etcd.rs 27.2 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
use crate::{CancellationToken, ErrorContext, Result, Runtime, error};
Ryan Olson's avatar
Ryan Olson committed
5
6
7
8
9

use async_nats::jetstream::kv;
use derive_builder::Builder;
use derive_getters::Dissolve;
use futures::StreamExt;
10
11
use std::collections::HashMap;
use std::sync::Arc;
12
use tokio::sync::{RwLock, mpsc};
Ryan Olson's avatar
Ryan Olson committed
13
14
use validator::Validate;

15
use etcd_client::{
16
17
18
    Certificate, Compare, CompareOp, DeleteOptions, GetOptions, Identity, LockClient, LockOptions,
    LockResponse, PutOptions, PutResponse, TlsOptions, Txn, TxnOp, TxnOpResponse, WatchOptions,
    Watcher,
19
};
Ryan Olson's avatar
Ryan Olson committed
20
pub use etcd_client::{ConnectOptions, KeyValue, LeaseClient};
21
use tokio::time::{Duration, interval};
Ryan Olson's avatar
Ryan Olson committed
22

23
mod connector;
Ryan Olson's avatar
Ryan Olson committed
24
mod lease;
25
mod lock;
26
27
mod path;

28
use connector::Connector;
Ryan Olson's avatar
Ryan Olson committed
29
use lease::*;
30
pub use lock::*;
31
pub use path::*;
Ryan Olson's avatar
Ryan Olson committed
32

33
34
use super::utils::build_in_runtime;

Ryan Olson's avatar
Ryan Olson committed
35
36
37
#[derive(Debug, Clone)]
pub struct Lease {
    /// ETCD lease ID
38
39
    /// Delivered as i64 by etcd because of documented gRPC limitations.
    id: u64,
Ryan Olson's avatar
Ryan Olson committed
40
41
42
43
44
45
46

    /// [`CancellationToken`] associated with the lease
    cancel_token: CancellationToken,
}

impl Lease {
    /// Get the lease ID
47
    pub fn id(&self) -> u64 {
Ryan Olson's avatar
Ryan Olson committed
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
        self.id
    }

    /// Get the primary [`CancellationToken`] associated with the lease.
    /// This token will revoke the lease if canceled.
    pub fn primary_token(&self) -> CancellationToken {
        self.cancel_token.clone()
    }

    /// Get a child [`CancellationToken`] from the lease's [`CancellationToken`].
    /// This child token will be triggered if the lease is revoked, but will not revoke the lease if canceled.
    pub fn child_token(&self) -> CancellationToken {
        self.cancel_token.child_token()
    }

    /// Revoke the lease triggering the [`CancellationToken`].
    pub fn revoke(&self) {
        self.cancel_token.cancel();
    }
67
68
69
70
71
72
73

    /// Check if the lease is still valid (not revoked)
    pub async fn is_valid(&self) -> Result<bool> {
        // A lease is valid if its cancellation token has not been triggered
        // We can use try_cancelled which returns immediately with a boolean
        Ok(!self.cancel_token.is_cancelled())
    }
Ryan Olson's avatar
Ryan Olson committed
74
75
}

76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
/// ETCD Client
#[derive(Clone)]
pub struct Client {
    connector: Arc<Connector>,
    primary_lease: u64,
    runtime: Runtime,
    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
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
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();

107
        let ((connector, lease_id), rt) = build_in_runtime(
108
            async move {
109
110
111
112
113
114
115
116
117
118
119
120
                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(", ")
                        )
                    })?;
121

122
                let lease_id = if config.attach_lease {
123
                    let lease = create_lease(connector.clone(), 10, token)
124
                        .await
125
126
127
128
129
130
                        .with_context(|| {
                            format!(
                                "Unable to create lease. Check etcd server status at {}",
                                config.etcd_url.join(", ")
                            )
                        })?;
131
132
133
134
135
136

                    lease.id
                } else {
                    0
                };

137
                Ok((connector, lease_id))
138
139
140
141
            },
            1,
        )
        .await?;
Ryan Olson's avatar
Ryan Olson committed
142
143

        Ok(Client {
144
            connector,
145
            primary_lease: lease_id,
146
            rt,
Ryan Olson's avatar
Ryan Olson committed
147
148
149
150
            runtime,
        })
    }

151
152
153
154
    /// Get a clone of the underlying [`etcd_client::Client`] instance.
    /// This returns a clone since the client is behind an RwLock.
    pub fn etcd_client(&self) -> etcd_client::Client {
        self.connector.get_client()
Ryan Olson's avatar
Ryan Olson committed
155
156
157
    }

    /// Get the primary lease ID.
158
    pub fn lease_id(&self) -> u64 {
Ryan Olson's avatar
Ryan Olson committed
159
160
161
162
163
164
165
166
167
168
169
170
171
        self.primary_lease
    }

    /// Primary [`Lease`]
    pub fn primary_lease(&self) -> Lease {
        Lease {
            id: self.primary_lease,
            cancel_token: self.runtime.primary_token(),
        }
    }

    /// Create a [`Lease`] with a given time-to-live (TTL).
    /// This [`Lease`] will be tied to the [`Runtime`], specifically a child [`CancellationToken`].
172
    pub async fn create_lease(&self, ttl: u64) -> Result<Lease> {
Ryan Olson's avatar
Ryan Olson committed
173
        let token = self.runtime.child_token();
174
        self.rt
175
            .spawn(create_lease(self.connector.clone(), ttl, token))
Ryan Olson's avatar
Ryan Olson committed
176
177
178
            .await?
    }

179
    // Revoke an etcd lease given its lease id. A wrapper over etcd_client::LeaseClient::revoke
180
    pub async fn revoke_lease(&self, lease_id: u64) -> Result<()> {
181
182
183
        self.rt
            .spawn(revoke_lease(self.connector.clone(), lease_id))
            .await?
184
185
    }

186
    pub async fn kv_create(&self, key: &str, value: Vec<u8>, lease_id: Option<u64>) -> Result<()> {
187
        let id = lease_id.unwrap_or(self.lease_id());
188
        let put_options = PutOptions::new().with_lease(id as i64);
Ryan Olson's avatar
Ryan Olson committed
189
190
191

        // Build the transaction
        let txn = Txn::new()
192
            .when(vec![Compare::version(key, CompareOp::Equal, 0)]) // Ensure the lock does not exist
Ryan Olson's avatar
Ryan Olson committed
193
            .and_then(vec![
194
                TxnOp::put(key, value, Some(put_options)), // Create the object
Ryan Olson's avatar
Ryan Olson committed
195
196
197
            ]);

        // Execute the transaction
198
        let result = self.connector.get_client().kv_client().txn(txn).await?;
199

200
201
202
203
        if result.succeeded() {
            Ok(())
        } else {
            for resp in result.op_responses() {
204
                tracing::warn!(response = ?resp, "kv_create etcd op response");
205
            }
206
            Err(error!("Unable to create key. Check etcd server status"))
207
208
209
210
211
212
213
214
        }
    }

    /// 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>,
215
        lease_id: Option<u64>,
216
    ) -> Result<()> {
217
        let id = lease_id.unwrap_or(self.lease_id());
218
        let put_options = PutOptions::new().with_lease(id as i64);
219
220
221
222
223
224

        // 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![
225
                TxnOp::put(key.as_str(), value.clone(), Some(put_options)), // Create it
226
227
228
229
230
231
232
233
234
235
236
            ])
            .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
237
        let result = self.connector.get_client().kv_client().txn(txn).await?;
238
239
240
241
242
243
244
245
246

        // 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(()),
247
248
249
                        false => Err(error!(
                            "Unable to create or validate key. Check etcd server status"
                        )),
250
                    },
251
252
253
                    _ => Err(error!(
                        "Unable to validate key operation. Check etcd server status"
                    )),
254
                },
255
256
257
                None => Err(error!(
                    "Unable to create or validate key. Check etcd server status"
                )),
258
259
260
261
262
263
264
265
            }
        }
    }

    pub async fn kv_put(
        &self,
        key: impl AsRef<str>,
        value: impl AsRef<[u8]>,
266
        lease_id: Option<u64>,
267
    ) -> Result<()> {
268
        let id = lease_id.unwrap_or(self.lease_id());
269
        let put_options = PutOptions::new().with_lease(id as i64);
270
        let _ = self
271
272
            .connector
            .get_client()
273
            .kv_client()
274
            .put(key.as_ref(), value.as_ref(), Some(put_options))
275
            .await?;
Ryan Olson's avatar
Ryan Olson committed
276
277
278
        Ok(())
    }

279
280
281
282
283
284
285
286
    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()
287
            .with_lease(self.primary_lease().id() as i64);
288
289
        self.connector
            .get_client()
290
291
292
293
294
295
296
297
298
299
300
            .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>> {
301
302
303
304
305
306
        let mut get_response = self
            .connector
            .get_client()
            .kv_client()
            .get(key, options)
            .await?;
307
308
309
310
311
312
313
        Ok(get_response.take_kvs())
    }

    pub async fn kv_delete(
        &self,
        key: impl Into<Vec<u8>>,
        options: Option<DeleteOptions>,
314
    ) -> Result<u64> {
315
316
        self.connector
            .get_client()
317
318
319
            .kv_client()
            .delete(key, options)
            .await
320
            .map(|del_response| del_response.deleted() as u64)
321
322
323
            .map_err(|err| err.into())
    }

Ryan Olson's avatar
Ryan Olson committed
324
325
    pub async fn kv_get_prefix(&self, prefix: impl AsRef<str>) -> Result<Vec<KeyValue>> {
        let mut get_response = self
326
327
            .connector
            .get_client()
Ryan Olson's avatar
Ryan Olson committed
328
329
330
331
332
333
334
            .kv_client()
            .get(prefix.as_ref(), Some(GetOptions::new().with_prefix()))
            .await?;

        Ok(get_response.take_kvs())
    }

335
336
337
338
339
    /// 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>>,
340
        lease_id: Option<u64>,
341
    ) -> Result<LockResponse> {
342
        let mut lock_client = self.connector.get_client().lock_client();
343
        let id = lease_id.unwrap_or(self.lease_id());
344
        let options = LockOptions::new().with_lease(id as i64);
345
346
347
348
349
350
351
352
        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<()> {
353
        let mut lock_client = self.connector.get_client().lock_client();
354
355
356
357
358
359
360
        lock_client
            .unlock(lock_key)
            .await
            .map_err(|err: etcd_client::Error| anyhow::anyhow!(err))?;
        Ok(())
    }

361
362
363
364
365
366
367
368
    /// 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
    }

369
370
371
    pub async fn kv_get_and_watch_prefix(
        &self,
        prefix: impl AsRef<str> + std::fmt::Display,
372
373
374
375
376
377
378
379
    ) -> Result<PrefixWatcher> {
        self.watch_internal(prefix, true).await
    }

    async fn watch_internal(
        &self,
        prefix: impl AsRef<str> + std::fmt::Display,
        include_existing: bool,
380
    ) -> Result<PrefixWatcher> {
381
382
383
        let client = self.connector.get_client();
        let mut kv_client = client.kv_client();
        let mut watch_client = client.watch_client();
Ryan Olson's avatar
Ryan Olson committed
384
385
386
387
388
389
390
391
392
393

        let mut get_response = kv_client
            .get(prefix.as_ref(), Some(GetOptions::new().with_prefix()))
            .await?;

        let start_revision = get_response
            .header()
            .ok_or(error!("missing header; unable to get revision"))?
            .revision();

394
        tracing::trace!("{prefix}: start_revision: {start_revision}");
Ryan Olson's avatar
Ryan Olson committed
395
396
        let start_revision = start_revision + 1;

Ryan Olson's avatar
Ryan Olson committed
397
398
399
400
401
402
        let (watcher, mut watch_stream) = watch_client
            .watch(
                prefix.as_ref(),
                Some(
                    WatchOptions::new()
                        .with_prefix()
403
404
                        .with_start_revision(start_revision)
                        .with_prev_key(),
Ryan Olson's avatar
Ryan Olson committed
405
406
407
408
                ),
            )
            .await?;

409
410
411
412
413
414
415
        let kvs = if include_existing {
            let kvs = get_response.take_kvs();
            tracing::trace!("initial kv count: {:?}", kvs.len());
            kvs
        } else {
            vec![]
        };
Ryan Olson's avatar
Ryan Olson committed
416
417
418

        let (tx, rx) = mpsc::channel(32);

419
        self.rt.spawn(async move {
420
421
422
423
424
425
            if include_existing {
                for kv in kvs {
                    if tx.send(WatchEvent::Put(kv)).await.is_err() {
                        // receiver is already closed
                        return;
                    }
Ryan Olson's avatar
Ryan Olson committed
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
            loop {
                tokio::select! {
                    maybe_resp = watch_stream.next() => {
                        // Early return for None or Err cases
                        let Some(Ok(response)) = maybe_resp else {
                            tracing::info!("kv watch stream closed");
                            return;
                        };

                        // Process events
                        for event in response.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;
                                    }
452
                                }
453
454
455
456
                                etcd_client::EventType::Delete => {
                                    if tx.send(WatchEvent::Delete(kv.clone())).await.is_err() {
                                        return;
                                    }
Ryan Olson's avatar
Ryan Olson committed
457
458
459
460
                                }
                            }
                        }
                    }
461
462
463
464
                    _ = tx.closed() => {
                        tracing::debug!("no more receivers, stopping watcher");
                        return;
                    }
Ryan Olson's avatar
Ryan Olson committed
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
                }
            }
        });
        Ok(PrefixWatcher {
            prefix: prefix.as_ref().to_string(),
            watcher,
            rx,
        })
    }
}

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

483
#[derive(Debug)]
Ryan Olson's avatar
Ryan Olson committed
484
485
486
487
488
489
490
491
492
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
493
    pub etcd_url: Vec<String>,
Ryan Olson's avatar
Ryan Olson committed
494
495

    #[builder(default)]
Ryan Olson's avatar
Ryan Olson committed
496
    pub etcd_connect_options: Option<ConnectOptions>,
497
498
499

    /// If true, the client will attach a lease to the primary [`CancellationToken`].
    #[builder(default = "true")]
Ryan Olson's avatar
Ryan Olson committed
500
    pub attach_lease: bool,
Ryan Olson's avatar
Ryan Olson committed
501
502
503
504
}

impl Default for ClientOptions {
    fn default() -> Self {
wxsm's avatar
wxsm committed
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
        let mut connect_options = None;

        if let (Ok(username), Ok(password)) = (
            std::env::var("ETCD_AUTH_USERNAME"),
            std::env::var("ETCD_AUTH_PASSWORD"),
        ) {
            // username and password are set
            connect_options = Some(ConnectOptions::new().with_user(username, password));
        } else if let (Ok(ca), Ok(cert), Ok(key)) = (
            std::env::var("ETCD_AUTH_CA"),
            std::env::var("ETCD_AUTH_CLIENT_CERT"),
            std::env::var("ETCD_AUTH_CLIENT_KEY"),
        ) {
            // 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
528
529
        ClientOptions {
            etcd_url: default_servers(),
wxsm's avatar
wxsm committed
530
            etcd_connect_options: connect_options,
531
            attach_lease: true,
Ryan Olson's avatar
Ryan Olson committed
532
533
534
535
536
537
538
539
540
541
542
543
544
        }
    }
}

fn default_servers() -> Vec<String> {
    match std::env::var("ETCD_ENDPOINTS") {
        Ok(possible_list_of_urls) => possible_list_of_urls
            .split(',')
            .map(|s| s.to_string())
            .collect(),
        Err(_) => vec!["http://localhost:2379".to_string()],
    }
}
545

546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
/// 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
583
        // we won't miss events between the initial push and the watcher starting because
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
        // 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();

616
                            tracing::trace!("KvCache update: {} = {:?}", key, value);
617
618
619
620
621
622
                            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();

623
                            tracing::trace!("KvCache delete: {}", key);
624
625
626
627
628
629
                            let mut cache_write = cache.write().await;
                            cache_write.remove(&key);
                        }
                    }
                }

630
                tracing::debug!("KvCache watcher for prefix '{}' stopped", prefix);
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
            });
        }

        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
651
    pub async fn put(&self, key: &str, value: Vec<u8>, lease_id: Option<u64>) -> Result<()> {
652
653
654
655
656
657
658
659
660
661
662
663
664
665
        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(())
    }

666
667
668
669
670
671
672
673
674
675
676
677
678
    /// 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(())
    }
679
680
}

681
682
683
#[cfg(feature = "integration")]
#[cfg(test)]
mod tests {
684
    use crate::{DistributedRuntime, distributed::DistributedConfig};
685
686
687
688
689
690
691

    use super::*;

    #[test]
    fn test_ectd_client() {
        let rt = Runtime::from_settings().unwrap();
        let rt_clone = rt.clone();
692
        let config = DistributedConfig::from_settings(false);
693
694
695
696
697
698
699
700
701
702
703

        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";

704
705
706
707
708
        let client = drt.etcd_client().expect("etcd client should be available");
        let lease_id = drt
            .primary_lease()
            .expect("primary lease should be available")
            .id();
709
710

        // Create the key
711
        let result = client.kv_create(key, value.to_vec(), Some(lease_id)).await;
712
713
714
        assert!(result.is_ok(), "");

        // Try to create the key again - this should fail
715
        let result = client.kv_create(key, value.to_vec(), Some(lease_id)).await;
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
        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(())
    }
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751

    #[test]
    fn test_kv_cache() {
        let rt = Runtime::from_settings().unwrap();
        let rt_clone = rt.clone();
        let config = DistributedConfig::from_settings(false);

        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();
752
        let prefix = format!("v1/test_kv_cache_{}/", test_id);
753
754
755
756
757
758
759

        // 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
760
        let kv_cache = KvCache::new(client.clone(), prefix.clone(), initial_values).await?;
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830

        // 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(())
    }
831
}