The goal of the ProvSQL project is to add support for (m-)semiring provenance and uncertainty management to PostgreSQL databases, in the form of a PostgreSQL extension/module/plugin. It is work in progress.

Table of contents

• Features
• Docker container
• Prerequisites for installation
• Installation
• Testing your installation
• Using ProvSQL
• Uninstalling
• License
• Contact

The ProvSQL system supports:

• computation of provenance of SQL queries, in the form of a provenance circuit, for the following forms of provenance:
  • Boolean provenance
  • semiring provenance in arbitrary semirings
  • m-semiring provenance in arbitrary semirings with monus
  • semimodule provenance of aggregate queries
  • where-provenance
• probability computation from the Boolean provenance for query evaluation over probabilistic databases, through the following methods:
  • naïve evaluation
  • Monte-Carlo sampling
  • building a d-DNNF representation of the provenance from a tree decomposition of the Boolean circuit
  • compilation to a d-DNNF using an external tool (d4, c2d, minic2d or dsharp)
  • approximate weighted model counting using an external tool (weightmc)
• Shapley value computation
• expected Shapley value computation over probabilistic data

The following SQL features are currently supported.

• Regular SELECT-FROM-WHERE queries (aka conjunctive queries with multiset semantics)
• JOIN queries (regular joins only; outer, semijoins, and antijoins are not currently supported)
• SELECT queries with nested SELECT subqueries in the FROM clause
• GROUP BY queries
• SELECT DISTINCT queries (i.e., set semantics)
• UNION's or UNION ALL's of SELECT queries
• EXCEPT of SELECT queries
• aggregation on the final query

As an alternative to a ProvSQL installation (see below), you can try a demonstration version of ProvSQL (full-featured, except for minic2d support) as a Docker container. To deploy it, once Docker CE is installed, simply run:

docker run inriavalda/provsql

By following the instructions, you will be able to connect to the PostgreSQL server within the container using a PostgreSQL client, and to use a Web interface for simple visualization of where-provenance. The Docker container can also be built locally, using:

make docker-build

1. An install of PostgreSQL >= 9.6. The extension has currently been tested with versions from 9.6 to 16 (inclusive) of PostgreSQL, under Linux, Mac OS (both x86-64 and ARM architectures), and Windows Subsystem for Linux (if the extension does not work on a specific version or operating system, a bug report is appreciated).

2. A compilation environment for PostgreSQL, including the make tool, a C/C++ compiler (both can be obtained on Debian-based Linux distributions from the virtual build-essential package), and the headers for your PostgreSQL version (as can be obtained for instance from the postgresql-server-dev-xx package on Debian-based systems, or from the postgresql package on the Homebrew package manager for Mac OS X). The C++ compiler should support C++ 2017.

3. The uuid-ossp extension for PostgreSQL (on Debian-based systems, it is found in the postgresql-contrib-9.x package for PostgreSQL version 9.x, and is installed automatically for PostgreSQL version >= 10; on Homebrew, in the ossp-uuid package; if you compile PostgreSQL from source, make sure to also compile and install the additional modules in the contrib directory).

4. The Boost container library (on Debian-based systems, it is found in the libboost-dev package).

5. Optionally, for probability computation through knowledge compilation, any or all of the following software (note that some of them are not available under other OSs than Linux):

• c2d, from http://reasoning.cs.ucla.edu/c2d/download.php
• d4, from https://github.com/crillab/d4
• dsharp, from https://github.com/QuMuLab/dsharp
• minic2d, from http://reasoning.cs.ucla.edu/minic2d/
• weightmc, from https://bitbucket.org/kuldeepmeel/weightmc/src/master/

To be used, an executable with the name of this software must be available in the PATH of the PostgreSQL server user (e.g., in /usr/local/bin/). Using minic2d also requires the hgr2htree executable (it is provided with minic2d).

6. Optionally, for circuit visualization, the graph-easy executable from the Graph::Easy Perl library (that can be obtained from the libgraph-easy-perl package on Debian-based Linux distributions, or from CPAN).

1. Compile the code with make. If you have several installed versions of PostgreSQL, you can change the version the module is compiled against by changing the reference to pg_config in the Makefile.internal file.

2. Install it in the PostgreSQL extensions directory with make install (run as a user with rights to write to the PostgreSQL installation directories).

3. Add the line

   shared_preload_libraries = 'provsql'
   

   to the postgresql.conf configuration file (on Linux systems, it should be in /etc/postgresql/VERSION/main/postgresql.conf) and restart the PostgreSQL server (e.g., with service postgresql restart on systemd-based distributions). This is required because the extension includes hooks.

You can test your installation by running make test as a PostgreSQL administrator user. It will run all tests then, if tests fail, launch the pager command (usually less) on the diff between expected and actual output.

If you do not want to run this as the default administrator user, you can make yourself a PostgreSQL administrator with ALTER USER your_login WITH SUPERUSER. This assumes that your_login is a PostgreSQL user: on Debian-based Linux distributions, you can ensure this by running the command createuser your_login as the postgres user.

If your installation of PostgreSQL does not listen on the default (5432) port, you can add --port=xxxx to the EXTRA_REGRESS_OPTS line of Makefile.internal, where xxxx is the port number.

Note that the tests that depend on external software (c2d, d4, dsharp, minic2d, weightmc, graph-easy) will fail if no executable of that name can be found.

You can use ProvSQL in any PostgreSQL database by loading the provsql extension. See the file setup.sql for an example on how to do this.

You then need to add provenance to an existing table using the provsql.add_provenance(regclass) user-defined function. See add_provenance.sql for an example. The table will have an extra provsql column added. This column is handled in a special way and always represents, in query results, the provenance of each tuple as a UUID.

You can then use this provenance to run computation in various semirings. See security.sql and formula.sql for two examples.

See the other examples in test/sql for other use cases.

A demonstration of an early version of the ProvSQL system is available as a video, on https://youtu.be/iqzSNfGHbEE?vq=hd1080 The SQL commands used in this demonstration can be found in the doc/demo/ directory. An article describing this demonstration, presented at the VLDB 2018 conference, is available at http://pierre.senellart.com/publications/senellart2018provsql.pdf

Finally, a ProvSQL tutorial is provided, in the form of a crime mystery. It can be found in the doc/tutorial/ directory.

You can uninstall ProvSQL by running make uninstall (run as a user with rights to write to the PostgreSQL installation directories), and by removing the reference to provsql in the postgresql.conf configuration file.

ProvSQL is provided as open-source software under the MIT License. See LICENSE.

https://github.com/PierreSenellart/provsql

Pierre Senellart pierre@senellart.com

Bug reports and feature requests are preferably sent through the Issues feature of GitHub.