PrefectHQ/prefect - CodSpeed

prefect

Blog Docs Changelog

Performance History

Latest Results

Keep remaining non-Optional at the budget comparison Pyright's CI config rejects `deadline is not None and remaining <= 0`: it does not narrow `remaining` through the `deadline` correlation (reportOptionalOperand on both wait paths). Compute and compare the remaining budget inside the `deadline is not None` branch so the operand is always float there. Co-Authored-By: ZCode <zcode@bigmodel.ai>

Wilson0511:fix/memory-lock-acquire-toctou

3 hours ago

Share the acquire-timeout budget across stale-entry retries Codex review on efd6d8a (both P2s): - A stale-entry retry re-entered the wait with the caller's full acquire_timeout, so one call could wait nearly twice the requested maximum (and repeated replacements could extend it without bound). acquire_lock/aacquire_lock now compute a single monotonic deadline up front and the internal retry path re-derives the remaining budget each wait; a retry that finds the budget spent returns False instead of waiting again. - The replacement tests now synchronize on the waiter actually entering the stale object's acquire before the entry is expired (the async twin replaced its fixed sleep with the same handshake), so the stale-object branch is guaranteed to be exercised. - New regression: the retry spends the remaining budget, not a fresh one (total wait for a mid-budget replacement stays under the single acquire_timeout; a budget-restarting retry would run ~140% of it). Co-Authored-By: ZCode <zcode@bigmodel.ai>

Wilson0511:fix/memory-lock-acquire-toctou

3 hours ago

fix(utilities): scan pydantic models and dataclasses for ambiguous hash inputs Codex review on #23819 noted the ambiguity scan missed tuples hidden inside Pydantic models: the JSON serializer dumps models with model_dump(mode="json"), which normalizes tuple and list fields to the same array, so the collision survived inside the wrapper. Scan and encode models (and dataclasses, which the serializer flattens the same way) through their python-mode dump, which preserves container types. Unambiguous inputs still take the legacy path, so persistent hashes such as block schema checksums are unchanged. Signed-off-by: Shahzeb Ali <126076773+Shahzeb-DataEngineer@users.noreply.github.com>

Shahzeb-DataEngineer:fix/hash-objects-type-preserving

3 hours ago

fix parallel tasks failing when using kwargs

uinstinct:kwargs

6 hours ago

Keep hashes unchanged for JSON-native dicts using tag key names Co-authored-by: Alex Streed <desertaxle@users.noreply.github.com>

devin/1794201032-hash-objects-type-preserving

8 hours ago

Promote hashing to a subpackage and document its contracts Co-authored-by: Alex Streed <desertaxle@users.noreply.github.com>

devin/1794201032-hash-objects-type-preserving

8 hours ago

test(server): check db vacuum tasks run once across two scheduler cycles The two-cycle test now holds whichever task takes the slot first, waits for the second cycle's copies to park, and asserts one execution per key. The vacuum tests share one helper that stubs the maintenance connection. The scheduler docstring says that Docket drops the extra copy.

ShikhJohari:t3code/22912-serialize-db-vacuum-tasks

9 hours ago

Make nested state aggregation opt-in behind an experimental setting Recursive aggregation changes existing flow and task outcomes: a flow returning [[Failed(...)]] previously completed and now fails. Gate it behind the new `experiments.aggregate_nested_states` setting, disabled by default, following the default-off pattern of `runner.crash_on_cancellation_failure`. - Add `PREFECT_EXPERIMENTS_AGGREGATE_NESTED_STATES` (default false). When disabled, `return_value_to_state` and its sync counterpart keep the historical behavior: only a single state or a flat collection of states is aggregated. - When enabled, nested collections of states are aggregated for flows and tasks, with sync/async regressions. - `get_state_exception`/`aget_state_exception` keep understanding nested collections regardless of the setting, so a persisted aggregate state still reports its failure when read back under a different setting; covered by round-trip tests (in-memory and via persisted result metadata). - Document the collection boundaries (mixed data, empty containers, reference cycles, 1000-level depth cap) and a migration note (opt in via the setting; preserve intentional state-as-data returns with `quote()` or by leaving the setting off) on the flows concept page; regenerate the settings reference and schema.

Nacl-HCl:fix-nested-states-final-state-23137

11 hours ago

Latest Branches

+23%

Fix MemoryLockManager.acquire_lock silently revoking a holder after entry replacement#23789

3 hours ago

0a39a5f

Wilson0511:fix/memory-lock-acquire-toctou

+22%

fix(utilities): make hash_objects type-preserving so tuple/list inputs do not collide#23819

3 hours ago

179f4b7

Shahzeb-DataEngineer:fix/hash-objects-type-preserving

+23%

fix parallel tasks failing when using kwargs#23820

6 hours ago

432c6c1

uinstinct:kwargs

© 2026 CodSpeed Technology

Home Terms Privacy Docs