Before you start here, read INSTALL.Unix.md (or INSTALL.Windows.md) and follow the setup instructions including the installation of all the listed dependencies for your system.
Only follow these instructions if you are building from a source checkout.
If you're unsure what this means, ignore this document.
You need the following to run tests:
You need the following optionally to build documentation:
You need the following optionally to build releases:
You need the following optionally to build Fauxton:
You will need these optional dependencies installed if:
- You are working on the documentation, or
- You are preparing a distribution archive
However, you do not need them if:
- You are building from a distribution archive, or
- You don't care about building the documentation
If you intend to build Fauxton, you will also need to install its
dependencies. After running ./configure to download all of the
dependent repositories, you can read about required dependencies in
its readme. Typically,
installing npm and node.js are sufficient to enable a Fauxton
build.
Here is a list of optional dependencies for various operating systems. Installation will be easiest, when you install them all.
CouchDB maintains a Dockerfile based on Debian that includes all
the dependencies noted above in the .devcontainer
folder.
The Dockerfile can be used on its own, or together with the
associated devcontainer.json file to quickly provision a
development environment using GitHub Codespaces
or Visual Studio Code.
If you already have VS Code and Docker installed, you can click the badge above or here to get started. Clicking these links will cause VS Code to automatically install the Remote - Containers extension if needed, clone the source code into a container volume, and spin up a dev container for use.
This devcontainer will automatically run ./configure && make
the first time it is created. While this may take some extra time to
spin up, this tradeoff means you will be able to run things like
./dev/run, ./dev/run --admin=admin:admin, ./dev/run
--with-admin-party-please, and make check straight away.
Subsequent startups should be quick.
sudo apt-get install help2man python-sphinx gnupg nodejs npm \
python3 python3-venv
sudo emerge gnupg coreutils pkgconfig help2man sphinx python sudo pip install hypothesis requests nose
sudo yum install help2man python-sphinx python-docutils \
python-pygments gnupg nodejs npm
Install Homebrew, if you do not have it already.
Unless you want to install the optional dependencies, skip to the next section.
Install what else we can with Homebrew:
brew install help2man gnupg md5sha1sum node python elixir
If you don't already have pip installed, install it:
sudo easy_install pip
Now, install the required Python packages:
sudo pip install sphinx docutils pygments sphinx_rtd_theme
pkg install help2man gnupg py27-sphinx node pip install nose requests hypothesis
Follow the instructions in INSTALL.Windows.md and build all components from source, using the same Visual C++ compiler and runtime.
Configure the source by running:
./configure
If you intend to run the test suites with Clouseau:
./configure --with-clouseau
If you don't want to build Fauxton or documentation specify
--disable-fauxton and/or --disable-docs arguments for configure to
ignore their build and avoid any issues with their dependencies.
See ./configure --help for more information.
The erl files in src are formatted using erlfmt. The checks are run
for every PR in the CI. To run the checks locally, run make erlfmt-check.
To format the erl files in src, run make erlfmt-format.
To use erlfmt for specific files only, use the executable bin/erlfmt
that is installed by configure.
Python files throughout the repository should conform to (PEP
8-compliant) formatting rules as specified by black. Similarly to
erlfmt, the related checks are run for every PR in the CI. The
same checks could also be run locally via make python-black.
Files can be automatically formatted by make python-black-update.
To run all the tests use run:
make check
You can also run each test suite individually via the eunit, mango-test,
elixir, and weatherreport-test targets:
make eunit make mango-test make elixir make weatherreport-test
If you need to run specific Erlang tests, you can pass special "options" to make targets:
# Run tests only for couch and chttpd apps make eunit apps=couch,chttpd # Run only tests from couch_btree_tests suite make eunit apps=couch suites=couch_btree # Run only only specific tests make eunit tests=btree_open_test,reductions_test # Ignore tests for specified apps make eunit skip_deps=couch_log,couch_epi
The apps, suites, tests and skip_deps could be
combined in any way. These are mimics to rebar eunit arguments. If
you're not satisfied by these, you can use the EUNIT_OPTS variable
to specify exact rebar eunit options:
make eunit EUNIT_OPTS="apps=couch,chttpd"
If you have GNU make (gmake) installed, you can supply a -jN parameter to run as many jobs in parallel. To avoid clobbering output, use the --output-sync option.
gmake eunit -j2 --output-sync=target
This runs two test modules in parallel at a time. On a machine with two or more CPUs, this should reduce the total wall clock time of testing to up to one half.
Higher N might also work, but the test suite is still undergoing hardening to support full parallelism, so spurious errors that do not occur with -j1 might show up.
All the Elixir-based integration tests could be by the elixir target:
make elixir
There is an additional suite for Dreyfus, which is not run
automatically by either the elixir or the check target
but it could be done manually via the corresponding target:
make elixir-search
Note that this requires Clouseau to be configured for running, see above.
Tests for the Mango interface can be run individually with the help of
the mango-test target and they can be narrowed down to specific
test suites via the MANGO_TEST_OPTS variable:
make mango-test \ MANGO_TEST_OPTS="--pretty-assert --verbose 03-operator-test"
The value of the MANGO_TEST_OPTS variable will be passed down to
the Nose 2 testing framework which is used for
the implementation. Consult its documentation for more information.
Tests that rely on text indexes are run only if the search feature
is reported to be available (i.e. a working Clouseau instance is
configured and working), otherwise they will be skipped.
Note that the databases that are created during the tests will be all
removed after each of the suites completed. However, with the help of
the MANGO_TESTS_KEEP_DBS environment variable, it can be requested
to keep those databases around for further investigation:
MANGO_TESTS_KEEP_DBS=please \ make mango-test MANGO_TEST_OPTS='03-operator-test'
When configured with the ./configure script, the ./dev/run
script is capable of launching Clouseau instances alongside the
CouchDB nodes and hooking them up. This is what the mango-test
and elixir-search targets also use to run their respective test
suites, and let Clouseau automatically manage them.
Although the ./configure and the ./dev/run scripts try to take
care of the details of the Clouseau deployment, it is still the
responsibility of the user to provide a suitable Java environment for
running. Depending on the version, Clouseau can run with either JRE
1.7 and 1.8 only, when 2.x is selected, or JRE 21 for 3.x. Also, when
Nouveau is installed, which might require a different Java
environment, the one specific for Clouseau has to be installed
separately and the CLOUSEAU_JAVA_HOME environment variable has to
be set to point its location.
Fortunately, the `asdf tool <https://asdf-vm.com/>` provides a
convenient way to install different versions of JDK through its
`java plugin <https://github.com/halcyon/asdf-java>`:
asdf plugin add java
Then use asdf to install it, for example:
asdf install java openjdk-21
Finally, use asdf to set the CLOUSEAU_JAVA_HOME environment
variable:
export CLOUSEAU_JAVA_HOME=$(asdf where java openjdk-21)
If the use of asdf is not an option, any standard Java installer
would do it, and the Zulu site <https://cdn.azul.com/zulu/bin/>` could
be used directly to get the distribution package for the older JRE and
JDK versions, for example.
Once both Clouseau and the corresponding Java environment are set,
they are not put in use automatically. In order to do so, the
./dev/run script needs to be run with Clouseau enabled as
follows:
dev/run --with-clouseau
When a specific Erlang cookie string is needed, the
--erlang-cookie flag could be used to configure CouchDB and
Clouseau to work with that:
dev/run --with-clouseau --erlang-cookie=brumbrum
It is possible to override Clouseau's location per invocation of
./dev/run in case some other version needs to be exercised for the
moment. This can be done with the help of the --clouseau-dir
flag. The specified location could be either an unpacked bundle of
JAR files or a git clone of the Clouseau source code repository:
dev/run --with-clouseau --clouseau-dir $HOME/git/clouseau.wip
Through the CLOUSEAU_DIR variable the same could be forwarded to the
respective test targets, e.g. mango-test:
make mango-test CLOUSEAU_DIR=$HOME/git/clouseau.wip
This can even be done if there was no local Clouseau deployment
configured previously. Mind that this will require building Clouseau
from source, which causes the nodes start up somewhat slower. It also
requires JDK 11 (or later) and SBT 1.10 (or later) to be present.
That is why it is important the set the CLOUSEAU_JAVA_HOME and the
CLOUSEAU_SBT_HOME environment variables accordingly, for
instance:
asdf install java openjdk-21 asdf plugin add sbt asdf install sbt 1.11.7 export CLOUSEAU_JAVA_HOME=$(asdf where java openjdk-21) export CLOUSEAU_SBT_HOME=$(asdf where sbt 1.11.7)
Code analyzer could be run by:
make dialyze
If you need to analyze only specific apps, you can specify them in familiar way
make dialyze apps=couch,couch_epi
See make help for more info and useful commands.
Please report any problems to the developer's mailing list.
The release procedure is documented here:
https://cwiki.apache.org/confluence/display/COUCHDB/Release+Procedure
A release tarball can be built by running:
make dist
An Erlang CouchDB release includes the full Erlang Run Time System and all dependent applications necessary to run CouchDB, standalone. The release created is completely relocatable on the file system, and is the recommended way to distribute binaries of CouchDB. A release can be built by running:
make release
The release can then be found in the rel/couchdb directory.
The release tarball and Erlang CouchDB release commands work on
Microsoft Windows the same as they do on Unix-like systems. To create
a full installer, the separate couchdb-glazier repository is required. Full
instructions are available in that repository's README file.