[rust-scheduler] Replace round-robin ClusterFeed with priority-queue scheduling#2317
Draft
ramonfigueiredo wants to merge 3 commits into
Draft
Conversation
…scheduling Fixes AcademySoftwareFoundation#2315. The ClusterFeed previously iterated clusters with a fixed-step round-robin and applied lap-based 10ms/100ms/5s sleeps between rounds. Every cluster received equal air-time regardless of work, and a sleeping cluster added delay to neighboring clusters. Replace the Vec<Cluster> + AtomicUsize index with a BinaryHeap<Scheduled> ordered by (next_eligible_at asc, last_dispatched_jobs desc). The feed now pops the highest-priority cluster, sleeps until eligible (preemptible via Notify), emits, and waits for a Done message that re-inserts the cluster with updated stats. Lap-based sleep tiers and the parallel sleep_map are removed; eligibility now lives on each heap entry. Cluster feed changes (cluster.rs): - Added Scheduled { cluster, next_eligible_at, last_dispatched_jobs } with Ord prioritizing earliest eligibility, then busiest cluster on ties (productivity bias). - ClusterFeed now holds Arc<Mutex<BinaryHeap<Scheduled>>> + Notify. - FeedMessage::Sleep / Stop() collapsed into FeedMessage::Done { cluster, processed_jobs, sleep } + unit-variant Stop. - stream() dispatch loop now pops, sleeps with preemption, and emits. - Control loop re-inserts on Done and notifies the dispatch loop. - Added ClusterFeed::load_from_clusters shim for tests. Consumer (pipeline/entrypoint.rs): - Always sends FeedMessage::Done with processed_jobs and optional back-off. - Preserves should_stop / empty_job_cycles logic. Config (config/mod.rs): - Added cluster_empty_back_off: Duration (default 3s), replacing hardcoded entrypoint back-off. - Added cluster_productivity_bias: bool (default true), toggling busy-cluster tie-breaking. Observability (metrics/mod.rs): - Added scheduler_cluster_polls_total {show_id, facility_id} CounterVec. - Added scheduler_cluster_rounds_total Counter (mirrors CLUSTER_ROUNDS). - Added scheduler_cluster_last_dispatched_jobs {show_id, facility_id} Gauge, updated from the Done handler to expose the productivity-bias signal. Tests: - Added 6 unit tests covering Scheduled ordering, sleep-back-off timing, Done round-trip, and a 500ms mixed-workload proportionality test asserting busy clusters receive >2x the polls of idle ones. - Updated tests/util.rs for the two new QueueConfig fields.
Contributor
📝 WalkthroughWalkthroughThe scheduler's core dispatch mechanism is redesigned from round-robin iteration with a sleep-map to a priority-queue architecture. A new ChangesPriority-Queue Cluster Scheduler
Sequence Diagram(s)sequenceDiagram
participant Dispatch as Dispatch Task
participant Heap as BinaryHeap
participant Control as Control Task
participant Sender as Cluster Consumer
participant RecvMsg as FeedMessage Receiver
Dispatch->>Heap: Pop best Scheduled (by next_eligible_at, then productivity bias)
Dispatch->>Dispatch: Sleep until next_eligible_at with Notify preemption
Dispatch->>Sender: Send selected Cluster
Dispatch->>Dispatch: Loop or stop
Control->>RecvMsg: Recv FeedMessage
alt FeedMessage::Done
Control->>Control: Compute next_eligible_at from optional sleep duration
Control->>Control: Record processed_jobs via metrics
Control->>Heap: Insert updated Scheduled with new state
Control->>Dispatch: Notify preemption to wake early
else FeedMessage::Stop
Control->>Control: Set stop_flag
end
Estimated code review effort🎯 4 (Complex) | ⏱️ ~60 minutes Possibly related issues
Poem
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 4
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@rust/crates/scheduler/src/cluster.rs`:
- Around line 539-583: The control loop currently only sets stop_flag_ctrl and
notifies on receiving FeedMessage::Stop, but it does not handle the case where
control_receiver.recv().await returns None (all senders dropped), allowing the
dispatcher to spin; update the async loop that awaits
control_receiver.recv().await to detect the None (channel closed) case and in
that branch call stop_flag_ctrl.store(true, Ordering::Relaxed) and
notify_ctrl.notify_one(), then break the loop; keep the existing handling for
FeedMessage::Done (pushing Scheduled into queue_ctrl) and FeedMessage::Stop
unchanged.
- Around line 98-123: The bug is that Ord::cmp can return Equal while
PartialEq::eq considers last_dispatched_jobs, breaking the Ord/Eq contract; fix
by making cmp consider the exact same fields used in eq (next_eligible_at,
last_dispatched_jobs, cluster) in some deterministic order so Ordering::Equal
implies equality. Concretely, change the logic in Scheduled::cmp to always
compare next_eligible_at first, then compare last_dispatched_jobs, then cluster;
if CONFIG.queue.cluster_productivity_bias should increase priority of
last_dispatched_jobs, implement that by moving the last_dispatched_jobs
comparison earlier (but still include it regardless), otherwise keep it as the
second comparison — do not skip comparing last_dispatched_jobs when bias is
false. Ensure the final tie-breaker remains self.cluster.cmp(&other.cluster).
In `@rust/crates/scheduler/src/pipeline/entrypoint.rs`:
- Around line 105-108: The Err(err) branch currently returns (0, true) which
sets should_stop and triggers FeedMessage::Stop; change this to avoid halting
the whole scheduler by not setting should_stop on a single fetch error: log the
error with cluster.id via error!("Failed to fetch job for cluster {}: {}",
cluster.id, err), return (0, false) to skip this cluster for now, and either
implement simple retry logic in the fetch path or add a per-cluster error
counter (e.g. in the cluster state used by the loop) and only set
should_stop/emit FeedMessage::Stop when that counter exceeds a threshold; update
the handling around the fetch call (the code that expects (count, should_stop))
and ensure tests cover transient failures.
- Around line 91-101: The shared atomic counter cycles_without_jobs is
read-after-write by separate tasks, causing TOCTOU races in the stop
calculation; replace the current fetch_add/store/load pattern inside the block
that checks CONFIG.queue.empty_job_cycles_before_quiting with an atomic update
that yields a consistent new value (e.g., use fetch_add(previous) and compute
new_count = prev + 1 when processed == 0, and use swap(0) or fetch_update to
reset to 0 when processed != 0) and base stop on that single atomic result
rather than a separate load; update uses of cycles_without_jobs, processed, and
the stop determination in the same branch (the code around the stop variable and
the for_each_concurrent section) and consider using Ordering::SeqCst for the
operations to ensure strict ordering across threads.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: 49579147-8cbd-4442-b499-869b9539a741
📒 Files selected for processing (5)
rust/crates/scheduler/src/cluster.rsrust/crates/scheduler/src/config/mod.rsrust/crates/scheduler/src/metrics/mod.rsrust/crates/scheduler/src/pipeline/entrypoint.rsrust/crates/scheduler/tests/util.rs
- Scheduled::cmp now always compares last_dispatched_jobs as a final tiebreaker so Ord agrees with PartialEq regardless of the cluster_productivity_bias toggle. Required by BinaryHeap. - Control loop now sets stop_flag and notifies the dispatcher when the control channel closes (e.g. consumer panic), preventing the dispatch loop from spinning on the empty-queue wake cycle. - Added ord_and_eq_stay_consistent_for_differing_jobs regression test.
Contributor
There was a problem hiding this comment.
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
rust/crates/scheduler/src/pipeline/entrypoint.rs (1)
71-123:⚠️ Potential issue | 🟠 Major | ⚡ Quick winCount fetch failures toward the empty-cycle stop threshold.
This branch now treats query errors as empty only for back-off, not for
empty_job_cycles_before_quiting:Err(err)returns(0, false), and the stop counter is only updated in theOk(jobs)path. If the DAO is down for every cluster, the scheduler will never trip the configured safety net and can loop forever.♻️ Minimal fix
- let (processed_count, should_stop) = match jobs { + let processed_count = match jobs { Ok(jobs) => { metrics::increment_jobs_queried(jobs.len()); let processed_jobs = AtomicUsize::new(0); stream::iter(jobs) @@ - let processed = processed_jobs.load(Ordering::Relaxed); - - let stop = match CONFIG.queue.empty_job_cycles_before_quiting { - Some(limit) => { - // Combine the update and read into a single atomic - // operation so concurrent clusters can't race between - // a fetch_add/store and a separate load. SeqCst gives - // us a strict total order across all clusters. - let new_count = if processed == 0 { - cycles_without_jobs.fetch_add(1, Ordering::SeqCst) + 1 - } else { - cycles_without_jobs.swap(0, Ordering::SeqCst); - 0 - }; - new_count >= limit - } - None => false, - }; - - (processed, stop) + processed_jobs.load(Ordering::Relaxed) } Err(err) => { // Don't halt the entire scheduler for a single cluster's // fetch error (network blip, transient DB load). Treat // the cluster as empty for this cycle so it gets the @@ error!( "Failed to fetch jobs for cluster {}: {}", cluster, err ); - (0, false) + 0 } }; + + let should_stop = match CONFIG.queue.empty_job_cycles_before_quiting { + Some(limit) => { + let new_count = if processed_count == 0 { + cycles_without_jobs.fetch_add(1, Ordering::SeqCst) + 1 + } else { + cycles_without_jobs.swap(0, Ordering::SeqCst); + 0 + }; + new_count >= limit + } + None => false, + };🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@rust/crates/scheduler/src/pipeline/entrypoint.rs` around lines 71 - 123, The Err branch currently returns (0, false) so fetch failures don't advance the empty-cycle safety counter; update the Err(err) handling to update cycles_without_jobs and compute the same stop boolean logic used in the Ok(jobs) path (i.e., increment cycles_without_jobs when treated-as-empty, reset it to 0 when not, and compare against CONFIG.queue.empty_job_cycles_before_quiting) and then return (0, stop) instead of (0, false); reference cycles_without_jobs, CONFIG.queue.empty_job_cycles_before_quiting, and the stop/limit logic in this function to mirror the behavior used when processed == 0.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Outside diff comments:
In `@rust/crates/scheduler/src/pipeline/entrypoint.rs`:
- Around line 71-123: The Err branch currently returns (0, false) so fetch
failures don't advance the empty-cycle safety counter; update the Err(err)
handling to update cycles_without_jobs and compute the same stop boolean logic
used in the Ok(jobs) path (i.e., increment cycles_without_jobs when
treated-as-empty, reset it to 0 when not, and compare against
CONFIG.queue.empty_job_cycles_before_quiting) and then return (0, stop) instead
of (0, false); reference cycles_without_jobs,
CONFIG.queue.empty_job_cycles_before_quiting, and the stop/limit logic in this
function to mirror the behavior used when processed == 0.
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: 2bd1a413-b156-4185-a997-14bb0ce83c7f
📒 Files selected for processing (2)
rust/crates/scheduler/src/cluster.rsrust/crates/scheduler/src/pipeline/entrypoint.rs
🚧 Files skipped from review as they are similar to previous changes (1)
- rust/crates/scheduler/src/cluster.rs
Collaborator
Author
|
Hi @DiegoTavares |
Collaborator
Author
|
This PR is ready for review, but kept in draft because there are other changes from @DiegoTavares in progress that conflict with this PR. |
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
1 participant
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.
Related Issues
Summarize your change.
The ClusterFeed previously iterated clusters with a fixed-step round-robin and applied lap-based 10ms/100ms/5s sleeps between rounds. Every cluster received equal air-time regardless of work, and a sleeping cluster added delay to neighboring clusters.
Replace the Vec + AtomicUsize index with a BinaryHeap ordered by (next_eligible_at asc, last_dispatched_jobs desc). The feed now pops the highest-priority cluster, sleeps until eligible (preemptible via Notify), emits, and waits for a Done message that re-inserts the cluster with updated stats. Lap-based sleep tiers and the parallel sleep_map are removed; eligibility now lives on each heap entry.
Cluster feed changes (cluster.rs):
Consumer (pipeline/entrypoint.rs):
Config (config/mod.rs):
Observability (metrics/mod.rs):
Tests:
Review feedback addressed: