PrefectHQ/prefect - CodSpeed

prefect

Blog Docs Changelog

Performance History

Latest Results

Regenerate api-ref docs, to update code line numbers Signed-off-by: Rich Scott <richscott@sent.com>

richscott:richscott/armada-integration

8 hours ago

Add missing conftest.py for prefect-armada integration tests. Signed-off-by: Rich Scott <richscott@sent.com>

richscott:richscott/armada-integration

8 hours ago

fix(runner): keep local changes when updating the shallow clone Review - `git reset --hard FETCH_HEAD` rewrote locally modified tracked files on every pull, even with no new remote commit, while `git pull` left them alone when the branch was already up to date. Change - Use `git reset --keep FETCH_HEAD`. It is a no-op when nothing changed upstream, applies the new commit while keeping local edits to files it doesn't touch, and refuses when a local edit would be overwritten, which falls back to the existing re-clone like the old pull path did. Tests - Real-git tests for a local edit with an unchanged remote and for an unrelated local edit with an advancing remote (both fail with `--hard`), for the conflicting-edit fallback, and for a submodule bump applied in place. - Mocked command expectations updated to `--keep`. Refs #16241

CassianoE:fix-16241-shallow-pull

9 hours ago

fix(runner): update shallow git clones with fetch + reset instead of pull Problem - GitRepository clones with `--depth 1` and then updated the clone with `git pull origin <branch> --depth 1`. Once the remote branch has a new commit, the shallow clone shares no history with it, so the pull fails on every update ("Need to specify how to reconcile divergent branches" on recent git, "refusing to merge unrelated histories" on older git) and `_pull_code_locked` falls back to rmtree + re-clone. - The re-clone deletes a directory that may be in use: `.serve()` runs flow runs from `storage.destination`, and process workers whose pull steps share a directory behave the same way. Change - Fetch the branch tip with `git fetch origin <branch> --depth 1` and move the checkout to it with `git reset --hard FETCH_HEAD`. - Run `git submodule update --init --recursive` after the reset when `include_submodules` is set, replacing `git pull --recurse-submodules`. - Keep the rmtree + re-clone fallback for real fetch or reset failures. Tests - New regression test with real git repositories: a new commit, a merge commit and a force push are applied in place, and an untracked file in the checkout survives the update. It fails on the previous implementation. - Updated the mocked command expectations that assumed `git pull`. Refs #16241

CassianoE:fix-16241-shallow-pull

10 hours ago

fix(runner): refresh _loop on every runner start A runner restarted under a new event loop kept the closed original loop, so SIGTERM scheduled astop on it. Refs OSS-8304 Co-authored-by: Alex Streed <desertaxle@users.noreply.github.com>

devin/OSS-8304-runner-sigterm-astop

12 hours ago

Fix `create_default_prefect_yaml` crash when `contents` is omitted Normalizes a missing or explicitly None contents argument to an empty dictionary, matching the docstring behavior. Adds regression tests.

Khadija-de:fix-23333-default-prefect-yaml

19 hours ago

Handle omitted project name in prefect.yaml

MuhammadWaleed12:fix/default-prefect-yaml-none-contents

1 day ago

Check for an existing work pool before Armada test setup When `--work-pool-name` named a pool that already existed, the Armada integration tests could create the Armada queue, and build and load the flow image, before the work pool fixture rejected the name. Do the check in a `pytest_sessionstart` hook instead, so the session stops before any fixture has run. Move the command line options and that hook to a `conftest.py` at the root of the integration tests. Pytest now loads them before it parses the command line, so `--work-pool-name foo` is accepted as well as `--work-pool-name=foo`. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: Rich Scott <richscott@sent.com>

richscott:richscott/armada-integration

1 day ago

Latest Branches

0%

Add work pool support for Armada (K8s multi-cluster) batch system#23232

8 hours ago

ad53c15

richscott:richscott/armada-integration

0%

Fix `GitRepository.pull_code()` re-cloning a shallow clone on every new commit#23351

10 hours ago

466b8ff

CassianoE:fix-16241-shallow-pull

0%

fix(runner): schedule `astop` on SIGTERM instead of dropping the coroutine#23346

12 hours ago

ef6408c

devin/OSS-8304-runner-sigterm-astop

© 2026 CodSpeed Technology

Home Terms Privacy Docs