Line data Source code
1 : //! This module implements the streaming side of replication protocol, starting
2 : //! with the "START_REPLICATION" message, and registry of walsenders.
3 :
4 : use crate::handler::SafekeeperPostgresHandler;
5 : use crate::metrics::RECEIVED_PS_FEEDBACKS;
6 : use crate::receive_wal::WalReceivers;
7 : use crate::safekeeper::{Term, TermLsn};
8 : use crate::timeline::Timeline;
9 : use crate::wal_service::ConnectionId;
10 : use crate::wal_storage::WalReader;
11 : use crate::GlobalTimelines;
12 : use anyhow::{bail, Context as AnyhowContext};
13 : use bytes::Bytes;
14 : use parking_lot::Mutex;
15 : use postgres_backend::PostgresBackend;
16 : use postgres_backend::{CopyStreamHandlerEnd, PostgresBackendReader, QueryError};
17 : use postgres_ffi::get_current_timestamp;
18 : use postgres_ffi::{TimestampTz, MAX_SEND_SIZE};
19 : use pq_proto::{BeMessage, WalSndKeepAlive, XLogDataBody};
20 : use serde::{Deserialize, Serialize};
21 : use tokio::io::{AsyncRead, AsyncWrite};
22 : use utils::failpoint_support;
23 : use utils::id::TenantTimelineId;
24 : use utils::pageserver_feedback::PageserverFeedback;
25 :
26 : use std::cmp::min;
27 : use std::net::SocketAddr;
28 : use std::str;
29 : use std::sync::Arc;
30 : use std::time::Duration;
31 : use tokio::sync::watch::Receiver;
32 : use tokio::time::timeout;
33 : use tracing::*;
34 : use utils::{bin_ser::BeSer, lsn::Lsn};
35 :
36 : // See: https://www.postgresql.org/docs/13/protocol-replication.html
37 : const HOT_STANDBY_FEEDBACK_TAG_BYTE: u8 = b'h';
38 : const STANDBY_STATUS_UPDATE_TAG_BYTE: u8 = b'r';
39 : // neon extension of replication protocol
40 : const NEON_STATUS_UPDATE_TAG_BYTE: u8 = b'z';
41 :
42 : type FullTransactionId = u64;
43 :
44 : /// Hot standby feedback received from replica
45 0 : #[derive(Debug, Clone, Copy, Serialize, Deserialize)]
46 : pub struct HotStandbyFeedback {
47 : pub ts: TimestampTz,
48 : pub xmin: FullTransactionId,
49 : pub catalog_xmin: FullTransactionId,
50 : }
51 :
52 : const INVALID_FULL_TRANSACTION_ID: FullTransactionId = 0;
53 :
54 : impl HotStandbyFeedback {
55 17976 : pub fn empty() -> HotStandbyFeedback {
56 17976 : HotStandbyFeedback {
57 17976 : ts: 0,
58 17976 : xmin: 0,
59 17976 : catalog_xmin: 0,
60 17976 : }
61 17976 : }
62 : }
63 :
64 : /// Standby status update
65 0 : #[derive(Debug, Clone, Copy, Serialize, Deserialize)]
66 : pub struct StandbyReply {
67 : pub write_lsn: Lsn, // The location of the last WAL byte + 1 received and written to disk in the standby.
68 : pub flush_lsn: Lsn, // The location of the last WAL byte + 1 flushed to disk in the standby.
69 : pub apply_lsn: Lsn, // The location of the last WAL byte + 1 applied in the standby.
70 : pub reply_ts: TimestampTz, // The client's system clock at the time of transmission, as microseconds since midnight on 2000-01-01.
71 : pub reply_requested: bool,
72 : }
73 :
74 : impl StandbyReply {
75 8 : fn empty() -> Self {
76 8 : StandbyReply {
77 8 : write_lsn: Lsn::INVALID,
78 8 : flush_lsn: Lsn::INVALID,
79 8 : apply_lsn: Lsn::INVALID,
80 8 : reply_ts: 0,
81 8 : reply_requested: false,
82 8 : }
83 8 : }
84 : }
85 :
86 0 : #[derive(Debug, Clone, Copy, Serialize, Deserialize)]
87 : pub struct StandbyFeedback {
88 : reply: StandbyReply,
89 : hs_feedback: HotStandbyFeedback,
90 : }
91 :
92 : /// WalSenders registry. Timeline holds it (wrapped in Arc).
93 : pub struct WalSenders {
94 : mutex: Mutex<WalSendersShared>,
95 : walreceivers: Arc<WalReceivers>,
96 : }
97 :
98 : impl WalSenders {
99 0 : pub fn new(walreceivers: Arc<WalReceivers>) -> Arc<WalSenders> {
100 0 : Arc::new(WalSenders {
101 0 : mutex: Mutex::new(WalSendersShared::new()),
102 0 : walreceivers,
103 0 : })
104 0 : }
105 :
106 : /// Register new walsender. Returned guard provides access to the slot and
107 : /// automatically deregisters in Drop.
108 0 : fn register(
109 0 : self: &Arc<WalSenders>,
110 0 : ttid: TenantTimelineId,
111 0 : addr: SocketAddr,
112 0 : conn_id: ConnectionId,
113 0 : appname: Option<String>,
114 0 : ) -> WalSenderGuard {
115 0 : let slots = &mut self.mutex.lock().slots;
116 0 : let walsender_state = WalSenderState {
117 0 : ttid,
118 0 : addr,
119 0 : conn_id,
120 0 : appname,
121 0 : feedback: ReplicationFeedback::Pageserver(PageserverFeedback::empty()),
122 0 : };
123 : // find empty slot or create new one
124 0 : let pos = if let Some(pos) = slots.iter().position(|s| s.is_none()) {
125 0 : slots[pos] = Some(walsender_state);
126 0 : pos
127 : } else {
128 0 : let pos = slots.len();
129 0 : slots.push(Some(walsender_state));
130 0 : pos
131 : };
132 0 : WalSenderGuard {
133 0 : id: pos,
134 0 : walsenders: self.clone(),
135 0 : }
136 0 : }
137 :
138 : /// Get state of all walsenders.
139 0 : pub fn get_all(self: &Arc<WalSenders>) -> Vec<WalSenderState> {
140 0 : self.mutex.lock().slots.iter().flatten().cloned().collect()
141 0 : }
142 :
143 : /// Get LSN of the most lagging pageserver receiver. Return None if there are no
144 : /// active walsenders.
145 0 : pub fn laggard_lsn(self: &Arc<WalSenders>) -> Option<Lsn> {
146 0 : self.mutex
147 0 : .lock()
148 0 : .slots
149 0 : .iter()
150 0 : .flatten()
151 0 : .filter_map(|s| match s.feedback {
152 0 : ReplicationFeedback::Pageserver(feedback) => Some(feedback.last_received_lsn),
153 0 : ReplicationFeedback::Standby(_) => None,
154 0 : })
155 0 : .min()
156 0 : }
157 :
158 : /// Returns total counter of pageserver feedbacks received and last feedback.
159 0 : pub fn get_ps_feedback_stats(self: &Arc<WalSenders>) -> (u64, PageserverFeedback) {
160 0 : let shared = self.mutex.lock();
161 0 : (shared.ps_feedback_counter, shared.last_ps_feedback)
162 0 : }
163 :
164 : /// Get aggregated hot standby feedback (we send it to compute).
165 0 : pub fn get_hotstandby(self: &Arc<WalSenders>) -> HotStandbyFeedback {
166 0 : self.mutex.lock().agg_hs_feedback
167 0 : }
168 :
169 : /// Record new pageserver feedback, update aggregated values.
170 0 : fn record_ps_feedback(self: &Arc<WalSenders>, id: WalSenderId, feedback: &PageserverFeedback) {
171 0 : let mut shared = self.mutex.lock();
172 0 : shared.get_slot_mut(id).feedback = ReplicationFeedback::Pageserver(*feedback);
173 0 : shared.last_ps_feedback = *feedback;
174 0 : shared.ps_feedback_counter += 1;
175 0 : drop(shared);
176 0 :
177 0 : RECEIVED_PS_FEEDBACKS.inc();
178 0 :
179 0 : // send feedback to connected walproposers
180 0 : self.walreceivers.broadcast_pageserver_feedback(*feedback);
181 0 : }
182 :
183 : /// Record standby reply.
184 0 : fn record_standby_reply(self: &Arc<WalSenders>, id: WalSenderId, reply: &StandbyReply) {
185 0 : let mut shared = self.mutex.lock();
186 0 : let slot = shared.get_slot_mut(id);
187 0 : match &mut slot.feedback {
188 0 : ReplicationFeedback::Standby(sf) => sf.reply = *reply,
189 : ReplicationFeedback::Pageserver(_) => {
190 0 : slot.feedback = ReplicationFeedback::Standby(StandbyFeedback {
191 0 : reply: *reply,
192 0 : hs_feedback: HotStandbyFeedback::empty(),
193 0 : })
194 : }
195 : }
196 0 : }
197 :
198 : /// Record hot standby feedback, update aggregated value.
199 0 : fn record_hs_feedback(self: &Arc<WalSenders>, id: WalSenderId, feedback: &HotStandbyFeedback) {
200 0 : let mut shared = self.mutex.lock();
201 0 : let slot = shared.get_slot_mut(id);
202 0 : match &mut slot.feedback {
203 0 : ReplicationFeedback::Standby(sf) => sf.hs_feedback = *feedback,
204 : ReplicationFeedback::Pageserver(_) => {
205 0 : slot.feedback = ReplicationFeedback::Standby(StandbyFeedback {
206 0 : reply: StandbyReply::empty(),
207 0 : hs_feedback: *feedback,
208 0 : })
209 : }
210 : }
211 0 : shared.update_hs_feedback();
212 0 : }
213 :
214 : /// Get remote_consistent_lsn reported by the pageserver. Returns None if
215 : /// client is not pageserver.
216 0 : fn get_ws_remote_consistent_lsn(self: &Arc<WalSenders>, id: WalSenderId) -> Option<Lsn> {
217 0 : let shared = self.mutex.lock();
218 0 : let slot = shared.get_slot(id);
219 0 : match slot.feedback {
220 0 : ReplicationFeedback::Pageserver(feedback) => Some(feedback.remote_consistent_lsn),
221 0 : _ => None,
222 : }
223 0 : }
224 :
225 : /// Unregister walsender.
226 0 : fn unregister(self: &Arc<WalSenders>, id: WalSenderId) {
227 0 : let mut shared = self.mutex.lock();
228 0 : shared.slots[id] = None;
229 0 : shared.update_hs_feedback();
230 0 : }
231 : }
232 :
233 : struct WalSendersShared {
234 : // aggregated over all walsenders value
235 : agg_hs_feedback: HotStandbyFeedback,
236 : // last feedback ever received from any pageserver, empty if none
237 : last_ps_feedback: PageserverFeedback,
238 : // total counter of pageserver feedbacks received
239 : ps_feedback_counter: u64,
240 : slots: Vec<Option<WalSenderState>>,
241 : }
242 :
243 : impl WalSendersShared {
244 4 : fn new() -> Self {
245 4 : WalSendersShared {
246 4 : agg_hs_feedback: HotStandbyFeedback::empty(),
247 4 : last_ps_feedback: PageserverFeedback::empty(),
248 4 : ps_feedback_counter: 0,
249 4 : slots: Vec::new(),
250 4 : }
251 4 : }
252 :
253 : /// Get content of provided id slot, it must exist.
254 0 : fn get_slot(&self, id: WalSenderId) -> &WalSenderState {
255 0 : self.slots[id].as_ref().expect("walsender doesn't exist")
256 0 : }
257 :
258 : /// Get mut content of provided id slot, it must exist.
259 0 : fn get_slot_mut(&mut self, id: WalSenderId) -> &mut WalSenderState {
260 0 : self.slots[id].as_mut().expect("walsender doesn't exist")
261 0 : }
262 :
263 : /// Update aggregated hot standy feedback. We just take min of valid xmins
264 : /// and ts.
265 4 : fn update_hs_feedback(&mut self) {
266 4 : let mut agg = HotStandbyFeedback::empty();
267 8 : for ws_state in self.slots.iter().flatten() {
268 8 : if let ReplicationFeedback::Standby(standby_feedback) = ws_state.feedback {
269 8 : let hs_feedback = standby_feedback.hs_feedback;
270 8 : // doing Option math like op1.iter().chain(op2.iter()).min()
271 8 : // would be nicer, but we serialize/deserialize this struct
272 8 : // directly, so leave as is for now
273 8 : if hs_feedback.xmin != INVALID_FULL_TRANSACTION_ID {
274 4 : if agg.xmin != INVALID_FULL_TRANSACTION_ID {
275 2 : agg.xmin = min(agg.xmin, hs_feedback.xmin);
276 2 : } else {
277 2 : agg.xmin = hs_feedback.xmin;
278 2 : }
279 4 : agg.ts = min(agg.ts, hs_feedback.ts);
280 4 : }
281 8 : if hs_feedback.catalog_xmin != INVALID_FULL_TRANSACTION_ID {
282 0 : if agg.catalog_xmin != INVALID_FULL_TRANSACTION_ID {
283 0 : agg.catalog_xmin = min(agg.catalog_xmin, hs_feedback.catalog_xmin);
284 0 : } else {
285 0 : agg.catalog_xmin = hs_feedback.catalog_xmin;
286 0 : }
287 0 : agg.ts = min(agg.ts, hs_feedback.ts);
288 8 : }
289 0 : }
290 : }
291 4 : self.agg_hs_feedback = agg;
292 4 : }
293 : }
294 :
295 : // Serialized is used only for pretty printing in json.
296 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
297 : pub struct WalSenderState {
298 : ttid: TenantTimelineId,
299 : addr: SocketAddr,
300 : conn_id: ConnectionId,
301 : // postgres application_name
302 : appname: Option<String>,
303 : feedback: ReplicationFeedback,
304 : }
305 :
306 : // Receiver is either pageserver or regular standby, which have different
307 : // feedbacks.
308 0 : #[derive(Debug, Clone, Copy, Serialize, Deserialize)]
309 : enum ReplicationFeedback {
310 : Pageserver(PageserverFeedback),
311 : Standby(StandbyFeedback),
312 : }
313 :
314 : // id of the occupied slot in WalSenders to access it (and save in the
315 : // WalSenderGuard). We could give Arc directly to the slot, but there is not
316 : // much sense in that as values aggregation which is performed on each feedback
317 : // receival iterates over all walsenders.
318 : pub type WalSenderId = usize;
319 :
320 : /// Scope guard to access slot in WalSenders registry and unregister from it in
321 : /// Drop.
322 : pub struct WalSenderGuard {
323 : id: WalSenderId,
324 : walsenders: Arc<WalSenders>,
325 : }
326 :
327 : impl Drop for WalSenderGuard {
328 0 : fn drop(&mut self) {
329 0 : self.walsenders.unregister(self.id);
330 0 : }
331 : }
332 :
333 : impl SafekeeperPostgresHandler {
334 : /// Wrapper around handle_start_replication_guts handling result. Error is
335 : /// handled here while we're still in walsender ttid span; with API
336 : /// extension, this can probably be moved into postgres_backend.
337 0 : pub async fn handle_start_replication<IO: AsyncRead + AsyncWrite + Unpin>(
338 0 : &mut self,
339 0 : pgb: &mut PostgresBackend<IO>,
340 0 : start_pos: Lsn,
341 0 : term: Option<Term>,
342 0 : ) -> Result<(), QueryError> {
343 0 : if let Err(end) = self
344 0 : .handle_start_replication_guts(pgb, start_pos, term)
345 0 : .await
346 : {
347 : // Log the result and probably send it to the client, closing the stream.
348 0 : pgb.handle_copy_stream_end(end).await;
349 0 : }
350 0 : Ok(())
351 0 : }
352 :
353 0 : pub async fn handle_start_replication_guts<IO: AsyncRead + AsyncWrite + Unpin>(
354 0 : &mut self,
355 0 : pgb: &mut PostgresBackend<IO>,
356 0 : start_pos: Lsn,
357 0 : term: Option<Term>,
358 0 : ) -> Result<(), CopyStreamHandlerEnd> {
359 0 : let appname = self.appname.clone();
360 0 : let tli =
361 0 : GlobalTimelines::get(self.ttid).map_err(|e| CopyStreamHandlerEnd::Other(e.into()))?;
362 :
363 : // Use a guard object to remove our entry from the timeline when we are done.
364 0 : let ws_guard = Arc::new(tli.get_walsenders().register(
365 0 : self.ttid,
366 0 : *pgb.get_peer_addr(),
367 0 : self.conn_id,
368 0 : self.appname.clone(),
369 0 : ));
370 :
371 : // Walsender can operate in one of two modes which we select by
372 : // application_name: give only committed WAL (used by pageserver) or all
373 : // existing WAL (up to flush_lsn, used by walproposer or peer recovery).
374 : // The second case is always driven by a consensus leader which term
375 : // must be supplied.
376 0 : let end_watch = if term.is_some() {
377 0 : EndWatch::Flush(tli.get_term_flush_lsn_watch_rx())
378 : } else {
379 0 : EndWatch::Commit(tli.get_commit_lsn_watch_rx())
380 : };
381 : // we don't check term here; it will be checked on first waiting/WAL reading anyway.
382 0 : let end_pos = end_watch.get();
383 0 :
384 0 : if end_pos < start_pos {
385 0 : warn!(
386 0 : "requested start_pos {} is ahead of available WAL end_pos {}",
387 : start_pos, end_pos
388 : );
389 0 : }
390 :
391 0 : info!(
392 0 : "starting streaming from {:?}, available WAL ends at {}, recovery={}, appname={:?}",
393 : start_pos,
394 : end_pos,
395 0 : matches!(end_watch, EndWatch::Flush(_)),
396 : appname
397 : );
398 :
399 : // switch to copy
400 0 : pgb.write_message(&BeMessage::CopyBothResponse).await?;
401 :
402 0 : let (_, persisted_state) = tli.get_state().await;
403 0 : let wal_reader = WalReader::new(
404 0 : self.conf.workdir.clone(),
405 0 : self.conf.timeline_dir(&tli.ttid),
406 0 : &persisted_state,
407 0 : start_pos,
408 0 : self.conf.is_wal_backup_enabled(),
409 0 : )?;
410 :
411 : // Split to concurrently receive and send data; replies are generally
412 : // not synchronized with sends, so this avoids deadlocks.
413 0 : let reader = pgb.split().context("START_REPLICATION split")?;
414 :
415 0 : let mut sender = WalSender {
416 0 : pgb,
417 0 : tli: tli.clone(),
418 0 : appname,
419 0 : start_pos,
420 0 : end_pos,
421 0 : term,
422 0 : end_watch,
423 0 : ws_guard: ws_guard.clone(),
424 0 : wal_reader,
425 0 : send_buf: [0; MAX_SEND_SIZE],
426 0 : };
427 0 : let mut reply_reader = ReplyReader {
428 0 : reader,
429 0 : ws_guard: ws_guard.clone(),
430 0 : tli,
431 0 : };
432 :
433 0 : let res = tokio::select! {
434 : // todo: add read|write .context to these errors
435 : r = sender.run() => r,
436 : r = reply_reader.run() => r,
437 : };
438 :
439 0 : let ws_state = ws_guard
440 0 : .walsenders
441 0 : .mutex
442 0 : .lock()
443 0 : .get_slot(ws_guard.id)
444 0 : .clone();
445 0 : info!(
446 0 : "finished streaming to {}, feedback={:?}",
447 : ws_state.addr, ws_state.feedback,
448 : );
449 :
450 : // Join pg backend back.
451 0 : pgb.unsplit(reply_reader.reader)?;
452 :
453 0 : res
454 0 : }
455 : }
456 :
457 : /// Walsender streams either up to commit_lsn (normally) or flush_lsn in the
458 : /// given term (recovery by walproposer or peer safekeeper).
459 : enum EndWatch {
460 : Commit(Receiver<Lsn>),
461 : Flush(Receiver<TermLsn>),
462 : }
463 :
464 : impl EndWatch {
465 : /// Get current end of WAL.
466 0 : fn get(&self) -> Lsn {
467 0 : match self {
468 0 : EndWatch::Commit(r) => *r.borrow(),
469 0 : EndWatch::Flush(r) => r.borrow().lsn,
470 : }
471 0 : }
472 :
473 : /// Wait for the update.
474 0 : async fn changed(&mut self) -> anyhow::Result<()> {
475 0 : match self {
476 0 : EndWatch::Commit(r) => r.changed().await?,
477 0 : EndWatch::Flush(r) => r.changed().await?,
478 : }
479 0 : Ok(())
480 0 : }
481 : }
482 :
483 : /// A half driving sending WAL.
484 : struct WalSender<'a, IO> {
485 : pgb: &'a mut PostgresBackend<IO>,
486 : tli: Arc<Timeline>,
487 : appname: Option<String>,
488 : // Position since which we are sending next chunk.
489 : start_pos: Lsn,
490 : // WAL up to this position is known to be locally available.
491 : // Usually this is the same as the latest commit_lsn, but in case of
492 : // walproposer recovery, this is flush_lsn.
493 : //
494 : // We send this LSN to the receiver as wal_end, so that it knows how much
495 : // WAL this safekeeper has. This LSN should be as fresh as possible.
496 : end_pos: Lsn,
497 : /// When streaming uncommitted part, the term the client acts as the leader
498 : /// in. Streaming is stopped if local term changes to a different (higher)
499 : /// value.
500 : term: Option<Term>,
501 : /// Watch channel receiver to learn end of available WAL (and wait for its advancement).
502 : end_watch: EndWatch,
503 : ws_guard: Arc<WalSenderGuard>,
504 : wal_reader: WalReader,
505 : // buffer for readling WAL into to send it
506 : send_buf: [u8; MAX_SEND_SIZE],
507 : }
508 :
509 : const POLL_STATE_TIMEOUT: Duration = Duration::from_secs(1);
510 :
511 : impl<IO: AsyncRead + AsyncWrite + Unpin> WalSender<'_, IO> {
512 : /// Send WAL until
513 : /// - an error occurs
514 : /// - receiver is caughtup and there is no computes (if streaming up to commit_lsn)
515 : ///
516 : /// Err(CopyStreamHandlerEnd) is always returned; Result is used only for ?
517 : /// convenience.
518 0 : async fn run(&mut self) -> Result<(), CopyStreamHandlerEnd> {
519 : loop {
520 : // Wait for the next portion if it is not there yet, or just
521 : // update our end of WAL available for sending value, we
522 : // communicate it to the receiver.
523 0 : self.wait_wal().await?;
524 0 : assert!(
525 0 : self.end_pos > self.start_pos,
526 0 : "nothing to send after waiting for WAL"
527 : );
528 :
529 : // try to send as much as available, capped by MAX_SEND_SIZE
530 0 : let mut chunk_end_pos = self.start_pos + MAX_SEND_SIZE as u64;
531 0 : // if we went behind available WAL, back off
532 0 : if chunk_end_pos >= self.end_pos {
533 0 : chunk_end_pos = self.end_pos;
534 0 : } else {
535 0 : // If sending not up to end pos, round down to page boundary to
536 0 : // avoid breaking WAL record not at page boundary, as protocol
537 0 : // demands. See walsender.c (XLogSendPhysical).
538 0 : chunk_end_pos = chunk_end_pos
539 0 : .checked_sub(chunk_end_pos.block_offset())
540 0 : .unwrap();
541 0 : }
542 0 : let send_size = (chunk_end_pos.0 - self.start_pos.0) as usize;
543 0 : let send_buf = &mut self.send_buf[..send_size];
544 : let send_size: usize;
545 : {
546 : // If uncommitted part is being pulled, check that the term is
547 : // still the expected one.
548 0 : let _term_guard = if let Some(t) = self.term {
549 0 : Some(self.tli.acquire_term(t).await?)
550 : } else {
551 0 : None
552 : };
553 : // Read WAL into buffer. send_size can be additionally capped to
554 : // segment boundary here.
555 0 : send_size = self.wal_reader.read(send_buf).await?
556 : };
557 0 : let send_buf = &send_buf[..send_size];
558 0 :
559 0 : // and send it
560 0 : self.pgb
561 0 : .write_message(&BeMessage::XLogData(XLogDataBody {
562 0 : wal_start: self.start_pos.0,
563 0 : wal_end: self.end_pos.0,
564 0 : timestamp: get_current_timestamp(),
565 0 : data: send_buf,
566 0 : }))
567 0 : .await?;
568 :
569 0 : if let Some(appname) = &self.appname {
570 0 : if appname == "replica" {
571 0 : failpoint_support::sleep_millis_async!("sk-send-wal-replica-sleep");
572 0 : }
573 0 : }
574 0 : trace!(
575 0 : "sent {} bytes of WAL {}-{}",
576 0 : send_size,
577 0 : self.start_pos,
578 0 : self.start_pos + send_size as u64
579 : );
580 0 : self.start_pos += send_size as u64;
581 : }
582 0 : }
583 :
584 : /// wait until we have WAL to stream, sending keepalives and checking for
585 : /// exit in the meanwhile
586 0 : async fn wait_wal(&mut self) -> Result<(), CopyStreamHandlerEnd> {
587 : loop {
588 0 : self.end_pos = self.end_watch.get();
589 0 : let have_something_to_send = (|| {
590 0 : fail::fail_point!(
591 0 : "sk-pause-send",
592 0 : self.appname.as_deref() != Some("pageserver"),
593 0 : |_| { false }
594 : );
595 0 : self.end_pos > self.start_pos
596 0 : })();
597 0 :
598 0 : if have_something_to_send {
599 0 : trace!("got end_pos {:?}, streaming", self.end_pos);
600 0 : return Ok(());
601 0 : }
602 :
603 : // Wait for WAL to appear, now self.end_pos == self.start_pos.
604 0 : if let Some(lsn) = self.wait_for_lsn().await? {
605 0 : self.end_pos = lsn;
606 0 : trace!("got end_pos {:?}, streaming", self.end_pos);
607 0 : return Ok(());
608 0 : }
609 0 :
610 0 : // Timed out waiting for WAL, check for termination and send KA.
611 0 : // Check for termination only if we are streaming up to commit_lsn
612 0 : // (to pageserver).
613 0 : if let EndWatch::Commit(_) = self.end_watch {
614 0 : if let Some(remote_consistent_lsn) = self
615 0 : .ws_guard
616 0 : .walsenders
617 0 : .get_ws_remote_consistent_lsn(self.ws_guard.id)
618 : {
619 0 : if self.tli.should_walsender_stop(remote_consistent_lsn).await {
620 : // Terminate if there is nothing more to send.
621 : // Note that "ending streaming" part of the string is used by
622 : // pageserver to identify WalReceiverError::SuccessfulCompletion,
623 : // do not change this string without updating pageserver.
624 0 : return Err(CopyStreamHandlerEnd::ServerInitiated(format!(
625 0 : "ending streaming to {:?} at {}, receiver is caughtup and there is no computes",
626 0 : self.appname, self.start_pos,
627 0 : )));
628 0 : }
629 0 : }
630 0 : }
631 :
632 0 : self.pgb
633 0 : .write_message(&BeMessage::KeepAlive(WalSndKeepAlive {
634 0 : wal_end: self.end_pos.0,
635 0 : timestamp: get_current_timestamp(),
636 0 : request_reply: true,
637 0 : }))
638 0 : .await?;
639 : }
640 0 : }
641 :
642 : /// Wait until we have available WAL > start_pos or timeout expires. Returns
643 : /// - Ok(Some(end_pos)) if needed lsn is successfully observed;
644 : /// - Ok(None) if timeout expired;
645 : /// - Err in case of error -- only if 1) term changed while fetching in recovery
646 : /// mode 2) watch channel closed, which must never happen.
647 0 : async fn wait_for_lsn(&mut self) -> anyhow::Result<Option<Lsn>> {
648 0 : let fp = (|| {
649 0 : fail::fail_point!(
650 0 : "sk-pause-send",
651 0 : self.appname.as_deref() != Some("pageserver"),
652 0 : |_| { true }
653 : );
654 0 : false
655 0 : })();
656 0 : if fp {
657 0 : tokio::time::sleep(POLL_STATE_TIMEOUT).await;
658 0 : return Ok(None);
659 0 : }
660 :
661 0 : let res = timeout(POLL_STATE_TIMEOUT, async move {
662 : loop {
663 0 : let end_pos = self.end_watch.get();
664 0 : if end_pos > self.start_pos {
665 0 : return Ok(end_pos);
666 0 : }
667 0 : if let EndWatch::Flush(rx) = &self.end_watch {
668 0 : let curr_term = rx.borrow().term;
669 0 : if let Some(client_term) = self.term {
670 0 : if curr_term != client_term {
671 0 : bail!("term changed: requested {}, now {}", client_term, curr_term);
672 0 : }
673 0 : }
674 0 : }
675 0 : self.end_watch.changed().await?;
676 : }
677 0 : })
678 0 : .await;
679 :
680 0 : match res {
681 : // success
682 0 : Ok(Ok(commit_lsn)) => Ok(Some(commit_lsn)),
683 : // error inside closure
684 0 : Ok(Err(err)) => Err(err),
685 : // timeout
686 0 : Err(_) => Ok(None),
687 : }
688 0 : }
689 : }
690 :
691 : /// A half driving receiving replies.
692 : struct ReplyReader<IO> {
693 : reader: PostgresBackendReader<IO>,
694 : ws_guard: Arc<WalSenderGuard>,
695 : tli: Arc<Timeline>,
696 : }
697 :
698 : impl<IO: AsyncRead + AsyncWrite + Unpin> ReplyReader<IO> {
699 0 : async fn run(&mut self) -> Result<(), CopyStreamHandlerEnd> {
700 : loop {
701 0 : let msg = self.reader.read_copy_message().await?;
702 0 : self.handle_feedback(&msg).await?
703 : }
704 0 : }
705 :
706 0 : async fn handle_feedback(&mut self, msg: &Bytes) -> anyhow::Result<()> {
707 0 : match msg.first().cloned() {
708 0 : Some(HOT_STANDBY_FEEDBACK_TAG_BYTE) => {
709 0 : // Note: deserializing is on m[1..] because we skip the tag byte.
710 0 : let hs_feedback = HotStandbyFeedback::des(&msg[1..])
711 0 : .context("failed to deserialize HotStandbyFeedback")?;
712 0 : self.ws_guard
713 0 : .walsenders
714 0 : .record_hs_feedback(self.ws_guard.id, &hs_feedback);
715 : }
716 0 : Some(STANDBY_STATUS_UPDATE_TAG_BYTE) => {
717 0 : let reply =
718 0 : StandbyReply::des(&msg[1..]).context("failed to deserialize StandbyReply")?;
719 0 : self.ws_guard
720 0 : .walsenders
721 0 : .record_standby_reply(self.ws_guard.id, &reply);
722 : }
723 : Some(NEON_STATUS_UPDATE_TAG_BYTE) => {
724 : // pageserver sends this.
725 : // Note: deserializing is on m[9..] because we skip the tag byte and len bytes.
726 0 : let buf = Bytes::copy_from_slice(&msg[9..]);
727 0 : let ps_feedback = PageserverFeedback::parse(buf);
728 0 :
729 0 : trace!("PageserverFeedback is {:?}", ps_feedback);
730 0 : self.ws_guard
731 0 : .walsenders
732 0 : .record_ps_feedback(self.ws_guard.id, &ps_feedback);
733 0 : self.tli
734 0 : .update_remote_consistent_lsn(ps_feedback.remote_consistent_lsn)
735 0 : .await;
736 : // in principle new remote_consistent_lsn could allow to
737 : // deactivate the timeline, but we check that regularly through
738 : // broker updated, not need to do it here
739 : }
740 0 : _ => warn!("unexpected message {:?}", msg),
741 : }
742 0 : Ok(())
743 0 : }
744 : }
745 :
746 : #[cfg(test)]
747 : mod tests {
748 : use utils::id::{TenantId, TimelineId};
749 :
750 : use super::*;
751 :
752 8 : fn mock_ttid() -> TenantTimelineId {
753 8 : TenantTimelineId {
754 8 : tenant_id: TenantId::from_slice(&[0x00; 16]).unwrap(),
755 8 : timeline_id: TimelineId::from_slice(&[0x00; 16]).unwrap(),
756 8 : }
757 8 : }
758 :
759 8 : fn mock_addr() -> SocketAddr {
760 8 : "127.0.0.1:8080".parse().unwrap()
761 8 : }
762 :
763 : // add to wss specified feedback setting other fields to dummy values
764 8 : fn push_feedback(wss: &mut WalSendersShared, feedback: ReplicationFeedback) {
765 8 : let walsender_state = WalSenderState {
766 8 : ttid: mock_ttid(),
767 8 : addr: mock_addr(),
768 8 : conn_id: 1,
769 8 : appname: None,
770 8 : feedback,
771 8 : };
772 8 : wss.slots.push(Some(walsender_state))
773 8 : }
774 :
775 : // form standby feedback with given hot standby feedback ts/xmin and the
776 : // rest set to dummy values.
777 8 : fn hs_feedback(ts: TimestampTz, xmin: FullTransactionId) -> ReplicationFeedback {
778 8 : ReplicationFeedback::Standby(StandbyFeedback {
779 8 : reply: StandbyReply::empty(),
780 8 : hs_feedback: HotStandbyFeedback {
781 8 : ts,
782 8 : xmin,
783 8 : catalog_xmin: 0,
784 8 : },
785 8 : })
786 8 : }
787 :
788 : // test that hs aggregation works as expected
789 : #[test]
790 2 : fn test_hs_feedback_no_valid() {
791 2 : let mut wss = WalSendersShared::new();
792 2 : push_feedback(&mut wss, hs_feedback(1, INVALID_FULL_TRANSACTION_ID));
793 2 : wss.update_hs_feedback();
794 2 : assert_eq!(wss.agg_hs_feedback.xmin, INVALID_FULL_TRANSACTION_ID);
795 2 : }
796 :
797 : #[test]
798 2 : fn test_hs_feedback() {
799 2 : let mut wss = WalSendersShared::new();
800 2 : push_feedback(&mut wss, hs_feedback(1, INVALID_FULL_TRANSACTION_ID));
801 2 : push_feedback(&mut wss, hs_feedback(1, 42));
802 2 : push_feedback(&mut wss, hs_feedback(1, 64));
803 2 : wss.update_hs_feedback();
804 2 : assert_eq!(wss.agg_hs_feedback.xmin, 42);
805 2 : }
806 : }
|
