Repository navigation
Notable Changes
Added
OtelCollectorOptions::histogram_aggregationselects how the OTLP exporter aggregates histograms.HistogramAggregation::Base2Exponentialexports every histogram, including custom ones, as an OpenTelemetry exponential histogram, which Prometheus-compatible backends store as a native histogram. Each series carriesBase2ExponentialHistogramOptions::name_suffix(_nativeby default) so it never shares a name with the fixed-bucket series it replaces, andhistogram_bucket_overridesis rejected in this mode. Explicit buckets remain the default.- Experimental envconfig extensions (requires the
experimentalfeature).ClientOptions::load_from_config_with_extensionscalls eachClientConfigExtensionin order with the loaded profile and environment, and registers any client plugin it returns. Extension config lives in[profile.<name>.ext.<extension>]TOML tables, exposed raw asClientConfigProfile::extand never rejected by strict mode. - Experimental external storage (requires the
experimentalfeature).DataConverter::with_external_storageoffloads payloads at or above a size threshold to a user-suppliedStorageDriver, replacing them in history with small references that workers and clients resolve transparently. Drivers are told the workflow or standalone activity each payload is stored for, and may receive concurrent requests. Tasks that fail on a storage error report theEXTERNAL_STORAGE_FAILUREcause andfailure_reason="ExternalStorageError". This API may change without notice. The newtemporalio-aws-s3drivercrate provides an S3StorageDriver, and its defaultaws-sdkfeature adds a client backed byaws-sdk-s3. - Experimental Event Groups support (requires the
experimentalfeature). Event Groups are workflow-level metadata that group related history events for UI and observability.EventGroup::newtakes a required ID (used verbatim; do not put secrets in it);EventGroup::with_labelsets an optional label that is stored as a codec-encoded payload. Attach groups withwith_event_group/with_event_groupsor theevent_groupsfield on command options. Signal and update handlers receive implicit inbound groups. This API may change without notice. Worker::run_untilconsumes a worker and runs it until a supplied future completes, then shuts it down and releases its worker registration before returning, including when running fails.- Local workflow environments and ephemeral server configurations support a
startup_timeoutsetting, defaulting to five seconds and excluding executable downloads. original_execution_run_id()on workflow context, the run ID recorded on theWorkflowExecutionStartedevent. Unlikerun_id(), this value is preserved across workflow resets.- OpenTelemetry client tracing can opt into encoding its own headers with the configured client codec using
OpenTelemetryPlugin::builder().encode_client_headers(true). User-provided client headers are unchanged.
Changed
- Poller pools now report their poller target to the server on each poll and in the worker heartbeat: the current target for autoscaling pools, the configured maximum for fixed-size pools. (We'll use this count on the server side for now for load balancing pollers between workers, and for creating a per-worker poller count metric.)
- Updated workflow and activity macros to syn 3 for improved Rust syntax compatibility.
💥 Breaking Changes
- Workflow replay now reports nondeterminism when a scheduled Nexus operation's service or operation differs from the command that produced it.
Fixed
-
temporal.api.sdk.v1messages are capable of serialization to proto JSON, so their JSON field names are camel case rather than snake case and the binary encoding is unchanged. They previously could not be deserialized from the camel case that other Temporal SDKs and protobuf tools emit, and both spellings are now accepted. -
Local activities still running when their workflow is evicted or the worker shuts down now always receive a cancellation. Local activities no longer retry ordinary failures or start-to-close timeouts after cancellation has been requested, matching non-local activities. Workflow code no longer observes a spurious local activity cancellation when its run is evicted or completes while the local activity is backing off between retries.
-
Resource-based worker tuning now reads CPU usage and quotas from standard cgroup v1 mounts, preventing host CPU measurements from understating container load when cgroup v2 files are unavailable.
-
Execute Update calls, including
Client::execute_update_with_start_workflow, now request completion in the initial RPC to avoid an unnecessary round trip to retrieve the result. -
Workflow tasks containing only rejected updates no longer start pending local activities while reconstructing workflow state.
-
Workflow code, including pending
wait_conditionfutures, is no longer polled in a workflow task that only delivers updates and rejects all of them. -
Evicting a workflow while some local activities are still running now preserves results already delivered to workflow code, rather than panicking or losing those results.
-
Workers using a custom metrics backend now report task-slot and poller counts correctly in worker heartbeats, rather than reporting zero when the backend does not expose its labels.
-
Avoid scanning the workflow cache when no workflow tasks are waiting for a cache slot, reducing workflow processing CPU overhead with large caches.
-
Worker::run()now waits for spawned activity tasks to finish before returning, including when running fails. -
Fixed queries timing out when several requests are queued while a workflow is being evicted from the worker cache.
-
Legacy queries no longer execute local activities while reconstructing workflow state, avoiding duplicate activity executions and stalled workflows.
-
Ephemeral dev and test servers now complete the automatic port reservation handshake on Linux and macOS before starting, avoiding intermittent address-in-use failures.
-
Downloading an ephemeral dev or test server binary now retries transient HTTP failures, such as a
503returned while a release is rolling out, with backoff instead of failing on the first attempt. -
Schedule updates preserve codec-encoded arguments, headers, and memo in retained workflow actions instead of resending decoded payloads. The schedule's response memo remains decoded.
-
Ephemeral server startup now enforces its elapsed-time deadline even when a connection stalls, and startup timeout messages include the last connection error when available.
-
Workflows paused before a scheduled workflow task starts now resume after unpause instead of failing during history replay.
-
A query or other workflow task that arrives while a workflow waits on a local activity no longer causes the current workflow task to be lost and time out, which could leave the workflow stuck on a nondeterminism error.
-
Prevent Nexus worker shutdown from hanging if a task completion future is dropped or its response is rejected as malformed. Dropping an in-flight completion logs an error and panics when debug assertions are enabled; callers must still await completion to preserve RPC delivery.
-
Workflows now run static summaries and details through a client's data converter and payload codec. Before they used a default PayloadConverter so describe() failed on summaries that a codec encoded (when using cross SDK workflows).
Security
- Update Wasmtime to 48.0.7 to fix a potential sandbox escape on aarch64 when running Wasm workflows (GHSA-mghf-wjfp-p5xm). Only the experimental
wasm-workflowsfeature now requires Rust 1.95 or later; the SDK's MSRV remains Rust 1.92.
New Contributors
- @Mahnoor-Zaffar made their first contribution in #1593
- @mckalexee made their first contribution in #1654
- @Anshuman-Prakash made their first contribution in #1677
- @moetemp made their first contribution in #1674
- @mbackschat made their first contribution in #1693
- @TiunovNN made their first contribution in #1702
- @johanngan made their first contribution in #1736
- @vishwakulkarni made their first contribution in #1771
Full Changelog: v1.1.0...v1.2.0