feat(sdk): auto-enroll default pollers into autoscaling#1425
Open
veeral-patel wants to merge 1 commit into
Open
feat(sdk): auto-enroll default pollers into autoscaling#1425veeral-patel wants to merge 1 commit into
veeral-patel wants to merge 1 commit into
Conversation
veeral-patel
force-pushed
the
feat/poller-autoscaling-auto-enroll-v2
branch
2 times, most recently
from
July 24, 2026 21:22
5954b11 to
d600242
Compare
Sushisource
reviewed
Jul 24, 2026
| /// `poller_autoscaling_auto_enroll` capability. Layers above core (the Rust SDK, the c-bridge) | ||
| /// set this; direct core users leave it `false`. | ||
| #[builder(default = false)] | ||
| pub workflow_task_poller_behavior_auto_enroll: bool, |
Member
There was a problem hiding this comment.
I don't think we need these. They can just explicitly set the option for whatever they do want, rather than disabling the enrollment explicitly.
Comment on lines
+100
to
+102
| impl TryFrom<&PollerBehavior> for Option<temporalio_sdk_core::PollerBehavior> { | ||
| type Error = anyhow::Error; | ||
| fn try_from(value: &PollerBehavior) -> Result<Self, Self::Error> { | ||
| fn try_from(value: &PollerBehavior) -> anyhow::Result<Self> { |
Member
There was a problem hiding this comment.
This won't compile (and doesn't, in CI).
| at_task_mgr: Option<WorkerActivityTasks>, | ||
| /// The poller-dependent subsystems (workflows, activity task manager, nexus manager). Built | ||
| /// lazily once namespace capabilities are known so effective poller behavior can be resolved. | ||
| pollers: std::sync::OnceLock<PollerSubsystems>, |
Member
There was a problem hiding this comment.
These really aren't (just) pollers. TaskSubsystems would be better I think.
Comment on lines
+1155
to
+1160
| // If shutdown was already initiated before the pollers were built (e.g. a worker that | ||
| // never ran validate() and only starts polling as part of draining during shutdown), | ||
| // propagate the shutdown signal to the freshly-built subsystems so they don't hang waiting | ||
| // for work that will never arrive. `initiate_shutdown` deliberately does not force a build, | ||
| // so it can't have done this for us. | ||
| if shutdown_token.is_cancelled() { |
Member
There was a problem hiding this comment.
If the task subsystems were never built, we shouldn't need to do anything at all I think?
| pollers: std::sync::OnceLock<PollerSubsystems>, | ||
| /// Builds [PollerSubsystems] on first access. Taken exactly once. | ||
| #[allow(clippy::type_complexity)] | ||
| poller_builder: parking_lot::Mutex<Option<Box<dyn FnOnce() -> PollerSubsystems + Send>>>, |
Member
There was a problem hiding this comment.
I think we could avoid needing this, and I think we can kill two birds with one stone and get rid of PollerInitInputs at the same time.
The pollers (now task_subsystems) field, should instead be a https://doc.rust-lang.org/beta/std/sync/struct.LazyLock.html
The closure given to the LazyLock constructor can be the same one you give now, and build_poller_subsystems can be inlined to eliminate the need for PollerInitInputs
Comment on lines
+2221
to
+2222
| /// Whether this worker polls for (non-local) activities at all. | ||
| activities_enabled: bool, |
Member
There was a problem hiding this comment.
We don't need a separate bool here. Simply not initializing the OnceLock is sufficient.
When a namespace advertises the `poller_autoscaling_auto_enroll` capability, workers automatically switch workflow, activity, and nexus pollers to autoscaling — but only for poller types the user left at their default (neither a fixed poller count nor an explicit behavior). Explicitly configured pollers are left unchanged. The effective `PollerBehavior` is resolved once, after the namespace capability is known. Poller construction is deferred out of the synchronous `Worker::new_with_pollers` into a lazily-built step that runs after `Worker::validate()` populates capabilities (and on first poll for paths that skip validate, e.g. replay). This keeps `poll_buffer.rs` / `PollScaler` free of any runtime mode-switching — the behavior is simply passed in already resolved. Eligibility (user left the poller at its default) is signalled from the layers above core: the Rust SDK exposes the poller options as `Option`, and the c-bridge treats unset (both FFI pointers null) as eligible. Ports temporalio/sdk-go#2442.
veeral-patel
force-pushed
the
feat/poller-autoscaling-auto-enroll-v2
branch
from
July 25, 2026 00:46
d600242 to
e32bf5f
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Ports temporalio/sdk-go#2442 to
sdk-core, so every SDK built on it (Rust directly; TS/Python/.NET/Ruby via the c-bridge) gets it.Supersedes #1406 — same feature, reworked per review feedback (see "How" below).
What it does
A worker's pollers can run in one of two modes: a fixed number, or autoscaling (the server tells the worker to add/remove pollers based on load).
This makes a worker automatically use autoscaling when the server says the namespace supports it (
poller_autoscaling_auto_enroll) — but only for pollers the user didn't configure. If you set a poller count or behavior yourself, it's left exactly as you asked.How it works
The worker figures out the right mode once, right after it checks in with the server (
validate()), and builds its pollers then. Previously the pollers were built before the check-in, so #1406 had to flip the mode later at runtime; this version avoids that entirely (poll_buffer.rsis unchanged here).worker/mod.rs: poller setup is deferred until after the server check-in, so the mode is known up front.validate()on startup (it never did before, so it was missing all namespace capabilities).Testing
Unit tests for the mode decision and the capability parsing; c-bridge conversion tests.
cargo check/fmt/clippyclean; sdk-core lib 368 pass, c-bridge 14 pass.