The :term:`Application Package` defines the internal script definition and configuration that will be executed by a :term:`Process`. This package is based on |CWL|_ (:term:`CWL`). Using the extensive |cwl-spec|_ as backbone for internal execution of the process allows it to run multiple type of applications, whether they are referenced to by :term:`Docker` image, scripts (bash, python, etc.), some remote :term:`Process` and more.
Note
The large community and use cases covered by :term:`CWL` makes it extremely versatile. If you encounter any issue running your :term:`Application Package` in Weaver (such as file permissions for example), chances are that there exists a workaround somewhere in the |cwl-spec|_. Most typical problems are usually handled by some flag or argument in the :term:`CWL` definition, so this reference should be explored first. Please also refer to :ref:`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 :term:`CWL` package definition will have their full package definition available with |pkg-req|_ request.
Note
The package request is a Weaver-specific implementation, and therefore, is not necessarily available on other :term:`ADES`/:term:`EMS` implementation as this feature is not part of |ogc-proc-api|_ specification.
Following :term:`CWL` package definition represents the :py:mod:`weaver.processes.builtin.jsonarray2netcdf` process.
.. literalinclude:: ../../weaver/processes/builtin/jsonarray2netcdf.cwl
:language: YAML
:linenos:
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-spec|_.
See Inputs/Outputs Type for more details.
When deploying a CommandLineTool that only needs to execute script or shell commands, it is recommended
to define an appropriate |cwl-docker-req|_ to containerize the :term:`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 :term:`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 :term:`Application Package` provider
defines a specific image to avoid unexpected failures if this auto-resolution changes across versions.
Below are minimalistic :term:`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 :term:`Docker` image.
The first example simply outputs the contents of a file input using the cat command.
Because the :term:`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
:term:`Application Package` when performing :term:`Job` executions.
.. literalinclude:: ../examples/docker-shell-script-cat.cwl
:language: yaml
:caption: Sample CWL definition of a shell script
The second example takes advantage of the |cwl-workdir-req|_ to generate a Python script dynamically
(i.e.: script.py), prior to executing it for processing the received inputs and produce the output file.
Because a Python runner is required, the |cwl-docker-req|_ specification defines a basic :term:`Docker` image that
meets our needs. Note that in this case, special interpretation of $(...) entries within the definition can be
provided to tell :term:`CWL` how to map :term:`Job` input values to the dynamically created script.
.. literalinclude:: ../examples/docker-python-script-report.cwl
:language: yaml
:caption: Sample CWL definition of a Python script
When advanced processing capabilities and more complicated environment preparation are required, it is recommended to package and push pre-built :term:`Docker` images to a remote registry. In this situation, just like for :ref:`app_pkg_script` examples, the |cwl-docker-req|_ 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 :term:`Docker` image reference is detected, Weaver will ensure that the application will be pulled using :term:`CWL` capabilities in order to run it.
Because :term:`Application Package` providers could desire to make use of :term:`Docker` images hosted on private registries, Weaver offers the capability to specify an authorization token through HTTP request headers during the :term:`Process` deployment. More specifically, the following definition can be provided during a :ref:`Deploy <proc_op_deploy>` request.
POST /processes HTTP/1.1
Host: weaver.example.com
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 (|auth-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>" | base64import base64
base64.b64encode(b"<username>:<password>")When the HTTP X-Auth-Docker header is detected in combination of a |cwl-docker-req|_ entry within
the :term:`Application Package` of the :term:`Process` being deployed, Weaver will parse the targeted :term:`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 :term:`Docker` image
whenever required for launching new :term:`Job` executions.
Note
Weaver only attempts to authenticate itself temporarily at the moment when the :term:`Job` is submitted to retrieve the :term:`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 :term:`Job` at later times, considering any retention time of cached :term:`Docker` images on the server. If the cache is cleaned, and the :term:`Docker` image is made unavailable, Weaver will attempt to authenticate itself again when receiving the new :term:`Job`. It is left up to the developer and :term:`Application Package` provider to manage expired tokens in Weaver according to their needs. To resolve such cases, the |update-token-req|_ request or an entire re-deployment of the :term:`Process` could be accomplished, whichever is more convenient for them.
.. versionadded:: 4.5.0
Specification and handling of the ``X-Auth-Docker`` header for providing an authentication token.
Weaver also supports :term:`CWL` class: Workflow. When an :term:`Application Package` is defined this way, the
|process-deploy-op|_ will attempt to resolve each step as another process. The reference to the :term:`CWL`
definition can be placed in any location supported as for the case of atomic processes
(see details about :ref:`supported package locations <WPS-REST>`).
The following :term:`CWL` definition demonstrates an example Workflow process that would resolve each step with
local processes of match IDs.
.. literalinclude:: ../../tests/functional/application-packages/WorkflowSubsetIceDays/package.cwl
:language: JSON
:linenos:
For instance, the jsonarray2netcdf (:ref:`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.
In order to resolve referenced processes as steps, Weaver supports 3 formats.
- Process ID explicitly given.Any visible process from |getcap-req|_ response should be resolved this way.(e.g.:
jsonarray2netcdfresolves to pre-deployed :py:mod:`weaver.processes.builtin.jsonarray2netcdf`). Full URL to the process description endpoint, provided that it also offers a |pkg-req|_ endpoint (Weaver-specific).
Full URL to the explicit CWL file (usually corresponding to (2) or the
hrefprovided in deployment body).
When an URL to the :term:`CWL` process "file" is provided with an extension, it must be one of the supported values defined in :py:data:`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 :ref:`Deploy <proc_op_deploy>` nor :ref:`Execute <proc_op_execute>`
a Workflow-flavored :term:`Application Package` until all referenced steps have themselves been deployed and
made visible.
Warning
Because Weaver needs to convert given :term:`CWL` documents into equivalent :term:`WPS` process definition,
embedded :term:`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.
.. seealso::
- :py:func:`weaver.processes.wps_package.get_package_workflow_steps`
- :ref:`Deploy <proc_op_deploy>` request details.
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 :ref:`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 :term:`CWL` and :term:`WPS` as long as it does
not introduce ambiguous results (see :ref:`File Format` for more details).
Because :term:`CWL` definition and :term:`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 :term:`Application Package` get its details merged with complementary :term:`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 :term:`WPS` that does
not have an exactly corresponding specification in :term:`CWL` (any-sized array). On the other hand, :term:`CWL` also
provides data preparation steps such as initial staging (i.e.: InitialWorkDirRequirement) that doesn't have an
equivalent under the :term:`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 :term:`CWL` specification will most of the time prevail over the :term:`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 :term:`WPS` specification help resolve some ambiguity or when
:term:`WPS` enforces the parametrisation of some elements, such as with maxOccurs field.
Note
Metadata merge operation between :term:`CWL` and :term:`WPS` is accomplished on per-mapped-field basis. In other
words, more explicit details such as maxOccurs could be obtained from :term:`WPS` and simultaneously the
same input's format could be obtained from the :term:`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 :term:`CWL` and :term:`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.
Warning
Merging of corresponding fields between :term:`CWL` and :term:`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 and outputs (:term:`I/O`) id from the :term:`CWL` context will be respectively matched against corresponding
id or identifier field from I/O of :term:`WPS` context. In the :term:`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-io-map|_ for more details). Regardless of array or mapping format,
:term:`CWL` requires that all I/O have unique id. On the :term:`WPS` side, a list of I/O is always expected.
This is because :term:`WPS` I/O with multiple values (array in :term:`CWL`) are specified by repeating the id with
each value instead of defining the value as a list of those values during :ref:`Execute <proc_op_execute>` request
(see also :ref:`Multiple Inputs`).
To summarize, the following :term:`CWL` and :term:`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.
{
"inputs": [
{
"id": "single-str",
"type": "string"
},
{
"id": "multi-file",
"type": "File[]"
}
],
"outputs": [
{
"id": "output-1",
"type": "File"
},
{
"id": "output-2",
"type": "File"
}
]
} |
{
"inputs": {
"single-str": {
"type": "string"
},
"multi-file": {
"type": "File[]"
}
},
"outputs": {
"output-1": {
"type": "File"
},
"output-2": {
"type": "File"
}
}
} |
{
"inputs": [
{
"id": "single-str"
},
{
"id": "multi-file",
"formats": []
}
],
"outputs": [
{
"id": "output-1",
"formats": []
},
{
"id": "output-2",
"formats": []
}
]
} |
The :term:`WPS` example above requires a format field for the corresponding :term:`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 :term:`CWL` and :term:`WPS` definitions can be specified in the :ref:`Deploy <proc_op_deploy>` request body with any of the following variations:
- Both are simultaneously fully specified (valid although extremely verbose).
- Both partially specified as long as sufficient complementary information is provided.
- Only :term:`CWL` :term:`I/O` is fully provided
(with empty or even unspecified
inputsoroutputssection from :term:`WPS`).
Warning
Weaver assumes that its main purpose is to eventually execute an :term:`Application Package` and will therefore
prioritize specification in :term:`CWL` over :term:`WPS`. Because of this, any unmatched id from the :term:`WPS`
context against provided :term:`CWL` ids of the same I/O section will be dropped, as they ultimately would
have no purpose during :term:`CWL` execution.
This does not apply in the case of referenced :ref:`WPS-1/2` processes since no :term:`CWL` is available in the first place.
In the :term:`CWL` context, the type field indicates the type of I/O. Available types are presented in the
|cwl-io-type|_ portion of the specification.
Warning
Weaver has two unsupported :term:`CWL` type, namely Any and Directory. This limitation is
intentional as :term:`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 :term:`Job` execution.
In the :term:`WPS` context, three data types exist, namely Literal, BoundingBox and Complex data.
As presented in the example of the previous section, :term:`I/O` in the :term:`WPS` context does not require an explicit
indication of the type from one of Literal, BoundingBox and Complex data. Instead, :term:`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 :term:`CWL`
or formats in :term:`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 :ref:`note <bbox-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.
{
"id": "input",
"literalDataDomains": [
{"dataType": {"name": "integer"}}
]
}Obviously, the equivalent :term:`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 :term:`CWL` definition and have the corresponding :term:`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.
Note
As of the current version of Weaver, :term:`WPS` data type BoundingBox is not supported. The schema definition
exists in :term:`WPS` context but is not handled by any :term:`CWL` type conversion yet. This feature is reflected
by issue #51. It is possible to use a Literal data of
type string corresponding to :term:`WKT` [1], [2] in the meantime.
| [1] | |wkt-example|_ |
| [2] | |wkt-format|_ |
An input or output resolved as :term:`CWL` File type, equivalent to a :term:`WPS` ComplexData, supports
format specification. Every mimeType field nested under formats entries of the :term:`WPS` definition
will be mapped against corresponding namespaced format of :term:`CWL`.
For example, the following input definitions are equivalent in both contexts.
{
"id": "input",
"formats": [
{"mimeType": "application/x-netcdf"},
{"mimeType": "application/json"}
]
} |
{
"inputs": [
{
"id": "input",
"format": [
"edam:format_3650",
"iana:application/json"
]
}
],
"$namespaces": {
"edam": "http://edamontology.org/",
"iana": "https://www.iana.org/assignments/media-types/"
}
} |
As demonstrated, both contexts accept multiple formats for inputs. These effectively represent supported formats by
the underlying application. The two :term:`MIME-types` selected for this example are chosen specifically to demonstrate
how :term:`CWL` formats must be specified. More precisely, :term:`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 :term:`CWL`
and :term:`WPS` is provided. In other words, an application developer could only specify the :term:`I/O`'s formats
in the :term:`WPS` portion during process deployment, and Weaver will take care to update the matching :term:`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 :term:`MIME-types` than full ontology references. Weaver has a large set of
commonly employed :term:`MIME-types` that it knows how to convert to corresponding ontologies. Also, Weaver will look
for new :term:`MIME-types` it doesn't explicitly know about onto either the :term:`IANA` or the :term:`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 :term:`WPS` side and iana:application/json on
the :term:`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 :term:`WPS` ComplexData definitions (see :ref:`Inputs/Outputs Type`) and
that :term:`MIME-types` are easier to provide in this context, it is recommended to provide all of them in the
:term:`WPS` definition.
Warning
Format specification differs between :term:`CWL` and :term:`WPS` in the case of outputs.
Although :term:`WPS` definition allows multiple supported formats for output that are later resolved to the applied
one onto the produced result of the job, :term:`CWL` only considers the output format that directly indicates the
applied schema. There is no concept of supported format in the :term:`CWL` world. This is simply because :term:`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 :term:`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 :term:`CWL` definition).
To ensure compatibility with multiple supported formats outputs of :term:`WPS`, any output that has more that one
format will have its format field dropped in the corresponding :term:`CWL` definition. Without any format on
the :term:`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 :term:`CWL`, but it will remove the additional validation layer
of the format (which especially deteriorates process resolution when chaining processes inside a :ref:`CWL Workflow`).
If the :term:`WPS` output only specifies a single MIME-type, then the equivalent format (after being resolved to a valid ontology) will be preserved on the :term:`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 in the context of :term:`WPS` LiteralData provides a mean for the application developer to restrict
inputs to a specific set of values. In :term:`CWL`, the same can be achieved using an enum definition. Therefore,
the following two variants are equivalent and completely interchangeable.
{
"id": "input",
"literalDataDomains": [
{"allowedValues": ["value-1", "value-2"]}
]
} |
{
"id": "input",
"type": {
"type": "enum",
"symbols": ["value-1", "value-2"]
}
} |
Weaver will ensure to propagate such definitions bidirectionally in order to update the :term:`CWL` or :term:`WPS`
correspondingly with the provided information in the other context if missing. The primitive type to apply to a missing
:term:`WPS` specification when resolving it from a :term:`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 :ref:`Multiple and Optional Values` definitions
presented next.
Inputs that take multiple values or references can be specified using minOccurs and maxOccurs in :term:`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, :term:`CWL` requires the type to specify null
or to use the shortcut ? character suffixed to the base type to indicate optional input. Resolution between
:term:`WPS` and :term:`CWL` for the merging strategy implies all corresponding parameter combinations and checks in
this case.
Because :term:`CWL` does not take an explicit amount of maximum occurrences, information in this case are not
necessarily completely interchangeable. In fact, :term:`WPS` is slightly more verbose and easier to define in this case
than :term:`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 :term:`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, :term:`CWL` will be
employed as deciding definition to resolve erroneous mismatches (as for any other corresponding fields).
.. todo:: update warning according to Weaver issue #25
Warning
Parameters minOccurs and maxOccurs are not permitted for outputs in the :term:`WPS` context. Native
:term:`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 :term:`CWL`
that does support array type as output (see issue #25).
Note
Although :term:`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 :ref:`Multiple Inputs` section of :ref:`Execute <proc_op_execute>` request.
Following are a few examples of equivalent :term:`WPS` and :term:`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.
{
"id": "input-multi-required",
"format": "application/json",
"minOccurs": 1,
"maxOccurs": "unbounded"
} |
{
"id": "input-multi-required",
"format": "iana:application/json",
"type": {
"type": "array", "items": "File"
}
} |
.. todo:: minOccurs/maxOccurs + array + WPS repeats IDs vs CWL as list
.. todo:: 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 :term:`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
:term:`CWL`, an array is always considered as unbounded, therefore :term:`WPS` is the only context that can limit
this amount.
.. todo:: (s:)keywords field, doc/label vs abstract/title per-I/O and overall process, etc?
Example: `cwl-metadata`_