Changes

Unreleased (latest)

Changes:

  • No change.

Fixes:

  • No change.

4.4.0 (2021-11-19)

Changes:

  • Add map_wps_output_location utility function to handle recurrent mapping of weaver.wps_output_dir back and forth with resolved weaver.wps_output_url.

  • Add more detection of map-able WPS output location to avoid fetching files unnecessarily. Common cases are Workflow running multiple steps on the same server or Application Package Process that reuses an output produced by a previous execution. Relates to #183.

  • Add pre-validation of file accessibility using HTTP HEAD request when a subsequent Workflow step employs an automatically mapped WPS output location from a previous step to verify that the file would otherwise be downloadable if it could not have been mapped. This is to ensure consistency and security validation of the reference WPS output location, although the unnecessary file download operation can be avoided.

  • Add functional Workflow tests to validate execution without the need of remote Weaver test application (relates to #141, relates to #281).

  • Add missing documentation details about Data Source and connect chapters with other relevant documentation details and updated Workflow tests.

  • Add handling of Content-Disposition header providing preferred filename or filename* parameters when fetching file references instead of the last URL fragment employed by default (resolves #364).

  • Add more security validation of the obtained file name from HTTP reference, whether generated from URL path fragment or other header specification.

Fixes:

  • Fix incorrect resolution of Process results endpoint to pass contents from one step to another during Workflow execution (resolves #358).

  • Fix logic of remotely and locally executed applications based on CWL requirements when attempting to resolve whether an input file reference should be fetched.

  • Fix resolution of WPS I/O provided as mapping instead of listing during deployment in order to properly parse them and merge their metadata with corresponding CWL I/O definitions.

  • Fix DataSource and OpenSearch typing definitions to more rapidly detect incorrect data structures during parsing.

4.3.0 (2021-11-16)

Changes:

  • Add support of type and processID query parameters for Job listing (resolves some tasks in #268).

  • Add type field to Job status information (resolves #351).

  • Add OGC API - Processes conformance references regarding supported operations for Job listing and filtering.

  • Add minDuration and maxDuration parameters to query Job listing filtered by specific execution time range (resolves #268). Range duration parameters are limited to single values each (relates to opengeospatial/ogcapi-processes#261).

  • Require minimally pymongo==3.12.0 and corresponding MongoDB 5.0 instance to process new filtering queries of minDuration and maxDuration. Please refer to Database Migration and MongoDB official documentation for migration methods.

  • Refactor Job search method to facilitate its extension in the event of future filter parameters.

  • Support contextual WPS output location using X-WPS-Output-Context header to store Job results. When a Job is executed by providing this header with a sub-directory, the resulting outputs of the Job will be placed and reported under the corresponding location relative to WPS outputs (path and URL).

  • Add weaver.wps_output_context setting as default contextual WPS output location when header is omitted.

  • Replace Job.execute_async getter/setter by simple property using more generic Job.execution_mode for storage in database. Provide Job.execute_async and Job.execute_sync properties based on stored mode.

  • Simplify execute_process function executed by Celery task into sub-step functions where applicable.

  • Simplify forwarding of Job parameters between PyWPS service WorkerService.execute_job method and Celery task instantiating it by reusing the Job object.

  • Provide corresponding Job log URL along already reported log file path to facilitate retrieval from server side.

  • Avoid Job.progress updates following failed or dismissed statuses to keep track of the last real progress percentage that was reached when that status was set.

  • Improve typing of database and store getter functions to infer correct types and facilitate code auto-complete.

  • Implement Job dismiss operation ensuring pending or running tasks are removed and output result artifacts are removed from disk.

  • Implement HTTP Gone (410) status from already dismissed Job when requested again or when fetching its artifacts.

Fixes:

  • Removes the need for specific configuration to handle public/private output directory settings using provided X-WPS-Output-Context header (fixes #110).

  • Fix retrieval of Pyramid Registry and application settings when available container is Werkzeug Request instead of Pyramid Request, as employed by underlying HTTP requests in PyWPS service.

  • Allow group query parameter to handle Job category listing with provider as service alias.

  • Improve typing of database and store getter functions to infer correct types and facilitate code auto-complete.

  • Fix incorrectly configured API views for batch Job dismiss operation with DELETE /jobs and corresponding endpoints for Process and Provider paths.

  • Fix invalid Job links sometimes containing duplicate / occurrences.

  • Fix invalid Job link URL for alternate relationship.

4.2.1 (2021-10-20)

Changes:

  • Add more frequent Job updates of execution checkpoint pushed to database in order to avoid inconsistent statuses between the parent Celery task and the underlying Application Package being executed, since both can update the same Job entry at different moments.

  • Add a Job log entry as "accepted" on the API side before calling the Celery task submission (Job not yet picked by a worker) in order to provide more detail between the submission time and initial execution time. This allows to have the first log entry not immediately set to "running" since both "started" and "running" statues are remapped to "running" within the task to be compliant with OGC status codes.

Fixes:

  • Fix an inconsistency between the final Job status and the reported “completed” message in logs due to missing push of a newer state prior re-fetch of the latest Job from the database.

4.2.0 (2021-10-19)

Changes:

  • Add execution endpoint POST /provider/{id}/process/{id}/execution corresponding to the OGC-API compliant endpoint for local Process definitions.

  • Add multiple additional relation links for Process and Job responses (resolves #234 and #267).

  • Add convenience DELETE /jobs endpoint with input list of Job UUIDs in order to dismiss multiple entries simultaneously. This is useful for quickly removing a set of Job returned by filtered GET /jobs contents.

  • Update conformance link list for dismiss and relevant relation links definitions (relates to #53 and #267).

  • Add better support and reporting of Job status dismissed when operation is called from API on running task.

  • Use explicit started status when Job has been picked up by a Celery worker instead of leaving it to accepted (same status that indicates the Job “pending”, although a worker is processing it). Early modification of status is done in case setup operations (send WPS request, prepare files, etc.) take some time which would leave users under the impression the Job is not getting picked up. Report explicit running status in Job once it has been sent to the remote WPS endpoint. The API will report running in both cases in order to support OGC API - Processes naming conventions, but internal Job status will have more detail.

  • Add updated timestamp to Job response to better track latest milestones saved to database (resolves #249). This avoids users having to compare many fields (created, started, finished) depending on latest status.

  • Apply stricter Deploy body schema validation and employ deserialized result directly. This ensures that preserved fields in the submitted content for deployment contain only known data elements with expected structures for respective schemas. Existing deployment body that contain invalid formats could start to fail or might generate inconsistent Process descriptions if not adjusted.

  • Add improved reporting of erroneous inputs during Process deployment whenever possible to identify the cause.

  • Add more documentation details about missing features such as EOImage inputs handled by OpenSearch requests.

  • Add weaver.celery flag to internal application settings when auto-detecting that current runner is celery. This bypasses redundant API-only operations during application setup and startup not needed by celery worker.

Fixes:

  • Fix OGC-API compliant execution endpoint POST /process/{id}/execution not registered in API.

  • Fix missing status for cancelled Jobs in order to properly support dismiss operation (resolves #145 and #228).

  • Fix all known OGC-specific link relationships with URI prefix (resolves #266).

  • Fix incorrect rendering of some table cells in the documentation.

4.1.2 (2021-10-13)

Changes:

  • No change.

Fixes:

  • Add celery worker task events flag (-E) to Docker command (weaver-worker) to help detect submitted delayed tasks when requesting job executions.

4.1.1 (2021-10-12)

Changes:

  • No change.

Fixes:

  • Fix handling of default format field of WPS input definition incorrectly resolved as default data by PyWPS for Process that allows optional (minOccurs=0) inputs of Complex type. Specific case is detected with relevant erroneous data and dropped silently because it should not be present (since omitted in WPS request) and should not generate a WPS input (relates to geopython/pywps#633).

  • Fix resolution of CWL field default value erroneously inserted as "null" literal string for inputs generated from WPS definition to avoid potential confusion with valid "null" input or default string. Default behaviour to drop or ignore omitted inputs are handled by "null" within type field in CWL definitions.

  • Fix Wps1Process job runner for dispatched execution of WPS-1 Process assuming all provided inputs contain data or reference. Skip omitted optional inputs that are resolved with None value following above fixes.

  • Resolve execution failure of WPS-1 Process ncdump under hummingbird Provider (fixes issue identified in output logs from notebook in PR pavics-sdi#230).

4.1.0 (2021-09-29)

Changes:

  • Improve reporting of mismatching Weaver configuration for Process and Application Package definitions that always require remote execution. Invalid combinations will be raised during execution with detailed problem.

  • Forbid Provider and applicable Process definitions to be deployed, executed or queried when corresponding remote execution is not supported according to Weaver instance configuration since Provider must be accessed remotely.

  • Refactor endpoint views and utilities referring to Provider operations into appropriate modules.

  • Apply weaver.configuration = HYBRID by default in example INI configuration since it is the most common use case. Apply same configuration by default in tests. Default resolution still employs DEFAULT for backward compatibility in case the setting was omitted completely from a custom INI file.

  • Add query parameter ignore to GET /providers listing in order to obtain full validation of remote providers (including XML contents parsing) to return 200. Invalid definitions will raise and return a [422] Unprocessable Entity HTTP error.

  • Add more explicit messages about the problem that produced an error (XML parsing, unreachable WPS, etc.) and which caused request failure when attempting registration of a remote Provider.

Fixes:

  • Fix reported links by processes nested under a provider Service. Generated URL references were omitting the /providers/{id} portion.

  • Fix documentation referring to incorrect setting name in some cases for WPS outputs configuration.

  • Fix strict XML parsing failing resolution of some remote WPS providers with invalid characters such as <, <= within process description fields. Although invalid, those easily recoverable errors will be handled by the parser.

  • Fix resolution and execution of WPS-1 remote Provider and validate it against end-to-end test procedure from scratch Service registration down to results retrieval (fixes #340).

  • Fix resolution of applicable Provider listing schema validation when none have been registered (fixes #339).

  • Fix incorrect schema definition of Process items for GET /processes response that did not report the alternative identifier-only listing when detail=false query is employed.

  • Fix incorrect reporting of documented OpenAPI reference definitions for query parameters with same names shared across multiple endpoints. Fix is directly applied on relevant reference repository that generates OpenAPI schemas (see fmigneault/cornice.ext.swagger@70eb702).

  • Fix weaver.exception definitions such that raising them directly will employ the corresponding HTTPException code (if applicable) to generate the appropriate error response automatically when raising them directly without further handling. The order of class inheritance were always using 500 due to WeaverException definition.

4.0.0 (2021-09-21)

Changes:

  • Apply conformance updates to better align with expected ProcessDescription schema from OGC-API - Processes v1.0-draft6. The principal change introduced in this case is that process description contents will be directly at the root of the object returned by /processes/{id} response instead of being nested under "process" field. Furthermore, inputs and outputs definitions are reported as mapping of {"<id>": {<parameters>}} as specified by OGP-API instead of old listing format [{"id": "<id-value>", <key:val parameters>}]. The old nested and listing format can still be obtained using request query parameter schema=OLD, and will otherwise use OGC-API by default or when schema=OGC. Note that some duplicated metadata fields are dropped regardless of selected format in favor of OGC-API names. Some examples are abstract that becomes description, processVersion that simply becomes version, mimeType that becomes mediaType, etc. Some of those changes are also reflected by ProcessSummary during listing of processes, as well as for corresponding provider-related endpoints (relates to #200).

  • Add backward compatibility support of some metadata fields (abstract, mimeType, etc.) for Deploy operation of pre-existing processes. When those fields are detected, they are converted inplace in favor of their corresponding new names aligned with OGC-API.

  • Update mimeType to mediaType as format type representation according to OGC-API (relates to #211).

  • Add explicit pattern validation (type/subtype) of format string definitions with MediaType schema.

  • Add sorting capability to generate mapping schemas for API responses using overrides of properties _sort_first and _sort_after using lists of desired ordered field names.

  • Improved naming of many ambiguous and repeated words across schema definitions that did not necessarily interact with each other although making use of similar naming convention, making their interpretation and debugging much more complicated. A stricter naming convention has been applied for consistent Deploy/Describe/Execute-related and Input/Output-related references.

  • Replace list_remote_processes function by method processes under the Service instance.

  • Replace get_capabilities function by reusing and extending method summary under the Service instance.

  • Improve generation of metadata and content validation of Service provider responses (relates to OGC #200 and #266).

  • Add query parameter detail to providers listing request to allow listing of names instead of their summary (similarly to the processes endpoint query parameter).

  • Add query parameter check to providers listing request to retrieve all registered Service regardless of their URL endpoint availability at the moment the request is executed (less metadata is retrieved in that case).

  • Add weaver.schema_url configuration parameter and weaver.wps_restapi.utils.get_schema_ref function to help generate $schema definition and return reference to expected/provided schema in responses (relates to #157) Only utilities are added, not all routes provide the information yet.

  • Add validation of schema field under Format schema (as per opengeospatial/ogcapi-processes schema format.yml) such that only URL formatted strings are allowed, or alternatively and explicit JSON definition. Previous definitions that would indicate an empty string schema are dropped since schema is optional.

  • Block unknown and builtin process types during deployment from the API (fixes #276). Type builtin can only be registered by Weaver itself at startup. Other unknown types that have no indication for mapping to an appropriate Process implementation are preemptively validated.

  • Add parsing and generation of additional literalDataDomains for specification of WPS I/O data constrains and provide corresponding definitions in process description responses (fixes #41, #211, #297).

  • Add additional maximumMegabyte metadata detail to formats of WPS I/O of complex type whenever available (requires geopython/OWSLib#796, future OWSLIB==0.26.0 release).

Fixes:

  • Revert an incorrectly removed schema deserialization operation during generation of the ProcessSummary employed for populating process listing.

  • Revert an incorrectly modified schema reference that erroneously replaced service provider ProcessSummary items during their listing by a single ProcessInputDescriptionSchema (introduced since 3.0.0).

  • Fix #203 with explicit validation test of ProcessSummary schema for providers response.

  • Fix failing minOccurs and maxOccurs generation from a remote provider Process to support OGC-API format (relates to #263).

  • Fix schemas references and apply deserialization to providers listing request.

  • Fix failing deserialization of variable children schema under mapping when this variable element is allowed to be undefined (i.e.: defined with missing=drop). Allows support of empty inputs mapping of OGC-API representation of ProcessDescription that permits such processes (constant or random output generator).

  • Fix some invalid definitions of execution inputs schemas under mapping with value sub-schema where key-based input IDs (using additionalProperties) where replaced by the variable <input-id> name instead of their original names in the request body (from #265 since 3.4.0).

  • Fix parsing error raised from wps_processes.yml configuration file when it can be found but contains neither a processes nor providers section. Also, apply more validation of specified name values.

  • Fix parsing of request_extra function/setting parameters for specifically zero values corresponding to retries and backoff options that were be ignored.

  • Fix incorrect parsing of default field within WPS input when literal data type is present and was assumed as complex (fixes #297).

  • Fix and test various invalid schema deserialization validation issues, notably regarding PermissiveMappingSchema, schema nodes ExtendedFloat, ExtendedInt and their handling strategies when combined in mappings or keywords.

  • Fix resolution of similar values that could be implicitly converted between ExtendedString, ExtendedFloat, ExtendedInt and ExtendedBool schema types to guarantee original data type explicitly defined are preserved.

  • Fix runningSeconds field reporting to be of float type although implicit int type conversion could occur.

  • Fix validation of Execute inputs schemas to adequately distinguish between optional inputs and incorrect formats.

  • Fix resolution of Accept-Language negotiation forwarded to local or remote WPS process execution.

  • Fix XML security issue flagged within dependencies to PyWPS and OWSLib by pinning requirements to versions pywps==4.5.0 and owslib==0.25.0, and apply the same fix in Weaver code (see following for details: geopython/pywps#616, geopython/pywps#618, geopython/pywps#624, CVE-2021-39371).

3.5.0 (2021-08-19)

Changes:

  • No change.

Fixes:

  • Fix weaver.datatype objects auto-resolution of fields using either attributes (accessed as dict) or properties (accessed as class) to ensure correct handling of additional operations on them.

  • Fix DuplicateKeyError that could sporadically arise during initial processes storage creation when builtin processes get inserted/updated on launch by parallel worker/threads running the application. Operation is relaxed only for default builtin to allow equivalent process replacement (upsert) instead of only explicit inserts, as they should be pre-validated for duplicate entries, and only new definitions should be registered during this operation (fixes #246).

3.4.0 (2021-08-11)

Changes:

  • Add missing processID detail in job status info response (relates to #270).

  • Add support for inputs under mapping for inline values and arrays in process execution (relates to #265).

Fixes:

  • Fix copy of headers when generating the WPS clients created for listing providers capabilities and processes.

3.3.0 (2021-07-16)

Changes:

  • Add support for array type as job inputs (relates to #233).

  • Remove automatic conversion of falsy/truthy string and integer type definitions to boolean type to align with OpenAPI boolean type definitions. Non explicit boolean values will not be automatically converted to bool anymore. They will require explicit false|true values.

Fixes:

  • Fix minOccurs and maxOccurs representation according to OGC-API (fixes #263).

  • Fixed the format of the output file URL. When the prefix / was not present, URL was incorrectly handled by not prepending the required base URL location.

3.2.1 (2021-06-08)

Changes:

  • No change.

Fixes:

  • Fix backward compatibility of pre-deployed processes that did not define jobControlOptions that is now required. Missing definition are substituted in-place by default ["execute-async"] mode.

3.2.0 (2021-06-08)

Changes:

  • Add reference link to ReadTheDocs URL of Weaver in API landing page.

  • Add references to OGC-API Processes requirements and recommendations for eventual conformance listing (relates to #231).

  • Add datetime query parameter for job searches queries (relates to #236).

  • Add limit query parameter validation and integration for jobs in retrieve queries (relates to #237).

Fixes:

  • Pin pywps==4.4.3 and fix incompatibility introduced by its refactor of I/O base classes in #602 (specifically commit 343d825), which broke the ComplexInput work-around to avoid useless of file URLs (see issue #526).

  • Fix default execution mode specification in process job control options (fixes #182).

  • Fix old OGC-API WPS REST bindings link in landing page for the more recent OGC-API Processes specification.

  • Fix invalid deserialization of schemas using not keyword that would result in all fields returned instead of limiting them to the expected fields from the schema definitions for LiteralInputType in process description.

  • Adjust InputType and OutputType schemas to use allOf instead of anyOf definition since all sub-schemas that define them must be combined, with their respectively required or optional fields.

3.1.0 (2021-04-23)

Changes:

  • Add caching of remote WPS requests according to request-options.yml and request header Cache-Control to allow reduced query of pre-fetched WPS client definition.

  • Add POST /processes/{}/execution endpoint that mimics its jobs counterpart to respect OGC-API Processes updates (see issue opengeospatial/ogcapi-processes#124 and PR opengeospatial/ogcapi-processes#159, resolves #235).

  • Add OpenAPI schema examples for some of the most common responses.

  • Add missing schema definitions for WPS XML requests and responses.

  • Improve schema self-validation with their specified default values.

  • Add explicit options usage and expected parsing results for all test variations of OpenAPI schemas generation and colander object arguments for future reference in tests.wps_restapi.test_colander_extras.

Fixes:

  • Fix erroneous tags in job inputs schemas.

  • Fix handling of deeply nested schema validator raising for invalid format within optional parent schema.

  • Fix retrieval of database connection from registry reference.

  • Fix test mock according to installed pyramid version to avoid error with modified mixin implementations.

3.0.0 (2021-03-16)

Changes:

  • Provide HTTP links to corresponding items of job in JSON body of status, inputs and outputs routes (#58, #86).

  • Provide Job.started datetime and calculate Job.duration from it to indicate the duration of the process execution instead of counting from the time the job was submitted (i.e.: Job.created).

  • Provide OGC compliant <job-uri>/results response schema as well as some expected code/description fields in case where the request fails.

  • Add <job-uri>/outputs providing the data/href formatted job results as well as <job-uri>/inputs to retrieve the inputs that were provided during job submission (#86).

  • Deprecate <job-uri>/result paths (indicated in OpenAPI schemas and UI) in favor of <job-uri>/outputs which provides the same structure with additional links references (#58). Result path requests are redirected automatically to outputs.

  • Add more reference/documentation links to WPS-1/2 and update conformance references (#53).

  • Add some minimal caching support of routes.

  • Adjust job creation route to return 201 (created) as it is now correctly defined by the OGC API specification (#14).

  • Add Job.link method that auto-generates all applicable links (inputs, outputs, logs, etc.).

  • Add image/jpeg, image/png, image/tiff formats to supported weaver.formats (relates to #100).

  • Handle additional trailing slash resulting in HTTPNotFound [404] to automatically resolve to corresponding valid route without the slash when applicable.

  • Provide basic conda environment setup through Makefile for Windows bash-like shell (ie: MINGW/MINGW64).

  • Update documentation for minimal adjustments needed to run under Windows.

  • Update OpenAPI template to not render the useless version selector since we only provide the current version.

  • Update Swagger definitions to reflect changes and better reuse existing schemas.

  • Update Swagger UI to provide the ReadTheDocs URL.

  • Add crim-ca/cwltool@docker-gpu as cwltool requirement to allow processing of GPU-enabled dockers with nvidia-docker.

  • Add fmigneault/cornice.ext.swagger@openapi-3 as cornice_swagger requirement to allow OpenAPI-3 definitions support of schema generation and deserialization validation of JSON payloads.

  • Disable default auto-generation of request-options.yml and wps_processes.yml configuration files from a copy of their respective .example files as these have many demo (and invalid values) that fail real execution of tests when no actual file was provided.

  • Add per-request caching support when using request_extra function, and caching control according to request headers and request-options.yml configuration.

Fixes:

  • Fix weaver.config.get_weaver_config_file called with empty path to be resolved just as requesting the default file path explicitly instead of returning an invalid directory.

  • Fix CWL package path resolution under Windows incorrectly parsed partition as URL protocol.

  • Fix AttributeError of pywps.inout.formats.Format equality check compared to null object (using getter patch on null since fix #507 not released at this point).

  • Fix potential invalid database state that could have saved an invalid process although the following ProcessSummary schema validation would fail and return HTTPBadRequest [400]. The process is now saved only after complete and successful schema validation.

2.2.0 (2021-03-03)

Changes:

  • Add weaver.wps.utils.get_wps_client function to handle the creation of owslib.wps.WebProcessingService client with appropriate request options configuration from application settings.

Fixes:

  • Fix job percent progress reported in logs to be more consistent with actual execution of the process (fixes #90).

  • Fix Job duration not stopped incrementing when its execution failed due to raised error (fixes #222).

  • Improve race condition handling of builtin process registration at application startup.

2.1.0 (2021-02-26)

Changes:

  • Ensure that configuration file definitions specified in processes and providers will override older database definitions respectively matched by id and name when starting Weaver if other parameters were modified.

  • Support dynamic instantiation of WPS-1/2 processes from remote WPS providers to accomplish job execution.

  • Remove previously flagged duplicate code to handle OWSLib processes conversion to JSON for OGC-API.

  • Replace GET HTTP request by HEAD for MIME-type check against IANA definitions (speed up).

  • Improve handling of CWL input generation in combination with minOccurs, maxOccurs, allowedValues and default empty ("null") value from WPS process from remote provider (fix #17).

  • Add HYBRID mode that allows Weaver to simultaneously run local Application Packages and remote WPS providers.

  • Rename ows2json_output to ows2json_output_data to emphasise its usage for parsing job result data rather than simple output definition as accomplished by ows2json_io.

  • Remove function duplicating operations accomplished by ows2json_io (previously marked with FIXME).

  • Improve typing definitions for CWL elements to help identify invalid parsing methods during development.

  • Improve listing speed of remote providers that require data fetch when some of them might have become unreachable.

Fixes:

  • Avoid failing WPS-1/2 processes conversion to corresponding OGC-API process if metadata fields are omitted.

  • Fix invalid function employed for GET /providers/{prov}/processes/{proc} route (some error handling was bypassed).

2.0.0 (2021-02-22)

Changes:

  • Add support of YAML format for loading weaver.data_sources definition.

  • Pre-install Docker CLI in worker image to avoid bad practice of mounting it from the host.

  • Adjust WPS request dispatching such that process jobs get executed by Celery worker as intended (see #21 and #126).

  • Move WPS XML endpoint functions under separate weaver.wps.utils and weaver.wps.views to remove the need to constantly handle circular imports issues due to processing related operations that share some code.

  • Move core processing of job operation by Celery worker under weaver.processes.execution in order to separate those components from functions specific for producing WPS-REST API responses.

  • Handle WPS-1/2 requests submitted by GET KVP or POST XML request with application/json in Accept header to return the same body content as if directly calling their corresponding WPS-REST endpoints.

  • Remove request parameter of every database store methods since they were not used nor provided most of the time.

  • Changed all forbidden access responses related to visibility status to return 403 instead of 401.

  • Add more tests for Docker applications and test suite execution with Github Actions.

  • Add more details in sample configurations and provide an example docker-compose.yml configuration that defines a typical Weaver API / Worker combination with docker-proxy for sibling container execution.

  • Add captured stdout and stderr details in job log following CWL execution error when retrievable.

  • Document the WPS KVP/XML endpoint within the generated OpenAPI specification.

  • Disable auto-generation of request_options.yml file from corresponding empty example file and allow application to start if no such configuration was provided.

  • Remove every Python 2 backward compatibility references and operations.

  • Drop Python 2 and Python 3.5 support.

Fixes:

  • Target PyWPS-4.4 to resolve multiple invalid dependency requirements breaking installed packages over builtin Python packages and other compatibility fixes (see geopython/pywps #568).

  • Fix retrieval of database connexion to avoid warning of MongoClient opened before fork of processes.

  • Fix indirect dependency oauthlib missing from esgf-compute-api (cwt) package.

  • Fix inconsistent python reference resolution of builtin applications when executed locally and in tests (using virtual/conda environment) compared to within Weaver Docker image (using OS python).

  • Fix many typing definitions.

1.14.0 (2021-01-11)

Changes:

  • Add data input support for CWL Workflow step referring to WPS-3 Process.

  • Add documentation example references to Application Package and Process Deploy/Execute repositories.

  • Add parsing of providers in wps_processes.yml to directly register remote WPS providers that will dynamically fetch underlying WPS processes, instead of static per-service processes stored locally.

  • Add field visible to wps_processes.yml entries to allow directly defining the registered processes visibility.

  • Adjust response of remote provider processes to return the same format as local processes.

Fixes:

  • Fix stdout/stderr log file not permitted directly within CWL Workflow (must be inside intermediate steps).

  • Fix missing S3 bucket location constraint within unittests.

1.13.1 (2020-07-17)

Changes:

  • No change.

Fixes:

  • Create an stdout.log or stderr.log file in case cwltool hasn’t created it.

1.13.0 (2020-07-15)

Changes:

  • Add AWS S3 bucket support for process input reference files.

  • Add weaver.wps_output_s3_bucket setting to upload results to AWS S3 bucket instead of local directory.

  • Add weaver.wps_output_s3_region setting to allow override parameter extracted from AWS profile otherwise.

  • Add more documentation about supported file reference schemes.

  • Add documentation references to ESGF-CWT Compute API.

  • Add conditional input file reference fetching (depending on ADES/EMS, process type from CWL hints) to take advantage of request-options and all supported scheme formats by Weaver, instead of relying on PyWPS and/or CWL wherever how far downstream the URL reference was reaching.

Fixes:

  • Adjust some docstrings to better indicate raised errors.

  • Adjust weaver.processes.wps_package.WpsPackage to use its internal logger when running the process in order to preserve log entries under its job execution. They were otherwise lost over time across all process executions.

1.12.0 (2020-07-03)

Changes:

  • Add multiple CWL ESGF processes and workflows, namely SubsetNASAESGF, SubsetNASAESGF and many more.

  • Add tests for ESGF processes and workflows.

  • Add documentation for ESGF-CWTRequirement processes.

  • Add file2string_array and metalink2netcdf builtins.

  • Add esgf_process Wps1Process extension, to handle ESGF-CWTRequirement processes and workflows.

Fixes:

  • Reset MongoDatabase connection when we are in a forked process.

1.11.0 (2020-07-02)

Changes:

  • Generate Weaver OpenAPI specification for readthedocs publication.

  • Add some sections for documentation (#61).

  • Add support of documentation RST file redirection to generated HTML for reference resolution in both Github source and Readthedocs served pages.

  • Improve documentation links, ReadTheDocs format and TOC references.

  • Avoid logging stdout/stderr in workflows.

  • Add tests to make sure processes stdout/stderr are logged.

  • Remove Python 2.7 version as not officially supported.

  • Move and update WPS status location and status check functions into weaver.wps module.

Fixes:

  • Fix reported WPS status location to handle when starting with / although not representing an absolute path.

1.10.1 (2020-06-03)

Changes:

  • No change.

Fixes:

  • Pin celery==4.4.2 to avoid import error on missing futures.utils called internally in following versions.

1.10.0 (2020-06-03)

Changes:

  • Add support of value-typed metadata fields for process description.

  • Enforce rel field when specifying an href JSON link to match corresponding XML requirement.

Fixes:

  • Add more examples of supported WPS endpoint metadata (fixes #84).

1.9.0 (2020-06-01)

Changes:

  • Add weaver.wps_workdir configuration setting to define the location where the underlying cwltool application should be executed under. This can allow more control over the scope of the mounted volumes for Application Package running a docker image.

  • Add mapping of WPS results from the Job’s UUID to generated PyWPS UUID for outputs, status and log locations.

  • Add experimental configuration settings weaver.cwl_euid and weaver.cwl_egid to provide effective user/group identifiers to employ when running the CWL Application Package. Using these require good control of the directory and process I/O locations as invalid permissions could break a previously working job execution.

  • Add more logging configuration and apply them to cwltool before execution of Application Package.

  • Enforce no_match_user=False and no_read_only=False of cwltool’s RuntimeContext to ensure that docker application is executed with same user as weaver and that process input files are not modified inplace (readonly) where potentially inaccessible (according to settings). Definition of CWL package will need to add InitialWorkDirRequirement as per defined by reference specification to stage those files if they need to be accessed with write permissions (see: example). Addresses some issues listed in #155.

  • Enforce removal of some invalid CWL hints/requirements that would break the behaviour offered by Weaver.

  • Use weaver.request_options for WPS GetCapabilities and WPS Check Status requests under the running job.

  • Change default DOCKER_REPO value defined in Makefile to point to reference mentioned in README.md and considered as official deployment location.

  • Add application/x-cwl MIME-type supported with updated EDAM 1.24 ontology.

  • Add application/x-yaml MIME-type to known formats.

  • Add application/x-tar and application/tar+gzip MIME-type (not official) but resolved as synonym application/gzip (official) to preserve compressed file support during CWL format validation.

Fixes:

  • Set get_cwl_file_format default argument must_exist=True instead of False to retrieve original default behaviour of the function. Since CWL usually doesn’t need to add File.format field when no corresponding reference actually exists, this default also makes more sense.

1.8.1 (2020-05-22)

Changes:

  • Add Travis-CI smoke test of built docker images for early detection of invalid setup or breaking code to boot them.

  • Add Travis-CI checks for imports. This check was not validated previously although available.

  • Adjust weaver.ini.example to reflect working demo server configuration (employed by smoke test).

  • Move weaver web application to weaver.app to reduce chances of breaking setup.py installation from import errors due to weaver dependencies not yet installed. Redirect to new location makes this change transparent when loaded with the usual weaver.ini configuration.

Fixes:

  • Fix base docker image to install Python 3 development dependencies in order to compile requirements with expected environment Python version. Package python-dev for Python 2 was being installed instead.

  • Fix failing docker image boot due to incorrectly placed yaml import during setup installation.

  • Fix imports according to Makefile targets check-imports and fix-imports.

  • Fix parsing of PyWPS metadata to correctly employ values provided by weaver.ini.

1.8.0 (2020-05-21)

Changes:

  • Modify weaver.utils.request_retry to weaver.utils.request_extra to include more requests functionality and reuse it across the whole code base.

  • Add requests_extra SSL verification option using specific URL regex(es) matches from configuration settings.

  • Add file:// transport scheme support directly to utility requests_extra to handle local file paths.

  • Add file weaver.request_options INI configuration setting to specify per-request method/URL options.

  • Add requests_extra support of Retry-After response header (if any available on 429 status) which indicates how long to wait until next request to avoid automatically defined response right after.

  • Add weaver.wps_workdir configuration setting with allow setting corresponding pywps.workdir directory.

Fixes:

  • Modify Dockerfile-manager to run web application using pserve as gunicorn doesn’t correctly handles worker options anymore when loaded form weaver.ini with --paste argument. Also simplifies the command which already required multiple patches such as reapplying the host/port binding from INI file.

  • Fix handling of Literal Data I/O type when retrieved from OWSLib.wps object with remote WPS XML body.

  • Adjust make start target to use new make install-run target which installs the dependencies and package in edition mode so that configuration files present locally can be employed for running the application. Previously, one would have to move their configurations to the site-package install location of the active Python.

  • Fix celery>4.2 not found because of application path modification.

  • Fix invalid handling of wps_processes.yml reference in weaver.ini when specified as relative path to configuration directory.

  • Fix handling of WPS<->CWL I/O merge of data_format field against supported_formats with pywps>=4.2.4.

  • Fix installation of yaml-related packages for Python 2 backward compatibility.

1.7.0 (2020-05-15)

Changes:

  • Add additional status log for EOImage input modification with OpenSearch during process execution.

  • Add captured stderr/stdout logging of underlying CWL application being executed to resulting Job logs (addresses first step of #131).

  • Use weaver.utils.request_retry in even more places and extend convenience arguments offered by it to adapt it to specific use cases.

Fixes:

  • Fix handling of WPS-REST output matching a JSON file for multiple-output format specified with a relative local path as specified by job output location. Only remote HTTP references where correctly parsed. Also avoid failing the job if the reference JSON parsing fails. It will simply return the original reference URL in this case without expanded data (relates to #25).

  • Fix CWL job logs to be timezone aware, just like most other logs that will report UTC time.

  • Fix JSON response parsing of remote provider processes.

  • Fix parsing of CWL ordered parsing when I/O is specified as shorthand "<id>":"<type>" directly under the inputs or outputs dictionary instead of extended JSON object variant such as {"input": {"type:" "<type>", "format": [...]}} (fixes #137).

1.6.0 (2020-05-07)

Changes:

  • Reuse weaver.utils.request_retry function across a few locations that where essentially reimplementing the core functionality.

  • Add even more failure-permissive request attempts when validating a MIME-type against IANA website.

  • Add auto-resolution of common extensions known under PyWPS as well as employing their specific encoding.

  • Add geotiff format type support via PyWPS (#100).

  • Make WPS status check more resilient to failing WPS outputs location not found in case the directory path can be resolved to a valid local file representing the XML status (i.e.: don’t depend as much on the HTTP WPS output route).

  • Ensure backward support of generic/default text/plain I/O when extracted from a referenced WPS-1/2 XML remote process which provides insufficient format details. For CWL output generated from it, replace the glob pattern to match anything (<id>.*) instead of <id>.txt extracted from text/plain to simulate MIME-type as */*. Issue log warning message for future use cases.

Fixes:

  • Fix invalid AllowedValue parsing when using LiteralData inputs that resulted in AnyValue being parsed as a "None" string. This was transparent in case of string inputs and breaking for other types like integer when they attempted conversion.

  • Fix erroneous Metadata keywords passed down to owslib.wps.Metadata objects in case of more verbose detailed not allowed by this implementation.

  • Fix parsing of explicitly-typed optional array CWL I/O notation that was not considered (i.e.: using type as list with additional "null" instead of type: "<type>?" shorthand).

  • Fix parsing of MIME-type from format field to exclude additional parameters (e.g.: ; charset=UTF-8 for remote IANA validation.

1.5.1 (2020-03-26)

Changes:

  • Add unittest of utility function fetch_file.

  • Split some unittest utility functions to allow more reuse.

Fixes:

  • Fix invalid retry parameter not handled automatically by request.

1.5.0 (2020-03-25)

Changes:

  • Adjust incorrectly parsed href file reference as WPS complex input which resulted in failing location retrieval.

  • Partially address unnecessary fetch of file that has to be passed down to CWL, which will in turn request the file as required. Need update from PyWPS to resolve completely (#91, geopython/pywps#526).

  • Adjust WPS output results to use relative HTTP path in order to recompose the output URL if server settings change.

  • Support WPS output results as value (WPS literal data). Everything was considered an href file beforehand.

  • Add additional timeout and retry during fetching of remote file for process jsonarray2netcdf to avoid unnecessary failures during edge case connexion problems.

  • Add support of title and version field of builtin processes.

Fixes:

  • Patch builtin process execution failing since cwltool 2.x update.

  • Avoid long fetch operation using streamed request that defaulted to chuck size of 1. Now, we use an appropriate size according to available memory.

1.4.0 (2020-03-18)

Changes:

  • Update owslib to 0.19.2

  • Drop support for python 3.5

1.3.0 (2020-03-10)

Changes:

  • Provide a way to override the external URL reported by WPS-1/2 and WPS-REST via configuration settings allowing for more advanced server-side results in response bodies.

1.2.0 (2020-03-06)

Changes:

  • Add WPS languages for other wps requests types: DescribeProcess and GetCapabilities.

Fixes:

  • Fix a bug where the validation of OneOf items was casting the value to the first valid possibility.

1.1.0 (2020-02-17)

Changes:

  • Simplify docker image generation and make base/manager/worker variants all available under the same docker repo docker-registry.crim.ca/ogc/weaver with different tags (#5).

  • Add planned future support of Accept-Language header for WPS-1/2 (geopython/OWSLib 0.20.0) (#74).

  • Improved job logs update with message and progress to allow better tracking of internal operations and/or problems.

  • Allow WPS builtin process jsonarray2netcdf to fetch a remote file.

  • Change doc to point to DockerHub pavics/weaver images.

  • Adjust CI rule long-lasting failures until it gets patched by original reference (gitleaks-actions#3).

Fixes:

  • Fix readthedocs documentation generation.

  • Fix .travis docker image build condition.

  • Fix geopython/OWSLib>=0.19.1 requirement for Python 3.8 support (#62).

  • Fix job update filling due to status location incorrectly resolved according to configured PyWPS output path.

1.0.0 (2020-01-28)

New Features:

  • Add notification_email field to Job datatype that stores an encrypted email (according to settings) when provided in the job submission body (#44).

  • Add ability to filter jobs with notification_email query parameter (#44).

  • Add jobs statistics grouping by specific fields using comma-separated list groups query parameter (#46).

  • Add some tests to evaluate new job search methods / grouping results and responses (#44, #46).

  • Add handling of multiple CWL field format for File type.

  • Add missing ontology reference support for CWL field format by defaulting to IANA namespace.

  • Add support for I/O array of enum (ie: multiple values of AllowedValues for a given input) (#30).

  • Add support of label synonym as title for inputs and process description (CWL specifying a label will set it in WPS process) (#31)

  • Add support of input minOccurs and maxOccurs as int while maintaining str support (#14).

  • Add conformance route with implementation links (#53).

  • Add additional landing page link details (#54).

  • Add weaver.wps_restapi.colander_extras.DropableNoneSchema to auto-handle some schema JSON deserialization.

  • Add weaver.wps_restapi.colander_extras.VariableMappingSchema to auto-handle some schema JSON deserialization.

  • Add more functional tests (#11, #17).

Changes:

  • Use bump2version and move all config under setup.cfg.

  • Remove enforced text/plain for CWL File when missing format field.

  • Replace bubbling up of too verbose unhandled exceptions (500 Internal Server Error) by summary message and additional internal logging for debugging the cause using an utility exception log decorator.

  • Use the same exception log decorator to simplify function definitions when HTTP exceptions are already handled.

  • Make null reference a singleton so that multiple instantiation calls all refer to the same instance and produce the expected behaviour of <x> is null instead of hard-to-identify errors because of english syntax.

  • Remove unused function weaver.utils.replace_caps_url and corresponding tests.

  • Remove weaver.processes.utils.jsonify_value duplicated by weaver.processes.wps_package.complex2json.

  • Use more JSON body schema validation using API schema definitions deserialization defined by weaver.datatype.

  • Enforce builtin processes registration on startup to receive applicable updates.

  • Provide 2 separate docker images for Weaver manager and worker, corresponding to the EMS/ADES API and the celery job runner respectively.

  • Update Apache license.

Fixes:

  • Adjust some typing definitions incorrectly specified.

  • Fix some failing functionality tests (#11, #17).

  • Fix I/O field ordering preserved as specified in payload or loaded reference file.

  • Fix setting minOccurs=0 when a default is specified in the corresponding CWL I/O (#17, #25).

  • Fix incorrectly overridden maxOccurs="unbounded" by maxOccurs="1" when a partial array input definition is specified without explicit maxOccurs in WPS payload (#17, #25).

  • Fix case where omitted format[s] in both CWL and WPS deploy bodies generated a process description with complex I/O (file) without required formats field. Default text/plain format is now automatically added.

  • Fix case where format[s] lists between CWL and WPS where incorrectly merged.

  • Fix metadata field within a WPS I/O incorrectly parsed when provided by a WPS-1/2 XML process definition.

  • Fix invalid JSON response formatting on failing schema validation of process deployment body.

  • Fix docker images to support pserve when using gunicorn>=20.x dropping support of --paste config feature.

  • Fix multiple Python 2/3 compatibility issues.

0.2.2 (2019-05-31)

  • Support notification email subject template.

0.2.1 (2019-05-29)

  • Add per-process email notification template.

0.2.0 (2019-03-26)

  • Fixes to handle invalid key characters "$" and "." during CWL package read/write operations to database.

  • Fixes some invalid CWL package generation from WPS-1 references.

  • More cases handled for WPS-1 to CWL WPS1Requirement conversion (AllowedValues, Default, SupportedFormats, minOccurs, maxOccurs).

  • Add file format validation to generated CWL package from WPS-1 MIME-types.

  • Allow auto-deployment of WPS-REST processes from WPS-1 references specified by configuration.

  • Add many deployment and execution validation tests for WPS1Requirement.

  • Add builtin application packages support for common operations.

0.1.3 (2019-03-07)

  • Add useful Makefile targets for deployment.

  • Add badges indications in README.rst for tracking from repo landing page.

  • Fix security issue of PyYAML requirement.

  • Fix some execution issues for Wps1Process.

  • Fix some API schema erroneous definitions.

  • Additional logging of unhandled errors.

  • Improve some typing definitions.

0.1.2 (2019-03-05)

  • Introduce WPS1Requirement and corresponding Wps1Process to run a WPS-1 process under CWL.

  • Remove mongodb requirement, assume it is running on an external service or docker image.

  • Add some typing definitions.

  • Fix some problematic imports.

  • Fix some PEP8 issues and PyCharm warnings.

0.1.1 (2019-03-04)

  • Modify Dockerfile to use lighter debian:latest instead of birdhouse/bird-base:latest.

  • Modify Dockerfile to reduce build time by reusing built image layers (requirements installation mostly).

  • Make some buildout dependencies optional to also reduce build time and image size.

  • Some additional striping of deprecated or invalid items from Twitcher.

0.1.0 (2019-02-26)

  • Initial Release. Based off Twitcher tag ogc-0.4.7.