Struct OpenSearchSink 

pub struct OpenSearchSink { /* private fields */ }

A Sink that writes directly to OpenSearch clusters over pooled HTTP.

Holds a ClusterPool per cluster, its own base URL and pooled HTTP/1.1 and HTTP/2 (prior-knowledge) clients. Each operation selects the client matching its resolved upstream Protocol (docs/04 §7), so the proxy can speak h2 to a cluster that supports it while defaulting to h1. Every dispatch is bounded by a per-request timeout so a stuck upstream fails fast (NFR-R7), and a per-cluster circuit breaker sheds a cluster that keeps failing.

Implementations

impl OpenSearchSink

pub fn new() -> Self

Builds an empty sink. Cluster pools are created on demand from the endpoint each routing target carries (the tenancy’s placement result is the source of truth for where every cluster lives); there is no static endpoint catalog.

pub fn with_timeout(self, timeout: Duration) -> Self

Sets the per-request upstream timeout (builder style).

pub fn with_breaker(self, failure_threshold: u32, cooldown: Duration) -> Self

Sets the circuit-breaker thresholds: open after failure_threshold consecutive failures, shed for cooldown before a half-open trial.

pub fn with_clock(self, clock: Arc<dyn Clock>) -> Self

Swaps the clock the breaker reads (tests inject a ManualClock).

pub fn pool_stats(&self, cluster: &ClusterId) -> Option<PoolStats>

A snapshot of a cluster’s connection-reuse counters, or None if no pool has been built for it yet. Lets operators (and tests) verify the pool is amortizing handshakes, connections opened far below requests dispatched (NFR-P; the docs/11 M4 “pool reuse rates verified” exit gate).

pub fn pool_stats_all(&self) -> Vec<(ClusterId, PoolStats)>

Pool-reuse counters for every pooled cluster, paired with its id, the fleet-/agent-facing readout behind the /metrics snapshot. Order is unspecified (a HashMap walk); callers that need stability sort by id.

Trait Implementations

impl Debug for OpenSearchSink

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for OpenSearchSink

fn default() -> Self

Returns the “default value” for a type.

impl Reader for OpenSearchSink

async fn get(&self, op: ReadOp) -> Result<ReadOutcome, SinkError>

Fetches a single document by physical id.

async fn search(&self, op: SearchOp) -> Result<SearchOutcome, SinkError>

Runs a search, returning the raw hits envelope.

async fn count(&self, op: SearchOp) -> Result<CountOutcome, SinkError>

Counts the documents matching a (partition-filtered) query.

async fn cursor(&self, op: CursorOp) -> Result<CursorOutcome, SinkError>

Forwards a raw cursor request to its pinned cluster (scroll/PIT continue, clear, close). The default is unsupported, a sink that cannot passthrough (the in-memory test sink, a write-only queue) rejects it; OpenSearchSink overrides it with a real upstream call.

async fn search_stream( &self, op: SearchOp, ) -> Result<StreamingSearch, SinkError>

Runs a search whose response streams back (ADR-014, final stage): the upstream hits envelope is piped to the engine’s hit transform without being collected, so a large response (e.g. heavy aggregations) never lands in memory. The default is unsupported; OpenSearchSink overrides it with a real streamed upstream call.

async fn forward_stream( &self, op: ForwardOp, body: ByteBody, ) -> Result<StreamingForward, SinkError>

Forwards a request to a cluster with the body supplied as a stream (ADR-014 stage 2): the verbatim-passthrough path pipes the downstream body straight upstream without buffering. The default is unsupported; OpenSearchSink overrides it with a real streamed upstream call.

impl Sink for OpenSearchSink

async fn write(&self, batch: WriteBatch) -> Result<WriteAck, SinkError>

Applies a batch of writes, returning a per-operation acknowledgement.