Contribute to the E-Maj development¶
Any contribution to the development and the improvement of the E-Maj extension is welcome. This page gives some information to make these contributions easier.
Build the E-Maj environment¶
The E-Maj extension repository is hosted on the github site: https://github.com/beaud76/emaj
Clone the E-Maj repository¶
So the first acction to perform is to locally clone this repository on his/her own computer. This can be done by using the functionnalities of the github web interface or by typing the shell command:
git clone https://github.com/beaud76/emaj.git
Description of the E-Maj tree¶
So one has a full directory tree (except the web clients). It contains all directories and files described in the appendix, except the doc directory content that is separately maintained (see below).
The main directory also contains the following components:
- the tar.index file that is used to build the tarball of the E-Maj version distributed on pgxn.org
- the docs directory with all sources of the online documentation
- in the sql directory:
- the file emaj- -devel.sql, source of the extension in its current version
- the source of the previous version emaj- -<previous_version>.sql
- a ppa.sql script that allows to prepare an E-Maj environment to test the phpPgAdmin plugin or the Emaj_web client
- a test directory containing all components used to test the extension
- a tools directory containing some … tools.
Setting tools parameters¶
The tools stored in the tools directory need some parameters to be set, depending on his/her own environment. A parameter system covers some tools. For the others, the tools/README file details the changes to apply.
Créating the emaj_tools.env file¶
The parameters that may be modified are grouped into the tools/emaj_tools.env file, which is called by tools/emaj_tools.profile.
The repository contains a file tools/emaj_tools.env-dist that may be used as a template to create the emaj_tools.env file.
The emaj_tools.env file must contain:
- the list of PostgreSQL versions that are supported by the current E-Maj version and for which a PostgreSQL instance exists for tests (EMAJ_USER_PGVER variable),
- for each PostgreSQL version used for the tests, 6 variables describing the location of binaries, the main directory of the related instance, the role and the ip-port to be used for the connection to the instance.
The version currently under development is named devel.
Regularly and when it is justified, a new version is created. Its name has a X.Y.Z pattern.
The tools/create_version.sh shell script assists in creating this version. It is only used by the E-Maj maintainers. So its use is not described here.
Coding the emaj- -devel.sql script must follow these rules:
- script structure: after some checks about the execution conditions that must be met, the objects are created in the following order: roles, enumerated types, sequences, tables (with their indexes and contraints), composite types, E-Maj parameters, low level functions, elementary functions that manage tables and sequences, functions that manage tables groups, general purpose functions, event triggers, grants, additional actions for the extensions. The script ends with some final operations.
- all objects are created in the emaj schema, except the _emaj_protection_event_trigger_fnct() function, created in the public schema,
- tables and sequences names are prefixed by emaj_
- functions names are prefixed by emaj_ when they are usable by end users, or by _ for internal functions,
- the internal tables and the functions callable by end users must have a comment,
- the language keywords are in upper case, objects names are in lower case,
- the code is indented with 2 space characters,
- lines must not contain tab characters and must not end with spaces,
- in the functions structure, the code delimiters must contain the function name surrounded with a $ character (or $do$ for code blocks),
- variables names are prefixed with v_ for simple variables or r_ for RECORD type variables,
- the code must be compatible with all PostgreSQL versions supported by the current E-Maj version. When this is striclty necessary, the code may be differenciated depending on the PostgreSQL version.
A perl script, tools/check_code.pl performs some checks on the code format of the script that creates the extension. It also detects unused variables. This script is directly called in non-regression tests scenarios.
Version upgrade script¶
E-Maj is installed into a database as an extension. The E-Maj administrator must be able to easily upgrade the extension version. So an upgrade script is provided for each version, that upgrades from the previous version to the next version. It is named emaj- -<previous_version>- -devel.sql.
The development of this script follows these rules:
- Develop/maintain the upgrade script at the same time as the main emaj- -devel.sql script, so that the tests of a change include upgrade version cases,
- Apply the same coding rules as for the main script,
- As far as possible, ensure that the upgrade operation is able to process tables groups in logging state, without loosing the capability to perform E-Maj rollbacks on marks set prior the version upgrade.
At the beginning of a version, the upgrade script is built using a template (the file tools/emaj_upgrade.template).
As the development goes on, a perl script helps to synchronize the creation/deletion/replacement of functions. It compares the emaj- -devel.sql script and the script that creates the previous version and updates the emaj- -<previous_version>- -devel.sql script. To let it work properly, it is essential to keep both tags that frame the part of the script that describes functions.
After having adapted the parameters (see the TOOLS/README file), just submit:
The other parts of the script must be coded manually. If the structure of an internal table is changed, the table content must be migrated (scripts for prior version upgrade can be used as examples).
Through the rollback functions, the E-Maj extension updates database content. So the reliability is a key characteristics. For this reason, it is essential to pay a great attention to the tests.
Create PostgreSQL instances¶
The ideal is to be able to test E-Maj with all PostgreSQL versions that are supported by the extension (currently from version 9.2 to version 11).
The tools/create_cluster.sh script helps in creating a test instance. Its content may show the characteristics of the instance to create. It can also be executed (after parameters setting as indicated in tools/README):
Install software dependancies¶
Testing the clients may require to install some additional software components:
- the php software, with its PostgreSQL interface,
- the perl software, with the DBI and DBD::Pg modules.
Execute non regression tests¶
A solid test environment is supplied in the repository. It contains:
- a test tool,
- test scenarios,
- expected results.
The test scenarios¶
The test system contains 3 scenarios:
- a full standart scenario,
- the same scenario but installing the extension from the previous version with an immediate upgrade into the current version,
- a shorter scenario but with an upgrade into the current version while tables groups are in logging state.
The 3 scenarios call psql scripts, all located into the test/sql directory. The scripts chain E-Maj function calls in different contexts, and SQL statements to prepare or check the results.
At the end of scripts, internal sequences are often reset, so that a single function call insertion does not produce impacts in the next scripts results.
The psql test scripts must be maintained in the same time as the extension source.
The expected results¶
For each psql script, the test tool produces a result file. These files are distinguished from a PostgreSQL version to another. They are located in the test/<PostgreSQL_version>/results directory.
At the end of a run, the test tool compares these files with a reference located into the test/<PostgreSQL_version>/expected directory.
Unlike for files in the test/<PostgreSQL_version>/results directory, files in the test/<PostgreSQL_version>/expected directory belong to the git repository. They must always remain consistent with the source of the extension and the psql test scripts.
The test tool¶
The test tool, regress.sh, combines all test functions.
Before using it, it is necessary to:
- have the PostgrSQL instances to be used already created and the tools/emaj_tools.env file already setup,
- manually create the test/<PostgreSQL_version>/results directories.
The test tool can be launched with the command:
As it starts with a copy of the emaj.control file into the SHAREDIR/extension directory of each configured PostgreSQL version, it may ask for the password of the Linux account to be able to execute sudo commands.
It then displays the list of test functions in a menu. Just enter the letter corresponding to the choosen test.
The test functions are:
- standart tests for each configured PostgreSQL version,
- the tests with the installation of the previous version followed by an upgrade,
- the tests with an E-Maj version upgrade while tables groups are in logging state,
- tests chaining a database save with pg_dump and a restore, with different PostgreSQL versions,
- a PostgreSQL upgrade version test using pg_upgrade with a database containing the E-Maj extension.
It is important to execute the four first sets of tests for each E-Maj change.
After having executed a psql script, regress.sh compares the outputs of the run with the expected outputs and reports the comparison result with the words ok or FAILED.
Here is an example of the display issued by the test tool (in this case with the scenario chaining the installation and a version upgrade, and with a detected difference):
Run regression test ============== dropping database "regression" ============== DROP DATABASE ============== creating database "regression" ============== CREATE DATABASE ALTER DATABASE ============== running regression test queries ============== test install_upgrade ... ok test setup ... ok test create_drop ... ok test start_stop ... ok test mark ... ok test rollback ... ok test misc ... ok test alter ... ok test alter_logging ... ok test viewer ... ok test adm1 ... ok test adm2 ... ok test client ... ok test check ... FAILED test cleanup ... ok ======================= 1 of 15 tests failed. ======================= The differences that caused some tests to fail can be viewed in the file "/home/postgres/proj/emaj/test/11/regression.diffs". A copy of the test summary that you see above is saved in the file "/home/postgres/proj/emaj/test/11/regression.out".
When at least one script fails, it is important to closely analyze the differences, by reviewing the test/<PostgreSQL_version>/regression.diff file content, and check that the differences are directly linked to changes applied in the extension source code or in the test scripts.
Once the reported differences are considered as valid, the content of the test/<PostgreSQL_version>/result directories must be copied into the test/<PostgreSQL_version>/expected directories. A shell script processes all PostgreSQL versions in a single command:
It may happen that some test outputs do not match the expected outputs, due to differences in the PostgreSQL behaviour from one run to another. Repeating the test allows to check these cases.
Functions test coverage¶
The PostgreSQL test instances are configured to count the functions executions. The check.sql test script displays the functions execution counters. It also displays E-Maj functions that have not been executed.
Error messages test coverage¶
A perl script extracts error and warning messages coded in the sql/emaj- -devel.sql file. It then extracts the messages from the files of the test/10/expected directory. It finally displays error or warning messages that are not covered by tests.
The script can be run with the command:
Some messages are known to not be covered by tests (for instance internal errors that are hard to reproduce). These messages, coded in the perl script, are excluded from the final report.
Evaluate the performances¶
The tools/performance directory contains some shell scripts helping in measuring performances. As the measurement results totally depend on the platform and the environment used, no reference results are supplied.
The scripts cover the following domains:
- log_overhead/pgbench.sh evaluates the log mechanism overhead, using pgbench,
- large_group/large_group.sh evaluates the behaviour of groups containing a large number of tables,
- rollback/rollback_perf.sh evaluates the E-Maj rollback performances with different tables profiles.
For all these files, some variables have to be configured at the begining of the scripts.
A LibreOffice format documentation is managed by the maintainers. It has its own github reporistory: emaj_doc. Thus the doc directory of the main repository remains empty.
The online documentation is managed by sphinx. It is located in the docs directory.
To install sphinx, refer to the docs/README.rst file.
The documentation exists in two languages, English and French. Depending on the languages, document sources are located in /docs/en and /docs/fr. These documents are in ReStructured Text format.
To compile the documentation for a language, set the current directory to docs/<language> and execute the command:
When there is no compilation error anymore, the documentation becomes available locally on a brower, by opening the docs/<language>/_build/html/index.html file.
The documentation on the readthedocs.org site is automatically updated as soon as the main github repository is updated.
Submitting a patch¶
Patches can be proposed to the E-Maj maintainers through Pull Requests on the github site.
Before submitting a patch, it may be useful to create an issue on github, in order to start a discussion with the maintainers and help in working on the patch.