lib.rs 32.2 KB
Newer Older
1
2
3
// SPDX-FileCopyrightText: Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
// SPDX-License-Identifier: Apache-2.0

4
5
6
use futures::StreamExt;
use once_cell::sync::OnceCell;
use pyo3::exceptions::PyStopAsyncIteration;
7
8
use pyo3::types::PyBytes;
use pyo3::types::{PyDict, PyList, PyString};
9
10
11
use pyo3::IntoPyObjectExt;
use pyo3::{exceptions::PyException, prelude::*};
use rs::pipeline::network::Ingress;
12
use std::path::PathBuf;
13
14
15
use std::{fmt::Display, sync::Arc};
use tokio::sync::Mutex;

Neelay Shah's avatar
Neelay Shah committed
16
use dynamo_runtime::{
Ryan Olson's avatar
Ryan Olson committed
17
    self as rs, logging,
18
    pipeline::{
19
20
        context::Context as RsContext, network::egress::push_router::RouterMode as RsRouterMode,
        EngineStream, ManyOut, SingleIn,
21
    },
22
    protocols::annotated::Annotated as RsAnnotated,
23
    traits::DistributedRuntimeProvider,
24
25
};

Neelay Shah's avatar
Neelay Shah committed
26
use dynamo_llm::{self as llm_rs};
27
28
use dynamo_llm::{entrypoint::RouterConfig, kv_router::KvRouterConfig};

29
30
use crate::llm::local_model::ModelRuntimeConfig;

31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
#[pyclass(eq, eq_int)]
#[derive(Clone, Debug, PartialEq)]
pub enum RouterMode {
    RoundRobin,
    Random,
    KV,
}

impl From<RouterMode> for RsRouterMode {
    fn from(mode: RouterMode) -> Self {
        match mode {
            RouterMode::RoundRobin => Self::RoundRobin,
            RouterMode::Random => Self::Random,
            RouterMode::KV => Self::KV,
        }
    }
}
48

49
mod context;
50
mod engine;
51
mod http;
52
mod llm;
53
mod parsers;
54
55
56
57
58
59
60
61
62
63
64
65
66

type JsonServerStreamingIngress =
    Ingress<SingleIn<serde_json::Value>, ManyOut<RsAnnotated<serde_json::Value>>>;

static INIT: OnceCell<()> = OnceCell::new();

const DEFAULT_ANNOTATED_SETTING: Option<bool> = Some(true);

/// A Python module implemented in Rust. The name of this function must match
/// the `lib.name` setting in the `Cargo.toml`, else Python will not be able to
/// import the module.
#[pymodule]
fn _core(m: &Bound<'_, PyModule>) -> PyResult<()> {
Ryan Olson's avatar
Ryan Olson committed
67
    logging::init();
Yan Ru Pei's avatar
Yan Ru Pei committed
68
    m.add_function(wrap_pyfunction!(llm::kv::compute_block_hash_for_seq_py, m)?)?;
Ryan Olson's avatar
Ryan Olson committed
69
    m.add_function(wrap_pyfunction!(log_message, m)?)?;
70
    m.add_function(wrap_pyfunction!(register_llm, m)?)?;
71
72
    m.add_function(wrap_pyfunction!(llm::entrypoint::make_engine, m)?)?;
    m.add_function(wrap_pyfunction!(llm::entrypoint::run_input, m)?)?;
73

74
75
76
77
78
79
    m.add_class::<DistributedRuntime>()?;
    m.add_class::<CancellationToken>()?;
    m.add_class::<Namespace>()?;
    m.add_class::<Component>()?;
    m.add_class::<Endpoint>()?;
    m.add_class::<Client>()?;
80
    m.add_class::<EtcdClient>()?;
81
    m.add_class::<AsyncResponseStream>()?;
82
    m.add_class::<llm::disagg_router::DisaggregatedRouter>()?;
83
84
85
    m.add_class::<llm::entrypoint::EntrypointArgs>()?;
    m.add_class::<llm::entrypoint::EngineConfig>()?;
    m.add_class::<llm::entrypoint::EngineType>()?;
86
87
    m.add_class::<llm::entrypoint::RouterConfig>()?;
    m.add_class::<llm::entrypoint::KvRouterConfig>()?;
88
    m.add_class::<llm::kv::WorkerMetricsPublisher>()?;
89
    m.add_class::<llm::model_card::ModelDeploymentCard>()?;
90
    m.add_class::<llm::local_model::ModelRuntimeConfig>()?;
91
92
    m.add_class::<llm::preprocessor::OAIChatPreprocessor>()?;
    m.add_class::<llm::backend::Backend>()?;
93
94
    m.add_class::<llm::kv::OverlapScores>()?;
    m.add_class::<llm::kv::KvIndexer>()?;
95
    m.add_class::<llm::kv::ApproxKvIndexer>()?;
96
97
98
    m.add_class::<llm::kv::EndpointKvMetrics>()?;
    m.add_class::<llm::kv::AggregatedMetrics>()?;
    m.add_class::<llm::kv::KvMetricsAggregator>()?;
99
    m.add_class::<llm::kv::KvEventPublisher>()?;
Yan Ru Pei's avatar
Yan Ru Pei committed
100
101
    m.add_class::<llm::kv::RadixTree>()?;
    m.add_class::<llm::kv::ZmqKvEventListener>()?;
102
103
    m.add_class::<llm::kv::ZmqKvEventPublisher>()?;
    m.add_class::<llm::kv::ZmqKvEventPublisherConfig>()?;
104
    m.add_class::<llm::kv::KvRecorder>()?;
105
    m.add_class::<llm::nats::NatsQueue>()?;
106
107
108
    m.add_class::<http::HttpService>()?;
    m.add_class::<http::HttpError>()?;
    m.add_class::<http::HttpAsyncEngine>()?;
109
    m.add_class::<context::Context>()?;
110
    m.add_class::<EtcdKvCache>()?;
111
    m.add_class::<ModelType>()?;
112
113
114
115
    m.add_class::<llm::kv::ForwardPassMetrics>()?;
    m.add_class::<llm::kv::WorkerStats>()?;
    m.add_class::<llm::kv::KvStats>()?;
    m.add_class::<llm::kv::SpecDecodeStats>()?;
116
117
    m.add_class::<llm::kv::KvPushRouter>()?;
    m.add_class::<llm::kv::KvPushRouterStream>()?;
118
    m.add_class::<RouterMode>()?;
119
120

    engine::add_to_module(m)?;
121
    parsers::add_to_module(m)?;
122

123
124
125
    #[cfg(feature = "block-manager")]
    llm::block_manager::add_to_module(m)?;

126
127
128
129
130
131
132
133
134
135
    Ok(())
}

pub fn to_pyerr<E>(err: E) -> PyErr
where
    E: Display,
{
    PyException::new_err(format!("{}", err))
}

Ryan Olson's avatar
Ryan Olson committed
136
137
138
139
140
141
142
/// Log a message from Python with file and line info
#[pyfunction]
#[pyo3(text_signature = "(level, message, module, file, line)")]
fn log_message(level: &str, message: &str, module: &str, file: &str, line: u32) {
    logging::log_message(level, message, module, file, line);
}

143
#[pyfunction]
144
#[pyo3(signature = (model_type, endpoint, model_path, model_name=None, context_length=None, kv_cache_block_size=None, router_mode=None, migration_limit=0, runtime_config=None, user_data=None, custom_template_path=None))]
145
#[allow(clippy::too_many_arguments)]
146
147
148
fn register_llm<'p>(
    py: Python<'p>,
    model_type: ModelType,
149
150
151
    endpoint: Endpoint,
    model_path: &str,
    model_name: Option<&str>,
152
153
    context_length: Option<u32>,
    kv_cache_block_size: Option<u32>,
154
    router_mode: Option<RouterMode>,
155
    migration_limit: u32,
156
    runtime_config: Option<ModelRuntimeConfig>,
157
    user_data: Option<&Bound<'p, PyDict>>,
158
    custom_template_path: Option<&str>,
159
160
161
162
163
) -> PyResult<Bound<'p, PyAny>> {
    let model_type_obj = match model_type {
        ModelType::Chat => llm_rs::model_type::ModelType::Chat,
        ModelType::Completion => llm_rs::model_type::ModelType::Completion,
        ModelType::Backend => llm_rs::model_type::ModelType::Backend,
164
        ModelType::Embedding => llm_rs::model_type::ModelType::Embedding,
165
166
    };

167
168
    let inner_path = model_path.to_string();
    let model_name = model_name.map(|n| n.to_string());
169
170
171
    let router_mode = router_mode.unwrap_or(RouterMode::RoundRobin);
    let router_config = RouterConfig::new(router_mode.into(), KvRouterConfig::default());

172
173
174
175
176
177
178
179
180
181
182
183
184
    // Early validation of custom template path
    let custom_template_path_owned = custom_template_path
        .map(|s| {
            let path = PathBuf::from(s);
            if !path.exists() {
                return Err(PyErr::new::<pyo3::exceptions::PyFileNotFoundError, _>(
                    format!("Custom template file does not exist: {}", path.display()),
                ));
            }
            Ok(path)
        })
        .transpose()?;

185
186
187
188
189
190
191
    let user_data_json = user_data
        .map(|dict| pythonize::depythonize(dict))
        .transpose()
        .map_err(|err| {
            PyErr::new::<PyException, _>(format!("Failed to convert user_data: {}", err))
        })?;

192
    pyo3_async_runtimes::tokio::future_into_py(py, async move {
193
194
195
196
197
        let mut builder = dynamo_llm::local_model::LocalModelBuilder::default();
        builder
            .model_path(Some(PathBuf::from(inner_path)))
            .model_name(model_name)
            .context_length(context_length)
198
            .kv_cache_block_size(kv_cache_block_size)
199
            .router_config(Some(router_config))
200
            .migration_limit(Some(migration_limit))
201
            .runtime_config(runtime_config.unwrap_or_default().inner)
202
203
            .user_data(user_data_json)
            .custom_template_path(custom_template_path_owned);
204
        // Download from HF, load the ModelDeploymentCard
205
        let mut local_model = builder.build().await.map_err(to_pyerr)?;
206
207
208
209
210
211
212
213
214
215
        // Advertise ourself on etcd so ingress can find us
        local_model
            .attach(&endpoint.inner, model_type_obj)
            .await
            .map_err(to_pyerr)?;

        Ok(())
    })
}

216
217
218
219
220
221
#[pyclass]
#[derive(Clone)]
struct EtcdKvCache {
    inner: Arc<rs::transports::etcd::KvCache>,
}

222
223
#[pyclass]
#[derive(Clone)]
Ryan Olson's avatar
Ryan Olson committed
224
pub struct DistributedRuntime {
225
226
227
228
    inner: rs::DistributedRuntime,
    event_loop: PyObject,
}

Ryan Olson's avatar
Ryan Olson committed
229
230
231
232
233
234
235
impl DistributedRuntime {
    #[allow(dead_code)]
    fn inner(&self) -> &rs::DistributedRuntime {
        &self.inner
    }
}

236
237
238
239
240
241
#[pyclass]
#[derive(Clone)]
struct EtcdClient {
    inner: rs::transports::etcd::Client,
}

242
#[pyclass]
243
#[derive(Clone)]
244
245
246
247
248
struct CancellationToken {
    inner: rs::CancellationToken,
}

#[pyclass]
249
#[derive(Clone)]
250
251
252
253
254
255
struct Namespace {
    inner: rs::component::Namespace,
    event_loop: PyObject,
}

#[pyclass]
256
#[derive(Clone)]
257
258
259
260
261
262
struct Component {
    inner: rs::component::Component,
    event_loop: PyObject,
}

#[pyclass]
263
#[derive(Clone)]
264
265
266
267
268
269
struct Endpoint {
    inner: rs::component::Endpoint,
    event_loop: PyObject,
}

#[pyclass]
270
#[derive(Clone)]
271
struct Client {
272
    router: rs::pipeline::PushRouter<serde_json::Value, RsAnnotated<serde_json::Value>>,
273
274
}

275
276
277
278
279
280
281
#[pyclass(eq, eq_int)]
#[derive(Clone, PartialEq)]
#[repr(i32)]
enum ModelType {
    Chat = 1,
    Completion = 2,
    Backend = 3,
282
    Embedding = 4,
283
284
}

285
286
287
#[pymethods]
impl DistributedRuntime {
    #[new]
288
    fn new(event_loop: PyObject, is_static: bool) -> PyResult<Self> {
289
        let worker = rs::Worker::from_settings().map_err(to_pyerr)?;
290
        INIT.get_or_try_init(|| {
291
            let primary = worker.tokio_runtime()?;
292
293
294
295
296
297
            pyo3_async_runtimes::tokio::init_with_runtime(primary)
                .map_err(|e| rs::error!("failed to initialize pyo3 static runtime: {:?}", e))?;
            rs::OK(())
        })
        .map_err(to_pyerr)?;

298
        let runtime = worker.runtime().clone();
299

300
301
302
303
304
305
306
307
308
309
310
        let inner =
            if is_static {
                runtime.secondary().block_on(
                    rs::DistributedRuntime::from_settings_without_discovery(runtime),
                )
            } else {
                runtime
                    .secondary()
                    .block_on(rs::DistributedRuntime::from_settings(runtime))
            };
        let inner = inner.map_err(to_pyerr)?;
311
312
313
314

        Ok(DistributedRuntime { inner, event_loop })
    }

Ryan Olson's avatar
Ryan Olson committed
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
    #[staticmethod]
    fn detached(py: Python) -> PyResult<Self> {
        let rt = rs::Worker::runtime_from_existing().map_err(to_pyerr)?;
        let handle = rt.primary();

        let inner = handle
            .block_on(rs::DistributedRuntime::from_settings(rt))
            .map_err(to_pyerr)?;

        Ok(DistributedRuntime {
            inner,
            event_loop: py.None(),
        })
    }

330
331
332
333
334
335
336
    fn namespace(&self, name: String) -> PyResult<Namespace> {
        Ok(Namespace {
            inner: self.inner.namespace(name).map_err(to_pyerr)?,
            event_loop: self.event_loop.clone(),
        })
    }

337
338
339
340
341
    fn etcd_client(&self) -> PyResult<Option<EtcdClient>> {
        match self.inner.etcd_client().clone() {
            Some(etcd_client) => Ok(Some(EtcdClient { inner: etcd_client })),
            None => Ok(None),
        }
342
343
    }

344
345
346
347
348
349
350
351
352
    fn shutdown(&self) {
        self.inner.runtime().shutdown();
    }

    fn event_loop(&self) -> PyObject {
        self.event_loop.clone()
    }
}

353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
#[pymethods]
impl EtcdKvCache {
    #[new]
    fn py_new(
        _etcd_client: &EtcdClient,
        _prefix: String,
        _initial_values: &Bound<'_, PyDict>,
    ) -> PyResult<Self> {
        // We can't create the KvCache here because it's async, so we'll return an error
        Err(PyErr::new::<pyo3::exceptions::PyRuntimeError, _>(
            "EtcdKvCache must be created using the 'new' class method",
        ))
    }

    #[staticmethod]
    #[allow(clippy::new_ret_no_self)]
    fn create<'p>(
        py: Python<'p>,
        etcd_client: &EtcdClient,
        prefix: String,
        initial_values: &Bound<'p, PyDict>,
    ) -> PyResult<Bound<'p, PyAny>> {
        let client = etcd_client.inner.clone();

        // Convert Python dict to Rust HashMap
        let mut rust_initial_values = std::collections::HashMap::new();
        for (key, value) in initial_values.iter() {
            let key_str = key.extract::<String>()?;

            // Handle both string and bytes values
            let value_bytes = if let Ok(bytes) = value.extract::<Vec<u8>>() {
                bytes
            } else if let Ok(string) = value.extract::<String>() {
                string.into_bytes()
            } else {
                return Err(PyErr::new::<pyo3::exceptions::PyTypeError, _>(
                    "Values must be either strings or bytes",
                ));
            };

            rust_initial_values.insert(key_str, value_bytes);
        }

        pyo3_async_runtimes::tokio::future_into_py(py, async move {
            let kv_cache = rs::transports::etcd::KvCache::new(client, prefix, rust_initial_values)
                .await
                .map_err(to_pyerr)?;

            Ok(EtcdKvCache {
                inner: Arc::new(kv_cache),
            })
        })
    }

    fn get<'p>(&self, py: Python<'p>, key: String) -> PyResult<Bound<'p, PyAny>> {
        let inner = self.inner.clone();

        pyo3_async_runtimes::tokio::future_into_py(py, async move {
            if let Some(value) = inner.get(&key).await {
                match Python::with_gil(|py| {
                    let py_obj = PyBytes::new(py, &value).into_pyobject(py)?;
                    Ok(py_obj.unbind().into_any())
                }) {
                    Ok(result) => Ok(result),
                    Err(e) => Err(e),
                }
            } else {
                Ok(Python::with_gil(|py| py.None()))
            }
        })
    }

    fn get_all<'p>(&self, py: Python<'p>) -> PyResult<Bound<'p, PyAny>> {
        let inner = self.inner.clone();

        pyo3_async_runtimes::tokio::future_into_py(py, async move {
            let all_values = inner.get_all().await;

            Python::with_gil(|py| {
                let dict = PyDict::new(py);
                for (key, value) in all_values {
                    // Strip the prefix from the key
                    let stripped_key = if let Some(stripped) = key.strip_prefix(&inner.prefix) {
                        stripped.to_string()
                    } else {
                        key
                    };
                    dict.set_item(stripped_key, PyBytes::new(py, &value))?;
                }
                let py_obj = dict.into_pyobject(py)?;
                Ok(py_obj.unbind().into_any())
            })
        })
    }

    #[pyo3(signature = (key, value, lease_id=None))]
    fn put<'p>(
        &self,
        py: Python<'p>,
        key: String,
        value: Vec<u8>,
        lease_id: Option<i64>,
    ) -> PyResult<Bound<'p, PyAny>> {
        let inner = self.inner.clone();

        pyo3_async_runtimes::tokio::future_into_py(py, async move {
            inner.put(&key, value, lease_id).await.map_err(to_pyerr)?;
            Ok(())
        })
    }
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497

    fn delete<'p>(&self, py: Python<'p>, key: String) -> PyResult<Bound<'p, PyAny>> {
        let inner = self.inner.clone();

        pyo3_async_runtimes::tokio::future_into_py(py, async move {
            inner.delete(&key).await.map_err(to_pyerr)?;
            Ok(())
        })
    }

    fn clear_all<'p>(&self, py: Python<'p>) -> PyResult<Bound<'p, PyAny>> {
        let inner = self.inner.clone();

        pyo3_async_runtimes::tokio::future_into_py(py, async move {
            // Get all keys with the prefix
            let all_keys = inner
                .get_all()
                .await
                .keys()
                .cloned()
                .collect::<Vec<String>>();

            // Delete each key
            for key in all_keys {
                // Strip the prefix from the key before deleting
                if let Some(stripped_key) = key.strip_prefix(&inner.prefix) {
                    inner.delete(stripped_key).await.map_err(to_pyerr)?;
                } else {
                    inner.delete(&key).await.map_err(to_pyerr)?;
                }
            }

            Ok(())
        })
    }
498
499
}

500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
#[pymethods]
impl CancellationToken {
    fn cancel(&self) {
        self.inner.cancel();
    }

    fn cancelled<'p>(&self, py: Python<'p>) -> PyResult<Bound<'p, PyAny>> {
        let token = self.inner.clone();
        pyo3_async_runtimes::tokio::future_into_py(py, async move {
            token.cancelled().await;
            Ok(())
        })
    }
}

#[pymethods]
impl Component {
    fn endpoint(&self, name: String) -> PyResult<Endpoint> {
        let inner = self.inner.endpoint(name);
        Ok(Endpoint {
            inner,
            event_loop: self.event_loop.clone(),
        })
    }

    fn create_service<'p>(&self, py: Python<'p>) -> PyResult<Bound<'p, PyAny>> {
        let builder = self.inner.service_builder();
        pyo3_async_runtimes::tokio::future_into_py(py, async move {
            let _ = builder.create().await.map_err(to_pyerr)?;
            Ok(())
        })
    }
}

#[pymethods]
impl Endpoint {
536
    #[pyo3(signature = (generator, graceful_shutdown = true, metrics_labels = None))]
537
538
539
540
    fn serve_endpoint<'p>(
        &self,
        py: Python<'p>,
        generator: PyObject,
541
        graceful_shutdown: Option<bool>,
542
        metrics_labels: Option<Vec<(String, String)>>,
543
544
545
546
547
548
    ) -> PyResult<Bound<'p, PyAny>> {
        let engine = Arc::new(engine::PythonAsyncEngine::new(
            generator,
            self.event_loop.clone(),
        )?);
        let ingress = JsonServerStreamingIngress::for_engine(engine).map_err(to_pyerr)?;
549
550
551
552
553
        let builder = self
            .inner
            .endpoint_builder()
            .metrics_labels(metrics_labels)
            .handler(ingress);
554
        let graceful_shutdown = graceful_shutdown.unwrap_or(true);
555
        pyo3_async_runtimes::tokio::future_into_py(py, async move {
556
557
558
559
560
            builder
                .graceful_shutdown(graceful_shutdown)
                .start()
                .await
                .map_err(to_pyerr)?;
561
562
563
564
            Ok(())
        })
    }

565
566
567
    fn client<'p>(&self, py: Python<'p>) -> PyResult<Bound<'p, PyAny>> {
        let inner = self.inner.clone();
        pyo3_async_runtimes::tokio::future_into_py(py, async move {
568
            let client = inner.client().await.map_err(to_pyerr)?;
569
570
571
572
573
574
            let push_router = rs::pipeline::PushRouter::<
                serde_json::Value,
                RsAnnotated<serde_json::Value>,
            >::from_client(client, Default::default())
            .await
            .map_err(to_pyerr)?;
575
576
577
            Ok(Client {
                router: push_router,
            })
578
579
        })
    }
580
581

    fn lease_id(&self) -> i64 {
582
583
584
585
586
        self.inner
            .drt()
            .primary_lease()
            .map(|l| l.id())
            .unwrap_or(0)
587
    }
588
589
590
591
592
593
594
595
596
597
598
599
600
}

#[pymethods]
impl Namespace {
    fn component(&self, name: String) -> PyResult<Component> {
        let inner = self.inner.component(name).map_err(to_pyerr)?;
        Ok(Component {
            inner,
            event_loop: self.event_loop.clone(),
        })
    }
}

601
602
#[pymethods]
impl EtcdClient {
603
604
605
606
607
608
609
610
611
612
613
    #[pyo3(signature = (key, value, lease_id=None))]
    fn kv_create<'p>(
        &self,
        py: Python<'p>,
        key: String,
        value: Vec<u8>,
        lease_id: Option<i64>,
    ) -> PyResult<Bound<'p, PyAny>> {
        let client = self.inner.clone();
        pyo3_async_runtimes::tokio::future_into_py(py, async move {
            client
614
                .kv_create(&key, value, lease_id)
615
616
617
618
619
620
                .await
                .map_err(to_pyerr)?;
            Ok(())
        })
    }

621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
    #[pyo3(signature = (key, value, lease_id=None))]
    fn kv_create_or_validate<'p>(
        &self,
        py: Python<'p>,
        key: String,
        value: Vec<u8>,
        lease_id: Option<i64>,
    ) -> PyResult<Bound<'p, PyAny>> {
        let client = self.inner.clone();
        pyo3_async_runtimes::tokio::future_into_py(py, async move {
            client
                .kv_create_or_validate(key, value, lease_id)
                .await
                .map_err(to_pyerr)?;
            Ok(())
        })
    }

639
640
641
642
    fn primary_lease_id(&self) -> i64 {
        self.inner.lease_id()
    }

643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
    #[pyo3(signature = (key, value, lease_id=None))]
    fn kv_put<'p>(
        &self,
        py: Python<'p>,
        key: String,
        value: Vec<u8>,
        lease_id: Option<i64>,
    ) -> PyResult<Bound<'p, PyAny>> {
        let client = self.inner.clone();
        pyo3_async_runtimes::tokio::future_into_py(py, async move {
            client
                .kv_put(key, value, lease_id)
                .await
                .map_err(to_pyerr)?;
            Ok(())
        })
    }

    fn kv_get_prefix<'p>(&self, py: Python<'p>, prefix: String) -> PyResult<Bound<'p, PyAny>> {
        let client = self.inner.clone();
        pyo3_async_runtimes::tokio::future_into_py(py, async move {
            let result = client
                .kv_get_prefix(prefix)
                .await
                .map_err(|e| pyo3::exceptions::PyRuntimeError::new_err(e.to_string()))?;

            // Convert Vec<KeyValue> to a list of dictionaries
            let py_list = Python::with_gil(|py| {
                let list = PyList::empty(py);
                for kv in result {
                    let dict = PyDict::new(py);
                    dict.set_item("key", String::from_utf8_lossy(kv.key()).to_string())?;
                    dict.set_item("value", PyBytes::new(py, kv.value()))?;
                    dict.set_item("create_revision", kv.create_revision())?;
                    dict.set_item("mod_revision", kv.mod_revision())?;
                    dict.set_item("version", kv.version())?;
                    dict.set_item("lease", kv.lease())?;
                    list.append(dict)?;
                }
                Ok::<Py<PyList>, PyErr>(list.into())
            })?;

            Ok(py_list)
        })
    }
688
689
690
691
692
693
694
695

    fn revoke_lease<'p>(&self, py: Python<'p>, lease_id: i64) -> PyResult<Bound<'p, PyAny>> {
        let client = self.inner.clone();
        pyo3_async_runtimes::tokio::future_into_py(py, async move {
            client.revoke_lease(lease_id).await.map_err(to_pyerr)?;
            Ok(())
        })
    }
696
697
}

698
699
#[pymethods]
impl Client {
700
701
702
703
    /// Get list of current instances.
    /// Replaces endpoint_ids.
    fn instance_ids(&self) -> Vec<i64> {
        self.router.client.instance_ids()
704
705
    }

706
707
708
    /// Wait for an instance to be available for work.
    /// Replaces wait_for_endpoints.
    fn wait_for_instances<'p>(&self, py: Python<'p>) -> PyResult<Bound<'p, PyAny>> {
709
        let inner = self.router.client.clone();
710
        pyo3_async_runtimes::tokio::future_into_py(py, async move {
711
            inner
712
                .wait_for_instances()
713
714
715
                .await
                .map(|v| v.into_iter().map(|cei| cei.id()).collect::<Vec<i64>>())
                .map_err(to_pyerr)
716
717
718
719
        })
    }

    /// Issue a request to the endpoint using the default routing strategy.
720
    #[pyo3(signature = (request, annotated=DEFAULT_ANNOTATED_SETTING, context=None))]
721
722
723
724
725
    fn generate<'p>(
        &self,
        py: Python<'p>,
        request: PyObject,
        annotated: Option<bool>,
726
        context: Option<context::Context>,
727
    ) -> PyResult<Bound<'p, PyAny>> {
728
        if self.router.client.is_static() {
729
            self.r#static(py, request, annotated, context)
730
        } else {
731
            self.random(py, request, annotated, context)
732
        }
733
734
735
    }

    /// Send a request to the next endpoint in a round-robin fashion.
736
    #[pyo3(signature = (request, annotated=DEFAULT_ANNOTATED_SETTING, context=None))]
737
738
739
740
741
    fn round_robin<'p>(
        &self,
        py: Python<'p>,
        request: PyObject,
        annotated: Option<bool>,
742
        context: Option<context::Context>,
743
744
745
746
747
    ) -> PyResult<Bound<'p, PyAny>> {
        let request: serde_json::Value = pythonize::depythonize(&request.into_bound(py))?;
        let annotated = annotated.unwrap_or(false);

        let (tx, rx) = tokio::sync::mpsc::channel(32);
748
        let client = self.router.clone();
749
750

        pyo3_async_runtimes::tokio::future_into_py(py, async move {
751
752
753
754
755
756
757
758
759
            let stream = match context {
                Some(context) => {
                    let request = RsContext::with_id(request, context.inner().id().to_string());
                    let stream = client.round_robin(request).await.map_err(to_pyerr)?;
                    context.inner().link_child(stream.context());
                    stream
                }
                _ => client.round_robin(request.into()).await.map_err(to_pyerr)?,
            };
760
761
762
763
764
765
766
767
768
            tokio::spawn(process_stream(stream, tx));
            Ok(AsyncResponseStream {
                rx: Arc::new(Mutex::new(rx)),
                annotated,
            })
        })
    }

    /// Send a request to a random endpoint.
769
    #[pyo3(signature = (request, annotated=DEFAULT_ANNOTATED_SETTING, context=None))]
770
771
772
773
774
    fn random<'p>(
        &self,
        py: Python<'p>,
        request: PyObject,
        annotated: Option<bool>,
775
        context: Option<context::Context>,
776
777
778
779
780
    ) -> PyResult<Bound<'p, PyAny>> {
        let request: serde_json::Value = pythonize::depythonize(&request.into_bound(py))?;
        let annotated = annotated.unwrap_or(false);

        let (tx, rx) = tokio::sync::mpsc::channel(32);
781
        let client = self.router.clone();
782
783

        pyo3_async_runtimes::tokio::future_into_py(py, async move {
784
785
786
787
788
789
790
791
792
            let stream = match context {
                Some(context) => {
                    let request = RsContext::with_id(request, context.inner().id().to_string());
                    let stream = client.random(request).await.map_err(to_pyerr)?;
                    context.inner().link_child(stream.context());
                    stream
                }
                _ => client.random(request.into()).await.map_err(to_pyerr)?,
            };
793
794
795
796
797
798
799
800
801
            tokio::spawn(process_stream(stream, tx));
            Ok(AsyncResponseStream {
                rx: Arc::new(Mutex::new(rx)),
                annotated,
            })
        })
    }

    /// Directly send a request to a specific endpoint.
802
    #[pyo3(signature = (request, instance_id, annotated=DEFAULT_ANNOTATED_SETTING, context=None))]
803
804
805
806
    fn direct<'p>(
        &self,
        py: Python<'p>,
        request: PyObject,
807
        instance_id: i64,
808
        annotated: Option<bool>,
809
        context: Option<context::Context>,
810
811
812
813
814
    ) -> PyResult<Bound<'p, PyAny>> {
        let request: serde_json::Value = pythonize::depythonize(&request.into_bound(py))?;
        let annotated = annotated.unwrap_or(false);

        let (tx, rx) = tokio::sync::mpsc::channel(32);
815
        let client = self.router.clone();
816
817

        pyo3_async_runtimes::tokio::future_into_py(py, async move {
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
            let stream = match context {
                Some(context) => {
                    let request = RsContext::with_id(request, context.inner().id().to_string());
                    let stream = client
                        .direct(request, instance_id)
                        .await
                        .map_err(to_pyerr)?;
                    context.inner().link_child(stream.context());
                    stream
                }
                _ => client
                    .direct(request.into(), instance_id)
                    .await
                    .map_err(to_pyerr)?,
            };
833
834
835
836
837
838
839
840
841

            tokio::spawn(process_stream(stream, tx));

            Ok(AsyncResponseStream {
                rx: Arc::new(Mutex::new(rx)),
                annotated,
            })
        })
    }
842
843

    /// Directly send a request to a pre-defined static worker
844
    #[pyo3(signature = (request, annotated=DEFAULT_ANNOTATED_SETTING, context=None))]
845
846
847
848
849
    fn r#static<'p>(
        &self,
        py: Python<'p>,
        request: PyObject,
        annotated: Option<bool>,
850
        context: Option<context::Context>,
851
852
853
854
855
    ) -> PyResult<Bound<'p, PyAny>> {
        let request: serde_json::Value = pythonize::depythonize(&request.into_bound(py))?;
        let annotated = annotated.unwrap_or(false);

        let (tx, rx) = tokio::sync::mpsc::channel(32);
856
        let client = self.router.clone();
857
858

        pyo3_async_runtimes::tokio::future_into_py(py, async move {
859
860
861
862
863
864
865
866
867
            let stream = match context {
                Some(context) => {
                    let request = RsContext::with_id(request, context.inner().id().to_string());
                    let stream = client.r#static(request).await.map_err(to_pyerr)?;
                    context.inner().link_child(stream.context());
                    stream
                }
                _ => client.r#static(request.into()).await.map_err(to_pyerr)?,
            };
868
869
870
871
872
873
874
875
876

            tokio::spawn(process_stream(stream, tx));

            Ok(AsyncResponseStream {
                rx: Arc::new(Mutex::new(rx)),
                annotated,
            })
        })
    }
877
878
879
}

async fn process_stream(
880
    stream: EngineStream<RsAnnotated<serde_json::Value>>,
881
882
883
884
885
    tx: tokio::sync::mpsc::Sender<RsAnnotated<PyObject>>,
) {
    let mut stream = stream;
    while let Some(response) = stream.next().await {
        // Convert the response to a PyObject using Python's GIL
886
        let annotated: RsAnnotated<serde_json::Value> = response;
887
888
889
890
891
892
893
894
895
896
897
898
        let annotated: RsAnnotated<PyObject> = annotated.map_data(|data| {
            let result = Python::with_gil(|py| match pythonize::pythonize(py, &data) {
                Ok(pyobj) => Ok(pyobj.into()),
                Err(e) => Err(e.to_string()),
            });
            result
        });

        let is_error = annotated.is_error();

        // Send the PyObject through the channel or log an error
        if let Err(e) = tx.send(annotated).await {
899
            tracing::error!("Failed to send response: {:?}", e);
900
            break;
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
        }

        if is_error {
            break;
        }
    }
}

#[pyclass]
struct AsyncResponseStream {
    rx: Arc<Mutex<tokio::sync::mpsc::Receiver<RsAnnotated<PyObject>>>>,
    annotated: bool,
}

#[pymethods]
impl AsyncResponseStream {
    /// This method is required to implement the `AsyncIterator` protocol.
    #[pyo3(name = "__aiter__")]
    fn aiter(slf: PyRef<Self>, py: Python) -> PyResult<Py<PyAny>> {
        slf.into_py_any(py)
    }
    /// This method is required to implement the `AsyncIterator` protocol.
    #[pyo3(name = "__anext__")]
    fn next<'p>(&self, py: Python<'p>) -> PyResult<Bound<'p, PyAny>> {
        let rx = self.rx.clone();
        let annotated = self.annotated;

        pyo3_async_runtimes::tokio::future_into_py(py, async move {
            loop {
                let value = rx.lock().await.recv().await;
                match value {
                    Some(pyobj) => {
                        let pyobj = match pyobj.ok() {
                            Ok(pyobj) => pyobj,
                            Err(e) => {
                                return Err(PyErr::new::<pyo3::exceptions::PyValueError, _>(e));
                            }
                        };

                        if annotated {
                            let object = Annotated { inner: pyobj };
                            #[allow(deprecated)]
                            let object = Python::with_gil(|py| object.into_py(py));
                            return Ok(object);
                        } else {
                            match pyobj.data {
                                Some(data) => return Ok(data),
                                None => continue,
                            }
                        }
                    }
                    None => return Err(PyStopAsyncIteration::new_err("Stream exhausted")),
                }
            }
        })
    }
}

#[pyclass]
struct Annotated {
    inner: RsAnnotated<PyObject>,
}

#[pymethods]
impl Annotated {
    #[new]
    fn new(data: PyObject) -> Self {
        Annotated {
            inner: RsAnnotated::from_data(data),
        }
    }

    fn is_error(&self) -> bool {
        self.inner.is_error()
    }

    fn data(&self) -> Option<PyObject> {
        self.inner.data.clone()
    }

    fn event(&self) -> Option<String> {
        self.inner.event.clone()
    }

    fn comments(&self) -> Option<Vec<String>> {
        self.inner.comment.clone()
    }

    fn id(&self) -> Option<String> {
        self.inner.id.clone()
    }

    #[pyo3(name = "__repr__")]
    fn _repr(&self, py: Python) -> String {
        let data = self.inner.data.clone().map(|obj| {
            obj.call_method0(py, "__repr__")
                .and_then(|repr_obj| repr_obj.extract::<Py<PyString>>(py))
                .map(|py_str| py_str.to_string_lossy(py).into_owned())
                .unwrap_or_else(|_| "<failed_repr>".to_string())
        });

        format!(
            "Annotated(data={}, event={}, comment={:?}, id={})",
            data.unwrap_or_else(|| "<no_data>".to_string()),
            self.inner.event.as_deref().unwrap_or("None"),
            self.inner.comment.as_deref().unwrap_or(&[]),
            self.inner.id.as_deref().unwrap_or("None")
        )
    }
}