I collect a list of Facebook friends from my users including First, Last, Gender and DOB. I am then attempting to compare that database of names (stored as a table in MySQL) to another database comprised of similar information.
What would be the best way to conceptually link these results, with the second database being the much larger set of records (>500k rows)?
Here was what I was proposing:
Iterate through Facebook names
Search Last + DOB - if they match, assume a "confident" match
Search Last + First - if they match, assume a "probable" match
Search Last + Lichtenstein(First) above a certain level, assume a "possible" match
Are there distributed computing concepts that I am missing that may make this faster than a sequential mySQL approach? What other pitfalls may spring up, noting that it is much more important to not have a false-positive rather than miss a record?
Yes, your idea seems like a better algorithm.
Assuming performance is your concern, you can use caching to store the values that are just being searched. You can also start indexing the results in a NoSQL database such that the results will be very faster, so that you will have better read performance. If you have to use MySQL, read about polyglot persistence.
Assuming simplicity is your concern, you can still use indexing in a NoSQL database, so over the time you don't have to do myriad of joins will spoil the experience of the user and the developer.
There could be much more concerns, but it all depends on where would you like to use it, to use in a website, or to such data analytic purpose.
If you want to operate on the entire set of data (as opposed to some interactive thing), this data set size might be small enough to simply slurp into memory and go from there. Use a List to hang on to the data then create a Map> that for each unique last name points (via integer index) to all the places in the list where it exists. You'll also set yourself up to be able to perform more complex matching logic without getting caught up trying to coerce SQL into doing it. Especially since you are spanning two different physical databases...
I'm designing a central search function in a PHP web application. It is focused around a single table and each result is exactly one unique ID out of that table. Unfortunately there are a few dozen tables related to this central one, most of them being 1:n relations. Even more unfortunate, I need to join quite a few of them. A couple to gather the necessary data for displaying the results, and a couple to filter according to the search criteria.
I have been mainly relying on a single query to do this. It has a lot of joins in there and, as there should be exactly one result displayed per ID, it also works with rather complex subqueries and group by uses. It also gets sorted according to a user-set sort method and there's pagination in play as well done by the use of LIMIT.
Anyways, this query has become insanely complex and while I nicely build it up in PHP it is a PITA to change or debug. I have thus been considering another approach, and I'm wondering just how bad (or not?) this is for performance before I actually develop it. The idea is as follows:
run one less complex query only filtering according the search parameters. This means less joins and I can completely ignore group by and similar constructs, I will just "SELECT DISTINCT item_id" on this and get a list of IDs
then run another query, this time only joining in the tables I need to display the results (only about 1/4 of the current total joins) using ... WHERE item_id IN (....), passing the list of "valid" IDs gathered in the first query.
Note: Obviously the IN () could actually contain the first query in full instead of relying on PHP to build up a comma-separated list).
How bad will the IN be performance-wise? And how much will it possibly hurt me that I can not LIMIT the first query at all? I'm also wondering if this is a common approach to this or if there are more intelligent ways to do it. I'd be thankful for any input on this :)
Note to clarify: We're not talking about a few simple joins here. There is even (simple) hierarchical data in there where I need to compare the search parameter against not only the items own data but also against its parent's data. In no other project I've ever worked on have I encountered a query close to this complexity. And before you even say it, yes, the data itself has this inherent complexity, which is why the data model is complex too.
My experience has shown that using the WHERE IN(...) approach tends to be slower. I'd go with the joins, but make sure you're joining on the smallest dataset possible first. Reduce down the simple main table, then join onto that. Make sure your most complex joins are saved to the end to minimize the rows required to search. Try to join on indexes wherever possible to improve speed, and ditch wildcards in JOINS where possible.
But I agree with Andomar, if you have the time build both and measure.
Why is SELECT * bad practice? Wouldn't it mean less code to change if you added a new column you wanted?
I understand that SELECT COUNT(*) is a performance problem on some DBs, but what if you really wanted every column?
There are really three major reasons:
Inefficiency in moving data to the consumer. When you SELECT *, you're often retrieving more columns from the database than your application really needs to function. This causes more data to move from the database server to the client, slowing access and increasing load on your machines, as well as taking more time to travel across the network. This is especially true when someone adds new columns to underlying tables that didn't exist and weren't needed when the original consumers coded their data access.
Indexing issues. Consider a scenario where you want to tune a query to a high level of performance. If you were to use *, and it returned more columns than you actually needed, the server would often have to perform more expensive methods to retrieve your data than it otherwise might. For example, you wouldn't be able to create an index which simply covered the columns in your SELECT list, and even if you did (including all columns [shudder]), the next guy who came around and added a column to the underlying table would cause the optimizer to ignore your optimized covering index, and you'd likely find that the performance of your query would drop substantially for no readily apparent reason.
Binding Problems. When you SELECT *, it's possible to retrieve two columns of the same name from two different tables. This can often crash your data consumer. Imagine a query that joins two tables, both of which contain a column called "ID". How would a consumer know which was which? SELECT * can also confuse views (at least in some versions SQL Server) when underlying table structures change -- the view is not rebuilt, and the data which comes back can be nonsense. And the worst part of it is that you can take care to name your columns whatever you want, but the next guy who comes along might have no way of knowing that he has to worry about adding a column which will collide with your already-developed names.
But it's not all bad for SELECT *. I use it liberally for these use cases:
Ad-hoc queries. When trying to debug something, especially off a narrow table I might not be familiar with, SELECT * is often my best friend. It helps me just see what's going on without having to do a boatload of research as to what the underlying column names are. This gets to be a bigger "plus" the longer the column names get.
When * means "a row". In the following use cases, SELECT * is just fine, and rumors that it's a performance killer are just urban legends which may have had some validity many years ago, but don't now:
SELECT COUNT(*) FROM table;
in this case, * means "count the rows". If you were to use a column name instead of * , it would count the rows where that column's value was not null. COUNT(*), to me, really drives home the concept that you're counting rows, and you avoid strange edge-cases caused by NULLs being eliminated from your aggregates.
Same goes with this type of query:
SELECT a.ID FROM TableA a
WHERE EXISTS (
SELECT *
FROM TableB b
WHERE b.ID = a.B_ID);
in any database worth its salt, * just means "a row". It doesn't matter what you put in the subquery. Some people use b's ID in the SELECT list, or they'll use the number 1, but IMO those conventions are pretty much nonsensical. What you mean is "count the row", and that's what * signifies. Most query optimizers out there are smart enough to know this. (Though to be honest, I only know this to be true with SQL Server and Oracle.)
The asterisk character, "*", in the SELECT statement is shorthand for all the columns in the table(s) involved in the query.
Performance
The * shorthand can be slower because:
Not all the fields are indexed, forcing a full table scan - less efficient
What you save to send SELECT * over the wire risks a full table scan
Returning more data than is needed
Returning trailing columns using variable length data type can result in search overhead
Maintenance
When using SELECT *:
Someone unfamiliar with the codebase would be forced to consult documentation to know what columns are being returned before being able to make competent changes. Making code more readable, minimizing the ambiguity and work necessary for people unfamiliar with the code saves more time and effort in the long run.
If code depends on column order, SELECT * will hide an error waiting to happen if a table had its column order changed.
Even if you need every column at the time the query is written, that might not be the case in the future
the usage complicates profiling
Design
SELECT * is an anti-pattern:
The purpose of the query is less obvious; the columns used by the application is opaque
It breaks the modularity rule about using strict typing whenever possible. Explicit is almost universally better.
When Should "SELECT *" Be Used?
It's acceptable to use SELECT * when there's the explicit need for every column in the table(s) involved, as opposed to every column that existed when the query was written. The database will internally expand the * into the complete list of columns - there's no performance difference.
Otherwise, explicitly list every column that is to be used in the query - preferably while using a table alias.
Even if you wanted to select every column now, you might not want to select every column after someone adds one or more new columns. If you write the query with SELECT * you are taking the risk that at some point someone might add a column of text which makes your query run more slowly even though you don't actually need that column.
Wouldn't it mean less code to change if you added a new column you wanted?
The chances are that if you actually want to use the new column then you will have to make quite a lot other changes to your code anyway. You're only saving , new_column - just a few characters of typing.
If you really want every column, I haven't seen a performance difference between select (*) and naming the columns. The driver to name the columns might be simply to be explicit about what columns you expect to see in your code.
Often though, you don't want every column and the select(*) can result in unnecessary work for the database server and unnecessary information having to be passed over the network. It's unlikely to cause a noticeable problem unless the system is heavily utilised or the network connectivity is slow.
If you name the columns in a SELECT statement, they will be returned in the order specified, and may thus safely be referenced by numerical index. If you use "SELECT *", you may end up receiving the columns in arbitrary sequence, and thus can only safely use the columns by name. Unless you know in advance what you'll be wanting to do with any new column that gets added to the database, the most probable correct action is to ignore it. If you're going to be ignoring any new columns that get added to the database, there is no benefit whatsoever to retrieving them.
In a lot of situations, SELECT * will cause errors at run time in your application, rather than at design time. It hides the knowledge of column changes, or bad references in your applications.
Think of it as reducing the coupling between the app and the database.
To summarize the 'code smell' aspect:
SELECT * creates a dynamic dependency between the app and the schema. Restricting its use is one way of making the dependency more defined, otherwise a change to the database has a greater likelihood of crashing your application.
If you add fields to the table, they will automatically be included in all your queries where you use select *. This may seem convenient, but it will make your application slower as you are fetching more data than you need, and it will actually crash your application at some point.
There is a limit for how much data you can fetch in each row of a result. If you add fields to your tables so that a result ends up being over that limit, you get an error message when you try to run the query.
This is the kind of errors that are hard to find. You make a change in one place, and it blows up in some other place that doesn't actually use the new data at all. It may even be a less frequently used query so that it takes a while before someone uses it, which makes it even harder to connect the error to the change.
If you specify which fields you want in the result, you are safe from this kind of overhead overflow.
I don't think that there can really be a blanket rule for this. In many cases, I have avoided SELECT *, but I have also worked with data frameworks where SELECT * was very beneficial.
As with all things, there are benefits and costs. I think that part of the benefit vs. cost equation is just how much control you have over the datastructures. In cases where the SELECT * worked well, the data structures were tightly controlled (it was retail software), so there wasn't much risk that someone was going to sneek a huge BLOB field into a table.
Reference taken from this article.
Never go with "SELECT *",
I have found only one reason to use "SELECT *"
If you have special requirements and created dynamic environment when add or delete column automatically handle by application code. In this special case you don’t require to change application and database code and this will automatically affect on production environment. In this case you can use “SELECT *”.
Generally you have to fit the results of your SELECT * ... into data structures of various types. Without specifying which order the results are arriving in, it can be tricky to line everything up properly (and more obscure fields are much easier to miss).
This way you can add fields to your tables (even in the middle of them) for various reasons without breaking sql access code all over the application.
Using SELECT * when you only need a couple of columns means a lot more data transferred than you need. This adds processing on the database, and increase latency on getting the data to the client. Add on to this that it will use more memory when loaded, in some cases significantly more, such as large BLOB files, it's mostly about efficiency.
In addition to this, however, it's easier to see when looking at the query what columns are being loaded, without having to look up what's in the table.
Yes, if you do add an extra column, it would be faster, but in most cases, you'd want/need to change your code using the query to accept the new columns anyways, and there's the potential that getting ones you don't want/expect can cause issues. For example, if you grab all the columns, then rely on the order in a loop to assign variables, then adding one in, or if the column orders change (seen it happen when restoring from a backup) it can throw everything off.
This is also the same sort of reasoning why if you're doing an INSERT you should always specify the columns.
Selecting with column name raises the probability that database engine can access the data from indexes rather than querying the table data.
SELECT * exposes your system to unexpected performance and functionality changes in the case when your database schema changes because you are going to get any new columns added to the table, even though, your code is not prepared to use or present that new data.
There is also more pragmatic reason: money. When you use cloud database and you have to pay for data processed there is no explanation to read data that you will immediately discard.
For example: BigQuery:
Query pricing
Query pricing refers to the cost of running your SQL commands and user-defined functions. BigQuery charges for queries by using one metric: the number of bytes processed.
and Control projection - Avoid SELECT *:
Best practice: Control projection - Query only the columns that you need.
Projection refers to the number of columns that are read by your query. Projecting excess columns incurs additional (wasted) I/O and materialization (writing results).
Using SELECT * is the most expensive way to query data. When you use SELECT *, BigQuery does a full scan of every column in the table.
Understand your requirements prior to designing the schema (if possible).
Learn about the data,
1)indexing
2)type of storage used,
3)vendor engine or features; ie...caching, in-memory capabilities
4)datatypes
5)size of table
6)frequency of query
7)related workloads if the resource is shared
8)Test
A) Requirements will vary. If the hardware can not support the expected workload, you should re-evaluate how to provide the requirements in the workload. Regarding the addition column to the table. If the database supports views, you can create an indexed(?) view of the specific data with the specific named columns (vs. select '*'). Periodically review your data and schema to ensure you never run into the "Garbage-in" -> "Garbage-out" syndrome.
Assuming there is no other solution; you can take the following into account. There are always multiple solutions to a problem.
1) Indexing: The select * will execute a tablescan. Depending on various factors, this may involve a disk seek and/or contention with other queries. If the table is multi-purpose, ensure all queries are performant and execute below you're target times. If there is a large amount of data, and your network or other resource isn't tuned; you need to take this into account. The database is a shared environment.
2) type of storage. Ie: if you're using SSD's, disk, or memory. I/O times and the load on the system/cpu will vary.
3) Can the DBA tune the database/tables for higher performance? Assumming for whatever reason, the teams have decided the select '*' is the best solution to the problem; can the DB or table be loaded into memory. (Or other method...maybe the response was designed to respond with a 2-3 second delay? --- while an advertisement plays to earn the company revenue...)
4) Start at the baseline. Understand your data types, and how results will be presented. Smaller datatypes, number of fields reduces the amount of data returned in the result set. This leaves resources available for other system needs. The system resources are usually have a limit; 'always' work below these limits to ensure stability, and predictable behaviour.
5) size of table/data. select '*' is common with tiny tables. They typically fit in memory, and response times are quick. Again....review your requirements. Plan for feature creep; always plan for the current and possible future needs.
6) Frequency of query / queries. Be aware of other workloads on the system. If this query fires off every second, and the table is tiny. The result set can be designed to stay in cache/memory. However, if the query is a frequent batch process with Gigabytes/Terabytes of data...you may be better off to dedicate additional resources to ensure other workloads aren't affected.
7) Related workloads. Understand how the resources are used. Is the network/system/database/table/application dedicated, or shared? Who are the stakeholders? Is this for production, development, or QA? Is this a temporary "quick fix". Have you tested the scenario? You'll be surprised how many problems can exist on current hardware today. (Yes, performance is fast...but the design/performance is still degraded.) Does the system need to performance 10K queries per second vs. 5-10 queries per second. Is the database server dedicated, or do other applications, monitoring execute on the shared resource. Some applications/languages; O/S's will consume 100% of the memory causing various symptoms/problems.
8) Test: Test out your theories, and understand as much as you can about. Your select '*' issue may be a big deal, or it may be something you don't even need to worry about.
There's an important distinction here that I think most answers are missing.
SELECT * isn't an issue. Returning the results of SELECT * is the issue.
An OK example, in my opinion:
WITH data_from_several_tables AS (
SELECT * FROM table1_2020
UNION ALL
SELECT * FROM table1_2021
...
)
SELECT id, name, ...
FROM data_from_several_tables
WHERE ...
GROUP BY ...
...
This avoids all the "problems" of using SELECT * mentioned in most answers:
Reading more data than expected? Optimisers in modern databases will be aware that you don't actually need all columns
Column ordering of the source tables affects output? We still select and
return data explicitly.
Consumers can't see what columns they receive from the SQL? The columns you're acting on are explicit in code.
Indexes may not be used? Again, modern optimisers should handle this the same as if we didn't SELECT *
There's a readability/refactorability win here - no need to duplicate long lists of columns or other common query clauses such as filters. I'd be surprised if there are any differences in the query plan when using SELECT * like this compared with SELECT <columns> (in the vast majority of cases - obviously always profile running code if it's critical).
I have to perform some serious data mining on very large data sets stored in MySQL db. However, queries that require a bit more than a basic SELECT * FROM X WHERE ... tend to become rather inefficient since they return results on the order of 10e6 or more, especially when JOIN on one or more tables is introduced - think of joining 2 or more tables containing several tens of millions rows (after filtering data), which is something that pretty much happens on every query. More than often we'd like to run aggregate functions on these (sum, avg, count, etc), but this is impossible since MySQL simply chokes.
I should note that many efforts were put to optimize the current performance - all tables are indexed properly and queries are tuned, the hardware is top notch, the storage engine was configured and so on. However, still each query takes very long - to the point where "let's run it before we go home and hope for the best when we come to work tomorrow." Not good.
This has to be a solvable problem - many large companies perform very data and computational intensive mining, and handle it well (without writing their own storage engines, google). I'm willing to accept time penalty to get the job done, but on the order of hours, not days. My question is - what do people use to counter problems like this? I've heard of storage engines geared to this type of problem (greenplum, etc.), but I wanted to hear how this problem is typically approached. Our current data store is obviously relational and should probably remain such, but any thoughts or suggestions are welcome. Thanks.
I suggest PostgreSQL, which I've been working with quite successfully on tables with ~0.5B rows that required some complex join operations. Oracle should be good for that too, but I don't have much experience with it.
It should be noted that switching an RDBMS isn't a magic solution, if you want to scale to those sizes there's a LOT of hard work to be done in optimizing your queries, optimizing the database structure and indexes, fine tuning the database configuration, using the right hardware for your usage, replication, using materialized views (which are extremely powerful when used correctly. see here and here - its postgres specific, but applies to other RDBMSs too)... and at some point, you just have to throw more money on the problem.
edited fixed some weird typos (useless android auto correct...) and added some resources about materialized views
We have used MS SqlServer to run analytics on financial data with ten of millions of rows and more using complex JOIN and aggregation. Several things that we have done other than what you have mentioned are:
We chunk the calculation into a lot of temporary tables instead of using sub-query. These tables then we apply proper keys, indexing and so on via the code. Query with sub-query just fails for us
In the temporary tables, we often apply the clustered index that makes sense for us. Note that this temporary tables are filtered results so applying the index on the fly is not expensive compared to use the sub query in place of this temporary tables. Note I am speaking from our experience and might not apply to all cases
As we have done a lot of aggregation function as well, we did a lot indexing on the group columns
We do a lot of query run planning using SQL Query Analyzer that shows us the execution plan. Based on the plan, we revised the query, change the index
We provide hints for the SQL Server that we think could help the execution such as the choice of JOIN Algorithm to take (Hash, Merged or Nested)
I have read that distinct() API call has some performance issues at times. I wanted to try to rewrite a query through the orm which avoided using distinct (at least profile the difference).
My understanding is that values() performs a Group By under the hood. When I test out the two methods, though, the Count of objects differs depending on whether I use distinct() or values()/annotate().
zip_codes = Location.objects.values('zip_code').annotate(zip_count=Count('zip_code')).exclude(zip_code=None).count()
VS.
zip_codes = Location.objects.values_list('zip_code', flat=True).exclude(zip_code=None).distinct()
any thoughts on what is wrong here?
Thanks!
I just quickly checked your queries against a database I have with a similar query. The counts was identical so I'm not sure what about your data is resulting in issues.
I'd also be HIGHLY skeptical of the premise though. DISTINCT is indeed a cpu intensive query. However, so is COUNT(*) and your second query is going to first run an count aggregate with a group by and then run a COUNT on the results. I'd be put money on the single DISTINCT call being faster (I'd also check with whichever database backend you're using to see). All of this has very little to do with django's ORM and a whole heck of a lot more to do with your database backend.
Also think about this. The distinct based query is an order of magnitude clearer as to what it's accomplishing compared to the annotate based one. Do you have evidence to support that DISTINCT is going to be slow in your situation, or better still that it's forming a bottlneck right now? If not you're well into the range of premature optimization and should heavily reconsider your path.
Premature Optimization.
Optimization matters only when it matters. When it matters, it matters a lot, but until you know that it matters, don't waste a lot of time doing it. Even if you know it matters, you need to know where it matters. Without performance data, you won't know what to optimize, and you'll probably optimize the wrong thing.
The result will be obscure, hard to write, hard to debug, and hard to maintain code that doesn't solve your problem. Thus it has the dual disadvantage of (a) increasing software development and software maintenance costs, and (b) having no performance effect at all.
In other words write your software clearly and then when you find a problem trace it to the source and fix it. Anything you do before that is counterproductive. Spend your time worrying about which indexes are going to matter on your db, and where to use select_related. Those are 10000% more effective than what you are worrying about here (unless you are counting zip codes all the time, in which case let me introduce you to caching)