1'
WHEN 2 THEN 'ts_2'
-- ..
WHEN n THEN 'ts_n'
END
WHERE id in (1, 2, ... n) -- here this is necessary I believe
Well, without knowing what data, I'm not sure whether the answer is yes or no.
It certainly is possible to update multiple rows at once:
update table table1 set field1='value' where field2='bar'
This will update every row in table2 whose field2 value is 'bar'.
update table1 set field1='value' where field2 in (1, 2, 3, 4)
This will update every row in the table whose field2 value is 1, 2, 3 or 4.
update table1 set field1='value' where field2 > 5
This will update every row in the table whose field2 value is greater than 5.
update table1 set field1=concat('value', id)
This will update every row in the table, setting the field1 value to 'value' plus the value of that row's id field.
You could do it with a case statement, but it wouldn't be pretty:
UPDATE table
SET ts = CASE id WHEN 1 THEN ts_1 WHEN 2 THEN ts_2 ... WHEN n THEN ts_n END
I think that you should expand the context of the problem. Why do you want/need all the updates to be done in one statement? What benefit does that give you? Perhaps there's another way to get that benefit.
Presumably you are interacting with sql via some code, so certainly you can simply make sure that the three updates all happen atomically by creating a function that performs all three of the updates.
e.g. pseudocode:
function update_all_three(val){
// all the updates in one function
}
The difference between a single function update and some kind of update that performs multiple updates at once is probably not a very useful distinction.
generate the statements:
select concat('update table set ts = ts_', id, ' where id = ', id, '; ')
from table
or generate the case conditions, then connect it to your update statement:
select concat('when ', id, ' then ts_', id) from table
You can use INSERT ... ON DUPLICATE KEY UPDATE. See this quesion: Multiple Updates in MySQL
ts_1, ts_2, ts_3, etc. are different fields on the same table? There's no way to do that with a single statement.