forked from scylladb/scylla-rust-driver
-
Notifications
You must be signed in to change notification settings - Fork 0
/
cluster.rs
605 lines (513 loc) · 21.7 KB
/
cluster.rs
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
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
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
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
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
/// Cluster manages up to date information and connections to database nodes
use crate::frame::response::event::{Event, StatusChangeEvent};
use crate::frame::value::ValueList;
use crate::load_balancing::TokenAwarePolicy;
use crate::routing::Token;
use crate::transport::host_filter::HostFilter;
use crate::transport::{
connection::{Connection, VerifiedKeyspaceName},
connection_pool::PoolConfig,
errors::QueryError,
node::Node,
partitioner::PartitionerName,
session::AddressTranslator,
topology::{Keyspace, Metadata, MetadataReader},
};
use arc_swap::ArcSwap;
use bytes::{BufMut, Bytes, BytesMut};
use futures::future::join_all;
use futures::{future::RemoteHandle, FutureExt};
use itertools::Itertools;
use scylla_cql::errors::BadQuery;
use std::collections::{BTreeMap, HashMap};
use std::net::SocketAddr;
use std::sync::Arc;
use std::time::Duration;
use tracing::{debug, warn};
/// Cluster manages up to date information and connections to database nodes.
/// All data can be accessed by cloning Arc<ClusterData> in the `data` field
pub struct Cluster {
// `ArcSwap<ClusterData>` is wrapped in `Arc` to support sharing cluster data
// between `Cluster` and `ClusterWorker`
data: Arc<ArcSwap<ClusterData>>,
refresh_channel: tokio::sync::mpsc::Sender<RefreshRequest>,
use_keyspace_channel: tokio::sync::mpsc::Sender<UseKeyspaceRequest>,
_worker_handle: RemoteHandle<()>,
}
/// Enables printing [Cluster] struct in a neat way, by skipping the rather useless
/// print of channels state and printing [ClusterData] neatly.
pub struct ClusterNeatDebug<'a>(pub &'a Cluster);
impl<'a> std::fmt::Debug for ClusterNeatDebug<'a> {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
let cluster = self.0;
f.debug_struct("Cluster")
.field("data", &ClusterDataNeatDebug(&cluster.data.load()))
.finish_non_exhaustive()
}
}
#[derive(Clone, Debug)]
pub struct Datacenter {
pub nodes: Vec<Arc<Node>>,
pub rack_count: usize,
}
#[derive(Clone)]
pub struct ClusterData {
pub(crate) known_peers: HashMap<SocketAddr, Arc<Node>>, // Invariant: nonempty after Cluster::new()
pub(crate) ring: BTreeMap<Token, Arc<Node>>, // Invariant: nonempty after Cluster::new()
pub(crate) keyspaces: HashMap<String, Keyspace>,
pub(crate) all_nodes: Vec<Arc<Node>>,
pub(crate) datacenters: HashMap<String, Datacenter>,
}
/// Enables printing [ClusterData] struct in a neat way, skipping the clutter involved by
/// [ClusterData::ring] being large and [Self::keyspaces] debug print being very verbose by default.
pub struct ClusterDataNeatDebug<'a>(pub &'a Arc<ClusterData>);
impl<'a> std::fmt::Debug for ClusterDataNeatDebug<'a> {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
let cluster_data = &self.0;
f.debug_struct("ClusterData")
.field("known_peers", &cluster_data.known_peers)
.field("ring", {
struct RingSizePrinter(usize);
impl std::fmt::Debug for RingSizePrinter {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
write!(f, "<size={}>", self.0)
}
}
&RingSizePrinter(cluster_data.ring.len())
})
.field("keyspaces", &cluster_data.keyspaces.keys())
.field("all_nodes", &cluster_data.all_nodes)
.field("datacenters", &cluster_data.datacenters)
.finish_non_exhaustive()
}
}
// Works in the background to keep the cluster updated
struct ClusterWorker {
// Cluster data to keep updated:
cluster_data: Arc<ArcSwap<ClusterData>>,
// Cluster connections
metadata_reader: MetadataReader,
pool_config: PoolConfig,
// To listen for refresh requests
refresh_channel: tokio::sync::mpsc::Receiver<RefreshRequest>,
// Channel used to receive use keyspace requests
use_keyspace_channel: tokio::sync::mpsc::Receiver<UseKeyspaceRequest>,
// Channel used to receive server events
server_events_channel: tokio::sync::mpsc::Receiver<Event>,
// Keyspace send in "USE <keyspace name>" when opening each connection
used_keyspace: Option<VerifiedKeyspaceName>,
// The host filter determines towards which nodes we should open
// connections
host_filter: Option<Arc<dyn HostFilter>>,
}
#[derive(Debug)]
struct RefreshRequest {
response_chan: tokio::sync::oneshot::Sender<Result<(), QueryError>>,
}
#[derive(Debug)]
struct UseKeyspaceRequest {
keyspace_name: VerifiedKeyspaceName,
response_chan: tokio::sync::oneshot::Sender<Result<(), QueryError>>,
}
impl Cluster {
pub async fn new(
initial_peers: &[SocketAddr],
pool_config: PoolConfig,
fetch_schema_metadata: bool,
address_translator: &Option<Arc<dyn AddressTranslator>>,
host_filter: &Option<Arc<dyn HostFilter>>,
) -> Result<Cluster, QueryError> {
let (refresh_sender, refresh_receiver) = tokio::sync::mpsc::channel(32);
let (use_keyspace_sender, use_keyspace_receiver) = tokio::sync::mpsc::channel(32);
let (server_events_sender, server_events_receiver) = tokio::sync::mpsc::channel(32);
let mut metadata_reader = MetadataReader::new(
initial_peers,
pool_config.connection_config.clone(),
pool_config.keepalive_interval,
server_events_sender,
fetch_schema_metadata,
address_translator,
host_filter,
);
let metadata = metadata_reader.read_metadata(true).await?;
let cluster_data = ClusterData::new(
metadata,
&pool_config,
&HashMap::new(),
&None,
host_filter.as_deref(),
);
cluster_data.wait_until_all_pools_are_initialized().await;
let cluster_data: Arc<ArcSwap<ClusterData>> =
Arc::new(ArcSwap::from(Arc::new(cluster_data)));
let worker = ClusterWorker {
cluster_data: cluster_data.clone(),
metadata_reader,
pool_config,
refresh_channel: refresh_receiver,
server_events_channel: server_events_receiver,
use_keyspace_channel: use_keyspace_receiver,
used_keyspace: None,
host_filter: host_filter.clone(),
};
let (fut, worker_handle) = worker.work().remote_handle();
tokio::spawn(fut);
let result = Cluster {
data: cluster_data,
refresh_channel: refresh_sender,
use_keyspace_channel: use_keyspace_sender,
_worker_handle: worker_handle,
};
Ok(result)
}
pub fn get_data(&self) -> Arc<ClusterData> {
self.data.load_full()
}
pub async fn refresh_metadata(&self) -> Result<(), QueryError> {
let (response_sender, response_receiver) = tokio::sync::oneshot::channel();
self.refresh_channel
.send(RefreshRequest {
response_chan: response_sender,
})
.await
.expect("Bug in Cluster::refresh_metadata sending");
// Other end of this channel is in ClusterWorker, can't be dropped while we have &self to Cluster with _worker_handle
response_receiver
.await
.expect("Bug in Cluster::refresh_metadata receiving")
// ClusterWorker always responds
}
pub async fn use_keyspace(
&self,
keyspace_name: VerifiedKeyspaceName,
) -> Result<(), QueryError> {
let (response_sender, response_receiver) = tokio::sync::oneshot::channel();
self.use_keyspace_channel
.send(UseKeyspaceRequest {
keyspace_name,
response_chan: response_sender,
})
.await
.expect("Bug in Cluster::use_keyspace sending");
// Other end of this channel is in ClusterWorker, can't be dropped while we have &self to Cluster with _worker_handle
response_receiver.await.unwrap() // ClusterWorker always responds
}
/// Returns nonempty list of working connections to all shards
pub async fn get_working_connections(&self) -> Result<Vec<Arc<Connection>>, QueryError> {
let cluster_data: Arc<ClusterData> = self.get_data();
let peers = &cluster_data.known_peers;
let mut result: Vec<Arc<Connection>> = Vec::with_capacity(peers.len());
let mut last_error: Option<QueryError> = None;
for node in peers.values() {
match node.get_working_connections() {
Ok(conns) => result.extend(conns),
Err(e) => last_error = Some(e),
}
}
if result.is_empty() {
return Err(last_error.unwrap()); // By invariant peers is nonempty
}
Ok(result)
}
}
impl ClusterData {
/// Returns an iterator to the sequence of ends of vnodes, starting at the vnode in which t
/// lies and going clockwise. Returned sequence has the same length as ring.
pub(crate) fn ring_range<'a>(&'a self, t: &Token) -> impl Iterator<Item = Arc<Node>> + 'a {
let before_wrap = self.ring.range(t..).map(|(_token, node)| node.clone());
let after_wrap = self.ring.values().cloned();
before_wrap.chain(after_wrap).take(self.ring.len())
}
// Updates information about rack count in each datacenter
fn update_rack_count(datacenters: &mut HashMap<String, Datacenter>) {
for datacenter in datacenters.values_mut() {
datacenter.rack_count = datacenter
.nodes
.iter()
.filter_map(|node| node.rack.clone())
.unique()
.count();
}
}
pub(crate) async fn wait_until_all_pools_are_initialized(&self) {
for node in self.all_nodes.iter() {
node.wait_until_pool_initialized().await;
}
}
/// Creates new ClusterData using information about topology held in `metadata`.
/// Uses provided `known_peers` hashmap to recycle nodes if possible.
pub(crate) fn new(
metadata: Metadata,
pool_config: &PoolConfig,
known_peers: &HashMap<SocketAddr, Arc<Node>>,
used_keyspace: &Option<VerifiedKeyspaceName>,
host_filter: Option<&dyn HostFilter>,
) -> Self {
// Create new updated known_peers and ring
let mut new_known_peers: HashMap<SocketAddr, Arc<Node>> =
HashMap::with_capacity(metadata.peers.len());
let mut ring: BTreeMap<Token, Arc<Node>> = BTreeMap::new();
let mut datacenters: HashMap<String, Datacenter> = HashMap::new();
let mut all_nodes: Vec<Arc<Node>> = Vec::with_capacity(metadata.peers.len());
for peer in metadata.peers {
// Take existing Arc<Node> if possible, otherwise create new one
// Changing rack/datacenter but not ip address seems improbable
// so we can just create new node and connections then
let node: Arc<Node> = match known_peers.get(&peer.address) {
Some(node) if node.datacenter == peer.datacenter && node.rack == peer.rack => {
node.clone()
}
_ => {
let is_enabled = host_filter.map_or(true, |f| f.accept(&peer));
Arc::new(Node::new(
peer.address,
pool_config.clone(),
peer.datacenter,
peer.rack,
used_keyspace.clone(),
is_enabled,
))
}
};
new_known_peers.insert(peer.address, node.clone());
if let Some(dc) = &node.datacenter {
match datacenters.get_mut(dc) {
Some(v) => v.nodes.push(node.clone()),
None => {
let v = Datacenter {
nodes: vec![node.clone()],
rack_count: 0,
};
datacenters.insert(dc.clone(), v);
}
}
}
for token in peer.tokens {
ring.insert(token, node.clone());
}
all_nodes.push(node);
}
Self::update_rack_count(&mut datacenters);
ClusterData {
known_peers: new_known_peers,
ring,
keyspaces: metadata.keyspaces,
all_nodes,
datacenters,
}
}
/// Access keyspaces details collected by the driver
/// Driver collects various schema details like tables, partitioners, columns, types.
/// They can be read using this method
pub fn get_keyspace_info(&self) -> &HashMap<String, Keyspace> {
&self.keyspaces
}
/// Access datacenter details collected by the driver
/// Returned `HashMap` is indexed by names of datacenters
pub fn get_datacenters_info(&self) -> &HashMap<String, Datacenter> {
&self.datacenters
}
/// Access ring details collected by the driver
pub fn get_ring_info(&self) -> &BTreeMap<Token, Arc<Node>> {
&self.ring
}
/// Access details about nodes known to the driver
pub fn get_nodes_info(&self) -> &Vec<Arc<Node>> {
&self.all_nodes
}
/// Compute token of a table partition key
pub fn compute_token(
&self,
keyspace: &str,
table: &str,
partition_key: impl ValueList,
) -> Result<Token, BadQuery> {
let partitioner = self
.keyspaces
.get(keyspace)
.and_then(|k| k.tables.get(table))
.and_then(|t| t.partitioner.as_deref())
.and_then(PartitionerName::from_str)
.unwrap_or_default();
let serialized_values = partition_key.serialized()?;
// Null values are skipped in computation; null values in partition key are unsound,
// but it is consistent with computation of prepared statements token.
let serialized_pk = match serialized_values.len() {
0 => None,
1 => serialized_values
.iter()
.next()
.unwrap()
.map(Bytes::copy_from_slice),
_ => {
let mut buf = BytesMut::new();
for value in serialized_values.iter().flatten() {
let value_size = value
.len()
.try_into()
.map_err(|_| BadQuery::ValuesTooLongForKey(value.len(), u16::MAX.into()))?;
buf.put_u16(value_size);
buf.extend_from_slice(value);
buf.put_u8(0);
}
Some(buf.into())
}
};
Ok(partitioner.hash(serialized_pk.unwrap_or_default()))
}
/// Access to replicas owning a given token
pub fn get_token_endpoints(&self, keyspace: &str, token: Token) -> Vec<Arc<Node>> {
TokenAwarePolicy::replicas_for_token(self, &token, Some(keyspace))
}
/// Access to replicas owning a given partition key (similar to `nodetool getendpoints`)
pub fn get_endpoints(
&self,
keyspace: &str,
table: &str,
partition_key: impl ValueList,
) -> Result<Vec<Arc<Node>>, BadQuery> {
Ok(self.get_token_endpoints(
keyspace,
self.compute_token(keyspace, table, partition_key)?,
))
}
}
impl ClusterWorker {
pub async fn work(mut self) {
use tokio::time::Instant;
let refresh_duration = Duration::from_secs(60); // Refresh topology every 60 seconds
let mut last_refresh_time = Instant::now();
loop {
let mut cur_request: Option<RefreshRequest> = None;
// Wait until it's time for the next refresh
let sleep_until: Instant = last_refresh_time
.checked_add(refresh_duration)
.unwrap_or_else(Instant::now);
let sleep_future = tokio::time::sleep_until(sleep_until);
tokio::pin!(sleep_future);
tokio::select! {
_ = sleep_future => {},
recv_res = self.refresh_channel.recv() => {
match recv_res {
Some(request) => cur_request = Some(request),
None => return, // If refresh_channel was closed then cluster was dropped, we can stop working
}
}
recv_res = self.server_events_channel.recv() => {
if let Some(event) = recv_res {
debug!("Received server event: {:?}", event);
match event {
Event::TopologyChange(_) => (), // Refresh immediately
Event::StatusChange(status) => {
// If some node went down/up, update it's marker and refresh
// later as planned.
match status {
StatusChangeEvent::Down(addr) => self.change_node_down_marker(addr, true),
StatusChangeEvent::Up(addr) => self.change_node_down_marker(addr, false),
}
continue;
},
_ => continue, // Don't go to refreshing
}
} else {
// If server_events_channel was closed, than TopologyReader was dropped,
// so we can probably stop working too
return;
}
}
recv_res = self.use_keyspace_channel.recv() => {
match recv_res {
Some(request) => {
self.used_keyspace = Some(request.keyspace_name.clone());
let cluster_data = self.cluster_data.load_full();
let use_keyspace_future = Self::handle_use_keyspace_request(cluster_data, request);
tokio::spawn(use_keyspace_future);
},
None => return, // If use_keyspace_channel was closed then cluster was dropped, we can stop working
}
continue; // Don't go to refreshing, wait for the next event
}
}
// Perform the refresh
debug!("Requesting topology refresh");
last_refresh_time = Instant::now();
let refresh_res = self.perform_refresh().await;
// Send refresh result if there was a request
if let Some(request) = cur_request {
// We can ignore sending error - if no one waits for the response we can drop it
let _ = request.response_chan.send(refresh_res);
}
}
}
fn change_node_down_marker(&mut self, addr: SocketAddr, is_down: bool) {
let cluster_data = self.cluster_data.load_full();
let node = match cluster_data.known_peers.get(&addr) {
Some(node) => node,
None => {
warn!("Unknown node address {}", addr);
return;
}
};
node.change_down_marker(is_down);
}
async fn handle_use_keyspace_request(
cluster_data: Arc<ClusterData>,
request: UseKeyspaceRequest,
) {
let result = Self::send_use_keyspace(cluster_data, &request.keyspace_name).await;
// Don't care if nobody wants request result
let _ = request.response_chan.send(result);
}
async fn send_use_keyspace(
cluster_data: Arc<ClusterData>,
keyspace_name: &VerifiedKeyspaceName,
) -> Result<(), QueryError> {
let mut use_keyspace_futures = Vec::new();
for node in cluster_data.known_peers.values() {
let fut = node.use_keyspace(keyspace_name.clone());
use_keyspace_futures.push(fut);
}
let use_keyspace_results: Vec<Result<(), QueryError>> =
join_all(use_keyspace_futures).await;
// If there was at least one Ok and the rest were IoErrors we can return Ok
// keyspace name is correct and will be used on broken connection on the next reconnect
// If there were only IoErrors then return IoError
// If there was an error different than IoError return this error - something is wrong
let mut was_ok: bool = false;
let mut io_error: Option<Arc<std::io::Error>> = None;
for result in use_keyspace_results {
match result {
Ok(()) => was_ok = true,
Err(err) => match err {
QueryError::IoError(io_err) => io_error = Some(io_err),
_ => return Err(err),
},
}
}
if was_ok {
return Ok(());
}
// We can unwrap io_error because use_keyspace_futures must be nonempty
Err(QueryError::IoError(io_error.unwrap()))
}
async fn perform_refresh(&mut self) -> Result<(), QueryError> {
// Read latest Metadata
let metadata = self.metadata_reader.read_metadata(false).await?;
let cluster_data: Arc<ClusterData> = self.cluster_data.load_full();
let new_cluster_data = Arc::new(ClusterData::new(
metadata,
&self.pool_config,
&cluster_data.known_peers,
&self.used_keyspace,
self.host_filter.as_deref(),
));
new_cluster_data
.wait_until_all_pools_are_initialized()
.await;
self.update_cluster_data(new_cluster_data);
Ok(())
}
fn update_cluster_data(&mut self, new_cluster_data: Arc<ClusterData>) {
self.cluster_data.store(new_cluster_data);
}
}