The components of the system interact with each other in a specific order. As well as over defined interfaces. The following section will outline the order in which the components interact with each other and the interfaces that are used for the interaction.
The following is the order in which the different components interact with each other:
- The Deployment Utility uploads an initial version of the Workflow Placement Decision, the Deployment Manager Resource of this workflow, as well as the Workflow Config for the Deployment Manager to the corresponding tables in the distributed key-value store. It aditionally uploads the Deployment Package (source code) of the workflow to the distributed blob store.
- The Deployment Manager is informed of a new workflow to be solved.
- The Deployment Manager triggers the Deployment Solver to solve the workflow.
- The Deployment Solver updates the workflow placement decision with the current placement of the function instances in a staging distributed key-value store.
- The Deployment Migrator checks the staging distributed key-value store for updates to the workflow placement decision and re-deploys function instances if necessary.
- The Deployment Manager uploads the updated workflow placement decision to the distributed key-value store.
During a workflow invocation, the initial function instance will download the workflow placement decision from the distributed key-value store and add the run_id.
Subsequent functions will receive the workflow placement decision from the previous function instance which updated the current_instance_name to the name of the current function instance.
The workflow placement decision is a dictionary of information with regards to current function instance placement. This information is used both to determine the current function instance, as well as to determine where the next function instance is to be called. Additonally, this information is used to deploy functions to new regions and providers. The dictionary contains the following information:
{
"run_id": "test_run_id",
"workflow_placement": {
"current_deployment": {
"instances": {
"function_name:entry_point:0": {
"provider_region": {
"provider": "aws",
"region": "region"
},
"identifier": "test_identifier",
"function_identifier": "test_function_identifier"
},
"function_name_2:function_name_0_0:1": {
"provider_region": {
"provider": "aws",
"region": "region"
},
"identifier": "test_identifier",
"function_identifier": "test_function_identifier"
},
},
"metrics": {
...
},
"expiry_time": "2021-05-01T00:00:00"
},
"home_deployment": {
"instances": { ... },
"metrics": { ... },
}
},
"current_instance_name": "function_name:entry_point:0",
"instances": {
"workflow_id-function_name:entry_point:0": {
"instance_name": "workflow_id-function_name:entry_point:0",
"succeeding_instances": ["function_name_2:function_name_0_0:1"],
"preceding_instances": [],
"dependent_sync_predecessors": [],
},
"function_name_2:function_name_0_0:1": {
"instance_name": "function_name_2:function_name_0_0:1",
"succeeding_instances": [],
"preceding_instances": ["function_name:entry_point:0"],
"dependent_sync_predecessors": [],
}
}
}Different parts of this dictionary are provided by different components of the system.
- The
run_idis set by the initial function of the workflow. - The
workflow_placementis set by the Deployment Solver and contains the current placement of the function instances.- The
provider_regionis either set at initial deployment or is set by the Deployment Solver in the staging area and moved over at deployment and contains the provider and region of the function instance. - The
identifieris a unique identifier for the messaging queue instance at a provider. This is used to identify the calling point of the function instance. This is provided and updated by the deployment utilities. - The
function_identifieris a unique identifier for the function instance.
- The
- The
current_instance_nameis initially set by the deployment utilities as the name of the entry point function. It is then updated by the function instances to identify the successor function instance. - The
instancesis set and updated by the deployment utilities and contains the information about the workflow DAG.- The
instance_nameis the name of the function instance. - The
function_nameis the name of the function. This name always contains the provider and region of the function that this instance is deployed to. Thus this also has to be updated upon re-deployment. - The
succeeding_instancesis a list of the names of the succeeding function instances. - The
preceding_instancesis a list of the names of the preceding function instances. - The
dependent_sync_predecessorsis a list of the names of the synchronization nodes that the function instance is required to update in case of a conditional call.
- The
The deployment manager resource is a dictionary of information with regards to the deployment of the workflow. This is used by the deployment utilities to facilitate the re-deployment of function instances. The structure of the deployment manager resource is as follows:
{
"workflow_id": "test_workflow_id",
"workflow_function_descriptions": "[{\"name\": \"workflow_id-function_name_2_provider-region\", \"entry_point\":...}]",
"deployment_config": "{\"workflow_name\": \"workflow_name\",...}",
"deployed_regions": "{\"workflow_id-function_name_2_provider-region\": {\"provider\": \"provider\", \"region\": \"region\"},...}",
}Different parts of this dictionary are provided by different components of the system.
- The
workflow_idis set at initial deployment by the deployment utilities. It is the concatenation of the workflow name and version. - The
workflow_function_descriptionsis a list of functions that are part of the workflow. It is set at initial deployment by the deployment utilities. At re-deployment, if a function was re-deployed to a new region, the new function (new in the sense that the name is updated to the new provider and region) is added to the list. - The
deployment_configis the deployment config of the workflow. It is set at initial deployment by the deployment utilities. - The
deployed_regionsis a dictionary of the deployed regions of the workflow. It is set at initial deployment by the deployment utilities. At re-deployment, if a function was re-deployed to a new region, the new region is added to the dictionary. This is theoretically not necessary as the information is also contained in theworkflow_function_descriptions, however, it is more efficient to have this information in a separate dictionary.
The solver manager checks whether to trigger the Deployment Solver to solve the workflow.
The information required for this component is stored in the solver_update_checker_resources_table.
The structure of the solver manager resource is as follows:
{
"workflow_id": "image_processing_light-0.0.1",
"workflow_config": "{\"workflow_name\": ..."
}This information is set by the deployment client and is used by the Deployment Manager to determine whether to trigger the Deployment Solver to solve the workflow.
The workflow config is a dictionary of information with regards to the workflow. An example of the structure of the workflow config is as follows:
{
"workflow_name": "image_processing_light",
"workflow_version": "0.0.1",
"workflow_id": "image_processing_light-0.0.1",
"instances": {
"image_processing_light-0_0_1-GetInput:entry_point:0": {
"instance_name": "image_processing_light-0_0_1-GetInput:entry_point:0",
"regions_and_providers": {
"allowed_regions": [
{ "provider": "aws", "region": "us-east-1" },
{ "provider": "aws", "region": "us-east-2" },
{ "provider": "aws", "region": "us-west-1" },
{ "provider": "aws", "region": "us-west-2" }
],
"disallowed_regions": null,
"providers": { "aws": { "config": { "timeout": 60, "memory": 128 } } }
},
"succeeding_instances": [
"image_processing_light-0_0_1-Flip:image_processing_light-0_0_1-GetInput_0_0:1"
],
"preceding_instances": [],
"dependent_sync_predecessors": []
},
"image_processing_light-0_0_1-Flip:image_processing_light-0_0_1-GetInput_0_0:1": {
"instance_name": "image_processing_light-0_0_1-Flip:image_processing_light-0_0_1-GetInput_0_0:1",
"regions_and_providers": {
"allowed_regions": [
{ "provider": "aws", "region": "us-east-1" },
{ "provider": "aws", "region": "us-east-2" },
{ "provider": "aws", "region": "us-west-1" },
{ "provider": "aws", "region": "us-west-2" }
],
"disallowed_regions": null,
"providers": { "aws": { "config": { "timeout": 60, "memory": 128 } } }
},
"succeeding_instances": [],
"preceding_instances": [
"image_processing_light-0_0_1-GetInput:entry_point:0"
],
"dependent_sync_predecessors": []
}
},
"home_region": { "provider": "aws", "region": "us-east-1" },
"estimated_invocations_per_month": 1000000,
"constraints": {
"hard_resource_constraints": {
"cost": { "type": "absolute", "value": 1000 },
"runtime": { "type": "absolute", "value": 1000 },
"carbon": { "type": "absolute", "value": 1000 }
},
"soft_resource_constraints": {
"cost": null,
"runtime": null,
"carbon": null
},
"priority_order": ["cost", "runtime", "carbon"]
},
"regions_and_providers": {
"allowed_regions": [
{ "provider": "aws", "region": "us-east-1" },
{ "provider": "aws", "region": "us-east-2" },
{ "provider": "aws", "region": "us-west-1" },
{ "provider": "aws", "region": "us-west-2" }
],
"disallowed_regions": [],
"providers": { "aws": { "config": { "timeout": 60, "memory": 128 } } }
},
"num_calls_in_one_month": 0,
}The deployment package is the source code of the workflow. It is packaged as a zip file. And is uploaded to the distributed blob store.