I have the following table:
CREATE TABLE dbo.TestSort
(
Id int NOT NULL IDENTITY (1, 1),
Value int NOT NULL
)
The Value column could (and is expected to) contain duplicates.
Let's also assume there are already 1000 rows in the table.
I am trying to prove a point about unstable sorting.
Given this query that returns a 'page' of 10 results from the first 1000 inserted results:
SELECT TOP 10 * FROM TestSort WHERE Id <= 1000 ORDER BY Value
My intuition tells me that two runs of this query could return different rows if the Value column contains repeated values.
I'm basing this on the facts that:
the sort is not stable
if new rows are inserted in the table between the two runs of the query, it could possibly create a re-balancing of B-trees (the Value column may be indexed or not)
EDIT: For completeness: I assume rows never change once inserted, and are never deleted.
In contrast, a query with stable sort (ordering also by Id) should always return the same results, since IDs are unique:
SELECT TOP 10 * FROM TestSort WHERE Id <= 1000 ORDER BY Value, Id
The question is: Is my intuition correct? If yes, can you provide an actual example of operations that would produce different results (at least "on your machine")? You could modify the query, add indexes on the Values column etc.
I don't care about the exact query, but about the principle.
I am using MS SQL Server (2014), but am equally satisfied with answers for any SQL database.
If not, then why?
Your intuition is correct. In SQL, the sort for order by is not stable. So, if you have ties, they can be returned in any order. And, the order can change from one run to another.
The documentation sort of explains this:
Using OFFSET and FETCH as a paging solution requires running the query
one time for each "page" of data returned to the client application.
For example, to return the results of a query in 10-row increments,
you must execute the query one time to return rows 1 to 10 and then
run the query again to return rows 11 to 20 and so on. Each query is
independent and not related to each other in any way. This means that,
unlike using a cursor in which the query is executed once and state is
maintained on the server, the client application is responsible for
tracking state. To achieve stable results between query requests using
OFFSET and FETCH, the following conditions must be met:
The underlying data that is used by the query must not change. That is, either the rows touched by the query are not updated or all
requests for pages from the query are executed in a single transaction
using either snapshot or serializable transaction isolation. For more
information about these transaction isolation levels, see SET
TRANSACTION ISOLATION LEVEL (Transact-SQL).
The ORDER BY clause contains a column or combination of columns that are guaranteed to be unique.
Although this specifically refers to offset/fetch, it clearly applies to running the query multiple times without those clauses.
If you have ties when ordering the order by is not stable.
LiveDemo
CREATE TABLE #TestSort
(
Id INT NOT NULL IDENTITY (1, 1) PRIMARY KEY,
Value INT NOT NULL
) ;
DECLARE #c INT = 0;
WHILE #c < 100000
BEGIN
INSERT INTO #TestSort(Value)
VALUES ('2');
SET #c += 1;
END
Example:
SELECT TOP 10 *
FROM #TestSort
ORDER BY Value
OPTION (MAXDOP 4);
DBCC DROPCLEANBUFFERS; -- run to clear cache
SELECT TOP 10 *
FROM #TestSort
ORDER BY Value
OPTION (MAXDOP 4);
The point is I force query optimizer to use parallel plan so there is no guaranteed that it will read data sequentially like Clustered index probably will do when no parallelism is involved.
You cannot be sure how Query Optimizer will read data unless you explicitly force to sort result in specific way using ORDER BY Id, Value.
For more info read No Seatbelt - Expecting Order without ORDER BY.
I think this post will answer your question:
Is SQL order by clause guaranteed to be stable ( by Standards)
The result is everytime the same when you are in a single-threaded environment. Since multi-threading is used, you can't guarantee.
Related
Posting two questions:
1.
Let's say there is a query:
SELECT C1, C2, C3 from TABLE;
When this query is fired for the first time,it retrieves all the values in a certain order.
Next time, when the same query is fired, will the previous order be retained?
There are 2 tables, TABLE1 and TABLE2, both of them have identical data.
Will (SELECT * from TABLE1) and (SELECT * from TABLE1) retrieve the same order of rows?
SQL tables represent unordered sets. Period. There is no ordering in a result set unless you explicitly include an ORDER BY.
It is that simple. If you want data in a particular order, then you need to use ORDER BY. That is how relational databases work.
The same query can return results in different orders each time the query is executed. There are no guarantees about the order -- unless the query has an ORDER BY for the outermost SELECT.
No, unless you are fetching data from result cache!
No, unless they are very small tables and your query runs with low parallelism.
Sorry for extra answer, but I see Tim claims that the query will return same result as long as the underlying table(s) is not modified, and the query has same execution plan.
Snowflake executes the queries in parallel, therefore the order of data is not predictable unless ORDER BY is used.
Let's create a table (big enough to be processed in parallel), and run a simple test case:
-- running on medium warehouse
create or replace table my_test_table ( id number, name varchar ) as
select seq4(), 'gokhan' || seq4() from table(generator(rowcount=>1000000000));
alter session set USE_CACHED_RESULT = false;
select * from my_test_table limit 10;
You will see that it will return different rows every time you run.
To answer both questions short: No.
If your query has no ORDER BY-clause, the SELECT statement always returns an unordered set. This means: Even if you query the same table twice and the data didnt change, SELECT without ORDER BY can retrieve different row-orders.
https://docs.snowflake.com/en/sql-reference/sql/select.html
I came across an old script that in essence does the following:
CREATE TABLE #T (ColA VARCHAR (20), ID INT)
INSERT INTO #T VALUES ('BBBBBBBB', 1), ('AAAAAAA', 4), ('RRRRRR', 3)
CREATE TABLE #S (ColA VARCHAR (100), ID INT)
INSERT INTO #S
SELECT * FROM #T
ORDER BY ID -- odd to do an order by in an insert statement, but that's the code as it is...
SELECT * FROM #S
DROP TABLE #T, #S
First, I want to mention that I am aware of the fact that tables such as the ones I created here do not have an actual order, we just order the resultset if we want.
However, if you run the script above on a SQL version 2008, you will get the results ordered in the order that was specified in the insert statement. On a 2016 machine, this is not the case. There it returns the rows in the order they were created in the first place. Does anyone know what changes cause this different behaviour?
Thanks a lot!
As to your example - nothing is changed. The relation in the relation theory is represented in the SQL with a table. And the relation is not ordered. So, you are not allowed to defined how rows are ordered when they are materialized - and you should not care about this.
If you want to SELECT the data in a ordered way each time, you must specified unique order by criteria.
Also, in your example - you can SELECT the data one billion times and the data can be returned as "you inserted" it each time, but on the very next time you can get different results. The engine returns the data in the "best" way according to it when there is no order specified, but this can change anytime.
As you know - unless order by is specified, the database engine returns the rows in an arbitrary order - How this order is generated has to do with the internal parts of the database engine - the algorithm may change between versions, even between service packs, without any need for documentation since it's known to be arbitrary.
Please note that arbitrary is not the same as random - meaning you should not expect to get different row order each time you run the query - in fact, you will probably get the same row order every time until something changes - that might be a restart to the server, a rebuild of an index, another row added to the table, an index created or removed - I can't say because it's not documented anywhere.
Moreover, unless you have an Identity column in your table, the optimizer will simply ignore the order by clause in the insert...select statement, exactly because what you already wrote in your question - Database tables have no intrinsic order.
Order the result set of a query by the specified column list and,
optionally, limit the rows returned to a specified range. The order
in which rows are returned in a result set are not guaranteed unless
an ORDER BY clause is specified.
MSSQL Docs
This question already has answers here:
Order by not working when insert in temp table
(4 answers)
Closed 5 years ago.
Below is my query:
CREATE TABLE #TEMP(CID INT,PID INT,STAT VARCHAR(20),TIN DATETIME, TOUT DATETIME)
INSERT INTO #TEMP(CID,STAT,PID,TIN,TOUT)
SELECT DISTINCT CID,STST,PID,TIN,TOUT
FROM CVTBL
WHERE STAT = 'YES'
AND PID = '12'
ORDER BY CID DESC;
select * from #temp
drop table #temp
This is a very straight forward query. However everytime when I run the select * from #temp it produced different result sets but the total number of rows is the same. How does that work?
I'm going to elaborate on this as an answer,but both Veljko89 and tarheel have hit the nail on the head in the comments they made on the OP's question.
Data, in SQL Server, is stored in unordered HEAPS. Regardless of the order you INSERT the data, regardless of if you have a CLUSTERED INDEX or not, performing a SELECT statement without an ORDER BY has no guarenteed order. Period.
The only (yes that's right ONLY) way to guarentee the order of a result set is to (unsurprisingly) use the ORDER BY clause. If you omit that clause SQL Server will return the rows in whatever order it processed that rows, which could be any order at all. For small tables, yes, you are likely to get the same order, and if you have a CLUSTERED INDEX then that improves that possibility, but that's just it, it's a possibility.
Once you get to larger tables, and start introducing multiple cores processing the information, then the order will become more and more randomised; as with larger datasets the data that is read first is more likely to vary, and with multiple cores one may finish processing its data first, however, had data from "further" in the table.
So, in summary: Add an ORDER BY clause (so that each column has a unique set) to ensure your queries always return data in the same order.
Through R I connect to a remotely held database. The issue I have is my hardware isn't so great and the dataset contains tens of millions of rows with about 10 columns per table. When I run the below code, at the df step, I get a "Not enough RAM" error from R:
library(DatabaseConnector)
conn <- connect(connectionDetails)
df <- querySql(conn,"SELECT * FROM Table1")
What I thought about doing was splitting the tables into two parts any filter/analyse/combine as needed going forward. I think because I use the conn JDBC conection I have to use SQL syntax to make it work. With SQL, I start with the below code:
df <- querySql(conn,"SELECT TOP 5000000 FROM Table1")
And then where I get stuck is how do I create a second dataframe starting with n - 5000000 rows and ending at the final row, retrieved from Table1.
I'm open to suggestions but I think there are two potential answers to this question. The first is to work within the querySql to get it working. The second is to use an R function other than querySql (no idea what this would look like). I'm limited to R due to work environment.
The SQL statement
SELECT TOP 5000000 * from Table1
is not doing what you think it's doing.
Relational tables are conceptually unordered.
A relation is defined as a set of n-tuples. In both mathematics and the relational database model, a set is an unordered collection of unique, non-duplicated items, although some DBMSs impose an order to their data.
Selecting from a table produces a result-set. Result-sets are also conceptually unordered unless and until you explicitly specify an order for them, which is generally done using an order by clause.
When you use a top (or limit, depending on the DBMS) clause to reduce the number of records to be returned by a query (let's call these the "returned records") below the number of records that could be returned by that query (let's call these the "selected records") and if you have not specified an order by clause, then it is conceptually unpredictable and random which of the selected records will be chosen as the returned records.
Since you have not specified an order by clause in your query, you are effectively getting 5,000,000 unpredictable and random records from your table. Every single time you run the query you might get a different set of 5,000,000 records (conceptually, at least).
Therefore, it doesn't make sense to ask about how to get a second result-set "starting with n - 5000000 and ending at the final row". There is no n, and there is no final row. The choice of returned records was not deterministic, and the DBMS does not remember such choices of past queries. The only conceivable way such information could be incorporated into a subsequent query would be to explicitly include it in the SQL, such as by using a not in condition on an id column and embedding id values from the first query as literals, or doing some kind of negative join, again, involving the embedding of id values as literals. But obviously that's unreasonable.
There are two possible solutions here.
1: order by with limit and offset
Take a look at the PostgreSQL documentation on limit and offset. First, just to reinforce the point about lack of order, take note of the following paragraphs:
When using LIMIT, it is important to use an ORDER BY clause that constrains the result rows into a unique order. Otherwise you will get an unpredictable subset of the query's rows. You might be asking for the tenth through twentieth rows, but tenth through twentieth in what ordering? The ordering is unknown, unless you specified ORDER BY.
The query optimizer takes LIMIT into account when generating query plans, so you are very likely to get different plans (yielding different row orders) depending on what you give for LIMIT and OFFSET. Thus, using different LIMIT/OFFSET values to select different subsets of a query result will give inconsistent results unless you enforce a predictable result ordering with ORDER BY. This is not a bug; it is an inherent consequence of the fact that SQL does not promise to deliver the results of a query in any particular order unless ORDER BY is used to constrain the order.
Now, this solution requires that you specify an order by clause that fully orders the result-set. An order by clause that only partially orders the result-set will not be enough, since it will still leave room for some unpredictability and randomness.
Once you have the order by clause, you can then repeat the query with the same limit value and increasing offset values.
Something like this:
select * from table1 order by id1, id2, ... limit 5000000 offset 0;
select * from table1 order by id1, id2, ... limit 5000000 offset 5000000;
select * from table1 order by id1, id2, ... limit 5000000 offset 10000000;
...
2: synthesize a numbering column and filter on it
It is possible to add a column to the select clause which will provide a full order for the result-set. By wrapping this SQL in a subquery, you can then filter on the new column and thereby achieve your own pagination of the data. In fact, this solution is potentially slightly more powerful, since you could theoretically select discontinuous subsets of records, although I've never seen anyone actually do that.
To compute the ordering column, you can use the row_number() partition function.
Importantly, you will still have to specify id columns by which to order the partition. This is unavoidable under any conceivable solution; there always must be some deterministic, predictable record order to guide stateless paging through data.
Something like this:
select * from (select *, row_number() over (id1, id2, ...) rn from table1) t1 where rn>0 and rn<=5000000;
select * from (select *, row_number() over (id1, id2, ...) rn from table1) t1 where rn>5000000 and rn<=10000000;
select * from (select *, row_number() over (id1, id2, ...) rn from table1) t1 where rn>10000000 and rn<=15000000;
...
Obviously, this solution is more complicated and verbose than the previous one. And the previous solution might allow for performance optimizations not possible under the more manual approach of partitioning and filtering. Hence I would recommend the previous solution.
My above discussion focuses on PostgreSQL, but other DBMSs should provide equivalent features. For example, for SQL Server, see Equivalent of LIMIT and OFFSET for SQL Server?, which shows an example of the synthetic numbering solution, and also indicates that (at least as of SQL Server 2012) you can use OFFSET {offset} ROWS and FETCH NEXT {limit} ROWS ONLY to achieve limit/offset functionality.
I'm trying to make an optimal SQL query for an iSeries database table that can contain millions of rows (perhaps up to 3 million per month). The only key I have for each row is its RRN (relative record number, which is the physical record number for the row).
My goal is to join the table with another small table to give me a textual description of one of the numeric columns. However, the number of rows involved can exceed 2 million, which typically causes the query to fail due to an out-of-memory condition. So I want to rewrite the query to avoid joining a large subset with any other table. So the idea is to select a single page (up to 30 rows) within a given month, and then join that subset to the second table.
However, I ran into a weird problem. I use the following query to retrieve the RRNs of the rows I want for the page:
select t.RRN2 -- Gives correct RRNs
from (
select row_number() over() as SEQ,
rrn(e2) as RRN2, e2.*
from TABLE1 as e2
where e2.UPDATED between '2013-05-01' and '2013-05-31'
order by e2.UPDATED, e2.ACCOUNT
) as t
where t.SEQ > 270 and t.SEQ <= 300 -- Paging
order by t.UPDATED, t.ACCOUNT
This query works just fine, returning the correct RRNs for the rows I need. However, when I attempted to join the result of the subquery with another table, the RRNs changed. So I simplified the query to a subquery within a simple outer query, without any join:
select rrn(e) as RRN, e.*
from TABLE1 as e
where rrn(e) in (
select t.RRN2 -- Gives correct RRNs
from (
select row_number() over() as SEQ,
rrn(e2) as RRN2, e2.*
from TABLE1 as e2
where e2.UPDATED between '2013-05-01' and '2013-05-31'
order by e2.UPDATED, e2.ACCOUNT
) as t
where t.SEQ > 270 and t.SEQ <= 300 -- Paging
order by t.UPDATED, t.ACCOUNT
)
order by e.UPDATED, e.ACCOUNT
The outer query simply grabs all of the columns of each row selected by the subquery, using the RRN as the row key. But this query does not work - it returns rows with completely different RRNs.
I need the actual RRN, because it will be used to retrieve more detailed information from the table in a subsequent query.
Any ideas about why the RRNs end up different?
Resolution
I decided to break the query into two calls, one to issue the simple subquery and return just the RRNs (rows-IDs), and the second to do the rest of the JOINs and so forth to retrieve the complete info for each row. (Since the table gets updated only once a day, and rows never get deleted, there are no potential timing problems to worry about.)
This approach appears to work quite well.
Addendum
As to the question of why an out-of-memory error occurs, this appears to be a limitation on only some of our test servers. Some can only handle up to around 2m rows, while others can handle much more than that. So I'm guessing that this is some sort of limit imposed by the admins on a server-by-server basis.
Trying to use RRN as a primary key is asking for trouble.
I find it hard to believe there isn't a key available.
Granted, there may be no explicit primary key defined in the table itself. But is there a unique key defined in the table?
It's possible there's no keys defined in the table itself ( a practice that is 20yrs out of date) but in that case there's usually a logical file with a unique key defined that is by the application as the de-facto primary key to the table.
Try looking for related objects via green screen (DSPDBR) or GUI (via "Show related"). Keyed logical files show in the GUI as views. So you'd need to look at the properties to determine if they are uniquely keyed DDS logicals instead of non-keyed SQL views.
A few times I've run into tables with no existing de-facto primary key. Usually, it was possible to figure out what could be defined as one from the existing columns.
When there truly is no PK, I simply add one. Usually a generated identity column. There's a technique you can use to easily add columns without having to recompile or test any heritage RPG/COBOL programs. (and note LVLCHK(*NO) is NOT it!)
The technique is laid out in Chapter 4 of the modernizing Redbook
http://www.redbooks.ibm.com/abstracts/sg246393.html
1) Move the data to a new PF (or SQL table)
2) create new LF using the name of the existing PF
3) repoint existing LF to new PF (or SQL table)
Done properly, the record format identifiers of the existing objects don't change and thus you don't have to recompile any RPG/COBOL programs.
I find it hard to believe that querying a table of mere 3 million rows, even when joined with something else, should cause an out-of-memory condition, so in my view you should address this issue first (or cause it to be addressed).
As for your question of why the RRNs end up different I'll take the liberty of quoting the manual:
If the argument identifies a view, common table expression, or nested table expression derived from more than one base table, the function returns the relative record number of the first table in the outer subselect of the view, common table expression, or nested table expression.
A construct of the type ...where something in (select somethingelse...) typically translates into a join, so there.
Unless you can specifically control it, e.g., via ALWCPYDTA(*NO) for STRSQL, SQL may make copies of result rows for any intermediate set of rows. The RRN() function always accesses physical record number, as contrasted with the ROW_NUMBER() function that returns a logical row number indicating the relative position in an ordered (or unordered) set of rows. If a copy is generated, there is no way to guarantee that RRN() will remain consistent.
Other considerations apply over time; but in this case it's as likely to be simple copying of intermediate result rows as anything.