etcd.rs 30.3 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
mod path;

30
use connector::Connector;
Ryan Olson's avatar
Ryan Olson committed
31
use lease::*;
32
pub use lock::*;
33
pub use path::*;
Ryan Olson's avatar
Ryan Olson committed
34

35
use super::utils::build_in_runtime;
36
use crate::config::environment_names::etcd as env_etcd;
37

38
39
40
41
42
43
/// ETCD Client
#[derive(Clone)]
pub struct Client {
    connector: Arc<Connector>,
    primary_lease: u64,
    runtime: Runtime,
44
45
46
    // 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
47
48
49
50
51
52
53
54
55
    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
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
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();

72
        let ((connector, lease_id), rt) = build_in_runtime(
73
            async move {
74
75
76
77
78
79
80
81
82
83
84
85
                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(", ")
                        )
                    })?;
86

87
                let lease_id = if config.attach_lease {
88
                    create_lease(connector.clone(), 10, token)
89
                        .await
90
91
92
93
94
                        .with_context(|| {
                            format!(
                                "Unable to create lease. Check etcd server status at {}",
                                config.etcd_url.join(", ")
                            )
95
                        })?
96
97
98
99
                } else {
                    0
                };

100
                Ok((connector, lease_id))
101
102
103
104
            },
            1,
        )
        .await?;
Ryan Olson's avatar
Ryan Olson committed
105
106

        Ok(Client {
107
            connector,
108
            primary_lease: lease_id,
109
            rt,
Ryan Olson's avatar
Ryan Olson committed
110
111
112
113
            runtime,
        })
    }

114
115
    /// Get a clone of the underlying [`etcd_client::Client`] instance.
    /// This returns a clone since the client is behind an RwLock.
116
    fn etcd_client(&self) -> etcd_client::Client {
117
        self.connector.get_client()
Ryan Olson's avatar
Ryan Olson committed
118
119
120
    }

    /// Get the primary lease ID.
121
    pub fn lease_id(&self) -> u64 {
Ryan Olson's avatar
Ryan Olson committed
122
123
124
        self.primary_lease
    }

125
126
127
128
129
130
131
    /// 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>> {
132
        let id = lease_id.unwrap_or(self.lease_id());
133
        let put_options = PutOptions::new().with_lease(id as i64);
Ryan Olson's avatar
Ryan Olson committed
134

135
        // Build transaction that creates key only if it doesn't exist
Ryan Olson's avatar
Ryan Olson committed
136
        let txn = Txn::new()
137
            .when(vec![Compare::version(key, CompareOp::Equal, 0)]) // Ensure the lock does not exist
Ryan Olson's avatar
Ryan Olson committed
138
            .and_then(vec![
139
                TxnOp::put(key, value, Some(put_options)), // Create the object
140
141
142
            ])
            .or_else(vec![
                TxnOp::get(key, None), // Key exists, get its info
Ryan Olson's avatar
Ryan Olson committed
143
144
145
            ]);

        // Execute the transaction
146
        let result = self.connector.get_client().kv_client().txn(txn).await?;
147

148
        // Created
149
        if result.succeeded() {
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
            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");
165
        }
166
        anyhow::bail!("Unable to create key. Check etcd server status")
167
168
169
170
171
172
173
    }

    /// 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>,
174
        lease_id: Option<u64>,
175
    ) -> Result<()> {
176
        let id = lease_id.unwrap_or(self.lease_id());
177
        let put_options = PutOptions::new().with_lease(id as i64);
178
179
180
181
182
183

        // 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![
184
                TxnOp::put(key.as_str(), value.clone(), Some(put_options)), // Create it
185
186
187
188
189
190
191
192
193
194
195
            ])
            .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
196
        let result = self.connector.get_client().kv_client().txn(txn).await?;
197
198
199
200
201
202
203
204
205

        // 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(()),
206
                        false => anyhow::bail!(
207
                            "Unable to create or validate key. Check etcd server status"
208
                        ),
209
                    },
210
211
212
                    _ => {
                        anyhow::bail!("Unable to validate key operation. Check etcd server status")
                    }
213
                },
214
                None => anyhow::bail!("Unable to create or validate key. Check etcd server status"),
215
216
217
218
219
220
221
222
            }
        }
    }

    pub async fn kv_put(
        &self,
        key: impl AsRef<str>,
        value: impl AsRef<[u8]>,
223
        lease_id: Option<u64>,
224
    ) -> Result<()> {
225
        let id = lease_id.unwrap_or(self.lease_id());
226
        let put_options = PutOptions::new().with_lease(id as i64);
227
        let _ = self
228
229
            .connector
            .get_client()
230
            .kv_client()
231
            .put(key.as_ref(), value.as_ref(), Some(put_options))
232
            .await?;
Ryan Olson's avatar
Ryan Olson committed
233
234
235
        Ok(())
    }

236
237
238
239
240
241
242
243
    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()
244
            .with_lease(self.lease_id() as i64);
245
246
        self.connector
            .get_client()
247
248
249
250
251
252
253
254
255
256
257
            .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>> {
258
259
260
261
262
263
        let mut get_response = self
            .connector
            .get_client()
            .kv_client()
            .get(key, options)
            .await?;
264
265
266
267
268
269
270
        Ok(get_response.take_kvs())
    }

    pub async fn kv_delete(
        &self,
        key: impl Into<Vec<u8>>,
        options: Option<DeleteOptions>,
271
    ) -> Result<u64> {
272
273
        self.connector
            .get_client()
274
275
276
            .kv_client()
            .delete(key, options)
            .await
277
            .map(|del_response| del_response.deleted() as u64)
278
279
280
            .map_err(|err| err.into())
    }

Ryan Olson's avatar
Ryan Olson committed
281
282
    pub async fn kv_get_prefix(&self, prefix: impl AsRef<str>) -> Result<Vec<KeyValue>> {
        let mut get_response = self
283
284
            .connector
            .get_client()
Ryan Olson's avatar
Ryan Olson committed
285
286
287
288
289
290
291
            .kv_client()
            .get(prefix.as_ref(), Some(GetOptions::new().with_prefix()))
            .await?;

        Ok(get_response.take_kvs())
    }

292
293
294
295
296
    /// 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>>,
297
        lease_id: Option<u64>,
298
    ) -> Result<LockResponse> {
299
        let mut lock_client = self.connector.get_client().lock_client();
300
        let id = lease_id.unwrap_or(self.lease_id());
301
        let options = LockOptions::new().with_lease(id as i64);
302
303
304
305
306
307
308
309
        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<()> {
310
        let mut lock_client = self.connector.get_client().lock_client();
311
312
313
314
315
316
317
        lock_client
            .unlock(lock_key)
            .await
            .map_err(|err: etcd_client::Error| anyhow::anyhow!(err))?;
        Ok(())
    }

318
319
320
321
322
323
324
325
    /// 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
    }

326
327
328
    pub async fn kv_get_and_watch_prefix(
        &self,
        prefix: impl AsRef<str> + std::fmt::Display,
329
330
331
332
    ) -> Result<PrefixWatcher> {
        self.watch_internal(prefix, true).await
    }

333
334
335
336
337
    /// 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.
338
339
340
341
    async fn watch_internal(
        &self,
        prefix: impl AsRef<str> + std::fmt::Display,
        include_existing: bool,
342
    ) -> Result<PrefixWatcher> {
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
370
371
        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
372

373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
        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
389
390
391
392
        let mut get_response = kv_client
            .get(prefix.as_ref(), Some(GetOptions::new().with_prefix()))
            .await?;

393
394
        // Get the start revision
        let mut start_revision = get_response
Ryan Olson's avatar
Ryan Olson committed
395
            .header()
396
            .ok_or(anyhow::anyhow!("missing header; unable to get revision"))?
Ryan Olson's avatar
Ryan Olson committed
397
            .revision();
398
        tracing::trace!("{prefix}: start_revision: {start_revision}");
399
        start_revision += 1;
Ryan Olson's avatar
Ryan Olson committed
400

401
402
        // Send existing KVs from response if requested
        if let Some(tx) = existing_kvs_tx {
403
404
            let kvs = get_response.take_kvs();
            tracing::trace!("initial kv count: {:?}", kvs.len());
405
406
407
408
            for kv in kvs.into_iter() {
                tx.send(WatchEvent::Put(kv)).await?;
            }
        }
Ryan Olson's avatar
Ryan Olson committed
409

410
411
        Ok(start_revision)
    }
Ryan Olson's avatar
Ryan Olson committed
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
    /// 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);
451
                    }
452
                    // continue - retry establishing the watch stream
Ryan Olson's avatar
Ryan Olson committed
453
454
                }
            }
455
456
        }
    }
Ryan Olson's avatar
Ryan Olson committed
457

458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
    /// 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
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
525
526
                    };

                    // 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
527
                    }
528
529
530
531
                }
                etcd_client::EventType::Delete => {
                    if tx.send(WatchEvent::Delete(kv.clone())).await.is_err() {
                        return Err(anyhow::anyhow!("failed to send WatchEvent::Delete"));
532
                    }
Ryan Olson's avatar
Ryan Olson committed
533
534
                }
            }
535
536
        }
        Ok(())
Ryan Olson's avatar
Ryan Olson committed
537
538
539
540
541
542
543
544
545
    }
}

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

546
#[derive(Debug)]
Ryan Olson's avatar
Ryan Olson committed
547
548
549
550
551
552
553
554
555
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
556
    pub etcd_url: Vec<String>,
Ryan Olson's avatar
Ryan Olson committed
557
558

    #[builder(default)]
Ryan Olson's avatar
Ryan Olson committed
559
    pub etcd_connect_options: Option<ConnectOptions>,
560
561
562

    /// If true, the client will attach a lease to the primary [`CancellationToken`].
    #[builder(default = "true")]
Ryan Olson's avatar
Ryan Olson committed
563
    pub attach_lease: bool,
Ryan Olson's avatar
Ryan Olson committed
564
565
566
567
}

impl Default for ClientOptions {
    fn default() -> Self {
wxsm's avatar
wxsm committed
568
569
570
        let mut connect_options = None;

        if let (Ok(username), Ok(password)) = (
571
572
            std::env::var(env_etcd::auth::ETCD_AUTH_USERNAME),
            std::env::var(env_etcd::auth::ETCD_AUTH_PASSWORD),
wxsm's avatar
wxsm committed
573
574
575
576
        ) {
            // username and password are set
            connect_options = Some(ConnectOptions::new().with_user(username, password));
        } else if let (Ok(ca), Ok(cert), Ok(key)) = (
577
578
579
            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
580
581
582
583
584
585
586
587
588
589
590
        ) {
            // 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
591
592
        ClientOptions {
            etcd_url: default_servers(),
wxsm's avatar
wxsm committed
593
            etcd_connect_options: connect_options,
594
            attach_lease: true,
Ryan Olson's avatar
Ryan Olson committed
595
596
597
598
599
        }
    }
}

fn default_servers() -> Vec<String> {
600
    match std::env::var(env_etcd::ETCD_ENDPOINTS) {
Ryan Olson's avatar
Ryan Olson committed
601
602
603
604
605
606
607
        Ok(possible_list_of_urls) => possible_list_of_urls
            .split(',')
            .map(|s| s.to_string())
            .collect(),
        Err(_) => vec!["http://localhost:2379".to_string()],
    }
}
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
644
645
/// 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
646
        // we won't miss events between the initial push and the watcher starting because
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
        // 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();

679
                            tracing::trace!("KvCache update: {} = {:?}", key, value);
680
681
682
683
684
685
                            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();

686
                            tracing::trace!("KvCache delete: {}", key);
687
688
689
690
691
692
                            let mut cache_write = cache.write().await;
                            cache_write.remove(&key);
                        }
                    }
                }

693
                tracing::debug!("KvCache watcher for prefix '{}' stopped", prefix);
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
            });
        }

        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
714
    pub async fn put(&self, key: &str, value: Vec<u8>, lease_id: Option<u64>) -> Result<()> {
715
716
717
718
719
720
721
722
723
724
725
726
727
728
        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(())
    }

729
730
731
732
733
734
735
736
737
738
739
740
741
    /// 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(())
    }
742
743
}

744
745
746
#[cfg(feature = "integration")]
#[cfg(test)]
mod tests {
747
    use crate::{DistributedRuntime, distributed::DistributedConfig};
748
749
750
751
752
753
754

    use super::*;

    #[test]
    fn test_ectd_client() {
        let rt = Runtime::from_settings().unwrap();
        let rt_clone = rt.clone();
755
        let config = DistributedConfig::from_settings();
756
757
758
759
760
761
762
763
764
765
766

        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";

767
        let client = drt.etcd_client().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
807
808
809
810
811

        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<()> {
        // Get the client and unwrap it
        let client = drt.etcd_client().expect("etcd client should be available");

        // Create a unique test prefix to avoid conflicts with other tests
        let test_id = uuid::Uuid::new_v4().to_string();
812
        let prefix = format!("v1/test_kv_cache_{}/", test_id);
813
814
815
816
817
818
819

        // 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
820
        let kv_cache = KvCache::new(client.clone(), prefix.clone(), initial_values).await?;
821
822
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

        // 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(())
    }
891
}