PrefectHQ
prefect
BlogDocsChangelog

Branches performance

Pull requests

chore: switch to ty type checker#19835
last run
9 hours ago
chore: switch to ty type checker ty is astral's new extremely fast python type checker written in rust. this replaces mypy in pre-commit with ty while keeping mypy for the pytest-mypy-plugins typesafety tests. starting with permissive rules - all existing errors are ignored for now and can be gradually enabled as we fix them. see https://astral.sh/blog/ty 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
9 hours ago
b26f62d
chore/switch-to-ty
CodSpeed Performance Gauge
0%
fix: cancel orphaned flow runs when deployment is deleted during concurrency wait#19829
last run
13 hours ago
fix: cancel orphaned flow runs when deployment is deleted during concurrency wait When a flow run is waiting in AwaitingConcurrencySlot state and its deployment is deleted, the worker would previously keep polling the run indefinitely. The server's SecureFlowConcurrencySlots rule returns an ABORT when the deployment doesn't exist, but the worker only logged this and returned False, leaving the run stuck. This fix cancels the run when we receive a "Deployment not found" abort, preventing the infinite polling loop and giving users visibility that the run was terminated due to deployment deletion. Also adds regression tests for: - Server correctly updating next_scheduled_start_time on concurrency rejection - Worker cancelling runs on deployment-not-found abort - Worker NOT cancelling runs for other abort reasons (like concurrency limit=0) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
14 hours ago
08bca0b
fix/cancel-orphaned-runs-with-deleted-deployments
CodSpeed Performance Gauge
0%
Fix deployment concurrency slot starvation bug When a flow run is rejected to AwaitingConcurrencySlot state, its scheduled_time was set to now + 30 seconds (concurrency_slot_wait_seconds default). However, workers poll for runs with scheduled_before = now + 10 seconds (prefetch_seconds default). This mismatch caused runs to be invisible to workers for 20+ seconds, and if the slot was still unavailable when finally polled, they'd be rescheduled another 30 seconds out - creating indefinite starvation. Fix: Reduce concurrency_slot_wait_seconds default from 30s to 5s, ensuring runs become visible within the worker's prefetch window. Adds regression tests: - test_enqueued_runs_acquire_slots_after_release - test_enqueued_runs_visible_in_worker_polling_query - test_awaiting_concurrency_slot_visible_within_worker_prefetch_window 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
15 hours ago
a88dcb7
fix/deployment-concurrency-slot-starvation
CodSpeed Performance Gauge
0%
Remove `parameters` and `context` fields from TaskRunResponse#19816
last run
17 hours ago
Regenerate UI TypeScript types 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
17 hours ago
837f110
remove-taskrun-context-parameters
CodSpeed Performance Gauge
0%
© 2025 CodSpeed Technology
Home Terms Privacy Docs