Repository navigation
Releases
Penelope v0.4.0
Penelope adds rewindable computation alongside the saga orchestration and
recoverable task execution available in 0.3.0. Applications can reconcile local
predictions with admitted state or input corrections, replay on isolated scratch,
and publish validated results. Existing saga/task v1 formats and recovery
semantics remain unchanged.
Added
- Optional
penelope-timelinecrate for bounded replay, candidate validation
and staged effect intents, with a defaultno_stdprofile and fixed buffers
that require no heap allocation. - Independent timeline scopes, logical steps, branches and revisions, correction
planning, and the versionedpenelope.timeline.v1binary metadata codec. - Fixed-capacity immutable input and full-snapshot history with byte limits,
branch ancestry, pinned replay, explicit eviction and branch materialization. - Atomic state/head/intent publication through an exclusive owner or optional
hosted shared lock, with stale-candidate fencing and non-reused branch IDs. - Separately authorized effect finality and idempotent durable-admission
contracts for handing off finalized intents. - Owned-state input application, scoped correction cleanup, closure validators
and captured input/snapshot adapters, with a runnable quickstart. - Integer simulation and optimistic invoice examples, including SQLite handoff
across lost admission/delivery acknowledgements, plus client prediction tests
covering delayed snapshots, unacknowledged inputs and continued prediction.
Security
- Timeline scope, revision, authority ordering and finalized-prefix checks
refuse stale or incompatible corrections before publication. - Correction authentication, snapshot completeness, deterministic computation
and effect authorization remain explicit application responsibilities.
Replay does not reverse already-issued external effects.
Changed
- The facade exposes rewindable computation as
penelope::timelinethrough the
optionaltimelinefeature; existing users do not acquire that dependency
by default. - Release verification covers all ten library packages, with timeline portable
checks, examples, property/fuzz campaigns and scoped performance measurements.
Penelope v0.3.0
Added
- Optional
penelope-sqliteadapter for durable task histories and blobs,
generation authority, atomic process commits, inbox deduplication, and leased
outbox/timer delivery. Includes scoped quotas and pending-input discovery. - Optional
penelope-process-runtimeadapter for protected Linux workers, with
sealed executable identity, bounded IPC, resource restrictions and durable
generation fencing for the documented single-host deployment profile. - Typed saga/task composition and bounded services for committed dispatch,
verified results, independent effect reconciliation, task-result relay and
durable timer delivery. - Shared allocation-free task supervision through
TaskControland caller-owned
TaskWorkspace, alongside the existing hosted supervisor. - Integration examples and deployment documentation for the new adapters and
services, with crash, concurrency, corruption and recovery verification.
Changed
- Reduce allocations and repeated work in identifier encoding/decoding,
linear/graph definition validation, saga completion and hosted task dispatch.
Performance remains workload-dependent; scoped measurements and regressions
are documented in the performance reports. - Add
ProcessInputKind::TaskEffectResultand five
ProcessAuthorizationOperationvariants:Dispatch,DriveTask,
ReconcileEffect,DeliverTimerandRelayTaskResult. Downstream exhaustive
matches must handle the new variants. Existing saga/task v1 encodings remain
compatible. - Re-export the shared controller error as the existing hosted
penelope_executor::task::SupervisorError.
Penelope v0.2.0
Penelope now provides recoverable task execution alongside deterministic saga
orchestration. Applications can declare recovery permissions, persist valid
checkpoints and reconcile ambiguous effects through runtime-independent APIs.
The existing saga engine remains independently useful, with its v1 history
representations and execution guarantees preserved.
Added
- Recoverable task definitions with explicit restart-from-input,
restart-from-checkpoint, reconcile-before-restart and never-restart policies. - Checkpoint-aware execution with immutable input and result records, committed
recovery boundaries and application-authorized checkpoint migration. - Task supervision with generation fencing, typed fault observations, bounded
recovery attempts, deadlines, restart delays and execution/resource budgets. - Independently versioned task history and deterministic reconstruction without
rerunning computation or repeating external effects during replay. - Backend-neutral synchronous journal and runtime contracts, plus a bridge to
existing saga effect execution. - Optional StateChronicle verification evidence bound to the exact task scope,
generation and checkpoint identity. - An allocation-free
no_stdtask core, optional portable Serde support,
distinct identity/time newtypes and caller-owned bounded execution buffers. - Recoverable computation examples and deployment contracts covering journal
durability, runtime enforcement, checkpoint schemas and external-effect evidence. - Recovery validation through process termination against filesystem and
transactional SQLite test journals, concurrent supervisor stress, property
testing and sanitizer-instrumented fuzz targets.
Security
- Task scope, definition, generation, sequence and monotonic time are checked
before execution history advances; stale or cross-scope results fail closed. - Input commits precede execution, and checkpoints/results commit before
acknowledgement. Ambiguous persistence outcomes require reconstruction. - Live generations retain capacity for a fault or completion; rejected resource
proposals fence execution before recovery proceeds. - Corrupt checkpoints, incompatible schemas and unknown effects cannot become
unsupported claims of completion or authorize blind retries. - Runtime isolation, CPU enforcement, storage authenticity and the correctness
of computation and physical side effects remain explicit application contracts.
Changed
- The facade exposes both saga orchestration and recoverable task execution
while preserving existing linear/graph decisions, retries, compensation and
reconciliation semantics. - Task definition fingerprints bind recovery permissions and resource budgets
to execution scope, and strict task wire decoding rejects malformed records. - The workspace supports Rust 1.85 and uses Rust 1.99 for development, with
hosted adapters separated from the portable task core. - Release verification checks all seven library packages; tag releases use the
matching dated changelog entry as their GitHub release notes.
Penelope 0.1.0
[0.1.0] - 2026-09-15
Initial release of Penelope, a foundation for reliable sagas and long-running
asynchronous workflows. It provides deterministic decisions and recovery
contracts while leaving storage, messaging, scheduling, and service adapters
to the consuming application.
Added
- Deterministic linear and graph saga orchestration with ordered replay after
restart. - Append-only outcome-log contracts that reject gaps, duplicates, reordering,
and cross-process records. - Reliable retry handling with bounded attempts, backoff, jitter, deadlines,
and durable timer decisions. - LIFO compensation, cancellation, unknown-outcome escalation, and manual
review flows for exceptional or ambiguous work. - Reconciliation contracts that distinguish committed, not-committed, and
unknown external outcomes without blind retries. - Typed ports for persistence, inbox/outbox delivery, leases, timers, external
effects, authorization, quotas, diagnostics, and review queues. - StateChronicle integration boundary for idempotent commands and verified
committed-event correlation. - Developer-facing constructors, facade exports, examples, documentation,
and release verification for building reliable async systems.
Security
- Tenant, process, definition, action, effect, and resource scopes are checked
before decisions advance. - Stale, duplicate, forged, or cross-scope results fail closed.
- Ambiguous external effects escalate for reconciliation or operator review
instead of being treated as failures.