Skip to main content

Wraps the execution of processes so that a service API endpoint (CloudReactor) can monitor and manage them. Also implements retries, timeouts, and secret injection from AWS into the environment.

Project description


CI Status Documentation Status Test coverage percentage

Poetry pre-commit

PyPI Version Supported Python versions License

Wraps the execution of processes so that an API server (CloudReactor) can monitor and manage them. Available as a standalone executable or as a python module.


  • Runs either processes started with a command line or a python function you supply
  • Implements retries and time limits
  • Injects secrets from AWS Secrets Manager, AWS S3, or local files and extracts them into the process environment (for command-lines) or configuration (for functions)
  • When used with the CloudReactor service:
    • Reports when a process/function starts and when it exits, along with the exit code and runtime metadata (if running in AWS ECS, AWS Lambda, or AWS CodeBuild)
    • Sends heartbeats, optionally with status information like the number of items processed
    • Prevents too many concurrent executions
    • Stops execution when manually stopped in the CloudReactor dashboard
    • Sends CloudReactor the data necessary to start the process / function if running in AWS ECS, AWS Lambda, or AWS CodeBuild

How it works

First, secrets and other configuration are fetched and resolved from providers like AWS Secrets Manager, AWS S3, or the local filesystem.

Just before your code runs, the module requests the API server to create a Task Execution associated with the Task name or UUID which you pass to the module. The API server may reject the request if too many instances of the Task are currently running, but otherwise records that a Task Execution has started. The module then passes control to your code.

While your code is running, it may report progress to the API server, and the API server may signal that your Task stop execution (due to user manually stopping the Task Execution), in which case the module terminates your code and exits.

After your code finishes, the module informs the API server of the exit code or result. CloudReactor monitors Tasks to ensure they are still responsive, and keeps a history of the Executions of Tasks, allowing you to view failures and run durations in the past.

Auto-created Tasks

Before your Task is run (including this module), the AWS ECS CloudReactor Deployer can be used to set it up in AWS ECS, and inform CloudReactor of details of your Task. That way CloudReactor can start and schedule your Task, and setup your Task as a service. See CloudReactor python ECS QuickStart for an example.

However, it may not be possible or desired to change your deployment process. Instead, you may configure the Task to be auto-created.

Auto-created Tasks are created the first time your Task runs. This means there is no need to inform the API server of the Task details (during deployment) before it runs. Instead, each time the module runs, it informs the API server of the Task details at the same time as it requests the creation of a Task Execution. One disadvantage of auto-created Tasks is that they are not available in the CloudReactor dashboard until the first time they run.

When configuring a Task to be auto-created, you must specify the name or UUID of the Run Environment in CloudReactor that the Task is associated with. The Run Environment must be created ahead of time, either by the Cloudreactor AWS Setup Wizard, or manually in the CloudReactor dashboard.

You can also specify more Task properties, such as Alert Methods and external links in the dashboard, by setting the environment variable PROC_WRAPPER_AUTO_CREATE_TASK_PROPS set to a JSON-encoded object that has the CloudReactor Task schema.

Execution Methods

CloudReactor currently supports four Execution Methods:

  1. AWS ECS (in Fargate)
  2. AWS Lambda
  3. AWS CodeBuild
  4. Unknown

If a Task is running in AWS ECS, CloudReactor is able to run additional Task Executions, provided the details of running the Task is provided during deployment with the AWS ECS CloudReactor Deployer, or if the Task is configured to be auto-created, and this module is run. In the second case, this module uses the ECS Metadata endpoint to detect the ECS Task settings, and sends them to the API server. CloudReactor can also schedule Tasks or setup long-running services using Tasks, provided they are run in AWS ECS.

If a Task is running in AWS Lambda, CloudReactor is able to run additional Task Executions after the first run of the function.

However, a Task may use the Unknown execution method if it is not running in AWS ECS, AWS Lambda, or AWS CodeBuild. If that is the case, CloudReactor won't be able to start the Task in the dashboard or as part of a Workflow, schedule the Task, or setup a service with the Task. But the advantage is that the Task code can be executed by any method available to you, such as bare metal servers, VM's, or Kubernetes. All Tasks in CloudReactor, regardless of execution method, have their history kept and are monitored.

This module detects the execution method your Task is running with and sends that information to the API server, provided you configure your Task to be auto-created.

Passive Tasks

Passive Tasks are Tasks that CloudReactor does not manage. This means scheduling and service setup must be handled by other means (cron jobs, supervisord, etc). However, Tasks marked as services or that have a schedule will still be monitored by CloudReactor, which will send notifications if a service Task goes down or a Task does not run on schedule.

The module reports to the API server that auto-created Tasks are passive, unless you specify the --force-task-passive commmand-line option or set the environment variable PROC_WRAPPER_TASK_IS_PASSIVE to FALSE. If a Task uses the Unknown Execution Method, it must be marked as passive, because CloudReactor does not know how to manage it.


If you just want to use this module to retry processes, limit execution time, or fetch secrets, you can use offline mode, in which case no CloudReactor API key is required. But CloudReactor offers a free tier so we hope you sign up for a free account to enable monitoring and/or management.

If you want CloudReactor to be able to start your Tasks, you should use the Cloudreactor AWS Setup Wizard to configure your AWS environment to run Tasks in ECS Fargate. You can skip this step if running in passive mode is OK for you.

If you want to use CloudReactor to manage or just monitor your Tasks, you need to create a Run Environment and an API key in the CloudReactor dashboard. The API key can be scoped to the Run Environment if you wish. The key must have at least the Task access level, but for an auto-created Task, it must have at least the Developer access level.



Standalone executables built by PyInstaller for 64-bit Linux and Windows are available, located in bin/pyinstaller. These executables bundle python so you don't need to have python installed on your machine. They also bundle all optional extras so you can fetch secrets from AWS Secrets Manager and extract them with jsonpath-ng, for example.

RHEL or derivatives

To download and run the wrapper on a RHEL/Fedora/Amazon Linux 2 machine:

RUN wget -nv
ENTRYPOINT ["proc_wrapper.bin"]

Example Dockerfiles of known working environments are available for Amazon Linux 2 and Fedora.

Fedora 27 or later are supported.

Debian based machines

On a Debian based (including Ubuntu) machine:

RUN wget -nv
ENTRYPOINT ["proc_wrapper.bin"]

See the example Dockerfile for a known working Debian environment.

Debian 10 (Buster) or later are supported.

Special thanks to wine and PyInstaller Docker Images for making it possible to cross-compile!

When python is available

Install this module via pip (or your favorite package manager):

pip install cloudreactor-procwrapper

cloudreactor-procwrapper doesn't have any required dependencies, but it can be installed with the following extras:

aws: Support for fetching secrets from AWS Secrets Manager or S3, and determining the assumed role, implemented by the boto3 library

jsonpath: Support for secret resolution using JSON Path, implemented by the jsonpath-ng library

yaml: Support for configuration files in YAML format, implemented by the pyyaml library

dotenv: Support for environment variables defined in the dotenv format, implemented by the dotenv library

mergedeep: Support for secret object value merging using alternative strategies, implemented by the mergedeep library

Use brackets after cloudreactor-procwrapper to enable support for the desired functionality. For example, to install AWS support, JSON Path secret resolution, and support for dotenv files:

pip install cloudreactor-procwrapper[aws,jsonpath,dotenv]

Or, to install support for everything:

pip install cloudreactor-procwrapper[allextras]


There are two ways of using the module: wrapped mode and embedded mode.

Wrapped mode

In wrapped mode, you pass a command line to the module which it executes in a child process. The command can be implemented in whatever programming language the running machine supports.

Instead of running

somecommand --somearg x

you would run

./proc_wrapper somecommand --somearg x

assuming that are using the PyInstaller standalone executable, and that you configure the program using environment variables.

Or, if you have python installed:

python -m proc_wrapper somecommand --somearg x

Here are all the options:

usage: proc_wrapper [-h] [-v] [-n TASK_NAME] [--task-uuid TASK_UUID] [-a] [--auto-create-task-run-environment-name AUTO_CREATE_TASK_RUN_ENVIRONMENT_NAME]
                    [--auto-create-task-run-environment-uuid AUTO_CREATE_TASK_RUN_ENVIRONMENT_UUID] [--auto-create-task-props AUTO_CREATE_TASK_PROPS]
                    [--force-task-active] [--task-execution-uuid TASK_EXECUTION_UUID] [--task-version-number TASK_VERSION_NUMBER]
                    [--task-version-text TASK_VERSION_TEXT] [--task-version-signature TASK_VERSION_SIGNATURE]
                    [--build-task-execution-uuid BUILD_TASK_EXECUTION_UUID] [--deployment-task-execution-uuid DEPLOYMENT_TASK_EXECUTION_UUID]
                    [--execution-method-props EXECUTION_METHOD_PROPS] [--task-instance-metadata TASK_INSTANCE_METADATA] [-s] [--schedule SCHEDULE]
                    [--max-concurrency MAX_CONCURRENCY] [--max-conflicting-age MAX_CONFLICTING_AGE] [--api-base-url API_BASE_URL] [-k API_KEY]
                    [--api-heartbeat-interval API_HEARTBEAT_INTERVAL] [--api-error-timeout API_ERROR_TIMEOUT]
                    [--api-final-update-timeout API_FINAL_UPDATE_TIMEOUT] [--api-retry-delay API_RETRY_DELAY] [--api-resume-delay API_RESUME_DELAY]
                    [--api-task-execution-creation-error-timeout API_TASK_EXECUTION_CREATION_ERROR_TIMEOUT]
                    [--api-task-execution-creation-conflict-timeout API_TASK_EXECUTION_CREATION_CONFLICT_TIMEOUT]
                    [--api-task-execution-creation-conflict-retry-delay API_TASK_EXECUTION_CREATION_CONFLICT_RETRY_DELAY]
                    [--api-request-timeout API_REQUEST_TIMEOUT] [-o] [-p] [-d DEPLOYMENT] [--send-pid] [--send-hostname] [--no-send-runtime-metadata]
                    [--runtime-metadata-refresh-interval RUNTIME_METADATA_REFRESH_INTERVAL] [-l {DEBUG,INFO,WARNING,ERROR,CRITICAL}] [--log-secrets]
                    [--exclude-timestamps-in-log] [-w WORK_DIR] [-c COMMAND_LINE] [--shell-mode {auto,enable,disable}] [--no-strip-shell-wrapping]
                    [--no-process-group-termination] [-t PROCESS_TIMEOUT] [-r PROCESS_MAX_RETRIES] [--process-retry-delay PROCESS_RETRY_DELAY]
                    [--process-check-interval PROCESS_CHECK_INTERVAL] [--process-termination-grace-period PROCESS_TERMINATION_GRACE_PERIOD]
                    [--enable-status-update-listener] [--status-update-socket-port STATUS_UPDATE_SOCKET_PORT]
                    [--status-update-message-max-bytes STATUS_UPDATE_MESSAGE_MAX_BYTES] [--status-update-interval STATUS_UPDATE_INTERVAL] [-e ENV_LOCATIONS]
                    [--overwrite-env-during-resolution] [--config-ttl CONFIG_TTL] [--no-fail-fast-config-resolution]
                    [--resolved-env-var-name-prefix RESOLVED_ENV_VAR_NAME_PREFIX] [--resolved-env-var-name-suffix RESOLVED_ENV_VAR_NAME_SUFFIX]
                    [--resolved-config-property-name-prefix RESOLVED_CONFIG_PROPERTY_NAME_PREFIX]
                    [--resolved-config-property-name-suffix RESOLVED_CONFIG_PROPERTY_NAME_SUFFIX] [--env-var-name-for-config ENV_VAR_NAME_FOR_CONFIG]
                    [--config-property-name-for-env CONFIG_PROPERTY_NAME_FOR_ENV] [--main-container-name MAIN_CONTAINER_NAME]
                    [--monitor-container-name MONITOR_CONTAINER_NAME] [--sidecar-container-mode] [--rollbar-access-token ROLLBAR_ACCESS_TOKEN]
                    [--rollbar-retries ROLLBAR_RETRIES] [--rollbar-retry-delay ROLLBAR_RETRY_DELAY] [--rollbar-timeout ROLLBAR_TIMEOUT]

Wraps the execution of processes so that a service API endpoint (CloudReactor) is optionally informed of the progress. Also implements retries, timeouts, and
secret injection into the environment.

positional arguments:

optional arguments:
  -h, --help            show this help message and exit
  -v, --version         Print the version and exit

  Task settings

  -n TASK_NAME, --task-name TASK_NAME
                        Name of Task (either the Task Name or the Task UUID must be specified
  --task-uuid TASK_UUID
                        UUID of Task (either the Task Name or the Task UUID must be specified)
  -a, --auto-create-task
                        Create the Task even if not known by the API server
  --auto-create-task-run-environment-name AUTO_CREATE_TASK_RUN_ENVIRONMENT_NAME
                        Name of the Run Environment to use if auto-creating the Task (either the name or UUID of the Run Environment must be specified if auto-
                        creating the Task). Defaults to the deployment name if the Run Environment UUID is not specified.
  --auto-create-task-run-environment-uuid AUTO_CREATE_TASK_RUN_ENVIRONMENT_UUID
                        UUID of the Run Environment to use if auto-creating the Task (either the name or UUID of the Run Environment must be specified if auto-
                        creating the Task)
  --auto-create-task-props AUTO_CREATE_TASK_PROPS
                        Additional properties of the auto-created Task, in JSON format. See for the
  --force-task-active   Indicates that the auto-created Task should be scheduled and made a service by the API server, if applicable. Otherwise, auto-created
                        Tasks are marked passive.
  --task-execution-uuid TASK_EXECUTION_UUID
                        UUID of Task Execution to attach to
  --task-version-number TASK_VERSION_NUMBER
                        Numeric version of the Task's source code
  --task-version-text TASK_VERSION_TEXT
                        Human readable version of the Task's source code
  --task-version-signature TASK_VERSION_SIGNATURE
                        Version signature of the Task's source code (such as a git commit hash)
  --build-task-execution-uuid BUILD_TASK_EXECUTION_UUID
                        UUID of Task Execution that built this Task's source code
  --deployment-task-execution-uuid DEPLOYMENT_TASK_EXECUTION_UUID
                        UUID of Task Execution that deployed this Task to the Runtime Environment
  --execution-method-props EXECUTION_METHOD_PROPS
                        Additional properties of the execution method, in JSON format. See
               for the schema.
  --task-instance-metadata TASK_INSTANCE_METADATA
                        Additional metadata about the Task instance, in JSON format
  -s, --service         Indicate that this is a Task that should run indefinitely
  --schedule SCHEDULE   Run schedule reported to the API server
  --max-concurrency MAX_CONCURRENCY
                        Maximum number of concurrent Task Executions of the same Task. Defaults to 1.
  --max-conflicting-age MAX_CONFLICTING_AGE
                        Maximum age of conflicting Tasks to consider, in seconds. -1 means no limit. Defaults to the heartbeat interval, plus 60 seconds for
                        services that send heartbeats. Otherwise, defaults to no limit.

  API client settings

  --api-base-url API_BASE_URL
                        Base URL of API server. Defaults to
  -k API_KEY, --api-key API_KEY
                        API key. Must have at least the Task access level, or Developer access level for auto-created Tasks.
  --api-heartbeat-interval API_HEARTBEAT_INTERVAL
                        Number of seconds to wait between sending heartbeats to the API server. -1 means to not send heartbeats. Defaults to 30 for concurrency
                        limited services, 300 otherwise.
  --api-error-timeout API_ERROR_TIMEOUT
                        Number of seconds to wait while receiving recoverable errors from the API server. Defaults to 300.
  --api-final-update-timeout API_FINAL_UPDATE_TIMEOUT
                        Number of seconds to wait while receiving recoverable errors from the API server when sending the final update before exiting. Defaults to
  --api-retry-delay API_RETRY_DELAY
                        Number of seconds to wait before retrying an API request. Defaults to 120.
  --api-resume-delay API_RESUME_DELAY
                        Number of seconds to wait before resuming API requests, after retries are exhausted. Defaults to 600. -1 means to never resume.
  --api-task-execution-creation-error-timeout API_TASK_EXECUTION_CREATION_ERROR_TIMEOUT
                        Number of seconds to keep retrying Task Execution creation while receiving error responses from the API server. -1 means to keep trying
                        indefinitely. Defaults to 300.
  --api-task-execution-creation-conflict-timeout API_TASK_EXECUTION_CREATION_CONFLICT_TIMEOUT
                        Number of seconds to keep retrying Task Execution creation while conflict is detected by the API server. -1 means to keep trying
                        indefinitely. Defaults to 1800 for concurrency limited services, 0 otherwise.
  --api-task-execution-creation-conflict-retry-delay API_TASK_EXECUTION_CREATION_CONFLICT_RETRY_DELAY
                        Number of seconds between attempts to retry Task Execution creation after conflict is detected. Defaults to 60 for concurrency-limited
                        services, 120 otherwise.
  --api-request-timeout API_REQUEST_TIMEOUT
                        Timeout for contacting API server, in seconds. Defaults to 30.
  -o, --offline-mode    Do not communicate with or rely on an API server
  -p, --prevent-offline-execution
                        Do not start processes if the API server is unavailable or the wrapper is misconfigured.
  -d DEPLOYMENT, --deployment DEPLOYMENT
                        Deployment name (production, staging, etc.)
  --send-pid            Send the process ID to the API server
  --send-hostname       Send the hostname to the API server
                        Do not send metadata about the runtime environment
  --runtime-metadata-refresh-interval RUNTIME_METADATA_REFRESH_INTERVAL
                        Refresh interval for runtime metadata, in seconds. The default value depends on the execution method.

  Logging settings

                        Log level
  --log-secrets         Log sensitive information
                        Exclude timestamps in log (possibly because the log stream will be enriched by timestamps automatically by a logging service like AWS
                        CloudWatch Logs)

  Process settings

  -w WORK_DIR, --work-dir WORK_DIR
                        Working directory. Defaults to the current directory.
  -c COMMAND_LINE, --command-line COMMAND_LINE
                        Command line to execute
  --shell-mode {auto,enable,disable}
                        Indicates if the process command should be executed in a shell. Executing in a shell allows shell scripts, commands, and expressions to be
                        used, with the disadvantage that termination signals may not be propagated to child processes. Options are: enable -- Force the command to
                        be executed in a shell; disable -- Force the command to be executed without a shell; auto -- Auto-detect the shell mode by analyzing the
                        Do not strip the command-line of shell wrapping like "/bin/sh -c" that can be introduced by Docker when using shell form of ENTRYPOINT and
                        Send termination and kill signals to the wrapped process only, instead of its process group (which is the default). Sending to the process
                        group allows all child processes to receive the signals, even if the wrapped process does not forward signals. However, if your wrapped
                        process manually handles and forwards signals to its child processes, you probably want to send signals to only your wrapped process.
                        Timeout for process completion, in seconds. -1 means no timeout, which is the default.
                        Maximum number of times to retry failed processes. -1 means to retry forever. Defaults to 0.
  --process-retry-delay PROCESS_RETRY_DELAY
                        Number of seconds to wait before retrying a process. Defaults to 60.
  --process-check-interval PROCESS_CHECK_INTERVAL
                        Number of seconds to wait between checking the status of processes. Defaults to 10.
  --process-termination-grace-period PROCESS_TERMINATION_GRACE_PERIOD
                        Number of seconds to wait after sending SIGTERM to a process, but before killing it with SIGKILL. Defaults to 30.

  Status update settings

                        Listen for status updates from the process, sent on the status socket port via UDP. If not specified, status update messages will not be
  --status-update-socket-port STATUS_UPDATE_SOCKET_PORT
                        The port used to receive status updates from the process. Defaults to 2373.
  --status-update-message-max-bytes STATUS_UPDATE_MESSAGE_MAX_BYTES
                        The maximum number of bytes status update messages can be. Defaults to 65536.
  --status-update-interval STATUS_UPDATE_INTERVAL
                        Minimum of number of seconds to wait between sending status updates to the API server. -1 means to not send status updates except with
                        heartbeats. Defaults to -1.

  Environment/configuration resolution settings

                        Location of either local file, AWS S3 ARN, or AWS Secrets Manager ARN containing properties used to populate the environment for embedded
                        mode, or the process environment for wrapped mode. By default, the file format is assumed to be dotenv. Specify multiple times to include
                        multiple locations.
                        Location of either local file, AWS S3 ARN, or AWS Secrets Manager ARN containing properties used to populate the configuration for
                        embedded mode. By default, the file format is assumed to be in JSON. Specify multiple times to include multiple locations.
                        Merge strategy for merging configurations. Defaults to 'DEEP', which does not require mergedeep. Besides the 'SHALLOW' strategy, all other
                        strategies require the mergedeep extra to be installed.
                        Overwrite existing environment variables when resolving them
  --config-ttl CONFIG_TTL
                        Number of seconds to cache resolved environment variables and configuration properties instead of refreshing them when a process restarts.
                        -1 means to never refresh. Defaults to -1.
                        Continue execution even if an error occurs resolving the configuration
  --resolved-env-var-name-prefix RESOLVED_ENV_VAR_NAME_PREFIX
                        Required prefix for names of environment variables that should resolved. The prefix will be removed in the resolved variable name.
                        Defaults to ''.
  --resolved-env-var-name-suffix RESOLVED_ENV_VAR_NAME_SUFFIX
                        Required suffix for names of environment variables that should resolved. The suffix will be removed in the resolved variable name.
                        Defaults to '_FOR_PROC_WRAPPER_TO_RESOLVE'.
  --resolved-config-property-name-prefix RESOLVED_CONFIG_PROPERTY_NAME_PREFIX
                        Required prefix for names of configuration properties that should resolved. The prefix will be removed in the resolved property name.
                        Defaults to ''.
  --resolved-config-property-name-suffix RESOLVED_CONFIG_PROPERTY_NAME_SUFFIX
                        Required suffix for names of configuration properties that should resolved. The suffix will be removed in the resolved property name.
                        Defaults to '__to_resolve'.
  --env-var-name-for-config ENV_VAR_NAME_FOR_CONFIG
                        The name of the environment variable used to set to the value of the JSON encoded configuration. Defaults to not setting any environment
  --config-property-name-for-env CONFIG_PROPERTY_NAME_FOR_ENV
                        The name of the configuration property used to set to the value of the JSON encoded environment. Defaults to not setting any property.

  Container settings

  --main-container-name MAIN_CONTAINER_NAME
                        The name of the container that is monitored
  --monitor-container-name MONITOR_CONTAINER_NAME
                        The name of the container that will monitor the main container
                        Indicates that the current container is a sidecar container that will monitor the main container

  Rollbar settings

  --rollbar-access-token ROLLBAR_ACCESS_TOKEN
                        Access token for Rollbar (used to report error when communicating with API server)
  --rollbar-retries ROLLBAR_RETRIES
                        Number of retries per Rollbar request. Defaults to 2.
  --rollbar-retry-delay ROLLBAR_RETRY_DELAY
                        Number of seconds to wait before retrying a Rollbar request. Defaults to 120.
  --rollbar-timeout ROLLBAR_TIMEOUT
                        Timeout for contacting Rollbar server, in seconds. Defaults to 30.

These environment variables take precedence over command-line arguments:

  • PROC_WRAPPER_TASK_MAX_CONCURRENCY (set to -1 to indicate no limit)
  • PROC_WRAPPER_ENV_LOCATIONS (comma-separated list of locations)
  • PROC_WRAPPER_CONFIG_LOCATIONS (comma-separated list of locations)

With the exception of the settings for Secret Fetching and Resolution, these environment variables are read after Secret Fetching so that they can come from secret values.

The command is executed with the same environment that the wrapper script gets, except that these properties are copied/overridden:


Wrapped mode is suitable for running in a shell on your own (virtual) machine or in a Docker container. It requires multi-process support, as the module runs at the same time as the command it wraps.

In AWS ECS, it is possible to run proc_wrapper in sidecar container that monitors another (main) container. In that case, proc_wrapper will report the status and runtime metadata of the main container. The advantage of running proc_wrapper in a sidecar container is that the main container doesn't need to be modified to run proc_wrapper. To configure proc_wrapper as a sidecar, set the main container name and enable sidecar container mode.

Embedded mode

You can use embedded mode to execute python code from inside a python program. Include the proc_wrapper package in your python project's dependencies. To run a task you want to be monitored:

from typing import Any, Mapping

from proc_wrapper import ProcWrapper, ProcWrapperParams

def fun(wrapper: ProcWrapper, cbdata: dict[str, int],
        config: Mapping[str, Any]) -> int:
    return cbdata['a']

# This is the function signature of a function invoked by AWS Lambda.
def entrypoint(event: Any, context: Any) -> int:
    params = ProcWrapperParams()
    params.auto_create_task = True

    # If the Task Execution is running in AWS Lambda, CloudReactor can make
    # the associated Task available to run (non-passive) in the CloudReactor
    # dashboard or by API, after the wrapper reports its first execution.
    proc_wrapper_params.task_is_passive = False

    params.task_name = 'embedded_test_production'
    params.auto_create_task_run_environment_name = 'production'

    # For example only, in the real world you would use Secret Fetching;
    # see below.
    params.api_key = 'YOUR_CLOUDREACTOR_API_KEY'

    # In an AWS Lambda environment, passing the context and event allows
    # CloudReactor to monitor and manage this Task.
    proc_wrapper = ProcWrapper(params=params, runtime_context=context,

    x = proc_wrapper.managed_call(fun, {'a': 1, 'b': 2})
    # Should print 1

    return x

This is suitable for running in single-threaded environments like AWS Lambda, or as part of a larger process that executes sub-routines that should be monitored. See cloudreactor-python-lambda-quickstart for an example project that uses proc_wrapper in a function run by AWS Lambda.

Embedded mode configuration

In embedded mode, besides setting properties of ProcWrapperParams in code, ProcWrapper can be also configured in two ways:

First, using environment variables, as in wrapped mode.

Second, using the configuration dictionary. If the configuration dictionary contains the key proc_wrapper_params and its value is a dictionary, the keys and values in the dictionary will be used to to set these attributes in ProcWrapperParams:

Key Type Mutable Uses Resolved Config
log_secrets bool No No
env_locations list[str] No No
config_locations list[str] No No
config_merge_strategy str No No
overwrite_env_during_resolution bool No No
max_config_resolution_depth int No No
max_config_resolution_iterations int No No
config_ttl int No No
fail_fast_config_resolution bool No No
resolved_env_var_name_prefix str No No
resolved_env_var_name_suffix str No No
resolved_config_property_name_prefix str No No
resolved_config_property_name_suffix str No No
schedule str No Yes
max_concurrency int No Yes
max_conflicting_age int No Yes
offline_mode bool No Yes
prevent_offline_execution bool No Yes
service bool No Yes
deployment str No Yes
api_base_url str No Yes
api_heartbeat_interval int No Yes
enable_status_listener bool No Yes
status_update_socket_port int No Yes
status_update_message_max_bytes int No Yes
status_update_interval int No Yes
log_level str No Yes
include_timestamps_in_log bool No Yes
api_key str Yes Yes
api_request_timeout int Yes Yes
api_error_timeout int Yes Yes
api_retry_delay int Yes Yes
api_resume_delay int Yes Yes
api_task_execution_creation_error_timeout int Yes Yes
api_task_execution_creation_conflict_timeout int Yes Yes
api_task_execution_creation_conflict_retry_delay int Yes Yes
process_timeout int Yes Yes
process_max_retries int Yes Yes
process_retry_delay int Yes Yes
command list[str] Yes Yes
command_line str Yes Yes
shell_mode bool Yes Yes
strip_shell_wrapping bool Yes Yes
work_dir str Yes Yes
process_termination_grace_period int Yes Yes
send_pid bool Yes Yes
send_hostname bool Yes Yes
send_runtime_metadata bool Yes Yes

Keys that are marked with "Mutable" -- "No" in the table above can be overridden when the configuration is reloaded after the config_ttl expires.

Keys that are marked as "Uses Resolved Config" -- "Yes" in the table above can come from the resolved configuration after secret resolution (see below).

Secret Fetching and Resolution

A common requirement is that deployed code / images do not contain secrets internally which could be decompiled. Instead, programs should fetch secrets from an external source in a secure manner. If your program runs in AWS, it can make use of AWS's roles that have permission to access data in Secrets Manager or S3. However, in many scenarios, having your program access AWS directly has the following disadvantages:

  1. Your program becomes coupled to AWS, so it is difficult to run locally or switch to another infrastructure provider
  2. You need to write code or use a library for each programming language you use, so secret fetching is done in a non-uniform way
  3. Writing code to merge and parse secrets from different sources is tedious

Therefore, proc_wrapper implements Secret Fetching and Resolution to solve these problems so your programs don't have to. Both usage modes can fetch secrets from AWS Secrets Manager, AWS S3, or the local filesystem, and optionally extract embedded data into the environment or a configuration dictionary. The environment is used to pass values to processes run in wrapped mode, while the configuration dictionary is passed to the callback function in embedded mode.

proc_wrapper parses secret location strings that specify the how to resolve a secret value. Each secret location string has the format:

[PROVIDER_CODE:]<Provider specific address>[!FORMAT][|JP:<JSON Path expression>]

Secret Providers

Providers indicate the raw source of the secret data. The table below lists the supported providers:

Provider Code Value Prefix Provider Example Address Required extras Notes
AWS_SM arn:aws:secretsmanager: AWS Secrets Manager arn:aws:secretsmanager:us-east-2:1234567890:secret:config aws Can also include version suffix like -PPrpY
AWS_S3 arn:aws:s3::: AWS S3 Object arn:aws:s3:::examplebucket/staging/app1/config.json aws
FILE file:// Local file file:///home/appuser/app/.env The default provider if no provider is auto-detected
ENV The process environment SOME_TOKEN The name of another environment variable
CONFIG The configuration dictionary $.db jsonpath JSON path expression to extract the data in the configuration
PLAIN Plaintext {"user": "postgres", "password": "badpassword"}

If you don't specify an explicit provider prefix in a secret location (e.g. AWS_SM:), the provider can be auto-detected from the address portion using the Value Prefix. For example the secret location arn:aws:s3:::examplebucket/staging/app1/config.json will be auto-detected to with the AWS_S3 provider because it starts with arn:aws:s3:::.

Secret Formats

Formats indicate how the raw string data is parsed into a secret value (which may be a string, number, boolean, dictionary, or array). The table below lists the supported formats:

Format Code Extensions MIME types Required extras Notes
dotenv .env None dotenv Also auto-detected if location includes .env.
json .json application/json, text/x-json
yaml .yaml, .yml application/x-yaml, application/yaml, text/vnd.yaml, text/yaml, text/x-yaml yaml pyyaml's safe_load() is used for security

The format of a secret value can be auto-detected from the extension or by the MIME type if available. Otherwise, you may need to an explicit format code (e.g. !yaml).

AWS Secrets Manager / S3 notes

boto3 is used to fetch secrets when the aws extra is installed. It will try to access to AWS Secrets Manager or S3 using environment variables AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY if they are set, or use the EC2 instance role, ECS task role, or Lambda execution role if available.

For Secrets Manager, you can also use "partial ARNs" (without the hyphened suffix) as keys. In the example above


could be used to fetch the same secret, provided there are no conflicting secret ARNs. This allows you to get the latest version of the secret.

If the secret was stored in Secrets Manager as binary, the corresponding value will be set to the Base-64 encoded value.

If you're deploying a python function using AWS Lambda, note that boto3 is already included in the available packages, so there's no need to include it (unless the bundled version isn't compatible). Also we strongly encourage you to add:


to your code if you are using proc_wrapper for secrets resolution. This prevent secrets from Secrets Manager from being leaked. For details, see this issue.

Secret Tranformation

Fetching secrets can be relatively expensive and it makes sense to group related secrets together. Therefore it is common to store dictionaries (formatted as JSON or YAML) as secrets. However, each desired environment variable or configuration property may only consist of a fragment of the dictionary. For example, given the JSON-formatted dictionary

  "username": "postgres",
  "password": "badpassword"

you may want to populate the environment variable DB_USERNAME with postgres.

To facilitate this, dictionary fragments can be extracted to individual environment variables using jsonpath-ng when the jsonpath extra is installed. To specify that a variable be extracted from a dictionary using a JSON Path expression, append |JP: followed by the JSON Path expression to the secret location string. For example, if the AWS Secrets Manager ARN


contains the dictionary above, then the secret location string


will resolve to postgres as desired.

If you do something similar to get the password from the same JSON value, proc_wrapper is smart enough to cache the fetched dictionary, so that the raw data is only fetched once.

Since JSON path expressions yield a list of results, the secrets fetcher implements the following rules to transform the list to the final value:

  1. If the list of results has a single value, that value is used as the final value, unless [*] is appended to the JSON path expression.
  2. Otherwise, the final value is the list of results

Fetching from another environment variable

In some deployment scenarios, multiple secrets can be injected into a single environment variable as a JSON encoded object. In that case, the module can extract secrets using the ENV secret source. For example, you may have arranged to have the environment variable DB_CONFIG injected with the JSON encoded value:

{ "username": "postgres", "password": "nohackme" }

Then to extract the username to the environment variable DB_USERNAME you you would add the environment variable DB_USER_FOR_PROC_WRAPPER_TO_RESOLVE set to


Secret injection into environment and configuration

Now let's use secret location strings to inject the values into the environment (for wrapped mode) and/or the the configuration dictionary (for embedded mode). proc_wrapper supports two methods of secret injection which can be combined together:

  • Top-level fetching
  • Secrets Resolution

Top-level fetching

Top-level fetching refers to fetching a dictionary that contains multiple secrets and populating the environment / configuration dictionary with it. To use top-level fetching, you specify the secret locations from which you want to fetch the secrets and the corresponding values are merged together into the environment / configuration.

To use top-level fetching in wrapped mode, populate the environment variables PROC_WRAPPER_ENV_LOCATIONS with a comma-separated list of secret locations, or use the command-line option --env-locations <secret_location> one or more times. Secret location strings passed in via PROC_WRAPPER_ENV_LOCATIONS or --env-locations will be parsed as dotenv files unless format is auto-detected or explicitly specified.

To use top-level fetching in embedded mode, set the ProcWrapperParams property config_locations to a list of secret locations. Alternatively, you can set the environment variable PROC_WRAPPER_CONFIG_LOCATIONS to a comma-separated list, and this will be picked up automatically. Secret location values will be parsed as JSON unless the format is auto-detected or explicitly specified. The config argument passed to the your callback function will contain a merged dictionary of all fetched and parsed dictionary values. For example:

def callback(wrapper: ProcWrapper, cbdata: str,
        config: Dict[str, Any]) -> str:
    return "super" + cbdata + config["username"]

def main():
    params = ProcWrapperParams()

    # Optional: you can set an initial configuration dictionary which will
    # have its values included in the final configuration unless overridden.
    params.initial_config = {
        "log_level": "DEBUG"

    # You can omit this if you set PROC_WRAPPER_CONFIG_LOCATIONS environment
    # variable to the same ARN
    params.config_locations = [
        # More secret locations can be added here, and their values will
        # be merged

    wrapper = ProcWrapper(params=params)

    # Returns "superduperpostgres"
    return wrapper.managed_call(callback, "duper")

Merging Secrets

Top-level fetching can potentially fetch multiple dictionaries which are merged together in the final environment / configuration dictionary. The default merge strategy (DEEP) merges recursively, even dictionaries in lists. The SHALLOW merge strategy just overwrites top-level keys, with later secret locations taking precedence. However, if you install the mergedeep library with the mergedeep extra, you can also set the merge strategy to one of:


so that nested lists can be appended to instead of replaced (in the case of the ADDITIVE strategies), or errors will be raised if incompatibly-typed values are merged (in the case of the TYPESAFE strategies). In wrapped mode, the merge strategy can be set with the --config-merge-strategy command-line argument or PROC_WRAPPER_CONFIG_MERGE_STRATEGY environment variable. In embedded mode, the merge strategy can be set in the config_merge_strategy string property of ProcWrapperParams.

Secret Resolution

Secret Resolution substitutes configuration or environment values that are secret location strings with the computed values of those strings. Compared to Secret Fetching, Secret Resolution is more useful when you want more control over the names of variables or when you have secret values deep inside your configuration.

In wrapped mode, if you want to set the environment variable MY_SECRET with a value fetched from AWS Secrets Manager, you would set the environment variable MY_SECRET_FOR_PROC_WRAPPER_TO_RESOLVE to a secret location string which is ARN of the secret, for example:


(The _FOR_PROC_WRAPPER_TO_RESOLVE suffix of environment variable names is removed during resolution. It can also be configured with the PROC_WRAPPER_RESOLVABLE_ENV_VAR_NAME_SUFFIX environment variable.)

In embedded mode, if you want the final configuration dictionary to look like:

  "db_username": "postgres",
  "db_password": "badpassword",

The initial configuration dictionary would look like:

  "db_username__to_resolve": "arn:aws:secretsmanager:us-east-2:1234567890:secret:db-PPrpY|JP:$.username",
  "db_password__to_resolve": "arn:aws:secretsmanager:us-east-2:1234567890:secret:db-PPrpY|JP:$.password",


(The __to_resolve suffix (with 2 underscores!) of keys is removed during resolution. It can also be configured with the resolved_config_property_name_suffix property of ProcWrapperParams.)

proc_wrapper can also resolve keys in embedded dictionaries, like:

  "db": {
    "username__to_resolve": "arn:aws:secretsmanager:us-east-2:1234567890:secret:config-PPrpY|JP:$.username",

up to a maximum depth that you can control with ProcWrapperParams.max_config_resolution_depth (which defaults to 5). That would resolve to

  "db": {
    "username": "postgres",
    "password": "badpassword"

You can also inject entire dictionaries, like:

  "db__to_resolve": "arn:aws:secretsmanager:us-east-2:1234567890:secret:config-PPrpY",

which would resolve to

  "db": {
    "username": "postgres",
    "password": "badpassword"

To enable secret resolution in wrapped mode, set environment variable PROC_WRAPPER_RESOLVE_SECRETS to TRUE. In embedded mode, secret resolution is enabled by default; set the max_config_resolution_iterations property of ProcWrapperParams to 0 to disable resolution.

Secret resolution is run multiple times so that if a resolved value contains a secret location string, it will be resolved on the next pass. By default, proc_wrapper limits the maximum number of resolution passes to 3 but you can control this with the environment variable PROC_WRAPPER_MAX_CONFIG_RESOLUTION_ITERATIONS in embedded mode, or by setting the max_config_resolution_iterations property of ProcWrapperParams in wrapped mode.

Environment Projection

During secret fetching and secret resolution, proc_wrapper internally maintains the computed environment as a dictionary which may have embedded lists and dictionaries. However, the final environment passed to the process is a flat dictionary containing only string values. So proc_wrapper converts all top-level values to strings using these rules:

  • Lists and dictionaries are converted to their JSON-encoded string value
  • Boolean values are converted to their upper-cased string representation (e.g. the string FALSE for the boolean value false)
  • The None value is converted to the empty string
  • All other values are converted using python's str() function

Secrets Refreshing

You can set a Time to Live (TTL) on the duration that secret values are cached. Caching helps reduce expensive lookups of secrets and bandwidth usage.

In wrapped mode, set the TTL of environment variables set from secret locations using the --config-ttl command-line argument or PROC_WRAPPER_CONFIG_TTL_SECONDS environment variable. If the process exits, you have configured the script to retry, and the TTL has expired since the last fetch, proc_wrapper will re-fetch the secrets and resolve them again, for the environment passed to the next invocation of your process.

In embedded mode, set the TTL of configuration dictionary values set from secret locations by setting the config_ttl property of ProcWrapperParams. If 1) your callback function raises an exception, 2) you have configured the script to retry; and 3) the TTL has expired since the last fetch, proc_wrapper will re-fetch the secrets and resolve them again, for the configuration passed to the next invocation of the callback function.

Status Updates

Status Updates in Wrapped Mode

While your process in running, you can send status updates to CloudReactor by using the StatusUpdater class. Status updates are shown in the CloudReactor dashboard and allow you to track the current progress of a Task and also how many items are being processed in multiple executions over time.

In wrapped mode, your application code would send updates to the proc_wrapper program via UDP port 2373 (configurable with the PROC_WRAPPER_STATUS_UPDATE_PORT environment variable). If your application code is in python, you can use the provided StatusUpdater class to do this:

from proc_wrapper import StatusUpdater

with StatusUpdater() as updater:
    updater.send_update(last_status_message="Starting ...")
    success_count = 0

    for i in range(100):
            success_count += 1
        except Exception:
            failed_count += 1


Status Updates in Embedded Mode

In embedded mode, your callback in python code can use the wrapper instance to send updates:

from typing import Any, Mapping

import proc_wrapper
from proc_wrapper import ProcWrapper

def fun(wrapper: ProcWrapper, cbdata: dict[str, int],
        config: Mapping[str, Any]) -> int:
    wrapper.update_status(last_status_message="Starting the fun ...")

    success_count = 0
    error_count = 0
    for i in range(100):
            success_count += 1
        except Exception:

            error_count += 1

    wrapper.update_status(last_status_message="The fun is over.")

    return cbdata["a"]

params = ProcWrapperParams()
params.auto_create_task = True
params.auto_create_task_run_environment_name = "production"
params.task_name = "embedded_test"
params.api_key = "YOUR_CLOUDREACTOR_API_KEY"

proc_wrapper = ProcWrapper(params=params)
proc_wrapper.managed_call(fun, {"a": 1, "b": 2})

Example Projects

These projects contain sample Tasks that use this library to report their execution status and results to CloudReactor


This software is dual-licensed under open source (MPL 2.0) and commercial licenses. See LICENSE for details.

Contributors ✨

Thanks goes to these wonderful people (emoji key):

Jeff Tsay

💻 📖 🚇 🚧

Mike Waldner


Bruno Alla

💻 🤔 📖

This project follows the all-contributors specification. Contributions of any kind welcome!


This package was created with Cookiecutter and the browniebroke/cookiecutter-pypackage project template.

Project details

Download files

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

Source Distribution

cloudreactor_procwrapper-5.3.2.tar.gz (88.8 kB view hashes)

Uploaded Source

Built Distribution

cloudreactor_procwrapper-5.3.2-py3-none-any.whl (74.7 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