Line data Source code
1 : //! Safekeeper communication endpoint to WAL proposer (compute node).
2 : //! Gets messages from the network, passes them down to consensus module and
3 : //! sends replies back.
4 :
5 : use crate::handler::SafekeeperPostgresHandler;
6 : use crate::safekeeper::AcceptorProposerMessage;
7 : use crate::safekeeper::ProposerAcceptorMessage;
8 : use crate::safekeeper::ServerInfo;
9 : use crate::timeline::Timeline;
10 : use crate::wal_service::ConnectionId;
11 : use crate::GlobalTimelines;
12 : use anyhow::{anyhow, Context};
13 : use bytes::BytesMut;
14 : use parking_lot::MappedMutexGuard;
15 : use parking_lot::Mutex;
16 : use parking_lot::MutexGuard;
17 : use postgres_backend::CopyStreamHandlerEnd;
18 : use postgres_backend::PostgresBackend;
19 : use postgres_backend::PostgresBackendReader;
20 : use postgres_backend::QueryError;
21 : use pq_proto::BeMessage;
22 : use serde::Deserialize;
23 : use serde::Serialize;
24 : use std::net::SocketAddr;
25 : use std::sync::Arc;
26 : use tokio::io::AsyncRead;
27 : use tokio::io::AsyncWrite;
28 : use tokio::sync::mpsc::channel;
29 : use tokio::sync::mpsc::error::TryRecvError;
30 : use tokio::sync::mpsc::Receiver;
31 : use tokio::sync::mpsc::Sender;
32 : use tokio::task;
33 : use tokio::task::JoinHandle;
34 : use tokio::time::Duration;
35 : use tokio::time::Instant;
36 : use tracing::*;
37 : use utils::id::TenantTimelineId;
38 : use utils::lsn::Lsn;
39 : use utils::pageserver_feedback::PageserverFeedback;
40 :
41 : const DEFAULT_FEEDBACK_CAPACITY: usize = 8;
42 :
43 : /// Registry of WalReceivers (compute connections). Timeline holds it (wrapped
44 : /// in Arc).
45 : pub struct WalReceivers {
46 : mutex: Mutex<WalReceiversShared>,
47 : pageserver_feedback_tx: tokio::sync::broadcast::Sender<PageserverFeedback>,
48 : }
49 :
50 : /// Id under which walreceiver is registered in shmem.
51 : type WalReceiverId = usize;
52 :
53 : impl WalReceivers {
54 0 : pub fn new() -> Arc<WalReceivers> {
55 0 : let (pageserver_feedback_tx, _) =
56 0 : tokio::sync::broadcast::channel(DEFAULT_FEEDBACK_CAPACITY);
57 0 :
58 0 : Arc::new(WalReceivers {
59 0 : mutex: Mutex::new(WalReceiversShared { slots: Vec::new() }),
60 0 : pageserver_feedback_tx,
61 0 : })
62 0 : }
63 :
64 : /// Register new walreceiver. Returned guard provides access to the slot and
65 : /// automatically deregisters in Drop.
66 0 : pub fn register(self: &Arc<WalReceivers>, conn_id: Option<ConnectionId>) -> WalReceiverGuard {
67 0 : let slots = &mut self.mutex.lock().slots;
68 0 : let walreceiver = WalReceiverState {
69 0 : conn_id,
70 0 : status: WalReceiverStatus::Voting,
71 0 : };
72 : // find empty slot or create new one
73 0 : let pos = if let Some(pos) = slots.iter().position(|s| s.is_none()) {
74 0 : slots[pos] = Some(walreceiver);
75 0 : pos
76 : } else {
77 0 : let pos = slots.len();
78 0 : slots.push(Some(walreceiver));
79 0 : pos
80 : };
81 0 : WalReceiverGuard {
82 0 : id: pos,
83 0 : walreceivers: self.clone(),
84 0 : }
85 0 : }
86 :
87 : /// Get reference to locked slot contents. Slot must exist (registered
88 : /// earlier).
89 0 : fn get_slot<'a>(
90 0 : self: &'a Arc<WalReceivers>,
91 0 : id: WalReceiverId,
92 0 : ) -> MappedMutexGuard<'a, WalReceiverState> {
93 0 : MutexGuard::map(self.mutex.lock(), |locked| {
94 0 : locked.slots[id]
95 0 : .as_mut()
96 0 : .expect("walreceiver doesn't exist")
97 0 : })
98 0 : }
99 :
100 : /// Get number of walreceivers (compute connections).
101 0 : pub fn get_num(self: &Arc<WalReceivers>) -> usize {
102 0 : self.mutex.lock().slots.iter().flatten().count()
103 0 : }
104 :
105 : /// Get state of all walreceivers.
106 0 : pub fn get_all(self: &Arc<WalReceivers>) -> Vec<WalReceiverState> {
107 0 : self.mutex.lock().slots.iter().flatten().cloned().collect()
108 0 : }
109 :
110 : /// Get number of streaming walreceivers (normally 0 or 1) from compute.
111 0 : pub fn get_num_streaming(self: &Arc<WalReceivers>) -> usize {
112 0 : self.mutex
113 0 : .lock()
114 0 : .slots
115 0 : .iter()
116 0 : .flatten()
117 0 : // conn_id.is_none skips recovery which also registers here
118 0 : .filter(|s| s.conn_id.is_some() && matches!(s.status, WalReceiverStatus::Streaming))
119 0 : .count()
120 0 : }
121 :
122 : /// Unregister walreceiver.
123 0 : fn unregister(self: &Arc<WalReceivers>, id: WalReceiverId) {
124 0 : let mut shared = self.mutex.lock();
125 0 : shared.slots[id] = None;
126 0 : }
127 :
128 : /// Broadcast pageserver feedback to connected walproposers.
129 0 : pub fn broadcast_pageserver_feedback(&self, feedback: PageserverFeedback) {
130 0 : // Err means there is no subscribers, it is fine.
131 0 : let _ = self.pageserver_feedback_tx.send(feedback);
132 0 : }
133 : }
134 :
135 : /// Only a few connections are expected (normally one), so store in Vec.
136 : struct WalReceiversShared {
137 : slots: Vec<Option<WalReceiverState>>,
138 : }
139 :
140 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
141 : pub struct WalReceiverState {
142 : /// None means it is recovery initiated by us (this safekeeper).
143 : pub conn_id: Option<ConnectionId>,
144 : pub status: WalReceiverStatus,
145 : }
146 :
147 : /// Walreceiver status. Currently only whether it passed voting stage and
148 : /// started receiving the stream, but it is easy to add more if needed.
149 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
150 : pub enum WalReceiverStatus {
151 : Voting,
152 : Streaming,
153 : }
154 :
155 : /// Scope guard to access slot in WalReceivers registry and unregister from
156 : /// it in Drop.
157 : pub struct WalReceiverGuard {
158 : id: WalReceiverId,
159 : walreceivers: Arc<WalReceivers>,
160 : }
161 :
162 : impl WalReceiverGuard {
163 : /// Get reference to locked shared state contents.
164 0 : fn get(&self) -> MappedMutexGuard<WalReceiverState> {
165 0 : self.walreceivers.get_slot(self.id)
166 0 : }
167 : }
168 :
169 : impl Drop for WalReceiverGuard {
170 0 : fn drop(&mut self) {
171 0 : self.walreceivers.unregister(self.id);
172 0 : }
173 : }
174 :
175 : pub const MSG_QUEUE_SIZE: usize = 256;
176 : pub const REPLY_QUEUE_SIZE: usize = 16;
177 :
178 : impl SafekeeperPostgresHandler {
179 : /// Wrapper around handle_start_wal_push_guts handling result. Error is
180 : /// handled here while we're still in walreceiver ttid span; with API
181 : /// extension, this can probably be moved into postgres_backend.
182 0 : pub async fn handle_start_wal_push<IO: AsyncRead + AsyncWrite + Unpin>(
183 0 : &mut self,
184 0 : pgb: &mut PostgresBackend<IO>,
185 0 : ) -> Result<(), QueryError> {
186 0 : let mut tli: Option<Arc<Timeline>> = None;
187 0 : if let Err(end) = self.handle_start_wal_push_guts(pgb, &mut tli).await {
188 : // Log the result and probably send it to the client, closing the stream.
189 0 : let handle_end_fut = pgb.handle_copy_stream_end(end);
190 : // If we managed to create the timeline, augment logging with current LSNs etc.
191 0 : if let Some(tli) = tli {
192 0 : let info = tli.get_safekeeper_info(&self.conf).await;
193 0 : handle_end_fut
194 0 : .instrument(info_span!("", term=%info.term, last_log_term=%info.last_log_term, flush_lsn=%Lsn(info.flush_lsn), commit_lsn=%Lsn(info.commit_lsn)))
195 0 : .await;
196 : } else {
197 0 : handle_end_fut.await;
198 : }
199 0 : }
200 0 : Ok(())
201 0 : }
202 :
203 0 : pub async fn handle_start_wal_push_guts<IO: AsyncRead + AsyncWrite + Unpin>(
204 0 : &mut self,
205 0 : pgb: &mut PostgresBackend<IO>,
206 0 : tli: &mut Option<Arc<Timeline>>,
207 0 : ) -> Result<(), CopyStreamHandlerEnd> {
208 0 : // Notify the libpq client that it's allowed to send `CopyData` messages
209 0 : pgb.write_message(&BeMessage::CopyBothResponse).await?;
210 :
211 : // Experiments [1] confirm that doing network IO in one (this) thread and
212 : // processing with disc IO in another significantly improves
213 : // performance; we spawn off WalAcceptor thread for message processing
214 : // to this end.
215 : //
216 : // [1] https://github.com/neondatabase/neon/pull/1318
217 0 : let (msg_tx, msg_rx) = channel(MSG_QUEUE_SIZE);
218 0 : let (reply_tx, reply_rx) = channel(REPLY_QUEUE_SIZE);
219 0 : let mut acceptor_handle: Option<JoinHandle<anyhow::Result<()>>> = None;
220 :
221 : // Concurrently receive and send data; replies are not synchronized with
222 : // sends, so this avoids deadlocks.
223 0 : let mut pgb_reader = pgb.split().context("START_WAL_PUSH split")?;
224 0 : let peer_addr = *pgb.get_peer_addr();
225 0 : let mut network_reader = NetworkReader {
226 0 : ttid: self.ttid,
227 0 : conn_id: self.conn_id,
228 0 : pgb_reader: &mut pgb_reader,
229 0 : peer_addr,
230 0 : acceptor_handle: &mut acceptor_handle,
231 0 : };
232 :
233 : // Read first message and create timeline if needed.
234 0 : let res = network_reader.read_first_message().await;
235 :
236 0 : let network_res = if let Ok((timeline, next_msg)) = res {
237 0 : let pageserver_feedback_rx: tokio::sync::broadcast::Receiver<PageserverFeedback> =
238 0 : timeline
239 0 : .get_walreceivers()
240 0 : .pageserver_feedback_tx
241 0 : .subscribe();
242 0 : *tli = Some(timeline.clone());
243 :
244 : tokio::select! {
245 : // todo: add read|write .context to these errors
246 : r = network_reader.run(msg_tx, msg_rx, reply_tx, timeline.clone(), next_msg) => r,
247 : r = network_write(pgb, reply_rx, pageserver_feedback_rx) => r,
248 : }
249 : } else {
250 0 : res.map(|_| ())
251 : };
252 :
253 : // Join pg backend back.
254 0 : pgb.unsplit(pgb_reader)?;
255 :
256 : // Join the spawned WalAcceptor. At this point chans to/from it passed
257 : // to network routines are dropped, so it will exit as soon as it
258 : // touches them.
259 0 : match acceptor_handle {
260 : None => {
261 : // failed even before spawning; read_network should have error
262 0 : Err(network_res.expect_err("no error with WalAcceptor not spawn"))
263 : }
264 0 : Some(handle) => {
265 0 : let wal_acceptor_res = handle.await;
266 :
267 : // If there was any network error, return it.
268 0 : network_res?;
269 :
270 : // Otherwise, WalAcceptor thread must have errored.
271 0 : match wal_acceptor_res {
272 0 : Ok(Ok(_)) => Ok(()), // can't happen currently; would be if we add graceful termination
273 0 : Ok(Err(e)) => Err(CopyStreamHandlerEnd::Other(e.context("WAL acceptor"))),
274 0 : Err(_) => Err(CopyStreamHandlerEnd::Other(anyhow!(
275 0 : "WalAcceptor task panicked",
276 0 : ))),
277 : }
278 : }
279 : }
280 0 : }
281 : }
282 :
283 : struct NetworkReader<'a, IO> {
284 : ttid: TenantTimelineId,
285 : conn_id: ConnectionId,
286 : pgb_reader: &'a mut PostgresBackendReader<IO>,
287 : peer_addr: SocketAddr,
288 : // WalAcceptor is spawned when we learn server info from walproposer and
289 : // create timeline; handle is put here.
290 : acceptor_handle: &'a mut Option<JoinHandle<anyhow::Result<()>>>,
291 : }
292 :
293 : impl<'a, IO: AsyncRead + AsyncWrite + Unpin> NetworkReader<'a, IO> {
294 0 : async fn read_first_message(
295 0 : &mut self,
296 0 : ) -> Result<(Arc<Timeline>, ProposerAcceptorMessage), CopyStreamHandlerEnd> {
297 : // Receive information about server to create timeline, if not yet.
298 0 : let next_msg = read_message(self.pgb_reader).await?;
299 0 : let tli = match next_msg {
300 0 : ProposerAcceptorMessage::Greeting(ref greeting) => {
301 0 : info!(
302 0 : "start handshake with walproposer {} sysid {} timeline {}",
303 : self.peer_addr, greeting.system_id, greeting.tli,
304 : );
305 0 : let server_info = ServerInfo {
306 0 : pg_version: greeting.pg_version,
307 0 : system_id: greeting.system_id,
308 0 : wal_seg_size: greeting.wal_seg_size,
309 0 : };
310 0 : GlobalTimelines::create(self.ttid, server_info, Lsn::INVALID, Lsn::INVALID).await?
311 : }
312 : _ => {
313 0 : return Err(CopyStreamHandlerEnd::Other(anyhow::anyhow!(
314 0 : "unexpected message {next_msg:?} instead of greeting"
315 0 : )))
316 : }
317 : };
318 0 : Ok((tli, next_msg))
319 0 : }
320 :
321 0 : async fn run(
322 0 : self,
323 0 : msg_tx: Sender<ProposerAcceptorMessage>,
324 0 : msg_rx: Receiver<ProposerAcceptorMessage>,
325 0 : reply_tx: Sender<AcceptorProposerMessage>,
326 0 : tli: Arc<Timeline>,
327 0 : next_msg: ProposerAcceptorMessage,
328 0 : ) -> Result<(), CopyStreamHandlerEnd> {
329 0 : *self.acceptor_handle = Some(WalAcceptor::spawn(
330 0 : tli,
331 0 : msg_rx,
332 0 : reply_tx,
333 0 : Some(self.conn_id),
334 0 : ));
335 0 :
336 0 : // Forward all messages to WalAcceptor
337 0 : read_network_loop(self.pgb_reader, msg_tx, next_msg).await
338 0 : }
339 : }
340 :
341 : /// Read next message from walproposer.
342 : /// TODO: Return Ok(None) on graceful termination.
343 0 : async fn read_message<IO: AsyncRead + AsyncWrite + Unpin>(
344 0 : pgb_reader: &mut PostgresBackendReader<IO>,
345 0 : ) -> Result<ProposerAcceptorMessage, CopyStreamHandlerEnd> {
346 0 : let copy_data = pgb_reader.read_copy_message().await?;
347 0 : let msg = ProposerAcceptorMessage::parse(copy_data)?;
348 0 : Ok(msg)
349 0 : }
350 :
351 0 : async fn read_network_loop<IO: AsyncRead + AsyncWrite + Unpin>(
352 0 : pgb_reader: &mut PostgresBackendReader<IO>,
353 0 : msg_tx: Sender<ProposerAcceptorMessage>,
354 0 : mut next_msg: ProposerAcceptorMessage,
355 0 : ) -> Result<(), CopyStreamHandlerEnd> {
356 : loop {
357 0 : if msg_tx.send(next_msg).await.is_err() {
358 0 : return Ok(()); // chan closed, WalAcceptor terminated
359 0 : }
360 0 : next_msg = read_message(pgb_reader).await?;
361 : }
362 0 : }
363 :
364 : /// Read replies from WalAcceptor and pass them back to socket. Returns Ok(())
365 : /// if reply_rx closed; it must mean WalAcceptor terminated, joining it should
366 : /// tell the error.
367 0 : async fn network_write<IO: AsyncRead + AsyncWrite + Unpin>(
368 0 : pgb_writer: &mut PostgresBackend<IO>,
369 0 : mut reply_rx: Receiver<AcceptorProposerMessage>,
370 0 : mut pageserver_feedback_rx: tokio::sync::broadcast::Receiver<PageserverFeedback>,
371 0 : ) -> Result<(), CopyStreamHandlerEnd> {
372 0 : let mut buf = BytesMut::with_capacity(128);
373 0 :
374 0 : // storing append_response to inject PageserverFeedback into it
375 0 : let mut last_append_response = None;
376 :
377 : loop {
378 : // trying to read either AcceptorProposerMessage or PageserverFeedback
379 0 : let msg = tokio::select! {
380 : reply = reply_rx.recv() => {
381 : if let Some(msg) = reply {
382 : if let AcceptorProposerMessage::AppendResponse(append_response) = &msg {
383 : last_append_response = Some(append_response.clone());
384 : }
385 : Some(msg)
386 : } else {
387 : return Ok(()); // chan closed, WalAcceptor terminated
388 : }
389 : }
390 :
391 : feedback = pageserver_feedback_rx.recv() =>
392 : match (feedback, &last_append_response) {
393 : (Ok(feedback), Some(append_response)) => {
394 : // clone AppendResponse and inject PageserverFeedback into it
395 : let mut append_response = append_response.clone();
396 : append_response.pageserver_feedback = Some(feedback);
397 : Some(AcceptorProposerMessage::AppendResponse(append_response))
398 : }
399 : _ => None,
400 : }
401 : };
402 :
403 0 : let Some(msg) = msg else {
404 0 : continue;
405 : };
406 :
407 0 : buf.clear();
408 0 : msg.serialize(&mut buf)?;
409 0 : pgb_writer.write_message(&BeMessage::CopyData(&buf)).await?;
410 : }
411 0 : }
412 :
413 : // Send keepalive messages to walproposer, to make sure it receives updates
414 : // even when it writes a steady stream of messages.
415 : const KEEPALIVE_INTERVAL: Duration = Duration::from_secs(1);
416 :
417 : /// Encapsulates a task which takes messages from msg_rx, processes and pushes
418 : /// replies to reply_tx; reading from socket and writing to disk in parallel is
419 : /// beneficial for performance, this struct provides writing to disk part.
420 : pub struct WalAcceptor {
421 : tli: Arc<Timeline>,
422 : msg_rx: Receiver<ProposerAcceptorMessage>,
423 : reply_tx: Sender<AcceptorProposerMessage>,
424 : conn_id: Option<ConnectionId>,
425 : }
426 :
427 : impl WalAcceptor {
428 : /// Spawn task with WalAcceptor running, return handle to it. Task returns
429 : /// Ok(()) if either of channels has closed, and Err if any error during
430 : /// message processing is encountered.
431 : ///
432 : /// conn_id None means WalAcceptor is used by recovery initiated at this safekeeper.
433 0 : pub fn spawn(
434 0 : tli: Arc<Timeline>,
435 0 : msg_rx: Receiver<ProposerAcceptorMessage>,
436 0 : reply_tx: Sender<AcceptorProposerMessage>,
437 0 : conn_id: Option<ConnectionId>,
438 0 : ) -> JoinHandle<anyhow::Result<()>> {
439 0 : task::spawn(async move {
440 0 : let mut wa = WalAcceptor {
441 0 : tli,
442 0 : msg_rx,
443 0 : reply_tx,
444 0 : conn_id,
445 0 : };
446 0 :
447 0 : let span_ttid = wa.tli.ttid; // satisfy borrow checker
448 0 : wa.run()
449 0 : .instrument(
450 0 : info_span!("WAL acceptor", cid = %conn_id.unwrap_or(0), ttid = %span_ttid),
451 : )
452 0 : .await
453 0 : })
454 0 : }
455 :
456 : /// The main loop. Returns Ok(()) if either msg_rx or reply_tx got closed;
457 : /// it must mean that network thread terminated.
458 0 : async fn run(&mut self) -> anyhow::Result<()> {
459 0 : // Register the connection and defer unregister.
460 0 : // Order of the next two lines is important: we want first to remove our entry and then
461 0 : // update status which depends on registered connections.
462 0 : let _compute_conn_guard = ComputeConnectionGuard {
463 0 : timeline: Arc::clone(&self.tli),
464 0 : };
465 0 : let walreceiver_guard = self.tli.get_walreceivers().register(self.conn_id);
466 0 : self.tli.update_status_notify().await?;
467 :
468 : // After this timestamp we will stop processing AppendRequests and send a response
469 : // to the walproposer. walproposer sends at least one AppendRequest per second,
470 : // we will send keepalives by replying to these requests once per second.
471 0 : let mut next_keepalive = Instant::now();
472 :
473 : loop {
474 0 : let opt_msg = self.msg_rx.recv().await;
475 0 : if opt_msg.is_none() {
476 0 : return Ok(()); // chan closed, streaming terminated
477 0 : }
478 0 : let mut next_msg = opt_msg.unwrap();
479 0 :
480 0 : // Update walreceiver state in shmem for reporting.
481 0 : if let ProposerAcceptorMessage::Elected(_) = &next_msg {
482 0 : walreceiver_guard.get().status = WalReceiverStatus::Streaming;
483 0 : }
484 :
485 0 : let reply_msg = if matches!(next_msg, ProposerAcceptorMessage::AppendRequest(_)) {
486 : // loop through AppendRequest's while it's readily available to
487 : // write as many WAL as possible without fsyncing
488 : //
489 : // Note: this will need to be rewritten if we want to read non-AppendRequest messages here.
490 : // Otherwise, we might end up in a situation where we read a message, but don't
491 : // process it.
492 0 : while let ProposerAcceptorMessage::AppendRequest(append_request) = next_msg {
493 0 : let noflush_msg = ProposerAcceptorMessage::NoFlushAppendRequest(append_request);
494 :
495 0 : if let Some(reply) = self.tli.process_msg(&noflush_msg).await? {
496 0 : if self.reply_tx.send(reply).await.is_err() {
497 0 : return Ok(()); // chan closed, streaming terminated
498 0 : }
499 0 : }
500 :
501 : // get out of this loop if keepalive time is reached
502 0 : if Instant::now() >= next_keepalive {
503 0 : break;
504 0 : }
505 0 :
506 0 : match self.msg_rx.try_recv() {
507 0 : Ok(msg) => next_msg = msg,
508 0 : Err(TryRecvError::Empty) => break,
509 0 : Err(TryRecvError::Disconnected) => return Ok(()), // chan closed, streaming terminated
510 : }
511 : }
512 :
513 : // flush all written WAL to the disk
514 0 : self.tli
515 0 : .process_msg(&ProposerAcceptorMessage::FlushWAL)
516 0 : .await?
517 : } else {
518 : // process message other than AppendRequest
519 0 : self.tli.process_msg(&next_msg).await?
520 : };
521 :
522 0 : if let Some(reply) = reply_msg {
523 0 : if self.reply_tx.send(reply).await.is_err() {
524 0 : return Ok(()); // chan closed, streaming terminated
525 0 : }
526 0 : // reset keepalive time
527 0 : next_keepalive = Instant::now() + KEEPALIVE_INTERVAL;
528 0 : }
529 : }
530 0 : }
531 : }
532 :
533 : /// Calls update_status_notify in drop to update timeline status.
534 : struct ComputeConnectionGuard {
535 : timeline: Arc<Timeline>,
536 : }
537 :
538 : impl Drop for ComputeConnectionGuard {
539 0 : fn drop(&mut self) {
540 0 : let tli = self.timeline.clone();
541 0 : tokio::spawn(async move {
542 0 : if let Err(e) = tli.update_status_notify().await {
543 0 : error!("failed to update timeline status: {}", e);
544 0 : }
545 0 : });
546 0 : }
547 : }
|
