PrefectHQ/prefect 3.7.3

3.7.3 - Scheduler go brrrr

on GitHub

Automatic dependency installation for pulled-code deployments is now opt-in. In 3.7.2, the runner would automatically prepare project dependencies when a deployment pulled code containing a pyproject.toml. This behavior is now disabled by default. Set PREFECT_RUNNER_AUTO_INSTALL_DEPENDENCIES=true (or runner.auto_install_dependencies = true in your profile) to re-enable it.

Enhancements ➕➕

• Expose WebSocket ping interval/timeout settings for uvicorn by @chuqCTC in #22091
• Activate scheduler partial index for faster scheduling queries by @zzstoatzz in #22111
• Lazify prefect.server package exports for faster imports by @devin-ai-integration[bot] in #22122

Bug Fixes 🐞

• Raise pydantic-settings lower bound to >=2.9.1 by @devin-ai-integration[bot] in #22075
• Cap reconnect_delay exponent to prevent OverflowError by @devin-ai-integration[bot] in #22086
• Guard ZeroDivisionError in RetryFailedTasks when retry_jitter_factor is set without retry_delay by @devteamaegis in #22071
• Redact API URL userinfo in client errors by @devin-ai-integration[bot] in #22098
• Execute local deployment paths in place by @devin-ai-integration[bot] in #22100
• Preserve persisted result metadata by @desertaxle in #22110
• Make runner dependency installation opt-in by @desertaxle in #22116
• Preserve TaskRunRecorder upsert order by @desertaxle in #22112
• Lowercase work pool name in block document names for storage configure by @devin-ai-integration[bot] in #22144
• Remove double client injection in _resolve_flow_for_hooks by @devin-ai-integration[bot] in #22176

Integrations & Dependencies 🤝

• Bump starlette to 1.2.0 for CVE-2026-48710 by @zzstoatzz in #22145

Development & Tidiness 🧹

• Add server-side cleanup queue storage for worker communication by @desertaxle in #22041
• Don't copy file-derived labels from linked issues by @zzstoatzz in #22115
• Support string keys in temporary_settings by @zzstoatzz in #22120
• Clean up WebSocket ping settings surface and tests by @devin-ai-integration[bot] in #22117
• Fix flaky tests on main — regenerate SettingAccessor and widen timing tolerance by @devin-ai-integration[bot] in #22133
• Add timeout and transport retries to hosted_api_client fixture by @devin-ai-integration[bot] in #22130
• Use direct API calls in suspension integration test by @devin-ai-integration[bot] in #22157
• Use direct API calls in suspension integration test by @devin-ai-integration[bot] in #22152
• Pin local worker registry in test_create_with_unsupported_type by @devin-ai-integration[bot] in #22153
• Remove unneeded clear_db markers from isolated test modules by @devin-ai-integration[bot] in #22102
• Update CODEOWNERS to remove obsolete entries by @desertaxle in #22178

Documentation 📓

• Fix work-pool concurrency command examples by @devin-ai-integration[bot] in #22093
• Clarify non-interactive Prefect Cloud login requirements by @devin-ai-integration[bot] in #22095
• Fix multi-worker server docs Redis requirements by @devin-ai-integration[bot] in #22107

Full Changelog: 3.7.2...3.7.3