Resources 27.0.0 2024-05-28¶

This release requires Python 3.11 or newer.

New Features¶

Added methods to ResourcePath for generating presigned HTTP URLs: generate_presigned_get_url and generate_presigned_put_url. These are currently only implemented for S3. (DM-41879)
The forceDirectory flag now has three states. As before True indicates that it is known that the URI refers to a directory-like entity. Now False indicates that it is known that the URI refers to a file-like entity. None is the new default and that indicates that the caller does not know and that the status should be inferred either by checking the file system or looking for a trailing /. It is now an error to create a ResourcePath which has a trailing / but with forceDirectory=False. (DM-42306)
S3ResourcePath now supports using multiple S3 endpoints simultaneously. This is configured using URIs in the form s3://profile@bucket/path and environment variables LSST_RESOURCES_S3_PROFILE_<profile>=https://<access key ID>:<secret key>@<s3 endpoint hostname>. (DM-42704)
Allow threading of S3 downloads to be turned off by setting either the LSST_S3_USE_THREADS environment variable or the S3ResourcePath.use_threads class member. (PREOPS-4765)

When the process’s filesystem umask set to a restrictive value like 0222, transfer_from, write, and mkdir no longer fail due to incorrect permissions on newly-created parent directories. (DM-41112)
Installing optional dependencies for s3 and https will no longer pull in libraries only required for running unit tests (moto and responses). (DM-41547)
HttpResourcePath.exists() and HttpResourcePath.size() now work for S3 HTTP URLs pre-signed for GET. (DM-42522)
ResourePath.root_uri() now strips query parameters and fragments from the URL. This fixes a memory leak where HttpResourcePath would create and cache a new HTTP session for each different set of query parameters. (DM-43739)

- Schemeless URIs no longer check the file system on construction.
- Both getExtension and relativeToPathRoot have been rewritten to no longer use pathlib.
- It is now possible to declare that a URI is file-like on construction. Use forceDirectory=False. (DM-42306)

Deprecated clean_test_environment, setAwsEnvCredentials, and unsetAwsEnvCredentials from the s3utils submodule. The new function clean_test_environment_for_s3 replaces these. (DM-41879)

This package now requires Python 3.10 and newer.

resource URI schemes now use importlib.resources (or importlib_resources) rather than the deprecated pkg_resources. Due to this change, resource URI schemes now also support walk and findFileResources. (DM-33528)
- Modified the way that a schemeless absolute URI behaves such that we now always convert it to a file URI.
- The root parameter can now use any ResourcePath scheme such that a relative URI can be treated as a URI relative to, for example, a S3 or WebDAV root. (DM-38552)
The LSST_DISABLE_BUCKET_VALIDATION environment variable can now be set to disable validation of S3 bucket names, allowing Ceph multi-tenant colon-separated names to be used. (DM-38742)
- Added support for as_local for Python package resource URIs.
- Added explicit isdir() implementation for Python package resources. (DM-39044)

Fixed problem where a fragment associated with a schemeless URI was erroneously being quoted. (DM-35695)
Fixed invalid endpoint error in the FileReadWriteTestCase test when the S3_ENDPOINT_URL environment variable is set to an invalid endpoint. (DM-37439)
- Fixed EOF detection with S3 and HTTP resource handles when using repeated read().
- Ensured that HTTP reads with resource handles using byte ranges correctly disable remote compression. (DM-38589)
Reorganized mexists() implementation to allow S3 codepath to ensure that a client object was created before using multi-threading. (DM-40762)

ResourcePathExpression can now be used in an isinstance call on Python 3.10 and newer. (DM-38492)

For file copies with transfer_from() an attempt is now made to make the copies atomic by using os.rename with a temporary intermediate. Moves now explicitly prefer os.rename and will fall back to an atomic copy before deletion if needed. This is useful if multiple processes are trying to copy to the same destination file. (DM-36412)
Added allow_redirects=True to WebDAV HEAD requests since the default is False. This is needed when interacting with WebDAV storage systems which have a frontend redirecting to backend servers. (DM-36799)

This package is now available on PyPI as lsst-resources.
The lsst.daf.butler.ButlerURI code has been extracted from the daf_butler package and made into a standalone package. It is now known as lsst.resources.ResourcePath and distributed in the lsst-resources package.
Add support for Google Cloud Storage access using the gs URI scheme. (DM-27355)
Builds using setuptools now calculate versions from the Git repository, including the use of alpha releases for those associated with weekly tags. (DM-32408)
Add an open method that returns a file-like buffer wrapped by a context manager. (DM-32842)
Major cleanup of the WebDAV interface:
- Improve client timeout and retries.
- Improve management of persistent connections to avoid exhausting server resources when there are thousands of simultaneous clients.
- Rename environment variables previously named LSST_BUTLER_* by:
  - LSST_HTTP_CACERT_BUNDLE
  - LSST_HTTP_AUTH_BEARER_TOKEN
  - LSST_HTTP_AUTH_CLIENT_CERT
  - LSST_HTTP_AUTH_CLIENT_KEY
  - LSST_HTTP_PUT_SEND_EXPECT_HEADER (DM-33769)

Reorganize test code to enhance code reuse and allow new schemes to make use of existing tests. (DM-33394)
Attempt to catch 429 Retry client error in S3 interface. This code is not caught by botocore itself since it is not part of the AWS standard but Google can generate it. (DM-33597)
When walking the local file system symlinks to directories are now followed. (DM-35446)