Let me start by saying a big “thank you” to the staff at Oracle for deciding to open source reducer.sh. It’s a tool I developed whilst I was working for them several years ago. Its sole purpose is to do one thing – but do it good: test-case simplification.
So, let’s say some customer just sent you 120,000 lines of SQL code and affirms that “it definitely causes a crash.” Or maybe you ran RQG (the Random Query Generator) for awhile (with the general query log turned on) and now you have a nice SQL trace which may just lead to that crash the run resulted in. Or you’re a DBA testing the company’s usual queries with Valgrind, and noticed that 2 in 1000 queries give a Valgrind warning in the mysqld error log – you’re just not sure which one. Or maybe you’re a developer, and during testing you saw that a SELECT query output did not look the way it should – the output was “7” where it should have been “5” – the only problem – you have 1000 lines of INSERT statements and are not sure which one caused it. In all of these cases reducer can help.
Here are some of it’s benefits/features:
- It can reduce large amount of SQL fast. 40K lines to just a few can usually be done in around 1 hour.
- Larger files scale even better – the chunking elimination method automatically adapt to file size.
- It can reduce crashes/asserts, Valgrind testcases, mysqld error log messages, and mysql CLI output testcases
- Also working (but with complex setup atm) is multi-threaded SQL test-case simplification (ALPHA)
- It can reduce sporadic testcases for all of the above (i.e. testcases where the issue does not reproduce every time)
- It can reduce sporadic testcases using multiple threads which significantly improves reduction time
- It is aware of how to establish a testcase is sporadic (and will report the same) – or not – and will change it’s behavior
- It is capable (turned on by default) of reducing actual DML/DDL query code after completing line-based reduction
- It is capable (turned on by default) of reducing testcases by eliminating columns from tables and INSERT queries
- By default reducer.sh uses tmpfs (highly recommended) to ensure testcases are “as reproducible as possible” (disk I/O)
- Additional options for mysqld (necessary to reproduce an issue) can easily be listed/added
- Regex syntax can be used in search strings (where applicable)
So, without further ado, let’s have a look at how to get it to do your simplification job
To get reducer.sh today, use these commands (yum example used, but this can easily be adapted to apt-get):
$ sudo yum install bzr
$ cd ~
$ bzr branch lp:randgen
$ cd randgen/util/reducer/
$ ls *
(You may also want to checkout ./status.sh in this directory which is a handy tool for seeing what reducer.sh is upto when it is doing it’s first/original attempt to reproduce a given issue.)
And you can get percona-qa (for parse_general_log.sh [and the prepare_reducer.sh code bit if you need it] as shown in the video):
$ cd ~
$ bzr branch lp:percona-qa
$ cd perconq-qa
$ ls *