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.
Note
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.
Note
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 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 | #!/usr/bin/env cwl-runner
cwlVersion: v1.0
class: CommandLineTool
# target the installed python pointing to weaver conda env to allow imports
baseCommand: python
arguments:
- "${WEAVER_ROOT_DIR}/weaver/processes/builtin/jsonarray2netcdf.py"
- "-o"
- "$(runtime.outdir)"
inputs:
input:
type: File
format: iana:application/json
inputBinding:
position: 1
prefix: "-i"
outputs:
output:
format: edam:format_3650
type:
type: array
items: File
outputBinding:
glob: "*.nc"
$namespaces:
iana: "https://www.iana.org/assignments/media-types/"
edam: "http://edamontology.org/"
|
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.
cwlVersion: v1.0
class: CommandLineTool
baseCommand: cat
requirements:
DockerRequirement:
dockerPull: "debian:stretch-slim"
inputs:
- id: file
type: File
inputBinding:
position: 1
outputs:
- id: output
type: File
outputBinding:
glob: stdout.log
The second example takes advantage of the InitialWorkDirRequirement 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 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.
cwlVersion: v1.0
class: CommandLineTool
baseCommand:
- python3
- script.py
inputs:
- id: amount
type: int
- id: cost
type: float
outputs:
- id: quote
type: File
outputBinding:
glob: report.txt
requirements:
DockerRequirement:
dockerPull: "python:3.7-alpine"
InitialWorkDirRequirement:
listing:
# below script is generated dynamically in the working directory, and then called by the base command
- entryname: script.py
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
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 (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
base64.b64encode(b"<username>:<password>")
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.
Note
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.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 | {
"cwlVersion": "v1.0",
"class": "Workflow",
"requirements": [
{
"class": "StepInputExpressionRequirement"
}
],
"inputs": {
"tasmax": {
"type": {
"type": "array",
"items": "File"
}
},
"lat0": "float",
"lat1": "float",
"lon0": "float",
"lon1": "float",
"freq": {
"default": "YS",
"type": {
"type": "enum",
"symbols": ["YS", "MS", "QS-DEC", "AS-JUL"]
}
}
},
"outputs": {
"output": {
"type": "File",
"outputSource": "ice_days/output_netcdf"
}
},
"steps": {
"subset": {
"run": "ColibriFlyingpigeon_SubsetBbox.cwl",
"in": {
"resource": "tasmax",
"lat0": "lat0",
"lat1": "lat1",
"lon0": "lon0",
"lon1": "lon1"
},
"out": ["output"]
},
"json2nc": {
"run": "jsonarray2netcdf",
"in": {
"input": "subset/output"
},
"out": ["output"]
},
"ice_days": {
"run": "Finch_IceDays.cwl",
"in": {
"tasmax": "json2nc/output",
"freq": "freq"
},
"out": ["output_netcdf"]
}
}
}
|
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.
- 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-deployedweaver.processes.builtin.jsonarray2netcdf
). Full URL to the process description endpoint, provided that it also offers a
GET {WEAVER_URL}/processes/{id}/package
(Package) endpoint (Weaver-specific).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.
Warning
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.
See also
Deploy request details.
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.
Note
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.
Warning
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.
|
|
|
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:
Both are simultaneously fully specified (valid although extremely verbose).
Both partially specified as long as sufficient complementary information is provided.
Only CWL I/O is fully provided (with empty or even unspecified
inputs
oroutputs
section from WPS).
Warning
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 id
s 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.
Warning
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.
1 2 3 4 5 6 | {
"id": "input",
"literalDataDomains": [
{"dataType": {"name": "integer"}}
]
}
|
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.
Note
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.
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.
|
|
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¶
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.
|
|
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).
Todo
update warning according to Weaver issue #25
Warning
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).
Note
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.
|
|
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 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.
Metadata¶
Todo
(s:)keywords field, doc/label vs abstract/title per-I/O and overall process, etc?
Example: cwl-metadata