worker_query.rs 8.21 KB
Newer Older
1
// SPDX-FileCopyrightText: Copyright (c) 2024-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
2
3
// SPDX-License-Identifier: Apache-2.0

4
use std::sync::Arc;
5
6
7

use anyhow::{Context, Result};
use dynamo_runtime::component::Component;
8
9
10
11
12
use dynamo_runtime::pipeline::{
    AsyncEngine, AsyncEngineContextProvider, ManyOut, PushRouter, ResponseStream, RouterMode,
    SingleIn, async_trait, network::Ingress,
};
use dynamo_runtime::protocols::maybe_error::MaybeError;
13
use tokio::sync::OnceCell;
14
use tokio_stream::StreamExt;
15

16
use crate::discovery::RuntimeConfigsSubscriber;
17
18
use crate::kv_router::WORKER_KV_INDEXER_QUERY_ENDPOINT;
use crate::kv_router::indexer::{LocalKvIndexer, WorkerKvQueryRequest, WorkerKvQueryResponse};
19
use crate::kv_router::protocols::WorkerId;
20
use dynamo_runtime::stream;
21
22
23

/// Router-side client for querying worker local KV indexers
///
24
/// Performs request/reply communication with workers via request plane endpoint routing.
25
/// (Only queries workers that have `enable_local_indexer=true` in their MDC user_data)
26
/// The client is spawned by KvRouter; it uses a subscriber from RuntimeConfigs.
27
28
pub struct WorkerQueryClient {
    component: Component,
29
30
    /// Subscriber for runtime configs (includes shared configs DashMap)
    subscriber: RuntimeConfigsSubscriber,
31
    router: OnceCell<Arc<PushRouter<WorkerKvQueryRequest, WorkerKvQueryResponse>>>,
32
33
34
}

impl WorkerQueryClient {
35
36
    /// Create a new WorkerQueryClient with a subscriber to runtime configs
    pub fn new(component: Component, subscriber: RuntimeConfigsSubscriber) -> Self {
37
38
        Self {
            component,
39
            subscriber,
40
            router: OnceCell::new(),
41
42
43
        }
    }

44
45
46
47
48
49
    /// Wait until at least one worker has a known runtime config (Some).
    /// Returns the list of worker IDs that have configs.
    pub async fn wait_for_ready(&mut self) -> Vec<WorkerId> {
        self.subscriber.wait_for_some().await
    }

50
51
    /// Check if a worker has local indexer enabled
    pub fn has_local_indexer(&self, worker_id: WorkerId) -> bool {
52
53
        self.subscriber
            .configs
54
            .get(&worker_id)
55
            .and_then(|entry| entry.value().as_ref().map(|c| c.enable_local_indexer))
56
57
58
59
60
61
62
63
64
65
66
67
68
69
            .unwrap_or(false)
    }

    /// Query a specific worker's local KV indexer and return its buffered events.
    /// Returns an error if the worker does not have enable_local_indexer=true.
    pub async fn query_worker(
        &self,
        worker_id: WorkerId,
        start_event_id: Option<u64>,
        end_event_id: Option<u64>,
    ) -> Result<WorkerKvQueryResponse> {
        // Check if worker has local indexer enabled
        if !self.has_local_indexer(worker_id) {
            anyhow::bail!(
70
                "Worker {worker_id} does not have local indexer enabled (enable_local_indexer=false or not set in MDC user_data)"
71
72
73
            );
        }

74
75
76
77
78
79
80
81
82
        let router = self
            .router
            .get_or_try_init(|| async {
                let endpoint = self.component.endpoint(WORKER_KV_INDEXER_QUERY_ENDPOINT);
                let client = endpoint.client().await?;
                let router = PushRouter::from_client(client, RouterMode::RoundRobin).await?;
                Ok::<_, anyhow::Error>(Arc::new(router))
            })
            .await?;
83
84
85
86
87
88

        let request = WorkerKvQueryRequest {
            worker_id,
            start_event_id,
            end_event_id,
        };
89
90
        let mut stream = router
            .direct(SingleIn::new(request), worker_id)
91
92
            .await
            .with_context(|| {
93
                format!("Failed to send worker KV query request to worker {worker_id} via endpoint")
94
95
            })?;

96
97
98
99
100
101
102
103
        let response = stream
            .next()
            .await
            .context("Worker KV query returned an empty response stream")?;

        if let Some(err) = response.err() {
            return Err(err).context("Worker KV query response error");
        }
104
105
106
107

        Ok(response)
    }
}
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253

// Worker-side endpoint registration for Router -> LocalKvIndexer query service
pub(crate) async fn start_worker_kv_query_endpoint(
    component: Component,
    worker_id: u64,
    local_indexer: Arc<LocalKvIndexer>,
) {
    let engine = Arc::new(WorkerKvQueryEngine {
        worker_id,
        local_indexer,
    });

    let ingress = match Ingress::for_engine(engine) {
        Ok(ingress) => ingress,
        Err(e) => {
            tracing::error!(
                "Failed to build WorkerKvQuery endpoint handler for worker {worker_id}: {e}"
            );
            return;
        }
    };

    tracing::info!(
        "WorkerKvQuery endpoint starting for worker {worker_id} on endpoint '{}'",
        WORKER_KV_INDEXER_QUERY_ENDPOINT
    );

    if let Err(e) = component
        .endpoint(WORKER_KV_INDEXER_QUERY_ENDPOINT)
        .endpoint_builder()
        .handler(ingress)
        .graceful_shutdown(true)
        .start()
        .await
    {
        tracing::error!("WorkerKvQuery endpoint failed for worker {worker_id}: {e}");
    }
}

struct WorkerKvQueryEngine {
    worker_id: u64,
    local_indexer: Arc<LocalKvIndexer>,
}

#[async_trait]
impl AsyncEngine<SingleIn<WorkerKvQueryRequest>, ManyOut<WorkerKvQueryResponse>, anyhow::Error>
    for WorkerKvQueryEngine
{
    async fn generate(
        &self,
        request: SingleIn<WorkerKvQueryRequest>,
    ) -> anyhow::Result<ManyOut<WorkerKvQueryResponse>> {
        let (request, ctx) = request.into_parts();

        tracing::debug!(
            "Received query request for worker {}: {:?}",
            self.worker_id,
            request
        );

        // This is a sanity check to ensure the request is for the correct worker.
        // In production, this should never happen since the router should only
        // send requests to the worker it is associated with.
        if request.worker_id != self.worker_id {
            let error_message = format!(
                "WorkerKvQueryEngine::generate worker_id mismatch: request.worker_id={} this.worker_id={}",
                request.worker_id, self.worker_id
            );
            let response = WorkerKvQueryResponse::Error(error_message);
            return Ok(ResponseStream::new(
                Box::pin(stream::iter(vec![response])),
                ctx.context(),
            ));
        }

        let response = self
            .local_indexer
            .get_events_in_id_range(request.start_event_id, request.end_event_id)
            .await;

        Ok(ResponseStream::new(
            Box::pin(stream::iter(vec![response])),
            ctx.context(),
        ))
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::kv_router::RouterEvent;
    use crate::kv_router::indexer::KvIndexerMetrics;
    use crate::kv_router::protocols::{KvCacheEvent, KvCacheEventData};
    use tokio_stream::StreamExt;
    use tokio_util::sync::CancellationToken;

    #[tokio::test]
    async fn test_worker_kv_query_engine_returns_buffered_events() {
        let worker_id = 7u64;
        let token = CancellationToken::new();
        let metrics = Arc::new(KvIndexerMetrics::new_unregistered());
        let local_indexer = Arc::new(LocalKvIndexer::new(token, 4, metrics, 32));

        let event = RouterEvent::new(
            worker_id,
            KvCacheEvent {
                event_id: 1,
                data: KvCacheEventData::Cleared,
                dp_rank: 0,
            },
        );
        local_indexer
            .apply_event_with_buffer(event)
            .await
            .expect("apply_event_with_buffer should succeed");

        let engine = WorkerKvQueryEngine {
            worker_id,
            local_indexer,
        };

        let request = WorkerKvQueryRequest {
            worker_id,
            start_event_id: Some(1),
            end_event_id: Some(1),
        };

        let mut stream = engine
            .generate(SingleIn::new(request))
            .await
            .expect("generate should succeed");

        let response = stream
            .next()
            .await
            .expect("response stream should yield one item");

        match response {
            WorkerKvQueryResponse::Events(events) => {
                assert_eq!(events.len(), 1);
                assert_eq!(events[0].event.event_id, 1);
            }
            other => panic!("Unexpected response: {other:?}"),
        }
    }
}