Python Execution and Testing Environment

The directory tests_python contains:

  • A scripting API to write execution scenarios involving several node, bakers, endorsers,

  • a system testing environment based on the pytest package.

Code organization

It contains the following python packages.

  • daemons defines classes to run Tezos node and daemons,

  • client mainly defines the Client class, that provides a programmatic interface to a client,

  • codec defines a Codec` class, that provides a interface for tezos-codec binary,

  • launcher defines classes used to launch a nodes and daemons with specific settings,

  • tools contains utility functions and constants shared by the tests,

  • examples contains example of tests and scripts that run scenarios of interactions between tezos nodes and clients,

  • tests contains pytest tests,

  • scripts contains utility scripts.

They are organized in three layers.

  1. daemons, client and codec,

  2. launchers,

  3. tests, examples, tools.



  • A working environment (see documentation) with the binaries compiled,

  • A local copy of the tezos repository

  • python 3.8.2. It is recommended to use pyenv to manage the python versions. If pyenv is used, you can use pyenv install 3.8.2 followed by pyenv global 3.8.2 to set the python version to 3.8.2 globally. If you want to use python 3.8.2 only in the current shell, you can use pyenv shell 3.8.2. Be sure eval $(pyenv init -) has been executed first during the shell session.

  • poetry to manage the python dependencies and run the tests in a sandboxed python environment. All poetry commands are to be run in tests_python. Before running the tests for the first time, the dependencies must be installed. To achieve this, run poetry install.

Examples of test executions:

poetry run pytest examples/  # simple test example
poetry run pytest -m "not slow"  # run all tests not marked as slow
poetry run pytest -s tests/  # run a specific test with traces
poetry run pytest  # run all tests

A simple sandbox scenario

The following example runs a couple of nodes and performs a transfer operation.

import time
from tools import constants, paths, utils
from launchers.sandbox import Sandbox

def scenario():
    """ a private tezos network, initialized with network parameters
        and some accounts. """
    with Sandbox(paths.TEZOS_HOME,
                constants.GENESIS_PK) as sandbox:
        # Launch node running protocol Alpha
        # Launch a second node on the same private tezos network
        # Launch a baker associated to node 0, baking on behalf of delegate
        # bootstrap5
        sandbox.add_baker(0, 'bootstrap5', proto=constants.ALPHA_DAEMON)
        # first client tells node 0 to transfer money for an account to another
        # receipt is an object representing the client answer
        receipt = sandbox.client(0).transfer(500, 'bootstrap1', 'bootstrap3')
        transfer_hash = receipt.operation_hash
        # Wait for second node to update its protocol to Alpha, if not
        # it may not know yet the ``wait_for_inclusion`` operation which is
        # protocol specific
        # second client waits for inclusion of operation by the second node

if __name__ == "__main__":

This can be run with PYTHONPATH=./:$PYTHONPATH poetry run python examples/ It should display all the clients commands and their results.

The sandbox object allows users to add nodes, bakers or endorsers running in tezos sandboxed mode. Whenever a node has been added, one can access it using a client object.

The client object is a wrapper on the tezos-client command. It runs tezos-client with “administrative” parameters, plus the parameters determined by the method called by the user.

For instance

receipt = client.transfer(500, 'bootstrap1', 'bootstrap3')

will run something like

tezos-client -base-dir /tmp/tezos-client.be22ya16 -addr -port 18730 transfer 500 from bootstrap1 to bootstrap3

receipt is an object of type client_output.TransferResult which gives access to some data of the tezos-client output.

Alternatively, one can always construct the command manually:

client_output =['transfer', '500', 'from', 'bootstrap1', 'bootstrap3'])

In that case, client_output is the string returned by the client, such as

Node is bootstrapped, ready for injecting operations.
Estimated gas: 10100 units (will add 100 for safety)
Estimated storage: no bytes added
Operation successfully injected in the node.
Operation hash is 'op9K2VJjKJLaFnfQKzsoz9rzr5v1PrLjpefiPtVhuiiXYgkZes1'

The first method is more convenient and less error prone. But the second method is more generic and sometimes the only option if the specialized method isn’t implemented.

Test suite and pytest

Tests are located in the tests directory and rely on the pytest library.

Tests are divided into modules, and are furthermore subdivided into classes. A class defines a full testing scenario. A typical scenario is a sequence of client commands and assertions, operating on a set of Tezos nodes running in a private network (a.k.a sandbox mode).

Running tests

Useful options

pytest has a variety of launching options. Convenient options include:

  • -v display test names,

  • -x stop at first failure,

  • -s display output, including commands launched and stdout from client (by default, pytest captures all passing test output and show failed tests output),

  • --tb=short, --tb=long, --tb=no, set size of python trace back in case of failure. Default is long and is too verbose in most case. The python trace back is useful to detect bugs in the python scripts,

  • --log-dir=<dir> saves all servers log in the given dir (CREATE <DIR> FIRST).

  • -x --pdb, start python debugger at first failure, this allows interacting with the node in the same context of the test,

  • -m TAGS_EXPR, run all tests containing some combination of tags.

-v and --tb=short are set by default in pytest initialization file.


Tests can be classified with tags. Tags are added with the annotation


The configuration file pytest.ini defines the list of allowed tags. It includes vote, multinode, baker, endorser, contract, slow, multibranch.


There are typically two ways of using pytest:

  • run a subset of the tests (batch mode),

  • or run a specific test.

In batch mode, we usually don’t care about traces. No particular option is needed, but sometimes we want to stop at first failure using -x, and some tests require the server logs to be saved (--log-dir=tmp/) as they check some assertions in the logs at some point in the test.

To run a specific test, we usually want client and server traces (-s --log-dir=tmp/).

# Launch a simple test without capturing stdout
> poetry run pytest -s examples/
# run all tests about vote
> poetry run pytest -m "vote"
# run all vote and non-slow tests
> poetry run pytest -m "vote and not slow"
# run module, display all output, save server logs in tmp
> poetry run pytest -s tests/ --log-dir=tmp
# run all tests using a daemon
> poetry run pytest -m "endorser or baker"
# run everything
> poetry run pytest

Pre-commit hook

The pre-commit hook located in scripts/pre_commit/ executes modified python tests automatically. It looks for staged files (the default) or modified files (if --unstaged is passed) in tests_python/tests and calls pytest on those files. This avoids pushing commits that will break the CI. It is also handy to execute the relevant subset of tests by calling ./scripts/pre_commit/ [--unstaged] manually.

We refer to the header of and its --help flag for additional instructions.

Anatomy of a test

A typical testing scenario consists in:

  1. initializing the context (starting servers, setting up clients)

  2. running a sequence of commands and assertions

  3. releasing resources, terminating servers

This is done by grouping tests in a class, and managing the context in a fixture.

The following is the pytest counterpart of the first example.

import pytest
from tools import constants, paths, utils
from launchers.sandbox import Sandbox

def sandbox():
    """Example of sandbox fixture."""
    with Sandbox(paths.TEZOS_HOME,
                 constants.GENESIS_PK) as sandbox:
        sandbox.add_node(0, params=constants.NODE_PARAMS)
        sandbox.add_node(1, params=constants.NODE_PARAMS)
        sandbox.add_baker(0, 'bootstrap5', proto=constants.ALPHA_DAEMON)
        yield sandbox
        assert sandbox.are_daemons_alive()

def session():
    """Example of dictionary fixture. Used for keeping data between tests."""
    yield {}

class TestExample:

    def test_wait_sync_proto(self, sandbox, session):
        session['head_hash'] = sandbox.client(0).get_head()['hash']
        clients = sandbox.all_clients()
        for client in clients:
            proto = constants.ALPHA
            assert utils.check_protocol(client, proto)

    def test_transfer(self, sandbox, session):
        receipt = sandbox.client(0).transfer(500, 'bootstrap1', 'bootstrap3')
        session['operation_hash'] = receipt.operation_hash

    def test_inclusion(self, sandbox, session):
        operation_hash = session['operation_hash']

In this example, we defined the fixtures in the same module, but they are generally shared between tests and put in

Currently, all tests scenarios in the test suite are defined as classes, consisting of a sequence of methods that are run incrementally (as specified with the annotation @pytest.mark.incremental). Classes are used to define the scope of a fixture, and a unit of incremental testing sequence. We don’t directly instantiate them, or use self.

Data between methods are shared using a dictionary session. For instance, we save the result of the transfer operation, and retrieve it in the next method.


The list of fixtures available is given by

poetry run pytest --fixtures

Most fixtures are defined in The most general fixture is sandbox. It allows to instantiate an arbitrary number of nodes and daemons. Other fixtures, such as client, are specialized versions (slightly more convenient than using sandbox directly). Fixtures can be defined directly in a module defining a test, or they can be shared.

Skipping tests

Sometimes, a test can’t be run. For instance, it is known to fail, or it relies on some resources that may not be available. In that case, the test can be skipped (instead of failing).

For instance, if no log dir has been specified, the test_check_logs tests are skipped using pytest.skip().

def test_check_logs(self, sandbox):
        if not sandbox.log_dir:

Alternatively, one can use the skip annotation:

@pytest.mark.skip(reason="Not yet implemented")

Adding a test

  • By imitation, choose an existing test that looks similar,

  • use the proper tags,

  • say briefly what the test is supposed to test in the class docstring,

  • Run the linters and typechecker make lint_all, and make typecheck in tests_python/, or simple make test-python-lint from the Tezos home directory. Note that linting and typechecking are enforced by the CI in the build stage.

  • If you modify the API (launchers or daemons), make sure you maintain the layers structure. API shouldn’t rely testing constants (tools/ or tools/

Testing on a production branch (zeronet, mainnet,…)

On master, protocol Alpha is named ProtoALphaALphaALphaALphaALphaALphaALphaALphaDdp3zK, and daemons binary name are suffixed with alpha (tezos-baker-alpha, tezos-endorser-alpha…). However, on production branches, an actual hash of the protocol is used, and a shortened string is used to specify daemons.

For instance, on revision 816625bed0983f7201e4c369440a910f006beb1a of zeronet, protocol Alpha is named PsddFKi32cMJ2qPjf43Qv5GDWLDPZb3T3bF6fLKiF5HtvHNU7aP and daemons are suffixed by 003-PsddFKi3 (tezos-baker-003-PsddFKi3).

To reduce coupling between tests and the actual branch to be tested, tests refer to protocol Alpha using constants.ALPHA and constants.ALPHA_DAEMON rather than by hard-coded identifiers.

Tests based on fixed revisions (multibranch)

It is useful to test interactions between different server versions. There are currently two ways of doing this.

1. The Sandbox launcher can use binaries built from different revisions. Methods add_node, add_baker and add_endorser have an optional parameter branch that points to a subdirectory where binaries are to be looked for.

2. The SandboxMultibranch launcher is instantiated by map from ids to branches. Then every time we launch a node or a daemon the actual binary will be selected according to the map.

Tests using specific revisions are in tests/multibranch and aren’t run by default. They are not regression tests and are usually launched separately from the rest of the tests. To run these tests, you need to set up the TEZOS_BINARIES environment variable to a directory that contains the binaries for all revisions needed by test (see below). The tests will be skipped if this variable isn’t set, and fail if the binaries aren’t available.

Building binaries for several revisions

Before running the tests, the user has to build the binaries and copy them to the right location. This can be done by the scripts/ script.

For instance, suppose we want to build binaries for two different revisions of zeronet:

A = b8de4297db6a681eb13343d2773c6840969a5537
B = 816625bed0983f7201e4c369440a910f006beb1a

TEZOS_HOME=~/tezos  # TEZOS repo, read-only access from the script
TEZOS_BINARIES=~/tezos-binaries  # where the binaries will be stored
TEZOS_BUILD=~/tmp/tezos_tmp  # where the binaries will be built

The following command will generate binaries for each of the specified branches in TEZOS_BINARIES.

scripts/ --clone $TEZOS_HOME --build-dir $TEZOS_BUILD \
                        --bin-dir $TEZOS_BINARIES \
                        b8de4297db6a681eb13343d2773c6840969a5537 \

tezos-accuser-003-PsddFKi3  tezos-baker-004-Pt24m4xi    tezos-node
tezos-accuser-004-Pt24m4xi  tezos-client                tezos-protocol-compiler
tezos-admin-client          tezos-endorser-003-PsddFKi3 tezos-signer
tezos-baker-003-PsddFKi3    tezos-endorser-004-Pt24m4xi

tezos-accuser-003-PsddFKi3  tezos-baker-004-Pt24m4xi    tezos-node
tezos-accuser-004-Pt24m4xi  tezos-client                tezos-protocol-compiler
tezos-admin-client          tezos-endorser-003-PsddFKi3 tezos-signer
tezos-baker-003-PsddFKi3    tezos-endorser-004-Pt24m4xi

Note: One can specify a branch instead of a revision but this is error-prone. For instance, protocols may have different hashes on different revisions on the same branch, and these hashes are typically hard-coded in the tests to activate the protocols.

Example 1:

The test uses two different revisions.

the sandbox_multibranch fixtures (which uses the SandboxMultibranch launcher) parameterized by a map that alternates between the two revisions.

The executables will be selected from revisions A and B as specified by:

A = "d272059bf474018d0c39f5a6e60634a95f0c44aa" # MAINNET
B = "6718e80254d4cb8d7ad86bce8cf3cb692550c6e7"  # MAINNET SNAPSHOT
MAP = {i:A if i % 2 == 0 else B  for i in range(20)}
@pytest.mark.parametrize('sandbox_multibranch', [MAP], indirect=True)

Run the test with

# mkdir tmp
poetry run pytest tests/multibranch/ --log-dir=tmp

Example 2: A full voting scenario

This tests uses binaries from revision b8de4297db6a681eb13343d2773c6840969a5537 and implements a full voting scenario (voting, launching a test chain and a test chain baker, upgrading to a new protocol, performing operations on the new protocol). It uses two protocols implemented by this specific revision,

ALPHA = 'PsddFKi32cMJ2qPjf43Qv5GDWLDPZb3T3bF6fLKiF5HtvHNU7aP'
NEW_PROTO = 'Pt24m4xiPbLDhVgVfABUjirbmda3yohdN82Sp9FeuAXJ4eV9otd'

as well the corresponding bakers tezos-baker-003-PsddFKi3 tezos-baker-004-Pt24m4xi.

scripts/ --clone $TEZOS_HOME --build-dir $TEZOS_BUILD \
    --bin-dir $TEZOS_BINARIES \ b8de4297db6a681eb13343d2773c6840969a5537

It can be run with

poetry run pytest tests/multibranch/`

Note: this test uses only one revision but it can’t run on branch master as we need an extra protocol with bakers.

Regression testing

Some tests in the test suite are regression tests. Regression testing is a coarse-grained testing method for detecting unintended changes in the system under test. In addition to standard assertions, a regression test compares the “output” of the test to a stored test log. The regression test fails if the output and the stored test log do not match. We apply regression testing using the pytest-regtest plugin.

To simplify the writing of regression tests, we provide a specialized version of the client fixture, client_regtest. It registers all output of the tezos-client.

Output conversion

The output of the client might differ slightly from one test run to another, for instance due to timestamps. A specialized fixture client_regtest_scrubbed applies a series of conversions to the output. For example, a timestamp such as 2019-09-23T10:59:00Z is replaced by [TIMESTAMP]. These conversions are defined in the function client_output_converter of

Running regression tests

Regression tests are run during normal tests runs.

Updating regression tests

The test logs are stored in tests_python/tests/_regtest_outputs/. If the logs need to be updated, pass --regtest-reset to pytest:

poetry run pytest --regtest-reset <test-file>

The resulting changes should be committed after thoroughly verifying that they are as expected.

Writing regression tests

To write regression tests targeting the tezos-client, write a test as usual, but request the client_regtest (or client_regtest_scrubbed to enable output conversion) fixture instead of the client fixture.

In this example test, we test the output of the hash data command of tezos-client:

class TestDemonstrateRegtest:
    """Tests demonstrating regression testing."""

    def test_hash_regtest(self, client_regtest):
        assert client_regtest.hash('(Pair 1 "foo")', '(pair nat string)').blake2b == \

Before running the test we must generate the test log, that contains the expected output. This is done by passing the –regtest-reset flag as described above:

$ poetry run pytest --regtest-reset tests_python/tests/

We find the generated test log in tests_python/tests/_regtest_outputs/test_regtest.TestDemonstrateRegtest\:\:test_hash_regtest.out:

$ cat tests_python/tests/_regtest_outputs/test_regtest.TestDemonstrateRegtest\:\:test_hash_regtest.out
Raw packed data: 0x05070700010100000003666f6f
Script-expression-ID-Hash: exprvPNUJQXpct6VrbJQCazrDgh7pN8d8SH8P1UFHMrRPmQnxC16nr
Raw Script-expression-ID-Hash: 0xf65884dadd3a5ff1a6f8057fa442a2e8ecdbe1217f7759512509b36c016c5bce
Ledger Blake2b hash: Hadaf2hW4QwbgTdhtAfFTofrCbmnnPhkGy2Sa5ZneUDs
Raw Sha256 hash: 0xb01925b6b6180a31a17f74d92ac87e551ab08e1890211741abde5345b38cb61f
Raw Sha512 hash: 0x75547d33aca115154e5a0ec22e965237ec3c32a81b64f827668bbef3b3310d8c237ae06211ee63edf743fcf0a98a970bb159782c6b75fac42d6efc20b3fa5e82
Gas remaining: 799862 units remaining

This is exactly the output of the command that was executed by the test, namely tezos-client hash data '(Pair 1 "foo")' of type '(pair nat string)'.

As discussed below in the section Pitfalls to regression testing, regression tests cannot be put in a test class where the normal client fixture is used.

For other aspects of regression testing, we refer to the pytest-regtest documentation.

Typechecking python code

We also enforce the types on the python codebase. We use mypy, a typechecker for python. Code can be typechecked using the Makefile target make typecheck. It is also enforced in the CI with the job check_python_types.

Pitfalls to regression testing

The client and the client_regtest fixtures cannot be used in the same test class. If they are, then two nodes will be added to the sandbox. Their interference might cause unintended consequence disturbing the tests.


There are few simple possible improvements.

  • Many client methods and client_output classes haven’t been implemented yet,

  • Be more consistent in the use of retries, timeout, to make tests less sensitive on timing assumption,

  • Implement new launchers (i.e. zeronet),

  • Use parametric fixtures more consistently: one can relaunch the same tests, with different parameters such as the number of peers,

  • Finish porting bash scripts,

Known issues

  • On rare occasions, some servers may not be properly killed upon test termination,

  • One some occasions, the timeout marker doesn’t play well with blocking client commands. for instance, this may not stop the test if wait_for_inclusion is stuck.

def test_inclusion(self, sandbox, session):
    operation_hash = session['operation_hash']

The thread methods terminates the test but the resources aren’t properly cleaned up.

@pytest.mark.timeout(5, method='thread')

See discussion here.

To avoid this issue, one can use polling functions such as utils.check_contains_operations(client, [op_hash]) instead of using blocking commands.

Adding new dependencies

Dependencies are managed by poetry in the file pyproject.toml. See here. The file poetry.lock is generated by running poetry lock, and must never be changed manually. The resulting poetry.lock and its generator pyproject.toml must be copied in this repository.