Parameterized SQL queries

Mateusz Żółtak asked me to spread the word about his new R package for parameterized SQL queries. Below you can find the copy of package vignette. If you work with SQL in R you may find it useful.

The package RODBCext is an extension of the RODBC database connectivity package. It provides support for parameterized queries. This document describes what parameterized queries are and when they should be used. In addition some examples of ROBDCext usage are shown.

It is assumed that you already know the RODBC package and the basics of SQL and ODBC. If not, please read about them first, e.g. see the ODBC Connectivity vignette of the RODBC package.

1 What are parameterized queries for?

Parameterized queries (also known as prepared statements) are a technique of query execution which separates a query string from query parameters values. There are two main reasons for using this technique:

  • avoiding SQL injections,
  • speeding up query execution in some scenarios.

Both are discussed below.

2 SQL injections

SQL injection is an attack against your code which uses SQL queries. Malicious query parameter values are passed in order to modify and execute a query. If you use SQL data sources, it is highly likely that sooner or later your R code will experience a problem similar to an SQL injection (or an SQL injection itself). Consider the following:

  • Even data from trusted data sources (even SQL ones) can cause problems in SQL queries if use improper programming techniques.
  • Are you sure that your data came from a really trusted source?
  • All Shiny applications which process data from SQL data sources can be a target of an SQL injection attack.

2.1 Example – an apostrophe sign in data

Let us begin with a simple example illustrating how your own data can lead to problems similar to a SQL injections attack.

Imagine a simple SQL database with a table called cakes:

cake price
Chocolate cake 10
Strawberry cupcake 3
Kevin’s Cherry Tart 12.3

We receive a CSV data file containing the same database but with new prices. You are asked to update the database. So you write your R code as below:

Such a code will fail on a Kevin’s Cherry Tart because this name contains an apostrophe. The resulting query would be UPDATE cakes SET price = 12.3 WHERE cake = 'Kevin's Cherry Tart'; which is not a proper SQL query. To deal with the Kevin’s Cherry Tart we need to escape the apostrophe in the cake’s name so that the database knows that it doesn’t denote the end of the string value.

2.2 Example – simple SQL injection

There is a nice XKCD about that – see here. Let’s translate it into an example in R.

We have got a database of students with a table students

last_name first_name
Smith John
Lee Andrew
Wilson Linda

A new school year has begun and new students have come. We have just received a CSV file with the same structure as the table students and we are asked to add it to the database. So we prepare a simple script:

Unfortunately one of our new students’ name is:

last_name first_name
Smith Robert’); DROP TABLE students; –

For this student our query would be:

These are in fact two SQL queries and one SQL comment:

  • INSERT INTO students (last_name, first_name) VALUES ('Smith', 'Robert');
  • DROP TABLE students;
  • --')

Execution of such a query can lead to a serious data loss (hopefully we have made a backup copy or do not have sufficient rights to drop the students table). To avoid such problems we should properly escape parameters values in our SQL queries.

2.3 How to escape values in SQL queries?

At this point we already know that we should properly escape parameters values in our SQL queries. There are many techniques of doing that:

  • Manually checking the data types.
  • Using parameterized queries.
  • Using high-level functions which escape values for us.

2.3.1 Manually checking data types

You can escape your data manually, e.g.

  • cast numeric columns to numbers using as.numeric(column) or sprintf(“%d %f”, integerColumn, realColumn),
  • cast dates using as.character(as.Date(column)),
  • escape strings using gsub(“‘“,”’’”, column),
  • etc.

This is possible but is also very error prone, especially when escaping string values. Everyone knows that apostrophes have to be escaped, but:

  • Different database systems may use different escape sequences (e.g. C-style with a backslash or repeat-style a with double apostrophe).
  • our database system may handle HTML/XML entities or inserting characters by a Unicode value (or many, many other strange ways of data input), so e.g. my’value or my\U0027value will be converted into my’value and then lead to errors in your query.

It is almost impossible to remember all caveats by yourself, so it is strongly advised not to use this method.

2.3.2 Using parameterized queries

Another solution is to separate the query string from its parameters (data). In such case a query execution is divided into two steps:

  • query parsing and planing,
  • passing parameter values to query and query execution.

As query parameters are passed separately, parameter values cannot modify (and break) the query string. To indicate places in the query where parameters will be placed, a special character is used, typically a question mark.

Let us rewrite our cakes example using the sqlExecute(connHandle, queryString, data) function from the RODBCext package:

We replaced the parameter values in query with a question mark and passed query and data as separate function parameters. We made our code not only SQL injection resistant, but also easier to read.

Moreover, the function function sqlExecute() supports vectorized data, so we can make it even simpler:

2.3.3 Using high-level functions which deal with escaping values for us

This would be the most straightforward solution.

An excellent example is dplyr, which provides a complete R to SQL mapper and allows us to completely forget about the SQL. Another example are the sqlSave(), sqlUpdate(), sqlCopy() and sqlCopyTable() functions from the RODBC package which deal with escaping values for us.

The problem is that: * Dplyr escapes values rather naively. With respect to strings only simple ‘to’’ escaping is performed which is enough to prevent silly errors but will fail against more advanced SQL injections. * RODBC’s high-level functions escape values in a safe way (by internally using parameterized queries), but have very limited functionality. Interestingly, judging from the comments in the source code, the parameterized queries have been introduced to them not to make them safe but to improve speed.

2.4 Summary

When using SQL we must pay attention to escape query parameter values properly. The existing R database connectivity packages do not provide a completely reliable way of doing that. A set of SQL injections safe functions provides very limited functionality and more flexible functions are using naive escaping methods. That is why RODBCext is a preferred way to make your R code SQL injections safe.

I hope dplyr developers will switch to use parameterized queries internally at some point. This would provide R community with a brilliant and safe R to SQL mapper and to forget about a manual preparation of SQL queries.

3 Speeding up query execution using parameterized queries

SQL query execution is being performed in a few steps. The first two steps are

  • Parsing the query string into internal database query data structures.
  • Planning the query, e.g. deciding the order of joining the tables, indexes which should be used to execute a query, etc.

If we repeat the same query many times and only values of query parameters are changing, it will be faster to perform these steps only once and then reuse the already parsed and planed query. This can be achieved by using parameterized queries.

3.1 Example – big insert

A typical scenario is an insert of many rows to a table:

3.2 Example – speeding up a SELECT query

Also repeated execution of a SELECT query can benefit from using parameterized variant:

The longer query string, the more complicated query planning and the more query repetitions, the bigger amount of time can be saved.

4 Parameterized SQL queries in R

Unfortunately all known to me R packages providing support for SQL databases lacks support for parameterized queries. Even the R DBI interface doesn’t define any methods which would allow to implement parameterized queries. The main reason for that is probably that R packages developers used to see SQL databases as just another storage backend for data frames rather than powerful data processing engines (which modern SQL databases already are).

4.1 RODBCext

RODBCext package tries to fill this gap by introducing parameterized queries support on the top of the RODBC package. RODBCext provides only two functions, both of them using database connection handlers from RODBC:

  • sqlPrepare(connHandle, SQLquery, errors = TRUE)
  • sqlExecute(connHandle, SQLquery, data, fetch = FALSE, errors = TRUE, ...)

4.1.1 sqlExecute()

Allows execution of SQL queries separated from query parameters values, e.g.:

The nice thing is that sqlExecute() (in opposite to sqlQuery()) supports vectorization. In the example below data will contain results of all five queries bound by rows.

Results can be also fetched separately using RODBC’s sqlGetResults(). This also provides a way to fetch results in parts:

As sqlExecute() uses internally sqlGetResults() to fetch results of the query, it also accept all parameters of the sqlGetResults():

4.1.2 sqlPrepare()

Parses a query string and plans a query. Query can be executed later using sqlExecute() with a parameter query set too NULL. This can provide some performance gain when executing the same query multiple times (see the chapter Speeding up query execution using parameterized queries). Usage example:

4 myśli na temat “Parameterized SQL queries”

  1. Just a quick note: my package rpg (https://github.com/thk686/rpg and on CRAN) permits parameterized queries and prepared statements. Execution of prepared statements on matrix arguments is handled in C++ (Rcpp). It also wraps database cursors which is actually much more useful. I wrote the package as my students were tired of dealing with the overhead/maintenance of ODBC drivers, etc. The package is specific to PostgreSQL. Nearly every feature of libpq is exposed by the interface.

    1. @Tim,
      thanks for info, many nice features in rpg.
      I’m using RPostgreSQL to work with Amazon RedShift. Wonder if rpg (e.g. cursors) work with ResShift. Will check after holidays.

  2. @Tim
    Unfortunately I didn’t know about your package. Maybe it’s because it was published on CRAN on 24th July (and the RODBCext is 20 days “older” :-) ). I will have to update my vignette.
    Thread-safe cursors are great! If I have time, I will add support for them to RODBCext (in fact cursors are already used in RODBC’s fetch mechanism, but they are not directly exposed to the user and I’m not shure if the implementation is thread-safe).

    @Przemysław
    RPostgreSQL also uses libpq internally (you don’t need to have libpq-dev in your system because libpq source code has been included into the source code of RPostgresql), so you should be able to connect to Amazon RedShift with the rpg package as well.

Dodaj komentarz

Twój adres e-mail nie zostanie opublikowany. Pola, których wypełnienie jest wymagane, są oznaczone symbolem *

Możesz użyć następujących tagów oraz atrybutów HTML-a: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code class="" title="" data-url=""> <del datetime=""> <em> <i> <q cite=""> <strike> <strong> <pre class="" title="" data-url=""> <span class="" title="" data-url="">