Skip to main content
Version: 2.0.x


The data structures and operators in the zio-flow library only describe a ZIO Flow program. To execute them we need a persistent executor configured with a chosen set of dependencies. Most of the executor implementation is in the zio-flow-runtime module, with some backend-specific extra modules discussed in the backends section.

There are two main ways to execute a ZIO Flow program:

  • Embedding zio-flow-runtime in your own application, using the PersistentExecutor service
  • Use the default ZIO Flow Server which is an executable service application built on top of zio-flow-runtime

Embedding the persistent executor

To embed the ZIO Flow executor in your own application, you need to add the following dependency:

libraryDependencies += "dev.zio" %% "zio-flow-runtime" % "1.0.0-RC4"

Then you can initialize the PersistentExecutor service with the following method on its companion object:

def make(gcPeriod: Duration = 5.minutes): ZLayer[
DurableLog with
KeyValueStore with
Configuration with
OperationExecutor with
Serializer with
Deserializer, Nothing, ZFlowExecutor]

This is a ZLayer creating a ZFlowExecutor. Let's see first what this interface is capable of, then we'll see how to create the required dependencies.

The ZFlowExecutor interface has the following methods:


run submits a flow to the executor, and waits until it completes. The failure/success of the flow is represented as the failure/result of this ZIO effect.

To make this work ZIO Flow needs a Schema for both the error type and the result type. This is how the executor's result can be converted to the expected type.

The run method expects that the provided ZFlow has no input requirements (it's R parameter is Any). If you need to run a flow that requires an input, you can first use .provide on it to provide the input, and then run it.

Let's see an example, assuming that we have ZFlowExecutor in the environment of our ZIO program:

import zio._
import zio.flow._
import zio.flow.runtime._

val flow: ZFlow[Int, String, Int] = ZFlow.input[Int].flatMap { n =>
ZFlow.ifThenElse(n > 10)(ifTrue = ZFlow.succeed(1), ifFalse ="input is too low"))

def program: ZIO[ZFlowExecutor, String, Int] ="test1"), flow.provide(5))

The FlowId is a unique identifier for an execution of the flow. It is the caller's responsibility to generate a fresh identifier in case the flow has to be started from scratch. If a flow with the given identifier is already running, or it's persisted state is available, the execution of that flow will continue and the method will wait for the result of that execution.

start and poll

An alternative to run is to just start the execution of a flow with the start method, without waiting for its completion. The running workflow's state can be later queries with poll. The start method actually returns with a DurablePromise that can be directly used to wait for the result of the flow, but it is much simpler to use the poll method instead.

Another difference is that start and poll does not use the flow's error and result types. The polled result represents both failures and errors as DynamicValue values. This is a generic data type from the zio-schema library that can represent any value that has a Schema. By owning such a Schema you can convert back the dynamic value to its original type. You can learn more about how DynamicValue is used in the internals of ZIO Flow in the internals page.


When the executor is initialized, it is not running any flows, even if there were previously running flows persisted that could be resumed. To start running these persisted flows, you can call the restartAll method. This will load and immediately resume every persisted ZIO Flow program that was previously running.

If you don't want this, you can resume the flows one by one if you know their FlowId by calling run or start.


Once a flow is completed, it's last persisted state, as well as it's results is still stored in the database, so calling poll or getAll can return information about it. The delete method completely removes a completed flow from the executor's persisted state.

pause, resume and abort

A running flow can be paused with pause and resumed any time with resume. A paused flow remains paused in case the executor is restarted and restartAll is called.

A running (or paused) flow can be aborted with the abort method. An aborted flow's result is the ExecutorError.Interrupted error.


The getAll method returns a ZIO stream of FlowId and FlowStatus pairs, listing all the ZIO Flow programs the executor knows about. The FlowStatus is an enumeration with the following values:

  • Running - the flow is currently running
  • Paused - the flow is currently paused
  • Suspended - the flow is currently suspended in a transaction, waiting for a variable to change
  • Done - the flow finished running either with an error or with a success


Garbage collection in the context of a persistent executor is not about releasing items from the memory, but deleting old persistent state from the database backend. This is executed periodically by the executor, but you can also trigger it manually with the forceGarbageCollection method.


The persistent executor layer depends on the following other ZIO services:


DurableLog is a persistent event log used for waiting for variables to change or promises to be completed. Currently the only implementation is defined by the DurableLog.layer layer and it depends on an IndexedStore. IndexedStore is provided by the backend modules similar toKeyValueStore.


KeyValueStore is an interface for storing and retrieving information based on keys from a persistency solution. There are multiple implementations available in the backend modules.


The Configuration service stores the user defined configuration values, indexed by a ConfigKey, that can be accessed by the ZIO Flow programs using Remote.config. The following implementations are available:

  • Configuration.inMemory - stores the configuration in memory, useful mostly for testing. The initial state is empty.
  • Configuration.fromEnvironment - by providing a mapping from ConfigKey to system environment variable names, this implementation provides configuration values for ZIO Flow programs directly from environment variables
  • Configuration.fromConfig - uses ZIO's native configuration API to read configuration values from a given subsection of the provided configuration


The OperationExecutor service is responsible for executing Operatotions, described on the activities page.

There are two built-in OperatorExecutor implementations in ZIO Flow:

  • DefaultOperationExecutor.layer constructs an executor that supports the Operation.HTTP operations, implemented using the zio-http library
  • MockedOperationExecutor is useful for testing flow, more information about it can be found on the testing page

The default operation executor requires you to provide HTTP retry policy configuration, that describes how each HTTP requests are handling errors. More information about these policies can be found in the server configuration section below.

Serializer and Deserializer

These two services define how the persisted data (both variables and flow state) is serialized and deserialized. Serialization is based on codecs provided by the zio-schema library.

Currently we provide two implementations for these services:

  • Serializer.json and Deserializer.json - uses JSON serialization
  • Serializer.protobuf and Deserializer.protobuf - uses Protobuf serialization

ZIO Flow Server

ZIO Flow Server is a ready to use executable server application that wraps the persistent executor and provides a ( HOCON) file based configuration to set it up, and a REST API for running and querying flows.

Running the server

There is no packaged executable of the server at the moment, but later we are planning to provide a ready to use docker container.

Today to run the server the easiest way is to clone the repository and run:

export ZIO_FLOW_SERVER_CONFIG=custom-config.conf
sbt zioFlowServer/run


The configuration file pointed by ZIO_FLOW_SERVER_CONFIG is a HOCON file.

The file has the following sections and values:

portthe port the server will listen on
key-value-storeselects the backend to use for key value store, can be either rocksdb, cassandra, dynamodb or in-memory. See the backends page for more information. Currently the key-value-store is also used to store the flow templates (discussed below). This is expected to move to its own configuration key in next versions.
indexed-storeselects the backend to use for indexed store, most of the time it should be the same as key-value-store
metrics.intervaldefines the interval for collecting internal metrics
serialization-formatcan be either json or protobuf
gc-periodthe interval
flow-configurationa list of key-value pairs of user-defined configuration provided to flows via Remote.config
policiesdefinition of HTTP retry policies, explained in details below
rocksdb-key-value-storeconfiguration for the RocksDB key value store, if it was selected by key-value-store
rocksdb-indexed-storeconfiguration for the RocksDB indexed store, if it was selected by indexed-store
cassandra-key-value-storeconfiguration for the Cassandra key value store, if it was selected by key-value-store
cassandra-indexed-storeconfiguration for the Cassandra indexed store, if it was selected by indexed-store

For configuration of the DynamoDb store, check the documentation of the zio-aws library.

HTTP retry policies

The policies.http node contains two sub nodes:

  • default is the default policy for all HTTP requests that are not customized by the per-host settings
  • Items in per-host can override the default settings based on the request's host name

For each host (and the default settings) you can define the following settings:

  • max-parallel-request-count is the maximum number of parallel requests that can be sent to the host
  • host-override should only be used in the per-host configurations and it allows you to change the host name of the requests
  • retry-policies is a list of retry policies, described below
  • circuit-breaker-policy is an optional node describing how to reset the circuit breaker for the host after it gets opened
  • timeout is the maximum duration a request can take before get cancelled

Each element in retry-policies is a configuration object with the following properties:

condition specifies the condition when this specific retry policy should be used. The following condition types are supported:

  • always defines a retry policy that is going to match all requests
  • for-specific-status selects a single specific HTTP status code
  • for-4xx defines a retry policy for cases when the server responded with any HTTP status code between 400-499
  • for-5xx defines a retry policy for cases when the server responded with any HTTP status code between 500-599
  • open-circuit-breaker defines a retry policy for the case when a request is blocked by an open circuit breaker
  • or combines two conditions, defined in the first and second sub nodes

retry-policy defines how to retry the request; it is the same configuration structure as the one used for circuit-breaker-policy, and we are going to define it later.

break-circuit is a boolean option. If it is true, when the condition of this retry policy is triggered, it will not only retry the request but also report this as a failure for the circuit breaker. The circuit breaker will open the circuit, making all further requests fail (with the condition open-circuit-breaker) for a given period.

For both the retry-polcy of each retry policy and for the circuit-breaker-policy describing the resetting behavior of the circuit breaker the configuration structure looks the same:

  • fail-after defines the maximum number of retries, or the maximum elapsed time spent for retries
  • repetition defines how much time elapses between retries. it can be either a fixed time interval, or an exponential one
  • jitter is a boolean configuration enabling some jittering for these configured intervals


The ZIO Flow Server provides a HTTP REST API for working with ZIO Flow programs. This section defines all the available endpoints.

GET /healthcheck

Simple healthcheck endpoint to check that ZIO Flow server is running.

GET /metrics

Prometheus metrics. The list of available metrics is defined in the metrics section

GET /templates

Get all the available templates registered in the server. A template in ZIO Flow server is a stored ZIO Flow program that optionally can have an input parameter as well. By storing these on the server they get an associated template ID and you can refer to this ID when starting a new flow instead of sending the whole serialized ZIO Flow program every time.

The response JSON has the following structure:

"entries": [
"templateId": "xyz",
"template": {
"flow": {
"inputSchema": {

GET /templates/<templateId>

Gets a single stored flow template by its identifier.

PUT /templates/<templateId>

Stores a new flow template by providing an identifier and posting the flow and input schema in the request's body.

DELETE /templates/<templateId>

Deletes a flow template that was previously stored by the above defined PUT by its identifier.

POST /flows

Start executing a new ZIO Flow program.

The following JSON examples show the possibilities of this endpoint:

Starting execution of a ZIO Flow that has no parameters (ZFlow.succeed(1)) :

"Flow": {
"flow": {
"Return": {
"Literal": {
"Int": 1

Starting execution of a ZIO Flow that has a required input parameter by providing this parameter's type (schema) and the value as well (ZFlow.input[Int]):

"FlowWithParameter": {
"flow": {
"Input": {}
"schema": {
"Other": {
"toAst": {
"Value": {
"valueType": "int",
"path": [],
"optional": false
"value": 1

Starting a ZIO Flow program that is stored as a flow template and does not require any parameters:

"Template": {
"templateId": "template1"

Starting a ZIO Flow program that is stored as a flow template and requires a parameter as well:

"TemplateWithParameter": {
"templateId": "template2",
"value": 1

The response JSON contains a field called flowId containing the started flow's identifier. It can be used to poll for the result of the running flow, as well as pausing, resuming or aborting it.

"flowId": "xyz"

GET /flows

Gets a list of all the flows handled by the server's executor, together with their current status.

The response is a mapping from flow ID to status:

"flow1": "Running",
"flow2": "Paused",
"flow3": "Done",
"flow4": "Suspended"

(in reality the flow IDs are UUIDs)

GET /flows/<flowId>

Polls the status of a given flow. The following examples demonstrate the possible response JSONs for this request:

When the flow is still running:

"Running": {}

If the flow died with an internal error, or was aborted (could be any other ExecutionError, not just Interrupted):

"Died": {
"value": {
"Interrupted": {}

If the flow finished running and succeeded with a value:

"Succeeded": {
"value": 1

If the flow finished running and failed with a value:

"Failed": {
"value": "flow failed!"

DELETE /flows/<flowId>

Deletes an already completed flow from the executor.

POST /flows/<flowId>/pause

Pauses a running flow.

POST /flows/<flowId>/resume

Resumes a previously paused flow.

POST /flows/<flowId>/abort

Aborts a running flow.


Many components of ZIO Flow report metrics using ZIO's built-in metrics API. ZIO Flow Server exposes these metrics for Prometheus via the /metrics endpoint. In case of using the executor embedded in your own application, the metrics can be sent to any metrics backend that supports the ZIO metrics API.

The following list contains all the metrics reported by various components of the ZIO Flow runtime:

  • zioflow_remote_evals is a counter for (tracked Remotes)[remote#metrics]
  • zioflow_remote_eval_time_ms is a histogram for tracked Remotes
  • zioflow_started_total is a counter incremented every time a flow is started executing (either new or restarted)
  • zioflow_active_flows is a gauge containing the actual number of running or suspended flows
  • zioflow_operations_total is a counter for each primitive ZFlow operation that was executed
  • zioflow_transactions_total is a counter for the number of committed, failed or retried transactions
  • zioflow_finished_flows_total is a counter increased when a flow finishes with either success, failure or death
  • zioflow_executor_error_total is a counter for different executor errors
  • zioflow_state_size_bytes is a histogram for the serialized workflow state snapshots in bytes
  • zioflow_variable_access_total is a counter increased when a remote variable is accessed (read, write or delete)
  • zioflow_variable_size_bytes is a histogram of the serialized size of remote variables in bytes
  • zioflow_finished_flow_age_ms is a histogram of the duration between submitting the workflow and completing it
  • zioflow_total_execution_time_ms is a histogram of the total time a workflow was in either running or suspended state during its life
  • zioflow_suspended_time_ms is a histogram of time fragments a workflow spends in suspended state
  • zioflow_gc_time_ms is a histogram of the time a full persistent garbage collection run takes
  • zioflow_gc_deletion is a counter for the number of remote variables deleted by the garbage collector
  • zioflow_gc is a counter for the number of persistent garbage collector runs
  • zioflow_http_responses_total is a counter for the number of HTTP operations performed
  • zioflow_http_response_time_ms is a histogram of the HTTP operation response times
  • zioflow_http_failed_requests_total is a counter for the number of failed HTTP requests
  • zioflow_http_retried_requests_total is a counter for the number of retried HTTP requests per host- ``

The ZIO Flow Server also reports the default JVM metrics provided by ZIO Core.

Custom operation executor

Writing a custom implementation of OperationExecutor is the intended way to extend ZIO Flow with custom ways to interact with the outside world. An OperationExecutor gets an input value and an Operation[Input, Output] value, and it has to provide a ZIO effect that produces a Result or fails with an ActivityError.

As Operation is a sealed trait, custom operations are encoded by Operation.Custom with a payload represented by a DynamicValue. The payload can be converted to DynamicValue using a Schema.

The following example shows how to implement a custom operation that gets a static string (serialized as part of the operation) and a dynamic string from the operation's input, and pushes a concatenated string to a queue provided for the operation executor:

import zio.schema.{DeriveSchema, DynamicValue, Schema, TypeId}

final case class CustomOp(prefix: String)

object CustomOp {
val typeId: TypeId = TypeId.parse("zio.flow.runtime.internal.executor.CustomOperationExecutorSpec.CustomOp")
implicit val schema: Schema[CustomOp] = DeriveSchema.gen[CustomOp]

/** Helper for constructing the custom operation */
def customOp(prefix: String): Operation[String, Unit] =
Operation.Custom(CustomOp.typeId, DynamicValue(CustomOp(prefix)), Schema[String], Schema[Unit])

/** Example usage of the custom operation in an activity */
def customActivity(prefix: String): Activity[String, Unit] =
"test activity using custom operation",

/** Example implementation of the custom operation executor */
final class CustomOperationExecutor(queue: Queue[String]) extends OperationExecutor {
override def execute[Input, Result](
input: Input,
operation: Operation[Input, Result]
): ZIO[RemoteContext, ActivityError, Result] =
operation match {
case Operation.Custom(typeId, operation, _, _) if typeId == CustomOp.typeId =>
for {
op <-
ZIO.fromEither(operation.toTypedValue(CustomOp.schema)).mapError(failure => ActivityError(failure, None))
_ <- queue.offer(op.prefix + input.asInstanceOf[String])
} yield ().asInstanceOf[Result]
case _ =>"Unsupported operation", None))