etcd.rs 30.4 KB
Newer Older
1
2
// SPDX-FileCopyrightText: Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
// SPDX-License-Identifier: Apache-2.0
Ryan Olson's avatar
Ryan Olson committed
3

4
5
use crate::runtime::Runtime;
use anyhow::{Context, Result};
Ryan Olson's avatar
Ryan Olson committed
6
7
8
9
10

use async_nats::jetstream::kv;
use derive_builder::Builder;
use derive_getters::Dissolve;
use futures::StreamExt;
11
12
use std::collections::HashMap;
use std::sync::Arc;
13
use tokio::sync::{RwLock, mpsc};
Ryan Olson's avatar
Ryan Olson committed
14
15
use validator::Validate;

16
use etcd_client::{
17
18
    Certificate, Compare, CompareOp, DeleteOptions, GetOptions, Identity, LockClient, LockOptions,
    LockResponse, PutOptions, PutResponse, TlsOptions, Txn, TxnOp, TxnOpResponse, WatchOptions,
19
    WatchStream, Watcher,
20
};
Ryan Olson's avatar
Ryan Olson committed
21
pub use etcd_client::{ConnectOptions, KeyValue, LeaseClient};
22
use tokio::time::{Duration, interval};
23
use tokio_util::sync::CancellationToken;
Ryan Olson's avatar
Ryan Olson committed
24

25
mod connector;
Ryan Olson's avatar
Ryan Olson committed
26
mod lease;
27
mod lock;
28

29
use connector::Connector;
Ryan Olson's avatar
Ryan Olson committed
30
use lease::*;
31
pub use lock::*;
Ryan Olson's avatar
Ryan Olson committed
32

33
use super::utils::build_in_runtime;
34
use crate::config::environment_names::etcd as env_etcd;
35

36
37
38
39
40
41
/// ETCD Client
#[derive(Clone)]
pub struct Client {
    connector: Arc<Connector>,
    primary_lease: u64,
    runtime: Runtime,
42
43
44
    // Exclusive runtime for etcd lease keep-alive and watch tasks
    // Avoid those tasks from being starved when the main runtime is busy
    // WARNING: Do not await on main runtime from this runtime or deadlocks may occur
45
46
47
48
49
50
51
52
53
    rt: Arc<tokio::runtime::Runtime>,
}

impl std::fmt::Debug for Client {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        write!(f, "etcd::Client primary_lease={}", self.primary_lease)
    }
}

Ryan Olson's avatar
Ryan Olson committed
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
impl Client {
    pub fn builder() -> ClientOptionsBuilder {
        ClientOptionsBuilder::default()
    }

    /// Create a new discovery client
    ///
    /// This will establish a connection to the etcd server, create a primary lease,
    /// and spawn a task to keep the lease alive and tie the lifetime of the [`Runtime`]
    /// to the lease.
    ///
    /// If the lease expires, the [`Runtime`] will be shutdown.
    /// If the [`Runtime`] is shutdown, the lease will be revoked.
    pub async fn new(config: ClientOptions, runtime: Runtime) -> Result<Self> {
        let token = runtime.primary_token();

70
        let ((connector, lease_id), rt) = build_in_runtime(
71
            async move {
72
73
74
75
76
77
78
79
80
81
82
83
                let etcd_urls = config.etcd_url.clone();
                let connect_options = config.etcd_connect_options.clone();

                // Create the connector
                let connector = Connector::new(etcd_urls, connect_options)
                    .await
                    .with_context(|| {
                        format!(
                            "Unable to connect to etcd server at {}. Check etcd server status",
                            config.etcd_url.join(", ")
                        )
                    })?;
84

85
                let lease_id = if config.attach_lease {
86
                    create_lease(connector.clone(), 10, token)
87
                        .await
88
89
90
91
92
                        .with_context(|| {
                            format!(
                                "Unable to create lease. Check etcd server status at {}",
                                config.etcd_url.join(", ")
                            )
93
                        })?
94
95
96
97
                } else {
                    0
                };

98
                Ok((connector, lease_id))
99
100
101
102
            },
            1,
        )
        .await?;
Ryan Olson's avatar
Ryan Olson committed
103
104

        Ok(Client {
105
            connector,
106
            primary_lease: lease_id,
107
            rt,
Ryan Olson's avatar
Ryan Olson committed
108
109
110
111
            runtime,
        })
    }

112
113
    /// Get a clone of the underlying [`etcd_client::Client`] instance.
    /// This returns a clone since the client is behind an RwLock.
114
    fn etcd_client(&self) -> etcd_client::Client {
115
        self.connector.get_client()
Ryan Olson's avatar
Ryan Olson committed
116
117
118
    }

    /// Get the primary lease ID.
119
    pub fn lease_id(&self) -> u64 {
Ryan Olson's avatar
Ryan Olson committed
120
121
122
        self.primary_lease
    }

123
124
125
126
127
128
129
    /// Returns Ok(None) if value was created, Ok(Some(revision)) if the value already exists.
    pub async fn kv_create(
        &self,
        key: &str,
        value: Vec<u8>,
        lease_id: Option<u64>,
    ) -> Result<Option<u64>> {
130
        let id = lease_id.unwrap_or(self.lease_id());
131
        let put_options = PutOptions::new().with_lease(id as i64);
Ryan Olson's avatar
Ryan Olson committed
132

133
        // Build transaction that creates key only if it doesn't exist
Ryan Olson's avatar
Ryan Olson committed
134
        let txn = Txn::new()
135
            .when(vec![Compare::version(key, CompareOp::Equal, 0)]) // Ensure the lock does not exist
Ryan Olson's avatar
Ryan Olson committed
136
            .and_then(vec![
137
                TxnOp::put(key, value, Some(put_options)), // Create the object
138
139
140
            ])
            .or_else(vec![
                TxnOp::get(key, None), // Key exists, get its info
Ryan Olson's avatar
Ryan Olson committed
141
142
143
            ]);

        // Execute the transaction
144
        let result = self.connector.get_client().kv_client().txn(txn).await?;
145

146
        // Created
147
        if result.succeeded() {
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
            return Ok(None);
        }

        // Already exists
        if let Some(etcd_client::TxnOpResponse::Get(get_resp)) =
            result.op_responses().into_iter().next()
            && let Some(kv) = get_resp.kvs().first()
        {
            let version = kv.version() as u64;
            return Ok(Some(version));
        }

        // Error
        for resp in result.op_responses() {
            tracing::warn!(response = ?resp, "kv_create etcd op response");
163
        }
164
        anyhow::bail!("Unable to create key. Check etcd server status")
165
166
167
168
169
170
171
    }

    /// Atomically create a key if it does not exist, or validate the values are identical if the key exists.
    pub async fn kv_create_or_validate(
        &self,
        key: String,
        value: Vec<u8>,
172
        lease_id: Option<u64>,
173
    ) -> Result<()> {
174
        let id = lease_id.unwrap_or(self.lease_id());
175
        let put_options = PutOptions::new().with_lease(id as i64);
176
177
178
179
180
181

        // Build the transaction that either creates the key if it doesn't exist,
        // or validates the existing value matches what we expect
        let txn = Txn::new()
            .when(vec![Compare::version(key.as_str(), CompareOp::Equal, 0)]) // Key doesn't exist
            .and_then(vec![
182
                TxnOp::put(key.as_str(), value.clone(), Some(put_options)), // Create it
183
184
185
186
187
188
189
190
191
192
193
            ])
            .or_else(vec![
                // If key exists but values don't match, this will fail the transaction
                TxnOp::txn(Txn::new().when(vec![Compare::value(
                    key.as_str(),
                    CompareOp::Equal,
                    value.clone(),
                )])),
            ]);

        // Execute the transaction
194
        let result = self.connector.get_client().kv_client().txn(txn).await?;
195
196
197
198
199
200
201
202
203

        // We have to enumerate the response paths to determine if the transaction succeeded
        if result.succeeded() {
            Ok(())
        } else {
            match result.op_responses().first() {
                Some(response) => match response {
                    TxnOpResponse::Txn(response) => match response.succeeded() {
                        true => Ok(()),
204
                        false => anyhow::bail!(
205
                            "Unable to create or validate key. Check etcd server status"
206
                        ),
207
                    },
208
209
210
                    _ => {
                        anyhow::bail!("Unable to validate key operation. Check etcd server status")
                    }
211
                },
212
                None => anyhow::bail!("Unable to create or validate key. Check etcd server status"),
213
214
215
216
217
218
219
220
            }
        }
    }

    pub async fn kv_put(
        &self,
        key: impl AsRef<str>,
        value: impl AsRef<[u8]>,
221
        lease_id: Option<u64>,
222
    ) -> Result<()> {
223
        let id = lease_id.unwrap_or(self.lease_id());
224
        let put_options = PutOptions::new().with_lease(id as i64);
225
        let _ = self
226
227
            .connector
            .get_client()
228
            .kv_client()
229
            .put(key.as_ref(), value.as_ref(), Some(put_options))
230
            .await?;
Ryan Olson's avatar
Ryan Olson committed
231
232
233
        Ok(())
    }

234
235
236
237
238
239
240
241
    pub async fn kv_put_with_options(
        &self,
        key: impl AsRef<str>,
        value: impl AsRef<[u8]>,
        options: Option<PutOptions>,
    ) -> Result<PutResponse> {
        let options = options
            .unwrap_or_default()
242
            .with_lease(self.lease_id() as i64);
243
244
        self.connector
            .get_client()
245
246
247
248
249
250
251
252
253
254
255
            .kv_client()
            .put(key.as_ref(), value.as_ref(), Some(options))
            .await
            .map_err(|err| err.into())
    }

    pub async fn kv_get(
        &self,
        key: impl Into<Vec<u8>>,
        options: Option<GetOptions>,
    ) -> Result<Vec<KeyValue>> {
256
257
258
259
260
261
        let mut get_response = self
            .connector
            .get_client()
            .kv_client()
            .get(key, options)
            .await?;
262
263
264
265
266
267
268
        Ok(get_response.take_kvs())
    }

    pub async fn kv_delete(
        &self,
        key: impl Into<Vec<u8>>,
        options: Option<DeleteOptions>,
269
    ) -> Result<u64> {
270
271
        self.connector
            .get_client()
272
273
274
            .kv_client()
            .delete(key, options)
            .await
275
            .map(|del_response| del_response.deleted() as u64)
276
277
278
            .map_err(|err| err.into())
    }

Ryan Olson's avatar
Ryan Olson committed
279
280
    pub async fn kv_get_prefix(&self, prefix: impl AsRef<str>) -> Result<Vec<KeyValue>> {
        let mut get_response = self
281
282
            .connector
            .get_client()
Ryan Olson's avatar
Ryan Olson committed
283
284
285
286
287
288
289
            .kv_client()
            .get(prefix.as_ref(), Some(GetOptions::new().with_prefix()))
            .await?;

        Ok(get_response.take_kvs())
    }

290
291
292
293
294
    /// Acquire a distributed lock using etcd's native lock mechanism
    /// Returns a LockResponse that can be used to unlock later
    pub async fn lock(
        &self,
        key: impl Into<Vec<u8>>,
295
        lease_id: Option<u64>,
296
    ) -> Result<LockResponse> {
297
        let mut lock_client = self.connector.get_client().lock_client();
298
        let id = lease_id.unwrap_or(self.lease_id());
299
        let options = LockOptions::new().with_lease(id as i64);
300
301
302
303
304
305
306
307
        lock_client
            .lock(key, Some(options))
            .await
            .map_err(|err| err.into())
    }

    /// Release a distributed lock using the key from the LockResponse
    pub async fn unlock(&self, lock_key: impl Into<Vec<u8>>) -> Result<()> {
308
        let mut lock_client = self.connector.get_client().lock_client();
309
310
311
312
313
314
315
        lock_client
            .unlock(lock_key)
            .await
            .map_err(|err: etcd_client::Error| anyhow::anyhow!(err))?;
        Ok(())
    }

316
317
318
319
320
321
322
323
    /// Like kv_get_and_watch_prefix but only for new changes, does not include existing values.
    pub async fn kv_watch_prefix(
        &self,
        prefix: impl AsRef<str> + std::fmt::Display,
    ) -> Result<PrefixWatcher> {
        self.watch_internal(prefix, false).await
    }

324
325
326
    pub async fn kv_get_and_watch_prefix(
        &self,
        prefix: impl AsRef<str> + std::fmt::Display,
327
328
329
330
    ) -> Result<PrefixWatcher> {
        self.watch_internal(prefix, true).await
    }

331
332
333
334
335
    /// Core watch implementation that sets up a resilient watcher for a key prefix.
    ///
    /// Creates a background task that maintains a watch stream with automatic reconnection
    /// on recoverable errors. If `include_existing` is true, existing keys are included
    /// in the initial watch events.
336
337
338
339
    async fn watch_internal(
        &self,
        prefix: impl AsRef<str> + std::fmt::Display,
        include_existing: bool,
340
    ) -> Result<PrefixWatcher> {
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
        let (tx, rx) = mpsc::channel(32);

        // Get start revision and send existing KVs
        let mut start_revision = self
            .get_start_revision(
                prefix.as_ref(),
                if include_existing { Some(&tx) } else { None },
            )
            .await?;

        // Resilience watch stream in background
        let connector = self.connector.clone();
        let prefix_str = prefix.as_ref().to_string();
        self.rt.spawn(async move {
            let mut reconnect = true;
            while reconnect {
                // Start a new watch stream
                let watch_stream =
                    match Self::new_watch_stream(&connector, &prefix_str, start_revision).await {
                        Ok(stream) => stream,
                        Err(_) => return,
                    };

                // Watch the stream
                reconnect =
                    Self::monitor_watch_stream(watch_stream, &prefix_str, &mut start_revision, &tx)
                        .await;
            }
        });
Ryan Olson's avatar
Ryan Olson committed
370

371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
        Ok(PrefixWatcher {
            prefix: prefix.as_ref().to_string(),
            rx,
        })
    }

    /// Fetch the initial revision for watching and optionally send existing key-values.
    ///
    /// Returns the next revision to watch from. If `existing_kvs_tx` is provided,
    /// all existing keys with the prefix are sent through the channel first.
    async fn get_start_revision(
        &self,
        prefix: impl AsRef<str> + std::fmt::Display,
        existing_kvs_tx: Option<&mpsc::Sender<WatchEvent>>,
    ) -> Result<i64> {
        let mut kv_client = self.connector.get_client().kv_client();
Ryan Olson's avatar
Ryan Olson committed
387
388
389
390
        let mut get_response = kv_client
            .get(prefix.as_ref(), Some(GetOptions::new().with_prefix()))
            .await?;

391
392
        // Get the start revision
        let mut start_revision = get_response
Ryan Olson's avatar
Ryan Olson committed
393
            .header()
394
            .ok_or(anyhow::anyhow!("missing header; unable to get revision"))?
Ryan Olson's avatar
Ryan Olson committed
395
            .revision();
396
        tracing::trace!("{prefix}: start_revision: {start_revision}");
397
        start_revision += 1;
Ryan Olson's avatar
Ryan Olson committed
398

399
400
        // Send existing KVs from response if requested
        if let Some(tx) = existing_kvs_tx {
401
402
            let kvs = get_response.take_kvs();
            tracing::trace!("initial kv count: {:?}", kvs.len());
403
404
405
406
            for kv in kvs.into_iter() {
                tx.send(WatchEvent::Put(kv)).await?;
            }
        }
Ryan Olson's avatar
Ryan Olson committed
407

408
409
        Ok(start_revision)
    }
Ryan Olson's avatar
Ryan Olson committed
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
    /// Establish a new watch stream with automatic retry and reconnection.
    ///
    /// Attempts to create a watch stream, reconnecting to ETCD if necessary.
    /// Uses a 10-second timeout for reconnection attempts before giving up.
    async fn new_watch_stream(
        connector: &Arc<Connector>,
        prefix: &String,
        start_revision: i64,
    ) -> Result<WatchStream> {
        loop {
            match connector
                .get_client()
                .watch_client()
                .watch(
                    prefix.as_str(),
                    Some(
                        WatchOptions::new()
                            .with_prefix()
                            .with_start_revision(start_revision)
                            .with_prev_key(),
                    ),
                )
                .await
            {
                Ok((_, watch_stream)) => {
                    tracing::debug!("Watch stream established for prefix '{}'", prefix);
                    return Ok(watch_stream);
                }
                Err(err) => {
                    tracing::debug!(error = %err, "Failed to establish watch stream for prefix '{}'", prefix);
                    let deadline = std::time::Instant::now() + Duration::from_secs(10);
                    if let Err(err) = connector.reconnect(deadline).await {
                        tracing::error!(
                            "Failed to reconnect to ETCD within 10 secs for watching prefix '{}': {}",
                            prefix,
                            err
                        );
                        return Err(err);
449
                    }
450
                    // continue - retry establishing the watch stream
Ryan Olson's avatar
Ryan Olson committed
451
452
                }
            }
453
454
        }
    }
Ryan Olson's avatar
Ryan Olson committed
455

456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
    /// Monitor a watch stream and forward events to receivers.
    ///
    /// Returns `true` for recoverable errors (network issues, stream closure) that warrant
    /// reconnection attempts. Returns `false` for permanent failures (protocol violations,
    /// channel errors, no receivers) where watching should stop.
    async fn monitor_watch_stream(
        mut watch_stream: WatchStream,
        prefix: &String,
        start_revision: &mut i64,
        tx: &mpsc::Sender<WatchEvent>,
    ) -> bool {
        loop {
            tokio::select! {
                maybe_resp = watch_stream.next() => {
                    // Handle the watch response
                    let response = match maybe_resp {
                        Some(Ok(res)) => res,
                        Some(Err(err)) => {
                            tracing::warn!(error = %err, "Error watching stream for prefix '{}'", prefix);
                            return true; // Exit to reconnect
                        }
                        None => {
                            tracing::warn!("Watch stream unexpectedly closed for prefix '{}'", prefix);
                            return true; // Exit to reconnect
Ryan Olson's avatar
Ryan Olson committed
480
                        }
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
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
                    };

                    // Update revision for reconnect
                    *start_revision = match response.header() {
                        Some(header) => header.revision() + 1,
                        None => {
                            tracing::error!("Missing header in watch response for prefix '{}'", prefix);
                            return false;
                        }
                    };

                    // Process events
                    if Self::process_watch_events(response.events(), tx).await.is_err() {
                        return false;
                    };
                }
                _ = tx.closed() => {
                    tracing::debug!("no more receivers, stopping watcher");
                    return false;
                }
            }
        }
    }

    /// Process etcd events and forward them as Put/Delete watch events.
    ///
    /// Filters out events without key-values and transforms etcd events into
    /// appropriate WatchEvent types for channel transmission.
    async fn process_watch_events(
        events: &[etcd_client::Event],
        tx: &mpsc::Sender<WatchEvent>,
    ) -> Result<()> {
        for event in events {
            // Extract the KeyValue if it exists
            let Some(kv) = event.kv() else {
                continue; // Skip events with no KV
            };

            // Handle based on event type
            match event.event_type() {
                etcd_client::EventType::Put => {
                    if let Err(err) = tx.send(WatchEvent::Put(kv.clone())).await {
                        tracing::error!("kv watcher error forwarding WatchEvent::Put: {err}");
                        return Err(err.into());
Ryan Olson's avatar
Ryan Olson committed
525
                    }
526
527
528
529
                }
                etcd_client::EventType::Delete => {
                    if tx.send(WatchEvent::Delete(kv.clone())).await.is_err() {
                        return Err(anyhow::anyhow!("failed to send WatchEvent::Delete"));
530
                    }
Ryan Olson's avatar
Ryan Olson committed
531
532
                }
            }
533
534
        }
        Ok(())
Ryan Olson's avatar
Ryan Olson committed
535
536
537
538
539
540
541
542
543
    }
}

#[derive(Dissolve)]
pub struct PrefixWatcher {
    prefix: String,
    rx: mpsc::Receiver<WatchEvent>,
}

544
#[derive(Debug)]
Ryan Olson's avatar
Ryan Olson committed
545
546
547
548
549
550
551
552
553
pub enum WatchEvent {
    Put(KeyValue),
    Delete(KeyValue),
}

/// ETCD client configuration options
#[derive(Debug, Clone, Builder, Validate)]
pub struct ClientOptions {
    #[validate(length(min = 1))]
Ryan Olson's avatar
Ryan Olson committed
554
    pub etcd_url: Vec<String>,
Ryan Olson's avatar
Ryan Olson committed
555
556

    #[builder(default)]
Ryan Olson's avatar
Ryan Olson committed
557
    pub etcd_connect_options: Option<ConnectOptions>,
558
559
560

    /// If true, the client will attach a lease to the primary [`CancellationToken`].
    #[builder(default = "true")]
Ryan Olson's avatar
Ryan Olson committed
561
    pub attach_lease: bool,
Ryan Olson's avatar
Ryan Olson committed
562
563
564
565
}

impl Default for ClientOptions {
    fn default() -> Self {
wxsm's avatar
wxsm committed
566
567
568
        let mut connect_options = None;

        if let (Ok(username), Ok(password)) = (
569
570
            std::env::var(env_etcd::auth::ETCD_AUTH_USERNAME),
            std::env::var(env_etcd::auth::ETCD_AUTH_PASSWORD),
wxsm's avatar
wxsm committed
571
572
573
574
        ) {
            // username and password are set
            connect_options = Some(ConnectOptions::new().with_user(username, password));
        } else if let (Ok(ca), Ok(cert), Ok(key)) = (
575
576
577
            std::env::var(env_etcd::auth::ETCD_AUTH_CA),
            std::env::var(env_etcd::auth::ETCD_AUTH_CLIENT_CERT),
            std::env::var(env_etcd::auth::ETCD_AUTH_CLIENT_KEY),
wxsm's avatar
wxsm committed
578
579
580
581
582
583
584
585
586
587
588
        ) {
            // TLS is set
            connect_options = Some(
                ConnectOptions::new().with_tls(
                    TlsOptions::new()
                        .ca_certificate(Certificate::from_pem(ca))
                        .identity(Identity::from_pem(cert, key)),
                ),
            );
        }

Ryan Olson's avatar
Ryan Olson committed
589
590
        ClientOptions {
            etcd_url: default_servers(),
wxsm's avatar
wxsm committed
591
            etcd_connect_options: connect_options,
592
            attach_lease: true,
Ryan Olson's avatar
Ryan Olson committed
593
594
595
596
597
        }
    }
}

fn default_servers() -> Vec<String> {
598
    match std::env::var(env_etcd::ETCD_ENDPOINTS) {
Ryan Olson's avatar
Ryan Olson committed
599
600
601
602
603
604
605
        Ok(possible_list_of_urls) => possible_list_of_urls
            .split(',')
            .map(|s| s.to_string())
            .collect(),
        Err(_) => vec!["http://localhost:2379".to_string()],
    }
}
606

607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
/// A cache for etcd key-value pairs that watches for changes
pub struct KvCache {
    client: Client,
    pub prefix: String,
    cache: Arc<RwLock<HashMap<String, Vec<u8>>>>,
    watcher: Option<PrefixWatcher>,
}

impl KvCache {
    /// Create a new KV cache for the given prefix
    pub async fn new(
        client: Client,
        prefix: String,
        initial_values: HashMap<String, Vec<u8>>,
    ) -> Result<Self> {
        let mut cache = HashMap::new();

        // First get all existing keys with this prefix
        let existing_kvs = client.kv_get_prefix(&prefix).await?;
        for kv in existing_kvs {
            let key = String::from_utf8_lossy(kv.key()).to_string();
            cache.insert(key, kv.value().to_vec());
        }

        // For any keys in initial_values that don't exist in etcd, write them
        // TODO: proper lease handling, this requires the first process that write to a prefix atomically
        // create a lease and write the lease to etcd. Later processes will attach to the lease and
        // help refresh the lease.
        for (key, value) in initial_values.iter() {
            let full_key = format!("{}{}", prefix, key);
            if let std::collections::hash_map::Entry::Vacant(e) = cache.entry(full_key.clone()) {
                client.kv_put(&full_key, value.clone(), None).await?;
                e.insert(value.clone());
            }
        }

        // Start watching for changes
644
        // we won't miss events between the initial push and the watcher starting because
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
        // client.kv_get_and_watch_prefix() will get all kv pairs and put them back again
        let watcher = client.kv_get_and_watch_prefix(&prefix).await?;

        let cache = Arc::new(RwLock::new(cache));
        let mut result = Self {
            client,
            prefix,
            cache,
            watcher: Some(watcher),
        };

        // Start the background watcher task
        result.start_watcher().await?;

        Ok(result)
    }

    /// Start the background watcher task
    async fn start_watcher(&mut self) -> Result<()> {
        if let Some(watcher) = self.watcher.take() {
            let cache = self.cache.clone();
            let prefix = self.prefix.clone();

            tokio::spawn(async move {
                let mut rx = watcher.rx;

                while let Some(event) = rx.recv().await {
                    match event {
                        WatchEvent::Put(kv) => {
                            let key = String::from_utf8_lossy(kv.key()).to_string();
                            let value = kv.value().to_vec();

677
                            tracing::trace!("KvCache update: {} = {:?}", key, value);
678
679
680
681
682
683
                            let mut cache_write = cache.write().await;
                            cache_write.insert(key, value);
                        }
                        WatchEvent::Delete(kv) => {
                            let key = String::from_utf8_lossy(kv.key()).to_string();

684
                            tracing::trace!("KvCache delete: {}", key);
685
686
687
688
689
690
                            let mut cache_write = cache.write().await;
                            cache_write.remove(&key);
                        }
                    }
                }

691
                tracing::debug!("KvCache watcher for prefix '{}' stopped", prefix);
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
            });
        }

        Ok(())
    }

    /// Get a value from the cache
    pub async fn get(&self, key: &str) -> Option<Vec<u8>> {
        let full_key = format!("{}{}", self.prefix, key);
        let cache_read = self.cache.read().await;
        cache_read.get(&full_key).cloned()
    }

    /// Get all key-value pairs in the cache
    pub async fn get_all(&self) -> HashMap<String, Vec<u8>> {
        let cache_read = self.cache.read().await;
        cache_read.clone()
    }

    /// Update a value in both the cache and etcd
712
    pub async fn put(&self, key: &str, value: Vec<u8>, lease_id: Option<u64>) -> Result<()> {
713
714
715
716
717
718
719
720
721
722
723
724
725
726
        let full_key = format!("{}{}", self.prefix, key);

        // Update etcd first
        self.client
            .kv_put(&full_key, value.clone(), lease_id)
            .await?;

        // Then update local cache
        let mut cache_write = self.cache.write().await;
        cache_write.insert(full_key, value);

        Ok(())
    }

727
728
729
730
731
732
733
734
735
736
737
738
739
    /// Delete a key from both the cache and etcd
    pub async fn delete(&self, key: &str) -> Result<()> {
        let full_key = format!("{}{}", self.prefix, key);

        // Delete from etcd first
        self.client.kv_delete(full_key.clone(), None).await?;

        // Then remove from local cache
        let mut cache_write = self.cache.write().await;
        cache_write.remove(&full_key);

        Ok(())
    }
740
741
}

742
743
744
#[cfg(feature = "integration")]
#[cfg(test)]
mod tests {
745
    use crate::{DistributedRuntime, distributed::DistributedConfig};
746
747
748
749
750
751
752

    use super::*;

    #[test]
    fn test_ectd_client() {
        let rt = Runtime::from_settings().unwrap();
        let rt_clone = rt.clone();
753
        let config = DistributedConfig::from_settings();
754
755
756
757
758
759
760
761
762
763
764

        rt_clone.primary().block_on(async move {
            let drt = DistributedRuntime::new(rt, config).await.unwrap();
            test_kv_create_or_validate(drt).await.unwrap();
        });
    }

    async fn test_kv_create_or_validate(drt: DistributedRuntime) -> Result<()> {
        let key = "__integration_test_key";
        let value = b"test_value";

765
766
767
        let client = Client::new(ClientOptions::default(), drt.runtime().clone())
            .await
            .expect("etcd client should be available");
768
        let lease_id = drt.connection_id();
769
770

        // Create the key
771
        let result = client.kv_create(key, value.to_vec(), Some(lease_id)).await;
772
773
774
        assert!(result.is_ok(), "");

        // Try to create the key again - this should fail
775
        let result = client.kv_create(key, value.to_vec(), Some(lease_id)).await;
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
        assert!(result.is_err());

        // Create or validate should succeed as the values match
        let result = client
            .kv_create_or_validate(key.to_string(), value.to_vec(), Some(lease_id))
            .await;
        assert!(result.is_ok());

        // Try to create the key with a different value
        let different_value = b"different_value";
        let result = client
            .kv_create_or_validate(key.to_string(), different_value.to_vec(), Some(lease_id))
            .await;
        assert!(result.is_err(), "");

        Ok(())
    }
793
794
795
796
797

    #[test]
    fn test_kv_cache() {
        let rt = Runtime::from_settings().unwrap();
        let rt_clone = rt.clone();
798
        let config = DistributedConfig::from_settings();
799
800
801
802
803
804
805
806

        rt_clone.primary().block_on(async move {
            let drt = DistributedRuntime::new(rt, config).await.unwrap();
            test_kv_cache_operations(drt).await.unwrap();
        });
    }

    async fn test_kv_cache_operations(drt: DistributedRuntime) -> Result<()> {
807
808
809
810
        // Make the client and unwrap it
        let client = Client::new(ClientOptions::default(), drt.runtime().clone())
            .await
            .expect("etcd client should be available");
811
812
813

        // Create a unique test prefix to avoid conflicts with other tests
        let test_id = uuid::Uuid::new_v4().to_string();
814
        let prefix = format!("v1/test_kv_cache_{}/", test_id);
815
816
817
818
819
820
821

        // Initial values
        let mut initial_values = HashMap::new();
        initial_values.insert("key1".to_string(), b"value1".to_vec());
        initial_values.insert("key2".to_string(), b"value2".to_vec());

        // Create the KV cache
822
        let kv_cache = KvCache::new(client.clone(), prefix.clone(), initial_values).await?;
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892

        // Test get
        let value1 = kv_cache.get("key1").await;
        assert_eq!(value1, Some(b"value1".to_vec()));

        let value2 = kv_cache.get("key2").await;
        assert_eq!(value2, Some(b"value2".to_vec()));

        // Test get_all
        let all_values = kv_cache.get_all().await;
        assert_eq!(all_values.len(), 2);
        assert_eq!(
            all_values.get(&format!("{}key1", prefix)),
            Some(&b"value1".to_vec())
        );
        assert_eq!(
            all_values.get(&format!("{}key2", prefix)),
            Some(&b"value2".to_vec())
        );

        // Test put - using None for lease_id
        kv_cache.put("key3", b"value3".to_vec(), None).await?;

        // Allow some time for the update to propagate
        tokio::time::sleep(std::time::Duration::from_millis(100)).await;

        // Verify the new value
        let value3 = kv_cache.get("key3").await;
        assert_eq!(value3, Some(b"value3".to_vec()));

        // Test update
        kv_cache
            .put("key1", b"updated_value1".to_vec(), None)
            .await?;

        // Allow some time for the update to propagate
        tokio::time::sleep(std::time::Duration::from_millis(100)).await;

        // Verify the updated value
        let updated_value1 = kv_cache.get("key1").await;
        assert_eq!(updated_value1, Some(b"updated_value1".to_vec()));

        // Test external update (simulating another client updating a value)
        client
            .kv_put(
                &format!("{}key2", prefix),
                b"external_update".to_vec(),
                None,
            )
            .await?;

        // Allow some time for the update to propagate
        tokio::time::sleep(std::time::Duration::from_millis(100)).await;

        // Verify the cache was updated
        let external_update = kv_cache.get("key2").await;
        assert_eq!(external_update, Some(b"external_update".to_vec()));

        // Clean up - delete the test keys
        let etcd_client = client.etcd_client();
        let _ = etcd_client
            .kv_client()
            .delete(
                prefix,
                Some(etcd_client::DeleteOptions::new().with_prefix()),
            )
            .await?;

        Ok(())
    }
893
}