BigBash is a SQL parser that converts select statements to Bash one-liners that are executable directly on csv, log files, and other flat files. BigBash doesn't use a database, so the generated script should run on almost any *nix device. It is optimized for speed and should outperform simple databases when the data is large.
You might find BigBash useful if:
- you don't have access to a database
- you're a sysop who wants to do simple aggregations without installing a database on your machine
- you're feeling uncomfortable to spin up a Big Data stack just to crunch a couple gigabytes of data
Head over to BigBash It! to try the converter online!
BigBash is currently not meant for use in production or in a commercial setting. It can produce unexpected or incomplete results. The project originated during Zalando Hack Week and should be treated as an experiment.
That said, if you'd like to contribute to enhancing the project's useability please see the developers section.
- Java JDK >= 1.7
- Maven >= 3.0
BigBash is free of dependencies, so after installing the above you should be able to run it instantly on any operating system.
Download the source code. To launch the converter, use ./bigbash.sh. This should automatically start the build process.
You can change which *nix programs BigBash uses by editing the bigbash.conf file. BigBash first looks in the
current directory for the config file. If it finds none, it chooses the paths ~/.config/bigbash.conf and /etc/bigbash.conf.
Let's have BigBash query the Movielens dataset, which contains user-generated movie ratings. Download the dataset and extract it via:
wget "http://files.grouplens.org/datasets/movielens/ml-1m.zip"
unzip ml-1m.zip
You should now have a directory ml-1m that contains three files:
- movies.dat: Title and genre information
- ratings.dat: The ratings (from one to five) and timestamps
- users.dat: Additional user information
Open an editor and type:
CREATE TABLE movies (id INT UNIQUE, title TEXT, genres TEXT);
MAP movies TO 'movies.dat' DELIMITER '::';
SELECT title FROM movies ORDER BY title LIMIT 10;
Then save it under ml_test1.sql. In your terminal, type:
./bigbash.sh -f ml_test1.sql
The program should output a Bash one-liner like this:
(trap "kill 0" SIGINT; export LC_ALL=C; cat movies.dat|sed 's/::/\t/g'|cut -d $'\t' -f2|sort -S2G --parallel=4 -t$'\t' -k 1,1|head -n10|awk -F '\t' '{print $1}')
Executing this one-liner in the Movielens directory should produce an alphabetically sorted list of the first 10 movies in the dataset.
Let's take a look at the SQL file. The first line uses a typical CREATE TABLE statement to define the table movies with three columns. The next line:
MAP movies TO 'movies.dat' DELIMITER '::';
states that the table should be mapped to the file movies.dat. The column delimiter used in this file are two colons. Because we are using a shell script to execute the query, we can also use a general globbing pattern instead of a simple filename.
BigBash also supports gzipped files as well as quotations. Check the MAP command for more details.
The third line is a standard SQL SELECT statement. It outputs the movie names, which are then translated to the Bash one-liner. If you are familiar with Bash scriptings and UNIX tools, understanding this Bash code shouldn't be a problem.
The above example doesn't show BigBash's full power. Let's create a more complicated statement on the dataset: Finding the top ten movies sorted by the average ratings submitted by all male users age 30 or above.
As in the previous example, we first create the tables, mappings, and the SELECT statement. Open an editor and type:
CREATE TABLE movies (id INT UNIQUE, title TEXT, genres TEXT);
CREATE TABLE ratings (user_id int, movie_id int, rating int, ratingtime LONG);
CREATE TABLE users (id int UNIQUE, gender TEXT, age int, occupation int, zipcode Text);
MAP movies TO 'movies.dat' DELIMITER '::';
MAP ratings TO 'ratings.dat' DELIMITER '::';
MAP users TO 'users.dat' DELIMITER '::';
SELECT title, sum(rating)/count(*) from ratings AS r
hash JOIN movies on movies.id=r.movie_id
hash JOIN users on users.id=r.user_id
WHERE age >= 30 and gender ='M'
GROUP BY title
HAVING count(*) > 10
ORDER BY sum(rating)/count(*) DESC
LIMIT 10;
Save it as ml_test2.sql. Executing it leads to a rather long one-liner similar to this:
(trap "kill 0" SIGINT; export LC_ALL=C;
awk 'BEGIN{FS=OFS="\t"}NR==FNR{map[$1]=$0; next}{c=map[$1]; if (c) print $0,c; }'
<(cat users.dat|sed 's/::/\t/g'|cut -d $'\t' -f1,2,3|awk -F '\t' '($3 >= 30 && $2 == "M") {print}')
<(awk 'BEGIN{FS=OFS="\t"}NR==FNR{map[$1]=$0; next}{c=map[$2]; if (c) print $0,c; }'
<(cat movies.dat|sed 's/::/\t/g'|cut -d $'\t' -f1,2) <(cat ratings.dat|sed 's/::/\t/g'|cut -d $'\t' -f1,2,3))
|awk -F '\t' '($8 >= 30 && $7 == "M") {print}'
|awk -F '\t' 'BEGIN {OFS="\t"} {k=$5; row[k]=$0;sum_0[k] +=$3;count_1[k] +=1;} END{for (k in row) print row[k],sum_0[k],count_1[k]}'
|awk -F '\t' '($10 > 5) {print}'|awk 'BEGIN{FS=OFS="\t"}{print $0,$9 / $10}'
|sort -S2G --parallel=4 -t$'\t' -k 11,11rn|head -n10|awk -F '\t' '{print $5"\t"$9 / $10}')
It's cumbersome to work with such a long script, so you can save the output directly into a file via:
echo "#!/usr/bin/env bash" > ml_test2.sh
./bigbash.sh -f ml_test2.sql >> ml_test2.sh
chmod +x ml_test2.sh
Executing this in our directory leads to the following output (piped through column -t -s $'\t'):
Seven Samurai (The Magnificent Seven) (Shichinin no samurai) (1954) 4.57915
Carmen (1984) 4.57143
For All Mankind (1989) 4.5625
Godfather, The (1972) 4.55674
Casablanca (1942) 4.54377
Sanjuro (1962) 4.53846
Wrong Trousers, The (1993) 4.52804
To Kill a Mockingbird (1962) 4.52548
Schindler's List (1993) 4.5022
Dr. Strangelove or: How I Learned to Stop Worrying and Love the Bomb (1963) 4.50093
Here we see the HASH JOIN operator at work. We also add a UNIQUE constraint to the id column in the movie table; this, like HASH JOIN, creates a more performant script. Replacing it via a normal join would produce the same output, but usually takes longer.
java -jar big-bash.jar [OPTION]... [SQL commands]...
Options:
--help : Print this message (default: false)
--noAnsiC : Use this if you have problems with the default Ansi-C
decoding (default: false)
--sortAggregation : Uses sort instead of hash-based aggregation (default:
false)
-d VAL : Output delimiter (default: \t)
-f (--file) FILE : SQL file that should be processed
BigBash supports a subset of SQL92 syntax for select and create table statements. Please note that the * operator, e.g. in "SELECT * FROM ...", sometimes shows unexpected results.
- Sub-selects, UNION, IN
- LIKE (use REGEXP instead)
- DISTINCT (you can use GROUP BY, though this will not work in some aggregation functions)
- BETWEEN
- Column Aliases, AS operator (table aliases work though)
- Joins that are not equi-joins
- Implicit joins (use explicit joins instead)
- Queries without a FROM part
- Possibly various other stuff you might expect
In addition to a standard JOIN operator BigBash also supports a special HASH JOIN operator. HASH JOIN acts like a normal join, but uses a hash map internally to store the values of the right-hand side. It's faster/more performant in situations where the right side of the join fits completely into memory. Note that only inner and left hash joins are supported; for right or outer joins, use the normal join operation instead.
If the join-key of the right table is marked "unique" in the corresponding CREATE TABLE statement, the hash join operation is even faster.
MAP table_name TO 'filename[s]'|'command' [DELIMITER 'delimiter'] [QUOTE 'quote_char'] [TYPE 'type'] [REMOVEHEADER]
- table_name: The name of an existing table
- filename[s]: A globbing expression that denotes one or more files. Examples: *.gz, and access.log.2015-03-0[123].gz. You can also use Bash-extended globbing patterns, if enabled via shopt -s extglob.
- command (If TYPE is set to
RAW): A Bash command that outputs to the table the stream you wish to map. Example: bzip2 -dc *.bz. - DELIMITER: Specifies the character that separates the columns. The default is a tabulator.
- QUOTE: Specifies the character used to quote column expressions. The default is no-quotation. BigBash removes all quotations, so you have to choose an output delimiter that isn't used in the mapped data.
- TYPE: Allows you to specify the input type. 'type' must be one of these: 'FILE' (default), 'GZ' or 'RAW'.
- REMOVEHEADER: Use this if the files contain a header; otherwise they will be ignored.
- BigBash currently doesn't support escaping of delimiters in input files.
- Count (with DISTINCT support)
- Sum
- Min
- Max
- group_concat (with DISTINCT support)
Note that when using the DISTINCT operator, it is assumed that all distinct values fit into memory.
- <,>,>=,<=,<>,!=,=,==
- AND, OR
- +,-,*,/,% (modulo)
BigBash supports all awk functions. For more details, please refer to the awk documentation.
Q: Bigbash is slow on my machine. What can I do to make it faster?
A: Using the --parallel=<nr of cores> switch for the sort makes a big difference in most cases (change it in your bigbash.conf).
The performance of the queries depends on the tools used. Make sure you have updated sort, awk, sed and join to
the latest versions. In particular on Mac OS, the default installed core utils are really outdated
(look here
for details how to update). Also try different awk implementations, e.g., mawk or gawk.
Q: I can't use BigBash with JSON files, can I? A: BigBash does not support JSON out of the box, but in some cases you can convert JSON to csv using sed or jq. An example:
cat persons.json | jq -r '[.name,.address.street,.address.city,.gender]|@csv'
which could be used together with the RAW mapping type.
Q: Can I directly invoke BigBash with files on S3? A: Yes, using the AWS tools. One way is to use a one-liner like this as an input command:
AWS_BUCKET="s3://<bucketname>"; AWS_PATH_PREFIX="<path to look for files>"; AWS_PATTERN="<the regexp file pattern to match>"; aws s3 ls "$AWS_BUCKET/$AWS_PATH_PREFIX" --recursive | sed -n -E 's/.* +[0-9]+ +(.*)/\1/p' | grep "$AWS_PATTERN" | xargs -i aws s3 cp "$AWS_BUCKET/{}" -
There is an extensive list of integration tests that you can find in the file
src/test/resources/integrationTests. To run the tests on your local machine, execute
mvn verify -Pintegration-tests
If these test fail, it is very likely that BigBash will not work correctly on your machine, most likely due to incompatibility issues with your #nix tools.
We welcome code contributions. Please just clone the repository and create pull requests.
To contribute to BigBash, submit a pull request using the usual method. Here are some desired enhancements you could work on:
- enable BigBash to use GNU parallel for parallel execution
- provide aliases support
- ensure correct formatting of the data/error-checking