Processes¶
Type of Processes¶
Weaver supports multiple type of processes, as listed below. Each one of them are accessible through the same API interface, but they have different implications.
WPS-REST (a.k.a.: WPS-3, OGC API - Processes)
Builtin¶
These processes come pre-packaged with Weaver. They will be available directly on startup of the application and re-updated on each boot to make sure internal database references are updated with any source code changes.
Theses processes typically correspond to utility operations. They are specifically useful when employed as step within a Workflow process that requires data-type conversion between input/output of similar, but not perfectly, compatible definitions.
For example, process weaver.processes.builtin.jsonarray2netcdf
takes a single input JSON file which its
content contains an array-list of NetCDF file references, and returns them directly as the corresponding list of output
files. These two different file formats (single JSON to multiple NetCDF) can then be used to map two processes with
these respective output and inputs.
As of the latest release, following builtin processes are available:
All builtin processes are marked with weaver.processes.constants.CWL_REQUIREMENT_APP_BUILTIN
in the CWL
hints* section.
WPS-1/2¶
This kind of process corresponds to a traditional WPS XML or JSON endpoint (depending of supported version) prior to WPS-REST specification. When the WPS-REST process is deployed in Weaver using an URL reference to an WPS-1/2 process, Weaver parses and converts the XML or JSON body of the response and registers the process locally using this definition. This allows a remote server offering limited functionalities (e.g.: no REST bindings supported) to provide them through Weaver.
A minimal Deploy request body for this kind of process could be as follows:
{
"processDescription": {
"process": {
"id": "my-process-reference"
}
},
"executionUnit": [
{
"href": "https://example.com/wps?service=WPS&request=DescribeProcess&identifier=my-process&version=1.0.0"
}
]
}
This would tell Weaver to locally deploy the my-process-reference
process using the WPS-1 URL reference that is
expected to return a DescribeProcess
XML schema. Provided that this endpoint can be resolved and parsed according
to typical WPS specification, this should result into a successful process registration. The deployed process would
then be accessible with DescribeProcess requests.
The above deployment procedure can be automated on startup using Weaver’s wps_processes.yml
configuration file.
Please refer to Configuration of WPS Processes section for more details on this matter.
Warning
Because Weaver creates a snapshot of the reference process at the moment it was deployed, the local process definition could become out-of-sync with the remote reference where the Execute request will be sent. Refer to Remote Provider section for more details to work around this issue.
See also
WPS-REST¶
This process type is the main component of Weaver. All other process types are converted to this one either through some parsing (e.g.: WPS-1/2) or with some requirement indicators (e.g.: Builtin, Workflow) for special handling.
When deploying one such process directly, it is expected to have a reference to a CWL Application Package. This is most of the time employed to wrap a reference docker image process. The reference package can be provided in multiple ways as presented below.
Note
When a process is deployed with any of the below supported Application Package formats, additional parsing of this CWL as well as complementary details directly within the WPS deployment body is accomplished. See Correspondance between CWL and WPS fields section for more details.
Package as Literal Unit Block¶
In this situation, the CWL definition is provided as is using JSON-formatted package embedded within the
POST {WEAVER_URL}/processes
(Deploy) request. The request payload would take the following shape:
{
"processDescription": {
"process": {
"id": "my-process-reference"
}
},
"executionUnit": [
{
"unit": {
"cwlVersion": "v1.0",
"class": "CommandLineTool",
"inputs": ["<...>"],
"outputs": ["<...>"],
"<...>": "<...>"
}
}
]
}
ESGF-CWT¶
For traditional WPS-1 process type, Weaver adds default values to CWL definition. As we can see in
weaver/processes/wps_package.py
, the following default values for the CWL package are:
cwl_package = OrderedDict([
("cwlVersion", "v1.0"),
("class", "CommandLineTool"),
("hints", {
CWL_REQUIREMENT_APP_WPS1: {
"provider": get_url_without_query(wps_service_url),
"process": process_id,
}}),
])
In ESGF-CWT, ESGF-CWTRequirement
hint is used instead of default WPS1Requirement
, contained in the
weaver.processes.constants.CWL_REQUIREMENT_APP_WPS1
variable up here. The handling of this technicality is
handled in weaver/processes/wps_package.py
. We can define ESGF-CWT processes using this syntax:
"cwlVersion": "v1.0",
"class": "CommandLineTool",
"hints": {
"ESGF-CWTRequirement": {
"provider": "https://edas.nccs.nasa.gov/wps/cwt",
"process": "xarray.subset"
}
}
Workflow¶
Processes categorized as Workflow
are very similar to WPS-REST processes. From the API standpoint, they
actually look exactly the same as an atomic process when calling DescribeProcess or Execute requests.
The difference lies within the referenced Application Package which uses a CWL Workflow instead of
typical CWL CommandLineTool, and therefore, modifies how the process is internally executed.
For Workflow
processes to be deploy-able and executable, it is mandatory that Weaver is configured as EMS
(see: Configuration Settings). This requirement is due to the nature of workflows that chain processes that
need to be dispatched to known remote ADES servers (see: Configuration of Data Sources and
Workflow Operations).
Given that a Workflow
process was successfully deployed and that all process steps can be resolved, calling
its Execute request will tell Weaver to parse the chain of operations and send step process execution requests
to relevant ADES picked according to data sources. Each step’s job will then gradually be monitored from the remote
ADES until completion, and upon successful result, the EMS will retrieve the data references to pass it down to
the following step. When the complete chain succeeds, the final results of the last step will be provide as
Workflow
output as for atomic processes. In case of failure, the error will be indicated in the log with the
appropriate step and message where the error occurred.
Note
Although chaining sub-workflow(s) within a bigger scoped workflow is technically possible, this have not yet been fully explored (tested) in Weaver. There is a chance that data-source resolution fails to identify where to dispatch the step in this situation. If this impacts you, please vote and indicate your concern on issue #171.
Remote Provider¶
Remote provider correspond to a remote service that provides similar interfaces as supported by Weaver (WPS-like). For example, a remote WPS-1 XML endpoint can be referenced as a provider. When an API Providers-scoped request is executed, for example to list is processes capabilities (see GetCapabilities), Weaver will send the corresponding request using the registered reference URL to access the remote server and reply with parsed response, as if they its processes were registered locally.
Since remote providers obviously require access to the remote service, Weaver will only be able to provide results if the service is accessible with respect to standard implementation features and supported specifications.
The main advantage of using Weaver’s endpoint rather than directly accessing the referenced remote provider processes is in the case of limited functionality offered by the service. For instance, WPS-1 do not always offer GetStatus feature, and there is no extensive job monitoring availability. Since Weaver wraps the original reference with its own endpoints, these features indirectly become employable. Similarly, although WPS-1 offer XML-only endpoints, the parsing operation accomplished by Weaver makes theses services available as WPS-REST JSON endpoints. On top of that, registering remote providers into Weaver allows the user to use it as a central hub to keep references to all his accessible services and dispatch jobs from a common location.
A remote provider differs from previously presented WPS-1/2 processes such that the underlying processes of the service are not registered locally. For example, if a remote service has two WPS processes, only top-level service URL will be registered locally (in Weaver’s database) and the application will have no explicit knowledge of these remote processes. When calling process-specific requests (e.g.: DescribeProcess or Execute), Weaver will re-send the corresponding request directly to the remote provider each time and return the result accordingly. On the other hand, a WPS-1/2 reference would be parsed and saved locally with the response at the time of deployment. This means that a deployed WPS-1/2 reference would act as a snapshot of the reference (which could become out-of-sync), while Remote Provider will dynamically update according to the re-fetched response from the remote service. If our example remote service was extended to have a third WPS process, it would immediately be reflected in GetCapabilities and DescribeProcess retrieved via Weaver Providers-scoped requests. This would not be the case for the WPS-1/2 reference that would need manual update (deploy the third process to register it in Weaver).
An example body of the register provider request could be as follows:
{
"id": "my-service",
"url": "https://example.com/wps",
"public": true
}
Then, processes of this registered remote provider will be accessible. For example, if the referenced service by the above URL add a WPS process identified by my-process, its JSON description would be obtained with following request (DescribeProviderProcess):
GET {WEAVER_URL}/providers/my-service/processes/my-process
Note
Process my-process in the example is not registered locally. From the point of view of Weaver’s processes
(i.e.: route /processes/{id}), it does NOT exist. You must absolutely use the prefixed /providers/{id}
route.
Warning
API requests scoped under Providers are Weaver-specific implementation. These are not part of OGC API - Processes specification.
Managing processes included in Weaver ADES/EMS¶
Following steps represent the typical steps applied to deploy a process, execute it and retrieve the results.
Register a new process (Deploy)¶
Deployment of a new process is accomplished through the POST {WEAVER_URL}/processes
POST {WEAVER_URL}/processes
(Deploy) request.
The request body requires mainly two components:
processDescription
:Defines the process identifier, metadata, inputs, outputs, and some execution specifications. This mostly corresponds to information that corresponds to a traditional WPS definition.executionUnit
:Defines the core details of the Application Package. This corresponds to the explicit CWL definition that indicates how to execute the given application.
Upon deploy request, Weaver will either respond with a successful result, or with the appropriate error message, whether caused by conflicting ID, invalid definitions or other parsing issues. A successful process deployment will result in this process to become available for following steps.
Warning
When a process is deployed, it is not necessarily available immediately. This is because process visibility also
needs to be updated. The process must be made public to allow its discovery. For updating visibility, please
refer to the PUT {WEAVER_URL}/processes/{id}/visibility
(Visibility) request.
After deployment and visibility preconditions have been met, the corresponding process should become available through DescribeProcess requests and other routes that depend on an existing process.
Note that when a process is deployed using the WPS-REST interface, it also becomes available through the WPS-1/2 interface with the same identifier and definition. Because of compatibility limitations, some parameters in the WPS-1/2 might not be perfectly mapped to the equivalent or adjusted WPS-REST interface, although this concerns mostly only new features such as status monitoring. For most traditional use cases, properties are mapped between the two interfaces, but it is recommended to use the WPS-REST one because of the added features.
Access registered process(es) (GetCapabilities, DescribeProcess)¶
Available processes can all be listed using GET {WEAVER_URL}/processes
(GetCapabilities) request. This request will return all locally registered
process summaries. Other return formats and filters are also available according to provided request query parameters.
Note that processes not marked with public visibility will not be listed in this result.
For more specific process details, the GET {WEAVER_URL}/processes/{id}
(DescribeProcess) request should be used. This will return all information
that define the process references and expected inputs/outputs.
Note
For remote processes (see: Remote Provider), Provider requests are also available for more fine-grained
search of underlying processes. These processes are not necessarily listed as local processes, and will therefore
sometime not yield any result if using the typical DescribeProcess
endpoint.
All routes listed under Process requests should normally be applicable for remote processes by prefixing
them with /providers/{id}
.
Execution of a process (Execute)¶
Process execution (i.e.: submitting a job) is accomplished using the POST {WEAVER_URL}/processes/{id}/jobs
(Execute) request. This section will first
describe the basics of this request format, and after go into details for specific use cases and parametrization of
various input/output combinations.
Todo
detail execute I/O (basic example)
Todo
detail returned location + example
This location can then be employed to call GetStatus monitoring request.
Process Operations¶
Todo
detail ‘operations’ accomplished (stage-in, exec-cwl, stage-out)
Workflow Operations¶
Todo
same as prev + ‘operations’ (deploy, visibility, exec-remote for each step)
File Reference Types¶
Most inputs can be categorized into two of the most commonly employed types, namely LiteralData
and ComplexData
.
The former represents basic values such as integers or strings, while the other represents a file reference.
Files in Weaver (and WPS in general) can be specified with any formats
as MIME-type.
As for standard WPS, remote file references are usually limited to http(s)
scheme, unless the process takes
an input string and parses the unusual reference from the literal data to process it by itself. On the other hand,
Weaver supports all following reference schemes.
http(s)://
file://
opensearchfile://
[experimental]s3://
[experimental]
The method in which Weaver will handle such references depends on its configuration, in other words, whether it is
running as ADES or EMS (see: Configuration), as well as depending on some other CWL
package requirements.
These use-cases are described below.
Warning
Missing schemes in URL reference are considered identical as if file://
was used. In most cases, if not always,
an execution request should not employ this scheme unless the file is ensured to be at the specific location where
the running Weaver application can find it. This scheme is usually only employed as byproduct of the fetch
operation that Weaver uses to provide the file locally to underlying CWL application package to be executed.
When Weaver is able to figure out that the process needs to be executed locally in ADES mode, it will fetch all necessary files prior to process execution in order to make them available to the CWL package. When Weaver is in EMS configuration, it will always forward the references (regardless of scheme) exactly as provided as input of the process execution request, since it assumes it needs to dispatch the execution to another ADES remote server, and therefore only needs to verify that the file reference is reachable remotely. In this case, it becomes the responsibility of this remote instance to handle the reference appropriately. This also avoids potential problems such as if Weaver as EMS doesn’t have authorized access to a link that only the target ADES would have access to.
When CWL
package defines WPS1Requirement
under hints
for corresponding WPS-1/2 remote processes being
monitored by Weaver, it will skip fetching of http(s)
-based references since that would otherwise lead to useless
double downloads (one on Weaver and the other on the WPS side). It is the same in case of ESGF-CWTRequirement
employed for ESGF-CWT processes. Because these processes do not normally
Note
When Weaver is fetching remote files with http(s)://
, it can take advantage of additional request options to
support unusual or server-specific handling of remote reference as necessary. This could be employed for instance
to attribute access permissions only to some given ADES server by providing additional authorization tokens to
the requests. Please refer to Configuration of Request Options for this matter.
When using S3 references, Weaver will attempt to retrieve the file using server configuration and credentials.
Provided that the corresponding S3 bucket can be accessed by the running Weaver application, it will fetch the file
and store it locally temporarily for CWL
execution.
Note
When using S3 buckets, authorization are handled through typical AWS credentials and role permissions. This means that AWS access must be granted to the application in order to allow it fetching the file. There are also different formats of S3 reference formats handled by Weaver. Please refer to Configuration of AWS S3 Buckets for more details.
When using OpenSearch references, additional parameters are necessary to handle retrieval of specific file URL. Please refer to OpenSearch Data Source for more details.
Following table summarize the default behaviour of input file reference handling of different situations when received as input argument of process execution. For simplification, keyword <any> is used to indicate that any other value in the corresponding column can be substituted for a given row when applied with conditions of other columns, which results to same operational behaviour. Elements that behave similarly are also presented together in rows to reduce displayed combinations.
Configuration |
Process Type |
File Scheme |
Applied Operation |
---|---|---|---|
<any> |
<any> |
|
Query and re-process 1 |
ADES |
|
Convert to |
|
|
Nothing (left unmodified) |
||
|
Fetch and convert to |
||
|
Nothing (file already local) |
||
|
Fetch and convert to |
||
|
|||
EMS |
|
Convert to |
|
|
Nothing (left unmodified) |
||
|
Footnotes
- 1
References defined by
opensearch://
will trigger an OpenSearch query using the provided URL as well as other input additional parameters (see OpenSearch Data Source). After processing of this query, retrieved file references will be re-processed using the summarized logic in the table for the given use case.- 2(1,2,3)
When a
file://
(or empty scheme) maps to a local file that needs to be exposed externally for another remote process, the conversion tohttp(s)://
scheme employs settingweaver.wps_outputs_url
to form the result URL reference. The file is placed inweaver.wps_outputs_dir
to expose it as HTTP(S) endpoint.- 3(1,2)
When the process refers to a remote WPS-REST process (i.e.: remote WPS instance that supports REST bindings but that is not necessarily an ADES), Weaver simply wraps and monitor its remote execution, therefore files are handled just as for any other type of remote WPS-like servers. When the process contains an actual CWL Application Package that defines a
CommandLineTool
(including docker images), files are fetched as it will be executed locally. See CWL CommandLineTool, WPS-REST and Remote Providers for further details.- 4
When an
s3://
file is fetched, is gets downloaded to a temporaryfile://
location, which is NOT necessarily exposed ashttp(s)://
. If execution is transferred to a remove process that is expected to not support S3 references, only then the file gets converted as in 2.- 5
Workflows are only available on EMS instances. Since they chain processes, no fetch is needed as the first sub-step process will do it instead. See section about workflows as well as CWL Workflow for more details.
Todo
method to indicate explicit fetch to override these? (https://github.com/crim-ca/weaver/issues/183)
Todo
add tests that validate each combination of operation
OpenSearch Data Source¶
Todo
EOImage with AOI/TOI/CollectionId for OpenSearch
Multiple Outputs¶
Although CWL
allows output arrays, WPS
does not support it directly, as only single values are allowed for
WPS
outputs according to original specification. To work around this, Metalink files can be used to provide
a single output reference that embeds other references. This approach is also employed and preferred as described
in PyWPS Multiple Outputs.
Todo
fix doc when Multiple Output is supported with metalink (https://github.com/crim-ca/weaver/issues/25)
Todo
add example of multi-output process definition
Todo
and how CWL maps them with WPS
Warning
This feature is being worked on (Weaver Issue #25). Direct support between
See also
Email Notification¶
When submitting a job for execution, it is possible to provide the notification_email
field.
Doing so will tell Weaver to send an email to the specified address with successful or failure details upon job
completion. The format of the email is configurable from weaver.ini.example file with email-specific settings
(see: Configuration).
Monitoring of a process (GetStatus)¶
Todo
job status body example (success vs fail)
Obtaining output results, logs or errors¶
Todo
job logs/exceptions body example
Any job executed on Weaver will provide minimal log information, such as process job setup, moment when it started execution and final status. The extent of other log entries will more often than not depend on the verbosity of the underlying process being executed. When executing an Application Package, Weaver tries as best as possible to collect standard output and error steams to report them through log and exception lists.
Since Weaver can only report as much details as provided by the running application, it is recommended to provide
progressive status updates when developing applications in order to help understand problematic steps in event of
process execution failures. In the case of remote WPS processes monitored by Weaver, this means gradually reporting
process status updates (e.g.: calling WPSResponse.update_status
if you are using |pywps|_, see: Progress and Status Report),
using print
and/or logging
operation in scripts or docker images executed through CWL CommandLineTool
,
etc.
Note
Job logs and exceptions are a Weaver-specific implementation. They are not part of traditional OGC API - Processes.
Special Weaver EMS use-cases¶
This section highlight the additional behaviour available only through an EMS-configured Weaver instance. Some other points are already described in other sections, but are briefly indicated here for conciseness.
ADES dispatching using Data Sources¶
Todo
add details, data-source defines where to send request of known ADES
Todo
reference config weaver.data_sources