Skip to main content


Project description

Adhesive is a micro BPMN runner written in Python.

You can easily model complex logic in BPMN, and Adhesive will execute it for you taking care of parallelism, joining, etc. using the standard BPMN notation.

Since it’s small, it can easily be embedded in containers, or replace complex scripts.


pip install adhesive

Getting Started

Simple Builds

To create a basic build you just create a file in your project named In it you then declare some tasks. For example:

import adhesive

@adhesive.task("Checkout Code")
def checkout_code(context):

@adhesive.task("Run Build")
def run_build(context):"mvn clean install")

Since no process was defined, adhesive takes the defined tasks, stitches them in order, and has a process defined as <start>Checkout CodeRun Build<end>.

To run it simply call adhesive in the terminal:


This is the equivalent of Jenkins stages. But we can do better:

Programmatic Builds

In order to use the full programmatic functionalities that adhesive offers, you are able to stitch your BPM process manually. You have sub processes, branching and looping available:

import adhesive
import uuid

@adhesive.task("Run in parallel item {loop.value}")
def context_to_run(context):
    if not = set()

data = adhesive.process_start()\
        .sub_process_start() \
            .task("Run in parallel",
                  loop="items") \
    .branch_end() \
    .branch_start() \
        .sub_process_start() \
            .task("Run in parallel item {loop.value}",
                  loop="items") \
        .sub_process_end() \
    .branch_end() \
    .build(initial_data={"items": [1, 2, 3, 4, 5]})

assert len(data.executions) == 10

Here you see the full BPMN power starting to unfold. We create a process that branches out, creates sub processes (sub processes can be looped as a single unit). Loops are creating execution tokens that also run in parallel in the same pool.

Note that you can pass initial_data into the process, and you can also get the from the last execution token.

BPMN Process

Last but not least, adhesive reads BPMN files, and builds the process graph from them. This is particularly good if the process is complex and has a lot of dependencies:

BPMN Editor

The build of adhesive is modeled as a BPMN process itself, so we load it from the file directly using: adhesive.build_bpmn("adhesive-self.bpmn")

import adhesive

@adhesive.task("Read Parameters")
def read_parameters(context) -> None: = False = True

@adhesive.task(re=r"^Ensure Tooling:\s+(.+)$")
def gbs_ensure_tooling(context, tool_name) -> None:
    ge_tooling.ensure_tooling(context, tool_name)

# ...


As you see steps are parametrizable, and use the data from the task name into the step definition.

Defining BPMN Tasks

For example here, we define an implementation of tasks using regex matching, and extracting values:

@adhesive.task(re=r"^Ensure Tooling:\s+(.+)$")
def gbs_ensure_tooling(context, tool_name) -> None:
    # ...

Or a user task (interactive form):

@adhesive.usertask('Publish to PyPI?')
def publish_to_pypi_confirm(context, ui):
            ("nexus", "Publish to Nexus"),
            ("pypitest", "Publish to PyPI Test"),
            ("pypi", "Publish to PyPI"),
        value=("pypitest", "pypi")

Don’t forget, the @adhesive.task and @adhesive.usertask are just defining mappings for implementations of the task names available in the process. Only the creates a linear process out of the declaration of the tasks.

As you notice, there’s always a first parameter named context. The context parameter contains the following information:

  1. task - the Task in the graph that’s currently matched against this execution.

  2. task_name - The resolved name, with the variables interpolated. Matching is attempted after the name is resolved.

  3. data - Data that the current execution token contains. This data is always cloned across executions, and `set`s and `dict`s are automatically merged if multiple execution tokens are merged. So you have a modifiable copy of the data that you’re allowed to change, and is propagated into the following execution tokens.

  4. loop - if the current task is in a loop, the entry contains its index, the key and value of the items that are iterating, and the expression that was evaluated. Note that loop execution happens in parallel since these are simple execution tokens.

  5. lane - the current lane where the tasks belongs. Implicitly it’s default.

  6. workspace - a way to interact with a system, and execute commands, create files, etc.

adhesive runs all the tasks on a parallel process pool for better performance. This happens automatically.

The tasks perform the actual work for the build. But in order to have that, we need to be able to execute commands, and create files. For that we have the workspace.

Start Event Messages

Adhesive supports also start events with messages in the process. Each message start event, is being processed in its own thread and yield results:

@adhesive.message('Generate Event')
def message_generate_event(context):
    for i in range(10):
        yield i

@adhesive.task('Process Event')
def process_event(context):
    print(f"event data: {}")

Each yield generates a new event that fires up the connected tasks. The data yielded is present in the event attribute in the token, for the following tasks.

Callback Messages

The other option to push messages into a process is to use callback messages:

@adhesive.message_callback('REST: /rest/process-resource')
def message_rest_rest_process_resource(context, callback):
    def create_resource():
            "type": "CREATE"

        return "Create event fired"

Using this we’re able to hook into other systems that have their own loop, such as in this case the Flask server, and push messages using the callback. This approach has also the advantage of not creating new threads for each message endpoint.


Tasks are linked using connections. In some cases, connections can have conditions. Conditions are expressions that when evaluated to True will allow the token to pass the connection. In the connection there is access to the task, task_name, data, loop, lane and context, as well as the variables defined in the

So if in a task there is defined a data field such as:

@adhesive.task('prepare data')
def prepare_data(context): = "forward"

The navigation_direction can be validated in the condition with any of the following:

  • == "forward"

  • data.navigation_direction == "forward"

  • navigation_direction == "forward"


Workspaces are just a way of interacting with a system, running commands, and writing/reading files. Currently there’s support for:

  • the local system

  • docker containers

  • kubernetes

  • remote SSH connections

When starting adhesive allocates a default workspace folder in the configured temp location (implicitly /tmp/adhesive). The Workspace API is an API that allows you to run commands, and create files, taking care of redirecting outputs, and even escaping the commands to be able to easily run them inside docker containers.

The workspace is available from the cotext directly from the context, by calling context.workspace.

For example calling…​) will run the command on the host where adhesive is running:

@adhesive.task("Run Maven")
def build_project(context) -> None:"mvn clean install")

If we’re interested in the program output we simply do a run with a capture_stdout that returns the output as a string:

def gbs_test_linux(context) -> None:
    content ="echo yay", capture_stdout=True)
    assert content == "yay"

or we can use the simplified call with run_output that guarantees a str as result, unlike the Optional[str] for run:

def gbs_test_linux(context) -> None:
    content = context.workspace.run_output("echo yay")
    assert content == "yay"

The run commands implicitly use /bin/sh, but a custom shell can be specified by passing the shell argument:

content = context.workspace.run_output("echo yay", shell="/bin/bash")

Docker Workspace

To create a docker workspace that runs inside a container with the tooling you just need to:

from adhesive.workspace import docker

Then to spin up a container that has the current folder mounted in, where you’re able to execute commands inside the container. You just need to:

def gbs_test_linux(context) -> None:
    image_name = 'some-custom-python'

    with docker.inside(context.workspace, image_name) as w:"python -m pytest -n 4")

This creates a container using our current context workspace, where we simply execute what we want, using the run() method. After the with statement the container will be teared down automatically.

SSH Workspace

In order to have ssh, make sure you installed adhesive with SSH support:

pip install -U adhesive[ssh]

To have a SSH Workspace, it’s again the same approach:

from adhesive.workspace import ssh

Then to connect to a host, you can just use the ssh.inside the same way like in the docker sample:

@adhesive.task("Run over SSH")
def run_over_ssh(context) -> None:
    with ssh.inside(context.workspace,
                    key_fileaname="/home/raptor/.ssh/id_rsa") as s:"python -m pytest -n 4")

The parameters are being passed to paramiko, that’s the implementation beneath the SshWorkspace.

Kubernetes Workspace

To run things in pods, it’s the same approach:

from adhesive.workspace import kube

Then we can create a workspace to run things in kubernetes pods. The workspace, as well as the API, will use the kubectl command internally.

@adhesive.task("Run things in the pod")
def run_in_the_pod(context) -> None:
    with kube.inside(context.workspace,
                     pod_name="nginx-container") as pod:"ps x")  # This runs in the pod

Kubernetes API

Adhesive also packs a kubernetes api, that’s available on the adhesive.kubeapi:

from adhesive.kubeapi import KubeApi

To use it, we need to create an instance against a workspace.

@adhesive.gateway('Determine action')
def determine_action(context):
    kubeapi = KubeApi(context.workspace,

Let’s create a namespace:


Or let’s create a service using the kubectl apply approach:

    apiVersion: v1
    kind: Service
        name: nginx-http
            app: {}
        type: ClusterIP
        - port: 80
          protocol: TCP
          name: http
          app: {}

Or let’s get some pods:

pod_definitions = kubeapi.getall(

These returns objects that allow navigating properties as regular python attributes:

new_pods = dict()
for pod in pod_definitions:
    if not
        raise Exception(f"Wrong definition {pod}")

    new_pods[] = pod.status.phase

You can also navigate properties that are not existing yet, for example to wait for the status of a pod to appear:

@adhesive.task('Wait For Pod Creation {loop.key}')
def wait_for_pod_creation_loop_value_(context):
    kubeapi = KubeApi(context.workspace,
    pod_name = context.loop.key
    pod_status = context.loop.value

    while pod_status != 'Running':
        pod = kubeapi.get(kind="pod", name=pod_name)

        pod_status = pod.status.phase

To get the actual data from the wrappers that the adhesive API creates, you can simply call the _raw property.

Workspace API

Here’s the full API for it:

class Workspace(ABC):
    A workspace is a place where work can be done. That means a writable
    folder is being allocated, that might be cleaned up at the end of the

    def write_file(
            file_name: str,
            content: str) -> None:

    def run(self,
            command: str,
            capture_stdout: bool = False) -> Union[str, None]:
        Run a new command in the current workspace.

        :param capture_stdout:
        :param command:

    def rm(self, path: Optional[str]=None) -> None:
        Recursively remove the file or folder given as path. If no path is sent,
        the whole workspace will be cleared.

        :param path:

    def mkdir(self, path: str=None) -> None:
        Create a folder, including all its needed parents.

        :param path:

    def copy_to_agent(self,
                      from_path: str,
                      to_path: str) -> None:
        Copy the files to the agent from the current disk.
        :param from_path:
        :param to_path:

    def copy_from_agent(self,
                        from_path: str,
                        to_path: str) -> None:
        Copy the files from the agent to the current disk.
        :param from_path:
        :param to_path:

    def temp_folder(self):
        Create a temporary folder in the current `pwd` that will be deleted
        when the `with` block ends.


    def chdir(self, target_folder: str):
        Temporarily change a folder, that will go back to the original `pwd`
        when the `with` block ends. To change the folder for the workspace
        permanently, simply assing the `pwd`.
        :param target_folder:

User Tasks

In order to create user interactions, you have user tasks. These define form elements that are populated in the, and available in subsequent tasks.

When a user task is encountered in the process flow, the user is prompted to fill in the parameters. Note that the other started tasks continue running, proceeding forward with the build.

The name used in the method call defines the name of the variable that’s in the

For example in here we define a checkbox group that allows us to pick where to publish the package:

@adhesive.usertask("Read User Data")
def read_user_data(context, ui) -> None:
            values=["admin", "cyborg", "anonymous"])
    ui.add_radio_group("disabled",  # title is optional
            values=["yes", "no"],
            values=(("any", "<any>"),
                    ("win", "Windows"),
                    ("lin", "Linux")))

This will prompt the user with this form:


This data is also available for edge conditions, so in the BPMN modeler we can define a condition such as "pypi" in, or since data is also available in the edge scope: "pypi" in data.roles.

The other option is simply reading what the user has selected in a following task:

@adhesive.task("Register User")
def publish_items(context):
    for role in
        # ...

User tasks support the following API, available on the ui parameter, the parameter after the context:

class UiBuilderApi(ABC):
    def add_input_text(self,
                       name: str,
                       title: Optional[str] = None,
                       value: str = '') -> None:

    def add_input_password(self,
                           name: str,
                           title: Optional[str] = None,
                           value: str = '') -> None:

    def add_combobox(self,
                     name: str,
                     title: Optional[str] = None,
                     value: Optional[str]=None,
                     values: Optional[Iterable[Union[Tuple[str, str], str]]]=None) -> None:

    def add_checkbox_group(
            name: str,
            title: Optional[str]=None,
            value: Optional[Iterable[str]]=None,
            values: Optional[Iterable[Union[Tuple[str, str], str]]]=None) -> None:

    def add_radio_group(self,
                        name: str,
                        title: Optional[str]=None,
                        value: Optional[str]=None,
                        values: Optional[List[Any]]=None) -> None:

    def add_default_button(self,
                           name: str,
                           title: Optional[str] = None,
                           value: Optional[Any] = True) -> None:

Custom Buttons

In order to allow navigation inside the process, the add_default_button API exists to permit creation of buttons. Implicitly a single button with an OK label is added to the User Task, that when pressed fills the in the outgoing execution token.

With add_default_button we create custom buttons such as Back and Forward, or whatever we need in our process. Unlike the default OK button, when these are called, they also set in the the value that’s assigned to them. This value we use then further in a Gateway, or simple as a condition on the outgoing edges.

The title is optional, and only if missing it’s build either from the name if all the buttons in the form have unique names, since they assign a different variable in the, or from the value if they have overlapping names.


Secrets are files that contain sensitive information are not checked in the project. In order to make them available to the build, we need to define them in either ~/.adhesive/secrets/SECRET_NAME or in the current folder as .adhesive/secrets/SECRET_NAME.

In order to make them available, we just use the secret function that creates the file in the current workspace and deletes it when exiting. For example here’s how we’re doing the actual publish, creating the secret inside a docker container:

@adhesive.task('^PyPI publish to (.+?)$')
def publish_to_pypi(context, registry):
    with docker.inside(context.workspace, as w:
        with secret(w, "PYPIRC_RELEASE_FILE", "/germanium/.pypirc"):
  "python bdist_wheel upload -r {registry}")

Note the docker.inside that creates a different workspace.


Adhesive supports configuration via its config files, or environment variables. The values are read in the following order:

  1. environment variables: ADHESIVE_XYZ, then

  2. values that are in the project config yml file: .adhesive/config.yml, then

  3. values configured in the global config yml file: $HOME/.adhesive/config.yml.

Currently the following values are defined for configuration:


default value /tmp/adhesive, environment var: ADHESIVE_TEMP_FOLDER.

Is where all the build files will be stored.


default value [], environment var: ADHESIVE_PLUGINS_LIST.

This contains a list of folders, that will be added to the sys.path. So to create a reusable plugin that will be reused by multiple builds, you need to simply create a folder with python files, then point to it in the ~/.adhesive/config.yml:

- /path/to/folder

Then in the python path you can simply do regular imports.


default value True, environment var: ADHESIVE_COLOR.

Marks if the logging should use ANSI colors in the terminal. Implicitly this is true, but if log parsing is needed, it can make sense to have it false.


default_value info, environment var: ADHESIVE_LOG_LEVEL.

How verbose should the logging be on the terminal. Possible values are trace, debug, info, warning, error and critical.


default value is empty, environment var: ADHESIVE_POOL_SIZE.

Sets the number of workers that adhesive will use. Defaults to the number of CPUs if unset.


default value is empty, environment var: ADHESIVE_STDOUT.

Implicitly for each task, the log is redirected in a different file, and only shown if the task failed. The redirection can be disabled.


default value is thread, environment var: ADHESIVE_PARALLEL_PROCESSING.

Implicitly tasks are scaled using multiple threads in order to alleviate waits for I/O. This is useful for times when remote ssh workspaces are defined in the lanes, so the same connection can be reused for multiple tasks.

This value can be set to process, in case the tasks are CPU intensive. This has the drawback of recreating the connections on workspaces’ each task execution.

Hacking Adhesive

Adhesive builds with itself. In order to do that, you need to checkout the adhesive-lib shared plugin, and configure your local config to use it:

- /path/to/adhesive-lib

Then simply run the build using adhesive itself:


Project details

Release history Release notifications | RSS feed

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distributions

No source distribution files available for this release.See tutorial on generating distribution archives.

Built Distribution

adhesive-2021.4.3-py3-none-any.whl (131.6 kB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page