Application Package

The Application Package defines the internal script definition and configuration that will be executed by a Process. This package is based on Common Workflow Language (CWL). Using the extensive CWL Specification as backbone for internal execution of the process allows it to run multiple type of applications, whether they are referenced to by Docker image, scripts (bash, python, etc.), some remote Process and more.


The large community and use cases covered by CWL makes it extremely versatile. If you encounter any issue running your Application Package in Weaver (such as file permissions for example), chances are that there exists a workaround somewhere in the CWL Specification. Most typical problems are usually handled by some flag or argument in the CWL definition, so this reference should be explored first. Please also refer to FAQ section as well as existing Weaver issues. Ultimately if no solution can be found, open an new issue about your specific problem.

All processes deployed locally into Weaver using a CWL package definition will have their full package definition available with GET {WEAVER_URL}/processes/{id}/package (Package) request.


The package request is a Weaver-specific implementation, and therefore, is not necessarily available on other ADES/EMS implementation as this feature is not part of OGC API - Processes specification.

Typical CWL Package Definition

CWL CommandLineTool

Following CWL package definition represents the weaver.processes.builtin.jsonarray2netcdf process.

 1#!/usr/bin/env cwl-runner
 2cwlVersion: v1.0
 3class: CommandLineTool
 4# target the installed python pointing to weaver conda env to allow imports
 5baseCommand: python
 7  - "${WEAVER_ROOT_DIR}/weaver/processes/builtin/"
 8  - "-o"
 9  - "$(runtime.outdir)"
11 input:
12   type: File
13   format: iana:application/json
14   inputBinding:
15     position: 1
16     prefix: "-i"
18 output:
19   format: edam:format_3650
20   type:
21     type: array
22     items: File
23   outputBinding:
24     glob: "*.nc"
26  iana: ""
27  edam: ""

The first main components is the class: CommandLineTool that tells Weaver it will be an atomic process (contrarily to CWL Workflow presented later).

The other important sections are inputs and outputs. These define which parameters will be expected and produced by the described application. Weaver supports most formats and types as specified by CWL Specification. See Inputs/Outputs Type for more details.

Script Application

When deploying a CommandLineTool that only needs to execute script or shell commands, it is recommended to define an appropriate DockerRequirement to containerize the Process, even though no advanced operation is needed. The reason for this is because there is no way for Weaver to otherwise know for sure how to provide all appropriate dependencies that this operation might need. In order to preserve processing environment and results separate between any Process and Weaver itself, the executions will either be automatically containerized (with some default image), or blocked entirely when Weaver cannot resolve the appropriate execution environment. Therefore, it is recommended that the Application Package provider defines a specific image to avoid unexpected failures if this auto-resolution changes across versions.

Below are minimalistic Application Package samples that make use of a shell command and a custom Python script for quickly running some operations, without actually needing to package any specialized Docker image.

The first example simply outputs the contents of a file input using the cat command. Because the Docker image debian:stretch-slim is specified, we can guarantee that the command will be available within its containerized environment. In this case, we also take advantage of the stdout.log which is always collected by Weaver (along with the stderr) in order to obtain traces produced by any Application Package when performing Job executions.

Sample CWL definition of a shell script
cwlVersion: v1.0
class: CommandLineTool
baseCommand: cat
    dockerPull: "debian:stretch-slim"
  - id: file
    type: File
      position: 1
  - id: output
    type: File
      glob: stdout.log

The second example takes advantage of the InitialWorkDirRequirement to generate a Python script dynamically (i.e.:, prior to executing it for processing the received inputs and produce the output file. Because a Python runner is required, the DockerRequirement specification defines a basic Docker image that meets our needs. Note that in this case, special interpretation of $(...) entries within the definition can be provided to tell CWL how to map Job input values to the dynamically created script.

Sample CWL definition of a Python script
cwlVersion: v1.0
class: CommandLineTool
  - python3
  - id: amount
    type: int
  - id: cost
    type: float
  - id: quote
    type: File
      glob: report.txt
    dockerPull: "python:3.7-alpine"
      # below script is generated dynamically in the working directory, and then called by the base command
      - entryname:
        entry: |
          amount = $(inputs.amount)
          cost = $(inputs.cost)
          with open("report.txt", "w") as report:
              report.write(f"Order Total: {amount * cost:0.2f}$\n")

Dockerized Applications

When advanced processing capabilities and more complicated environment preparation are required, it is recommended to package and push pre-built Docker images to a remote registry. In this situation, just like for Script Application examples, the DockerRequirement is needed. The definitions would also be essentially the same as previous examples, but with more complicated operations and possibly larger amount of inputs or outputs.

Whenever a Docker image reference is detected, Weaver will ensure that the application will be pulled using CWL capabilities in order to run it.

Because Application Package providers could desire to make use of Docker images hosted on private registries, Weaver offers the capability to specify an authorization token through HTTP request headers during the Process deployment. More specifically, the following definition can be provided during a Deploy request.

POST /processes HTTP/1.1
Content-Type: application/json;charset=UTF-8
X-Auth-Docker: Basic <base64_token>

{ "processDescription": { }, "executionUnit": { } }

The X-Auth-Docker header should be defined exactly like any typical Authorization headers (HTTP Authentication Schemes). The name X-Auth-Docker is inspired from existing implementations that employ X-Auth-Token in a similar fashion. The reason why Authorization and X-Auth-Token headers are not themselves employed in this case is to ensure that they do not interfere with any proxy or server authentication mechanism, which Weaver could be located behind.

For the moment, only Basic (RFC 7617) authentication is supported. To generate the base64 token, following methods can be used:

echo -n "<username>:<password>" | base64
import base64

When the HTTP X-Auth-Docker header is detected in combination of a DockerRequirement entry within the Application Package of the Process being deployed, Weaver will parse the targeted Docker registry defined in dockerPull and will attempt to identify it for later authentication towards it with the provided token. Given a successful authentication, Weaver should then be able to pull the Docker image whenever required for launching new Job executions.


Weaver only attempts to authenticate itself temporarily at the moment when the Job is submitted to retrieve the Docker image, and only if the image is not already available locally. Because of this, the provided authentication token should have a sufficient lifetime to run the Job at later times, considering any retention time of cached Docker images on the server. If the cache is cleaned, and the Docker image is made unavailable, Weaver will attempt to authenticate itself again when receiving the new Job. It is left up to the developer and Application Package provider to manage expired tokens in Weaver according to their needs. To resolve such cases, the Update Token request or an entire re-deployment of the Process could be accomplished, whichever is more convenient for them.

New in version 4.5.0: Specification and handling of the X-Auth-Docker header for providing an authentication token.

CWL Workflow

Weaver also supports CWL class: Workflow. When an Application Package is defined this way, the Process deployment operation will attempt to resolve each step as another process. The reference to the CWL definition can be placed in any location supported as for the case of atomic processes (see details about supported package locations).

The following CWL definition demonstrates an example Workflow process that would resolve each step with local processes of match IDs.

 2    "cwlVersion": "v1.0",
 3    "class": "Workflow",
 4    "requirements": [
 5        {
 6            "class": "StepInputExpressionRequirement"
 7        }
 8    ],
 9    "inputs": {
10        "tasmax": {
11            "type": {
12                "type": "array",
13                "items": "File"
14            }
15        },
16        "lat0": "float",
17        "lat1": "float",
18        "lon0": "float",
19        "lon1": "float",
20        "freq": {
21            "default": "YS",
22            "type": {
23                "type": "enum",
24                "symbols": ["YS", "MS", "QS-DEC", "AS-JUL"]
25            }
26        }
27    },
28    "outputs": {
29        "output": {
30            "type": "File",
31            "outputSource": "ice_days/output_netcdf"
32        }
33    },
34    "steps": {
35        "subset": {
36            "run": "ColibriFlyingpigeon_SubsetBbox.cwl",
37            "in": {
38                "resource": "tasmax",
39                "lat0": "lat0",
40                "lat1": "lat1",
41                "lon0": "lon0",
42                "lon1": "lon1"
43            },
44            "out": ["output"]
45        },
46        "json2nc": {
47            "run": "jsonarray2netcdf",
48            "in": {
49                "input": "subset/output"
50            },
51            "out": ["output"]
52        },
53        "ice_days": {
54            "run": "Finch_IceDays.cwl",
55            "in": {
56                "tasmax": "json2nc/output",
57                "freq": "freq"
58            },
59            "out": ["output_netcdf"]
60        }
61    }

For instance, the jsonarray2netcdf (Builtin) middle step in this example corresponds to the CWL CommandLineTool process presented in previous section. Other processes referenced in this Workflow can be found in Weaver Test Resources.

Steps processes names are resolved using the variations presented below. Important care also needs to be given to inputs and outputs definitions between each step.

Step Reference

In order to resolve referenced processes as steps, Weaver supports 3 formats.

  1. Process ID explicitly given.
    Any visible process from GET {WEAVER_URL}/processes (GetCapabilities) response should be resolved this way.
    (e.g.: jsonarray2netcdf resolves to pre-deployed weaver.processes.builtin.jsonarray2netcdf).
  2. Full URL to the process description endpoint, provided that it also offers a GET {WEAVER_URL}/processes/{id}/package (Package) endpoint (Weaver-specific).

  3. Full URL to the explicit CWL file (usually corresponding to (2) or the href provided in deployment body).

When an URL to the CWL process “file” is provided with an extension, it must be one of the supported values defined in weaver.processes.wps_package.PACKAGE_EXTENSIONS. Otherwise, Weaver will refuse it as it cannot figure out how to parse it.

Because Weaver and the underlying CWL executor need to resolve all steps in order to validate their input and output definitions correspond (id, format, type, etc.) in order to chain them, all intermediate processes MUST be available. This means that you cannot Deploy nor Execute a Workflow-flavored Application Package until all referenced steps have themselves been deployed and made visible.


Because Weaver needs to convert given CWL documents into equivalent WPS process definition, embedded CWL processes within a Workflow step are not supported currently. This is a known limitation of the implementation, but not much can be done against it without major modifications to the code base. See also issue #56.

Step Inputs/Outputs

Inputs and outputs of connected steps are required to match types and formats in order for the workflow to be valid. This means that a process that produces an output of type String cannot be directly chained to a process that takes as input a File, even if the String of the first process represents an URL that could be resolved to a valid file reference. In order to chain two such processes, an intermediate operation would need to be defined to explicitly convert the String input to the corresponding File output. This is usually accomplished using Builtin processes, such as in the previous example.

Since formats must also match (e.g.: a process producing application/json cannot be mapped to one producing application/x-netcdf), all mismatching formats must also be converted with an intermediate step if such operation is desired. This ensures that workflow definitions are always explicit and that as little interpretation, variation or assumptions are possible between each execution. Because of this, all application generated by Weaver will attempt to preserve and enforce matching input/output format definition in both CWL and WPS as long as it does not introduce ambiguous results (see File Format for more details).

Correspondence between CWL and WPS fields

Because CWL definition and WPS process description inherently provide “duplicate” information, many fields can be mapped between one another. In order to handle any provided metadata in the various supported locations by both specifications, as well as to extend details of deployed processes, each Application Package get its details merged with complementary WPS description.

In some cases, complementary details are only documentation-related, but some information directly affect the format or execution behaviour of some parameters. A common example is the maxOccurs field provided by WPS that does not have an exactly corresponding specification in CWL (any-sized array). On the other hand, CWL also provides data preparation steps such as initial staging (i.e.: InitialWorkDirRequirement) that doesn’t have an equivalent under the WPS process description. For this reason, complementary details are merged and reflected on both sides (as applicable), when non-ambiguous resolution is possible.

In case of conflicting metadata, the CWL specification will most of the time prevail over the WPS metadata fields simply because it is expected that a strict CWL specification is provided upon deployment. The only exceptions to this situation are when WPS specification help resolve some ambiguity or when WPS enforces the parametrisation of some elements, such as with maxOccurs field.


Metadata merge operation between CWL and WPS is accomplished on per-mapped-field basis. In other words, more explicit details such as maxOccurs could be obtained from WPS and simultaneously the same input’s format could be obtained from the CWL side. Merge occurs bidirectionally for corresponding information.

The merging strategy of process specifications also implies that some details can be omitted from one context if they can be inferred from corresponding elements in the other. For example, the CWL and WPS context both define keywords (with minor naming variation) as a list of strings. Specifying this metadata in both locations is redundant and only makes the process description longer. Therefore, the user is allowed to provide only one of the two and Weaver will take care to propagate the information to the lacking location.

In order to help understand the resolution methodology between the contexts, following sub-section will cover supported mapping between the two specifications, and more specifically, how each field impacts the mapped equivalent metadata.


Merging of corresponding fields between CWL and WPS is a Weaver-specific implementation. The same behaviour is not necessarily supported by other implementations. For this reason, any converted information between the two contexts will be transferred to the other context if missing in order for both specification to reflect the similar details as closely as possible, wherever context the metadata originated from.

Inputs/Outputs ID

Inputs and outputs (I/O) id from the CWL context will be respectively matched against corresponding id or identifier field from I/O of WPS context. In the CWL definition, all of the allowed I/O structures are supported, whether they are specified using an array list with explicit definitions, using “shortcut” variant, or using key-value pairs (see CWL Mapping for more details). Regardless of array or mapping format, CWL requires that all I/O have unique id. On the WPS side, a list of I/O is always expected. This is because WPS I/O with multiple values (array in CWL) are specified by repeating the id with each value instead of defining the value as a list of those values during Execute request (see also Multiple Inputs).

To summarize, the following CWL and WPS I/O definitions are all equivalent and will result into the same process definition after deployment. For simplification purpose, below examples omit all but mandatory fields (only of the inputs and outputs portion of the full deployment body) to produce the same result. Other fields are discussed afterward in specific sections.

CWL I/O objects array
 2  "inputs": [
 3    {
 4      "id": "single-str",
 5      "type": "string"
 6    },
 7    {
 8      "id": "multi-file",
 9      "type": "File[]"
10    }
11  ],
12  "outputs": [
13    {
14      "id": "output-1",
15      "type": "File"
16    },
17    {
18      "id": "output-2",
19      "type": "File"
20    }
21  ]
CWL I/O key-value mapping
 2  "inputs": {
 3    "single-str": {
 4      "type": "string"
 5    },
 6    "multi-file": {
 7      "type": "File[]"
 8    }
 9  },
10  "outputs": {
11    "output-1": {
12      "type": "File"
13    },
14    "output-2": {
15      "type": "File"
16    }
17  }
WPS I/O definition
 2  "inputs": [
 3    {
 4      "id": "single-str"
 5    },
 6    {
 7      "id": "multi-file",
 8      "formats": []
 9    }
10  ],
11  "outputs": [
12    {
13      "id": "output-1",
14      "formats": []
15    },
16    {
17      "id": "output-2",
18      "formats": []
19    }
20  ]

The WPS example above requires a format field for the corresponding CWL File type in order to distinguish it from a plain string. More details are available in Inputs/Outputs Type below about this requirement.

Finally, it is to be noted that above CWL and WPS definitions can be specified in the Deploy request body with any of the following variations:

  1. Both are simultaneously fully specified (valid although extremely verbose).

  2. Both partially specified as long as sufficient complementary information is provided.

  3. Only CWL I/O is fully provided (with empty or even unspecified inputs or outputs section from WPS).


Weaver assumes that its main purpose is to eventually execute an Application Package and will therefore prioritize specification in CWL over WPS. Because of this, any unmatched id from the WPS context against provided CWL ids of the same I/O section will be dropped, as they ultimately would have no purpose during CWL execution.

This does not apply in the case of referenced WPS-1/2 processes since no CWL is available in the first place.

Inputs/Outputs Type

In the CWL context, the type field indicates the type of I/O. Available types are presented in the CWLType Symbols portion of the specification.


Weaver has two unsupported CWL type, namely Any and Directory. This limitation is intentional as WPS does not offer equivalents. Furthermore, both of these types make the process description too ambiguous. For instance, most processes expect remote file references, and providing a Directory doesn’t indicate an explicit reference to which files to retrieve during stage-in operation of a Job execution.

In the WPS context, three data types exist, namely Literal, BoundingBox and Complex data.

As presented in the example of the previous section, I/O in the WPS context does not require an explicit indication of the type from one of Literal, BoundingBox and Complex data. Instead, WPS type is inferred using the matched API schema of the I/O. For instance, Complex I/O (i.e.: file reference) requires the formats field to distinguish it from a plain string. Therefore, specifying either format in CWL or formats in WPS immediately provides all needed information for Weaver to understand that this I/O is expected to be a file reference. A crs field would otherwise indicate a BoundingBox I/O (see note). If none of the two previous schemas are matched, the I/O type resolution falls back to Literal data of string type. To employ another primitive data type such as Integer, an explicit indication needs to be provided as follows.

WPS Literal Data Type
2  "id": "input",
3  "literalDataDomains": [
4    {"dataType": {"name": "integer"}}
5  ]

Obviously, the equivalent CWL definition is simpler in this case (i.e.: only type: int is required). It is therefore recommended to take advantage of Weaver’s merging strategy in this case by providing only the details through the CWL definition and have the corresponding WPS I/O type automatically deduced by the generated process. If desired, literalDataDomains can still be explicitly provided as above to ensure that it gets parsed as intended type.


As of the current version of Weaver, WPS data type BoundingBox is not supported. The schema definition exists in WPS context but is not handled by any CWL type conversion yet. This feature is reflected by issue #51. It is possible to use a Literal data of type string corresponding to WKT 1, 2 in the meantime.


WKT Examples


WKT Formats

File Format

An input or output resolved as CWL File type, equivalent to a WPS ComplexData, supports format specification. Every mimeType field nested under formats entries of the WPS definition will be mapped against corresponding namespaced format of CWL.

For example, the following input definitions are equivalent in both contexts.

WPS Format with MIME-type
2  "id": "input",
3  "formats": [
4    {"mimeType": "application/x-netcdf"},
5    {"mimeType": "application/json"}
6  ]
CWL Format with Namespace
 2  "inputs": [
 3    {
 4      "id": "input",
 5      "format": [
 6        "edam:format_3650",
 7        "iana:application/json"
 8      ]
 9    }
10  ],
11  "$namespaces": {
12    "edam": "",
13    "iana": ""
14  }

As demonstrated, both contexts accept multiple formats for inputs. These effectively represent supported formats by the underlying application. The two MIME-types selected for this example are chosen specifically to demonstrate how CWL formats must be specified. More precisely, CWL requires a real schema definition referencing to an existing ontology to validate formats, specified through the $namespaces section. Each format entry is then defined as a mapping of the appropriate namespace to the identifier of the ontology. Alternatively, you can also provide the full URL of the ontology reference in the format string.

Like many other fields, this information can become quite rapidly redundant and difficult to maintain. For this reason, Weaver will automatically fill the missing detail if only one of the two corresponding information between CWL and WPS is provided. In other words, an application developer could only specify the I/O’s formats in the WPS portion during process deployment, and Weaver will take care to update the matching CWL definition without any user intervention. This makes it also easier for the user to specify supported formats since it is generally easier to remember names of MIME-types than full ontology references. Weaver has a large set of commonly employed MIME-types that it knows how to convert to corresponding ontologies. Also, Weaver will look for new MIME-types it doesn’t explicitly know about onto either the IANA or the EDAM ontologies in order to attempt automatically resolving them.

When formats are resolved between the two contexts, Weaver applies information in a complimentary fashion. This means for example that if the user provided application/x-netcdf on the WPS side and iana:application/json on the CWL side, both resulting contexts will have both of those formats combined. Weaver will not favour one location over the other, but will rather merge them if they can be resolved into different and valid entities.

Since format is a required field for WPS ComplexData definitions (see Inputs/Outputs Type) and that MIME-types are easier to provide in this context, it is recommended to provide all of them in the WPS definition.

Output File Format


Format specification differs between CWL and WPS in the case of outputs.

Although WPS definition allows multiple supported formats for output that are later resolved to the applied one onto the produced result of the job, CWL only considers the output format that directly indicates the applied schema. There is no concept of supported format in the CWL world. This is simply because CWL cannot predict nor reliably determine which output will be produced by a given application execution without running it, and therefore cannot expose consistent output specification before running the process. Because CWL requires to validate the full process integrity before it can be executed, this means that only a single output format is permitted in its context (providing many will raise a validation error when parsing the CWL definition).

To ensure compatibility with multiple supported formats outputs of WPS, any output that has more that one format will have its format field dropped in the corresponding CWL definition. Without any format on the CWL side, the validation process will ignore this specification and will effectively accept any type of file. This will not break any execution operation with CWL, but it will remove the additional validation layer of the format (which especially deteriorates process resolution when chaining processes inside a CWL Workflow).

If the WPS output only specifies a single MIME-type, then the equivalent format (after being resolved to a valid ontology) will be preserved on the CWL side since the result is ensured to be the unique one provided. For this reason, processes with specific single-format output are be preferred whenever possible. This also removes ambiguity in the expected output format, which usually requires a toggle input specifying the desired type for processes providing a multi-format output. It is instead recommended to produce multiple processes with a fixed output format for each case.

Allowed Values

Allowed values in the context of WPS LiteralData provides a mean for the application developer to restrict inputs to a specific set of values. In CWL, the same can be achieved using an enum definition. Therefore, the following two variants are equivalent and completely interchangeable.

WPS AllowedValues Input
2  "id": "input",
3  "literalDataDomains": [
4    {"allowedValues": ["value-1", "value-2"]}
5  ]
CWL Enum Values
2  "id": "input",
3  "type": {
4    "type": "enum",
5    "symbols": ["value-1", "value-2"]
6  }

Weaver will ensure to propagate such definitions bidirectionally in order to update the CWL or WPS correspondingly with the provided information in the other context if missing. The primitive type to apply to a missing WPS specification when resolving it from a CWL definition is automatically inferred with the best matching type from provided values in the enum list.

Note that enum such as these will also be applied on top of Multiple and Optional Values definitions presented next.

Multiple and Optional Values

Inputs that take multiple values or references can be specified using minOccurs and maxOccurs in WPS context, while they are specified using the array type in CWL. While the same minOccurs parameter with a value of zero (0) can be employed to indicate an optional input, CWL requires the type to specify null or to use the shortcut ? character suffixed to the base type to indicate optional input. Resolution between WPS and CWL for the merging strategy implies all corresponding parameter combinations and checks in this case.

Because CWL does not take an explicit amount of maximum occurrences, information in this case are not necessarily completely interchangeable. In fact, WPS is slightly more verbose and easier to define in this case than CWL because all details are contained within the same two parameters. Because of this, it is often preferable to provide the minOccurs and maxOccurs in the WPS context, and let Weaver infer the array and/or null type requirements automatically. Also, because of all implied parameters in this situation to specify the similar details, it is important to avoid providing contradicting specifications as Weaver will have trouble guessing the intended result when merging specifications. If unambiguous guess can be made, CWL will be employed as deciding definition to resolve erroneous mismatches (as for any other corresponding fields).


update warning according to Weaver issue #25


Parameters minOccurs and maxOccurs are not permitted for outputs in the WPS context. Native WPS therefore does not permit multiple output reference files. This can be worked around using a Metalink file, but this use case is not covered by Weaver yet as it requires special mapping with CWL that does support array type as output (see issue #25).


Although WPS multi-value inputs are defined as a single entity during deployment, special care must be taken to the format in which to specify these values during execution. Please refer to Multiple Inputs section of Execute request.

Following are a few examples of equivalent WPS and CWL definitions to represent multiple values under a given input. Some parts of the following definitions are purposely omitted to better highlight the concise details of multiple and optional information.

WPS Multi-Value Input (required)
2  "id": "input-multi-required",
3  "format": "application/json",
4  "minOccurs": 1,
5  "maxOccurs": "unbounded"
CWL Multi-Value Input (required)
2  "id": "input-multi-required",
3  "format": "iana:application/json",
4  "type": {
5    "type": "array", "items": "File"
6  }


minOccurs/maxOccurs + array + WPS repeats IDs vs CWL as list


example multi-value + enum

It can be noted from the examples that minOccurs and maxOccurs can be either an integer or a string representing one. This is to support backward compatibility of older WPS specification that always employed strings although representing numbers. Weaver understands and handles both cases. Also, maxOccurs can have the special string value "unbounded", in which case the input is considered to be allowed an unlimited amount if entries (although often capped by another implicit machine-level limitation such as memory capacity). In the case of CWL, an array is always considered as unbounded, therefore WPS is the only context that can limit this amount.



(s:)keywords field, doc/label vs abstract/title per-I/O and overall process, etc?

Example: cwl-metadata