Speeding Up SQL Query: Grouping & Joining

[pbuk]

the purpose of the query is to simply gather and format data to be imported into another third party software app, so no advantage to using Python or some other software.

But "gathering data" could be really slow without proper indexing, so there COULD be a big advantage in doing the indexing elsewhere if you can't do it in SQL Server, however

Also the database is MS SQLServer running in Azure

Ah, then transferring perhaps 10GB of unindexed data out each time is probably not an option (unless perhaps to a computational instance in the same Azure cluster).

 

[jack action]

If the table design from the third-party app is the culprit, then you could copy the data on your machine - in an optimized way - and update your copied database before each query.

But I seriously doubt this is your problem. I really would like to see your query (Telling us it is a "simple one with join" is not enough for us to help you further). The bold text is particularly a red flag for me:

and the query, which is not important to the question, just makes calculations based on classification and filtering of the category data

A query that makes calculations - as opposed to just gathering basic data - can take a lot of time.

Show us your query.

 

[whartung]

We need to see the SQL. You don't say how large the rows are, but, frankly, 10^7 rows on modern hardware is typically no big deal. You also don't say how large the other tables are. 10^7, 100 byte rows is 1G. Modern drives are pushing 100's of MB per second. Doesn't take that long to scan that kind of data. Not minutes, certainly.

Posting the output of the EXPLAIN will tell volumes. In fact, we can't really give, and you shouldn't really accept, any advice without the EXPLAIN. We don't know what it's doing.

You also don't say how often the query runs, how important it is to a workflow, etc. If it's for a weekly report then who cares? Meanwhile, if it's crushing the DB by voiding all of the cache memory with worthless data, then it has impact even beyond itself.

Simply, as with all performance questions , the answer is "it depends".

So, for best results, post the SQL query itself, and post the result of the EXPLAIN on it. Then, with some more context on how its run and expectations, perhaps better guidance can be given.

 

[harborsparrow]

Just out of curiosity, is anyone aware that Microsoft SQL Server Studio has a fantastic query speed optimizer? It saved me many times. I can't claim to understand the inner performance of Sql Server but MS SQL Studio had a 100% success rate in refactoring my complex stored procedures and queries. I don't recall how to access the feature but it was pretty easy to find 10 years ago and I assume it's still there. I'm talking about the free version that anyone can download and run.

 

[WWGD]

Only one I'm familiar with is running a trace. not sure that's the latest.

 

[jedishrfu]

The only way to speed up a sql query is to build an index table using the keys you are querying with in your sql query. This allows the query to use the index table to get its results.

 

[harborsparrow]

With T-SQL, the dev environment and tools are one's best friend. I have even had SQL Studio suggest and automate the addition of needed indexes. Learn to use that tool to the max, is my best advise. Of course, I had admin rights to the server. A lot of work environments force SQL programmers to work practically disconnected. I hated working like that and literally changed jobs to escape it. The sysadmins needed safety and control, I know, but I found them generally unhelpful. To use the tools fully, get a copy of the database and install it locally if you can. Or get admin rights so the dev environment can help you. People try to segregate programming from sysadmin on databases, which has a purpose sometimes but often leaves the programmer without enough information.

 

[harborsparrow]

The only way to speed up a sql query is to build an index table using the keys you are querying with in your sql query. This allows the query to use the index table to get its results.

I cannot agree that there is only one way. There might be a number of ways. I wrote T-SQL code for many years. For example, instead of querying a raw rable, you can write presorted views which are already compiled and optimized, and then query against those. And you can use completely different language semantics, sometimes, to accomplish the goal. Often there can be multiple approaches to the same problem. Indexes are one tool, and building an index is not free either. Everything has a cost.

 

[Vanadium 50]

So, I have some SQL code that I do not own, but is open source, and it is slower than I think it should be. There are probably 400,000 rows, and selecting rows containing a particular substring takes about a minute. The underlying technology is MariaDB. These queries cause the CPU to jump to 100% on one core.

Obviously, the best thing to do is to use multiple cores. but that seems not to be an option. Faster hardware helps, but not if this is CPU limited. Otherwise, I need to reduce the data - probably only 100-160K of these rows are of interest, although extracting them is non-trivial.

Am I missing something?

 

[jedishrfu]

It sounds like the query needs to parse a string for each row and that may be why you’re cpu bound.

 

[harborsparrow]

So, I have some SQL code that I do not own, but is open source, and it is slower than I think it should be. There are probably 400,000 rows, and selecting rows containing a particular substring takes about a minute. The underlying technology is MariaDB. These queries cause the CPU to jump to 100% on one core.

Obviously, the best thing to do is to use multiple cores. but that seems not to be an option. Faster hardware helps, but not if this is CPU limited. Otherwise, I need to reduce the data - probably only 100-160K of these rows are of interest, although extracting them is non-trivial.

Am I missing something?

Can you make this a new question, please? THIS thread is about T-Sql which is a Microsoft technology, whereas this interesting question concerns a Linux database using an entirely different dB engine and query language. This should get its own thread IMO.

 

[jedishrfu]

If one decides based on the title then it’s for any sql from any database.

 

[harborsparrow]

If one decides based on the title then it’s for any sql from any database.

The question itself specifies TSql and all my previous answers are specific to that.

 

[Vanadium 50]

Nevermind. I alraedy have frustration. I don't need to get any more here.

 

[pbuk]

selecting rows containing a particular substring takes about a minute.

Is the substring always in the same place? Create an index on just that part of the string. Is it always the same substring? Add a computed field to flag that the record contains the string.

 

[DrJohn]

Why not ignore the views and query the table they are formed from? A view is not a real table, so the view has to be created by its own query, and you are doing that several times. then joining them. Query the source of these views directly, so no join is required. However, are the views indexed? The index should be stored. But the index to the table of origin should be stored as well, so querying that should be faster as no join would be required.

Views are usually created to give office staff easy access to just the data their level of expertise or position in a company allows, or to avoid errors when they repeatedly enter a query with an extra column or two being required.

 

[pbuk]

Why not ignore the views and query the table they are formed from?

We are going round in circles:

The tables are actually views on a third party app, they are indexed, but I have no ability to alter them

The OP was fully addressed some time ago.