devpi-server 3.1.0

devpi-server: reliable private and pypi.python.org caching server

devpi-server: pypi server for caching and private indexes

• issue tracker, repo

• IRC: #devpi on freenode, mailing list

• compatibility: {win,unix}-py{26,27,34}

consistent robust pypi-cache

You can point pip or easy_install to the root/pypi/+simple/ index, serving as a self-updating transparent cache for pypi-hosted and external packages. Cache-invalidation uses the latest and greatest PyPI protocols. The cache index continues to serve when offline and will resume cache-updates once network is available.

user specific indexes

Each user (which can represent a person or a project, team) can have multiple indexes and upload packages and docs via standard setup.py invocations command. Users and indexes can be manipulated through a RESTful HTTP API.

index inheritance

Each index can be configured to merge in other indexes so that it serves both its uploads and all releases from other index(es). For example, an index using root/pypi as a parent is a good place to test out a release candidate before you push it to PyPI.

good defaults and easy deployment

Get started easily and create a permanent devpi-server deployment including pre-configured templates for nginx and cron.

separate tool for Packaging/Testing activities

The complimentary devpi-client tool helps to manage users, indexes, logins and typical setup.py-based upload and installation workflows.

See http://doc.devpi.net for getting started and documentation.

Changelog

3.1.0 (2016-04-22)

• fix issue208: Uncached mirrored files (PyPI) are streamed to the client while downloading. This prevents timeouts in pip etc. The files are only cached if there were no errors and in case there is a checksum, the content matches. Downloads on replicas won’t wait until they are in sync, but pass on what they get from the master.

• fix issue229: A replica talking to a master behind nginx decoded gzipped data, but left the Content-Encoding header unchanged. Now data is passed on unchanged. Thanks to Chad Wagner for the fix.

• fix issue317: When there is no data in the directory specified via --serverdir during export, then the process aborts instead of creating and exporting an empty database.

• fix issue210: When an external user authenticated by a plugin tries to create an index the required user object is now created automatically if the permissions allow it.

• address issue267: We unconditionally clean up the transaction if there was an exception in rollback or commit. This prevents issues in logging and a possible server lockup if at some point all threads contain a failed transaction object.

• fix issue321: All exceptions in the replica and event processing threads are caught now and can’t stop the threads anymore.

• fix issue338: Handle trailing slash in project listing for mirror indexes.

• Added checks on the index dependency tree built from bases during import.

• Every project is now imported together with all it’s release files on it’s own serial. Before the release files each got their own serial. This reduces the number of serials generated, especially when there are many projects and releases. That in turn improves import, as well as replication and event handling times (in particular devpi-web indexing).

3.0.2 (2016-03-03)

• fix setting of mirror_whitelist.

• normalize names when setting mirror_whitelist.

• fix handling of 404 in mirror indexes on replicas.

• include version in file paths in exported data to avoid possible name conflicts.

3.0.1 (2016-02-12)

• fix importing of uploaded files. Only the last index from exported data was processed.

3.0.0 (2016-02-12)

• dropped support for python2.6

• block most ascii symbols for user and index names except -.@_. unicode characters are fine.

• add --no-root-pypi option which prevents the creation of the root/pypi mirror instance on first startup.

• added optional title and description options to users and indexes.

• new indexes have no bases by default anymore. If you want to be able to install pypi packages, then you have to explicitly add root/pypi to the bases option of your index.

• added optional custom_data option to users.

• generalized mirroring to allow adding mirror indexes other than only PyPI

• renamed pypi_whitelist to mirror_whitelist

• speed up simple-page serving for private indexes. A private index with 200 release files should now be some 5 times faster.

• internally use normalized project names everywhere, simplifying code and slightly speeding up some operations.

• change {name} in route_urls to {project} to disambiguate. This is potentially incompatible for plugins which have registered on existing route_urls.

• use “project” variable naming consistently in APIs

• drop calling of devpi_pypi_initial hook in favor of the new “devpi_mirror_initialnames(stage, projectnames)” hook which is called when a mirror is initialized.

• introduce new “devpiserver_stage_created(stage)” hook which is called for each index which is created.

• simplify and unify internal mirroring code some more with “normal” stage handling.

• don’t persist the list of mirrored project names anymore but rely on a per-process RAM cache and the fact that neither the UI nor pip/easy_install typically need the projectnames list, anyway.

• introduce new “devpiserver_storage_backend” hook which allows plugins to provide custom storage backends. When there is more than one backend available, the “–storage” option becomes required for startup.

• introduce new “–requests-only” option to start devpi-server in “worker” mode. It can be used both for master and replica sites. It starts devpi-server without event processing and replication threads and thus depends on respective “main” instances (those not using “–request-only”) to perform event and hook processing. Each worker instance needs to share the filesystem with a main instance. Worker instances can not serve the “/+status” URL which must always be routed to the main instance.

2.6.1 (2016-03-03)

• add more info when importing data. Thanks Marc Abramowitz for the PR.

• include version in file paths in exported data to avoid possible name conflicts.