PrefectHQ
prefect
Blog
Docs
Changelog
Blog
Docs
Changelog
Overview
Branches
Benchmarks
Runs
Performance History
Latest Results
docs: Add release notes for 3.8.4 (#22942) Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> Co-authored-by: alex.s <alex.s@prefect.io> Co-authored-by: alex.s <ajstreed1@gmail.com>
main
11 hours ago
Fix ProcessWorker auto-uv after pull steps (#22857)
main
13 hours ago
Preserve workspace attribution metadata
codex/process-worker-pulled-auto-uv
14 hours ago
Preserve runner ownership during workspace preparation
codex/process-worker-pulled-auto-uv
15 hours ago
Count initial-connection attempts separately from the retry budget The first version raised the shared _send_with_retry budget to max(PREFECT_CLIENT_MAX_RETRIES, connection_attempts), which let the two budgets bleed into each other. With the default max retries of 5, setting connection attempts to 1 still made six connection calls, and setting max retries to 0 with connection attempts of 10 would retry a reachable server returning 503 ten times, resubmitting non-idempotent requests. Give initial ConnectErrors their own counter instead, so the connection budget can neither widen nor narrow the allowance for a server that is actually answering. The setting is now documented as a total attempt count: 5 means at most five connection attempts.
Chinmayrawat15:client-connection-attempts
24 hours ago
Add PREFECT_CLIENT_CONNECTION_ATTEMPTS for initial API connection retries Before the client has connected successfully even once, httpx.ConnectError is excluded from the retry set, so an API that is not up yet fails immediately with "Failed to reach API at ...". Neither PREFECT_CLIENT_MAX_RETRIES nor a flow's own retries help, because the flow run is never created. This adds an opt-in attempt budget for that window. It defaults to 0, keeping the existing fail-fast behaviour for a misconfigured API URL, and reuses the existing exponential backoff and jitter settings when enabled.
Chinmayrawat15:client-connection-attempts
1 day ago
Add PREFECT_CLIENT_CONNECTION_ATTEMPTS for initial API connection retries Before the client has connected successfully even once, httpx.ConnectError is excluded from the retry set, so an API that is not up yet fails immediately with "Failed to reach API at ...". Neither PREFECT_CLIENT_MAX_RETRIES nor a flow's own retries help, because the flow run is never created. This adds an opt-in attempt budget for that window. It defaults to 0, keeping the existing fail-fast behaviour for a misconfigured API URL, and reuses the existing exponential backoff and jitter settings when enabled.
Chinmayrawat15:client-connection-attempts
1 day ago
Prevent revocation of live filesystem concurrency leases Serialize the read-modify-write of the shared expiration index so concurrent renewals cannot lose each other's updates, verify candidate expirations against the authoritative lease file in `read_expired_lease_ids`, and re-check the expiration in `revoke_expired_lease` before slots are released. Closes #22935 (OSS-8173) Co-authored-by: Alex Streed <desertaxle@users.noreply.github.com>
devin1/oss-8173-filesystem-lease-storage-unsynchronised-expiration-index
1 day ago
Latest Branches
CodSpeed Performance Gauge
0%
Fix ProcessWorker auto-uv after pull steps
#22857
14 hours ago
7d2d335
codex/process-worker-pulled-auto-uv
CodSpeed Performance Gauge
0%
Add PREFECT_CLIENT_CONNECTION_ATTEMPTS for initial API connection retries
#22937
1 day ago
17a98c8
Chinmayrawat15:client-connection-attempts
CodSpeed Performance Gauge
0%
Prevent revocation of live filesystem concurrency leases
#22936
1 day ago
c04ef6e
devin1/oss-8173-filesystem-lease-storage-unsynchronised-expiration-index
© 2026 CodSpeed Technology
Home
Terms
Privacy
Docs