[ENH]: Dead letter queuing for compaction jobs (#5023)

## Description of changes

This change adds a dead letter queueing system to the compaction scheduler. If a compaction job on a collection fails `max_failure_count` times, it will be moved to a dead set that disables this collection from being compacted while it is in this set. As of this change, the only way to clear this set is by restarting the compaction process.

- Improvements & Bug fixes
  - Added a failing_jobs map in the CompactionManager to help keep track of jobs that have failed on consecutive attempts. 
  - Added a dead_jobs set in the CompactionManager to record "dead" jobs.
- New functionality
  - Described above.
  - Added a metric `compactor_dead_jobs_count` to track the size of the dead jobs set.

## Test plan

Added a test in scheduler.rs.

Also manually tested by injecting failures in certain compaction jobs and tracking the dead set size metric locally.

- [x] Tests pass locally with `pytest` for python, `yarn test` for js, `cargo test` for rust

## Documentation Changes

_Are all docstrings for user-facing APIs updated if required? Do we need to make documentation changes in the [docs section](https://github.com/chroma-core/chroma/tree/main/docs/docs.trychroma.com)?_

Author: tanujnay112

Cargo.lock

@@ -31,6 +31,7 @@ async-trait = { workspace = true }
 roaring = { workspace = true }
 figment = { workspace = true }
 futures = { workspace = true }
+opentelemetry = { workspace = true }
 parking_lot = { workspace = true }
 tracing = { workspace = true }
 tokio = { workspace = true }

rust/worker/Cargo.toml

@@ -49,14 +49,19 @@ use tracing::Instrument;
 use tracing::Span;
 use uuid::Uuid;
 
-type BoxedFuture =
-    Pin<Box<dyn Future<Output = Result<CompactionResponse, Box<dyn ChromaError>>> + Send>>;
+type CompactionOutput = Result<CompactionResponse, Box<dyn ChromaError>>;
+type BoxedFuture = Pin<Box<dyn Future<Output = CompactionOutput> + Send>>;
 
 struct CompactionTask {
     collection_id: CollectionUuid,
     future: BoxedFuture,
 }
 
+struct CompactionTaskCompletion {
+    collection_id: CollectionUuid,
+    result: CompactionOutput,
+}
+
 #[derive(Clone)]
 pub(crate) struct CompactionManagerContext {
     system: System,
@@ -85,7 +90,7 @@ pub(crate) struct CompactionManager {
     scheduler: Scheduler,
     context: CompactionManagerContext,
     compact_awaiter_channel: mpsc::Sender<CompactionTask>,
-    compact_awaiter_completion_channel: mpsc::UnboundedReceiver<CompactionResponse>,
+    compact_awaiter_completion_channel: mpsc::UnboundedReceiver<CompactionTaskCompletion>,
     compact_awaiter: tokio::task::JoinHandle<()>,
     on_next_memberlist_signal: Option<oneshot::Sender<()>>,
 }
@@ -129,7 +134,7 @@ impl CompactionManager {
         // Using unbounded channel for the completion channel as its size
         // is bounded by max_concurrent_jobs. It's far more important for the
         // completion channel to not block or drop messages.
-        let (completion_tx, completion_rx) = mpsc::unbounded_channel::<CompactionResponse>();
+        let (completion_tx, completion_rx) = mpsc::unbounded_channel::<CompactionTaskCompletion>();
         let compact_awaiter = tokio::spawn(async {
             compact_awaiter_loop(compact_awaiter_rx, completion_tx).await;
         });
@@ -237,11 +242,18 @@ impl CompactionManager {
         self.context.dispatcher = Some(dispatcher);
     }
 
-    fn process_completions(&mut self) -> Vec<CompactionResponse> {
+    fn process_completions(&mut self) -> Vec<CompactionTaskCompletion> {
         let compact_awaiter_completion_channel = &mut self.compact_awaiter_completion_channel;
         let mut completed_collections = Vec::new();
         while let Ok(resp) = compact_awaiter_completion_channel.try_recv() {
-            self.scheduler.complete_collection(resp.collection_id);
+            match resp.result {
+                Ok(_) => {
+                    self.scheduler.succeed_collection(resp.collection_id);
+                }
+                Err(_) => {
+                    self.scheduler.fail_collection(resp.collection_id);
+                }
+            }
             completed_collections.push(resp);
         }
         completed_collections
@@ -351,6 +363,7 @@ impl Configurable<(CompactionServiceConfig, System)> for CompactionManager {
             Box::<dyn AssignmentPolicy>::try_from_config(assignment_policy_config, registry)
                 .await?;
         let job_expiry_seconds = config.compactor.job_expiry_seconds;
+        let max_failure_count = config.compactor.max_failure_count;
         let scheduler = Scheduler::new(
             my_ip,
             log.clone(),
@@ -361,6 +374,7 @@ impl Configurable<(CompactionServiceConfig, System)> for CompactionManager {
             assignment_policy,
             disabled_collections,
             job_expiry_seconds,
+            max_failure_count,
         );
 
         let blockfile_provider = BlockfileProvider::try_from_config(
@@ -407,25 +421,31 @@ impl Configurable<(CompactionServiceConfig, System)> for CompactionManager {
 
 async fn compact_awaiter_loop(
     mut job_rx: mpsc::Receiver<CompactionTask>,
-    completion_tx: mpsc::UnboundedSender<CompactionResponse>,
+    completion_tx: mpsc::UnboundedSender<CompactionTaskCompletion>,
 ) {
     let mut futures = FuturesUnordered::new();
     loop {
         select! {
             Some(job) = job_rx.recv() => {
                 futures.push(async move {
-                    let _ = AssertUnwindSafe(job.future).catch_unwind().await;
-                    CompactionResponse {
-                        collection_id: job.collection_id,
+                    let result = AssertUnwindSafe(job.future).catch_unwind().await;
+                    match result {
+                        Ok(response) => CompactionTaskCompletion {
+                            collection_id: job.collection_id,
+                            result: response,
+                        },
+                        Err(_) => CompactionTaskCompletion {
+                            collection_id: job.collection_id,
+                            result: Err(Box::new(CompactionError::FailedToCompact)),
+                        },
                     }
                 });
             }
-            Some(compaction_response) = futures.next() => {
-                match completion_tx.send(compaction_response) {
+            Some(completed_job) = futures.next() => {
+                let collection_id = completed_job.collection_id;
+                match completion_tx.send(completed_job) {
                     Ok(_) => {},
-                    Err(_) => {
-                        tracing::error!("Failed to send compaction response");
-                    }
+                    Err(_) => tracing::error!("Failed to record compaction result for collection {}", collection_id),
                 }
             }
             else => {
@@ -773,6 +793,7 @@ mod tests {
         let fetch_log_batch_size = 100;
         let purge_dirty_log_timeout_seconds = 60;
         let job_expiry_seconds = 3600;
+        let max_failure_count = 3;
 
         // Set assignment policy
         let mut assignment_policy = Box::new(RendezvousHashingAssignmentPolicy::default());
@@ -788,6 +809,7 @@ mod tests {
             assignment_policy,
             HashSet::new(),
             job_expiry_seconds,
+            max_failure_count,
         );
         // Set memberlist
         scheduler.set_memberlist(vec![my_member.clone()]);
@@ -865,6 +887,7 @@ mod tests {
             completed_compactions.extend(
                 completed
                     .iter()
+                    .filter(|c| c.result.is_ok())
                     .map(|c| c.collection_id)
                     .collect::<Vec<CollectionUuid>>(),
             );

rust/worker/src/compactor/compaction_manager.rs

@@ -22,6 +22,8 @@ pub struct CompactorConfig {
     pub fetch_log_batch_size: u32,
     #[serde(default = "CompactorConfig::default_purge_dirty_log_timeout_seconds")]
     pub purge_dirty_log_timeout_seconds: u64,
+    #[serde(default = "CompactorConfig::default_max_failure_count")]
+    pub max_failure_count: u8,
 }
 
 impl CompactorConfig {
@@ -64,6 +66,10 @@ impl CompactorConfig {
     fn default_purge_dirty_log_timeout_seconds() -> u64 {
         60
     }
+
+    fn default_max_failure_count() -> u8 {
+        5
+    }
 }
 
 impl Default for CompactorConfig {
@@ -80,6 +86,7 @@ impl Default for CompactorConfig {
             fetch_log_batch_size: CompactorConfig::default_fetch_log_batch_size(),
             purge_dirty_log_timeout_seconds:
                 CompactorConfig::default_purge_dirty_log_timeout_seconds(),
+            max_failure_count: CompactorConfig::default_max_failure_count(),
         }
     }
 }