Line data Source code
1 : use std::collections::HashMap;
2 : use std::fmt;
3 : use std::io::Write as _;
4 : use std::str::FromStr;
5 : use std::sync::Arc;
6 :
7 : use http_utils::endpoint::{
8 : self, ChannelWriter, auth_middleware, check_permission_with, profile_cpu_handler,
9 : profile_heap_handler, prometheus_metrics_handler, request_span,
10 : };
11 : use http_utils::error::ApiError;
12 : use http_utils::failpoints::failpoints_handler;
13 : use http_utils::json::{json_request, json_response};
14 : use http_utils::request::{ensure_no_body, parse_query_param, parse_request_param};
15 : use http_utils::{RequestExt, RouterBuilder};
16 : use hyper::{Body, Request, Response, StatusCode};
17 : use pem::Pem;
18 : use postgres_ffi::WAL_SEGMENT_SIZE;
19 : use safekeeper_api::models::{
20 : AcceptorStateStatus, PullTimelineRequest, SafekeeperStatus, SkTimelineInfo, TenantDeleteResult,
21 : TermSwitchApiEntry, TimelineCopyRequest, TimelineCreateRequest, TimelineDeleteResult,
22 : TimelineStatus, TimelineTermBumpRequest,
23 : };
24 : use safekeeper_api::{ServerInfo, membership, models};
25 : use storage_broker::proto::{SafekeeperTimelineInfo, TenantTimelineId as ProtoTenantTimelineId};
26 : use tokio::sync::mpsc;
27 : use tokio::task;
28 : use tokio_stream::wrappers::ReceiverStream;
29 : use tokio_util::sync::CancellationToken;
30 : use tracing::{Instrument, info_span};
31 : use utils::auth::SwappableJwtAuth;
32 : use utils::id::{TenantId, TenantTimelineId, TimelineId};
33 : use utils::lsn::Lsn;
34 :
35 : use crate::debug_dump::TimelineDigestRequest;
36 : use crate::safekeeper::TermLsn;
37 : use crate::timelines_global_map::DeleteOrExclude;
38 : use crate::{
39 : GlobalTimelines, SafeKeeperConf, copy_timeline, debug_dump, patch_control_file, pull_timeline,
40 : };
41 :
42 : /// Healthcheck handler.
43 0 : async fn status_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
44 0 : check_permission(&request, None)?;
45 0 : let conf = get_conf(&request);
46 0 : let status = SafekeeperStatus { id: conf.my_id };
47 0 : json_response(StatusCode::OK, status)
48 0 : }
49 :
50 0 : fn get_conf(request: &Request<Body>) -> &SafeKeeperConf {
51 0 : request
52 0 : .data::<Arc<SafeKeeperConf>>()
53 0 : .expect("unknown state type")
54 0 : .as_ref()
55 0 : }
56 :
57 0 : fn get_global_timelines(request: &Request<Body>) -> Arc<GlobalTimelines> {
58 0 : request
59 0 : .data::<Arc<GlobalTimelines>>()
60 0 : .expect("unknown state type")
61 0 : .clone()
62 0 : }
63 :
64 0 : fn check_permission(request: &Request<Body>, tenant_id: Option<TenantId>) -> Result<(), ApiError> {
65 0 : check_permission_with(request, |claims| {
66 0 : crate::auth::check_permission(claims, tenant_id)
67 0 : })
68 0 : }
69 :
70 : /// Deactivates all timelines for the tenant and removes its data directory.
71 : /// See `timeline_delete_handler`.
72 0 : async fn tenant_delete_handler(mut request: Request<Body>) -> Result<Response<Body>, ApiError> {
73 0 : let tenant_id = parse_request_param(&request, "tenant_id")?;
74 0 : let only_local = parse_query_param(&request, "only_local")?.unwrap_or(false);
75 0 : check_permission(&request, Some(tenant_id))?;
76 0 : ensure_no_body(&mut request).await?;
77 0 : let global_timelines = get_global_timelines(&request);
78 0 : let action = if only_local {
79 0 : DeleteOrExclude::DeleteLocal
80 : } else {
81 0 : DeleteOrExclude::Delete
82 : };
83 0 : let delete_info = global_timelines
84 0 : .delete_all_for_tenant(&tenant_id, action)
85 0 : .await
86 0 : .map_err(ApiError::InternalServerError)?;
87 0 : let response_body: TenantDeleteResult = delete_info
88 0 : .iter()
89 0 : .map(|(ttid, resp)| (format!("{}", ttid.timeline_id), *resp))
90 0 : .collect::<HashMap<String, TimelineDeleteResult>>();
91 0 : json_response(StatusCode::OK, response_body)
92 0 : }
93 :
94 0 : async fn timeline_create_handler(mut request: Request<Body>) -> Result<Response<Body>, ApiError> {
95 0 : let request_data: TimelineCreateRequest = json_request(&mut request).await?;
96 :
97 0 : let ttid = TenantTimelineId {
98 0 : tenant_id: request_data.tenant_id,
99 0 : timeline_id: request_data.timeline_id,
100 0 : };
101 0 : check_permission(&request, Some(ttid.tenant_id))?;
102 :
103 0 : let server_info = ServerInfo {
104 0 : pg_version: request_data.pg_version,
105 0 : system_id: request_data.system_id.unwrap_or(0),
106 0 : wal_seg_size: request_data.wal_seg_size.unwrap_or(WAL_SEGMENT_SIZE as u32),
107 0 : };
108 0 : let global_timelines = get_global_timelines(&request);
109 0 : global_timelines
110 0 : .create(
111 0 : ttid,
112 0 : request_data.mconf,
113 0 : server_info,
114 0 : request_data.start_lsn,
115 0 : request_data.commit_lsn.unwrap_or(request_data.start_lsn),
116 0 : )
117 0 : .await
118 0 : .map_err(ApiError::InternalServerError)?;
119 :
120 0 : json_response(StatusCode::OK, ())
121 0 : }
122 :
123 0 : async fn utilization_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
124 0 : check_permission(&request, None)?;
125 0 : let global_timelines = get_global_timelines(&request);
126 0 : let utilization = global_timelines.get_timeline_counts();
127 0 : json_response(StatusCode::OK, utilization)
128 0 : }
129 :
130 : /// List all (not deleted) timelines.
131 : /// Note: it is possible to do the same with debug_dump.
132 0 : async fn timeline_list_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
133 0 : check_permission(&request, None)?;
134 0 : let global_timelines = get_global_timelines(&request);
135 0 : let res: Vec<TenantTimelineId> = global_timelines
136 0 : .get_all()
137 0 : .iter()
138 0 : .map(|tli| tli.ttid)
139 0 : .collect();
140 0 : json_response(StatusCode::OK, res)
141 0 : }
142 :
143 : impl From<TermSwitchApiEntry> for TermLsn {
144 0 : fn from(api_val: TermSwitchApiEntry) -> Self {
145 0 : TermLsn {
146 0 : term: api_val.term,
147 0 : lsn: api_val.lsn,
148 0 : }
149 0 : }
150 : }
151 :
152 : /// Report info about timeline.
153 0 : async fn timeline_status_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
154 0 : let ttid = TenantTimelineId::new(
155 0 : parse_request_param(&request, "tenant_id")?,
156 0 : parse_request_param(&request, "timeline_id")?,
157 : );
158 0 : check_permission(&request, Some(ttid.tenant_id))?;
159 :
160 0 : let global_timelines = get_global_timelines(&request);
161 0 : let tli = global_timelines.get(ttid).map_err(ApiError::from)?;
162 0 : let (inmem, state) = tli.get_state().await;
163 0 : let flush_lsn = tli.get_flush_lsn().await;
164 :
165 0 : let last_log_term = state.acceptor_state.get_last_log_term(flush_lsn);
166 0 : let term_history = state
167 0 : .acceptor_state
168 0 : .term_history
169 0 : .0
170 0 : .into_iter()
171 0 : .map(|ts| TermSwitchApiEntry {
172 0 : term: ts.term,
173 0 : lsn: ts.lsn,
174 0 : })
175 0 : .collect();
176 0 : let acc_state = AcceptorStateStatus {
177 0 : term: state.acceptor_state.term,
178 0 : epoch: last_log_term,
179 0 : term_history,
180 0 : };
181 0 :
182 0 : let conf = get_conf(&request);
183 : // Note: we report in memory values which can be lost.
184 0 : let status = TimelineStatus {
185 0 : tenant_id: ttid.tenant_id,
186 0 : timeline_id: ttid.timeline_id,
187 0 : mconf: state.mconf,
188 0 : acceptor_state: acc_state,
189 0 : pg_info: state.server,
190 0 : flush_lsn,
191 0 : timeline_start_lsn: state.timeline_start_lsn,
192 0 : local_start_lsn: state.local_start_lsn,
193 0 : commit_lsn: inmem.commit_lsn,
194 0 : backup_lsn: inmem.backup_lsn,
195 0 : peer_horizon_lsn: inmem.peer_horizon_lsn,
196 0 : remote_consistent_lsn: inmem.remote_consistent_lsn,
197 0 : peers: tli.get_peers(conf).await,
198 0 : walsenders: tli.get_walsenders().get_all_public(),
199 0 : walreceivers: tli.get_walreceivers().get_all(),
200 0 : };
201 0 : json_response(StatusCode::OK, status)
202 0 : }
203 :
204 : /// Deactivates the timeline and removes its data directory.
205 0 : async fn timeline_delete_handler(mut request: Request<Body>) -> Result<Response<Body>, ApiError> {
206 0 : let ttid = TenantTimelineId::new(
207 0 : parse_request_param(&request, "tenant_id")?,
208 0 : parse_request_param(&request, "timeline_id")?,
209 : );
210 0 : let only_local = parse_query_param(&request, "only_local")?.unwrap_or(false);
211 0 : check_permission(&request, Some(ttid.tenant_id))?;
212 0 : ensure_no_body(&mut request).await?;
213 0 : let global_timelines = get_global_timelines(&request);
214 0 : let action = if only_local {
215 0 : DeleteOrExclude::DeleteLocal
216 : } else {
217 0 : DeleteOrExclude::Delete
218 : };
219 0 : let resp = global_timelines
220 0 : .delete_or_exclude(&ttid, action)
221 0 : .await
222 0 : .map_err(ApiError::from)?;
223 0 : json_response(StatusCode::OK, resp)
224 0 : }
225 :
226 : /// Pull timeline from peer safekeeper instances.
227 0 : async fn timeline_pull_handler(mut request: Request<Body>) -> Result<Response<Body>, ApiError> {
228 0 : check_permission(&request, None)?;
229 :
230 0 : let data: PullTimelineRequest = json_request(&mut request).await?;
231 0 : let conf = get_conf(&request);
232 0 : let global_timelines = get_global_timelines(&request);
233 :
234 0 : let ca_certs = conf
235 0 : .ssl_ca_certs
236 0 : .iter()
237 0 : .map(Pem::contents)
238 0 : .map(reqwest::Certificate::from_der)
239 0 : .collect::<Result<Vec<_>, _>>()
240 0 : .map_err(|e| {
241 0 : ApiError::InternalServerError(anyhow::anyhow!("failed to parse CA certs: {e}"))
242 0 : })?;
243 :
244 0 : let resp =
245 0 : pull_timeline::handle_request(data, conf.sk_auth_token.clone(), ca_certs, global_timelines)
246 0 : .await
247 0 : .map_err(ApiError::InternalServerError)?;
248 0 : json_response(StatusCode::OK, resp)
249 0 : }
250 :
251 : /// Stream tar archive with all timeline data.
252 0 : async fn timeline_snapshot_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
253 0 : let destination = parse_request_param(&request, "destination_id")?;
254 0 : let ttid = TenantTimelineId::new(
255 0 : parse_request_param(&request, "tenant_id")?,
256 0 : parse_request_param(&request, "timeline_id")?,
257 : );
258 0 : check_permission(&request, Some(ttid.tenant_id))?;
259 :
260 0 : let global_timelines = get_global_timelines(&request);
261 0 : let tli = global_timelines.get(ttid).map_err(ApiError::from)?;
262 :
263 : // To stream the body use wrap_stream which wants Stream of Result<Bytes>,
264 : // so create the chan and write to it in another task.
265 0 : let (tx, rx) = mpsc::channel(1);
266 0 :
267 0 : let conf = get_conf(&request);
268 0 : task::spawn(pull_timeline::stream_snapshot(
269 0 : tli,
270 0 : conf.my_id,
271 0 : destination,
272 0 : tx,
273 0 : ));
274 0 :
275 0 : let rx_stream = ReceiverStream::new(rx);
276 0 : let body = Body::wrap_stream(rx_stream);
277 0 :
278 0 : let response = Response::builder()
279 0 : .status(200)
280 0 : .header(hyper::header::CONTENT_TYPE, "application/octet-stream")
281 0 : .body(body)
282 0 : .unwrap();
283 0 :
284 0 : Ok(response)
285 0 : }
286 :
287 : /// Error type for delete_or_exclude: either generation conflict or something
288 : /// internal.
289 : #[derive(thiserror::Error, Debug)]
290 : pub enum DeleteOrExcludeError {
291 : #[error("refused to switch into excluding mconf {requested}, current: {current}")]
292 : Conflict {
293 : requested: membership::Configuration,
294 : current: membership::Configuration,
295 : },
296 : #[error(transparent)]
297 : Other(#[from] anyhow::Error),
298 : }
299 :
300 : /// Convert DeleteOrExcludeError to ApiError.
301 : impl From<DeleteOrExcludeError> for ApiError {
302 0 : fn from(de: DeleteOrExcludeError) -> ApiError {
303 0 : match de {
304 : DeleteOrExcludeError::Conflict {
305 : requested: _,
306 : current: _,
307 0 : } => ApiError::Conflict(de.to_string()),
308 0 : DeleteOrExcludeError::Other(e) => ApiError::InternalServerError(e),
309 : }
310 0 : }
311 : }
312 :
313 : /// Remove timeline locally after this node has been excluded from the
314 : /// membership configuration. The body is the same as in the membership endpoint
315 : /// -- conf where node is excluded -- and in principle single ep could be used
316 : /// for both actions, but since this is a data deletion op let's keep them
317 : /// separate.
318 0 : async fn timeline_exclude_handler(mut request: Request<Body>) -> Result<Response<Body>, ApiError> {
319 0 : let ttid = TenantTimelineId::new(
320 0 : parse_request_param(&request, "tenant_id")?,
321 0 : parse_request_param(&request, "timeline_id")?,
322 : );
323 0 : check_permission(&request, Some(ttid.tenant_id))?;
324 :
325 0 : let global_timelines = get_global_timelines(&request);
326 0 : let data: models::TimelineMembershipSwitchRequest = json_request(&mut request).await?;
327 0 : let my_id = get_conf(&request).my_id;
328 0 : // If request doesn't exclude us, membership switch endpoint should be used
329 0 : // instead.
330 0 : if data.mconf.contains(my_id) {
331 0 : return Err(ApiError::Forbidden(format!(
332 0 : "refused to switch into {}, node {} is member of it",
333 0 : data.mconf, my_id
334 0 : )));
335 0 : }
336 0 : let action = DeleteOrExclude::Exclude(data.mconf);
337 :
338 0 : let resp = global_timelines
339 0 : .delete_or_exclude(&ttid, action)
340 0 : .await
341 0 : .map_err(ApiError::from)?;
342 0 : json_response(StatusCode::OK, resp)
343 0 : }
344 :
345 : /// Consider switching timeline membership configuration to the provided one.
346 0 : async fn timeline_membership_handler(
347 0 : mut request: Request<Body>,
348 0 : ) -> Result<Response<Body>, ApiError> {
349 0 : let ttid = TenantTimelineId::new(
350 0 : parse_request_param(&request, "tenant_id")?,
351 0 : parse_request_param(&request, "timeline_id")?,
352 : );
353 0 : check_permission(&request, Some(ttid.tenant_id))?;
354 :
355 0 : let global_timelines = get_global_timelines(&request);
356 0 : let tli = global_timelines.get(ttid).map_err(ApiError::from)?;
357 :
358 0 : let data: models::TimelineMembershipSwitchRequest = json_request(&mut request).await?;
359 0 : let my_id = get_conf(&request).my_id;
360 0 : // If request excludes us, exclude endpoint should be used instead.
361 0 : if !data.mconf.contains(my_id) {
362 0 : return Err(ApiError::Forbidden(format!(
363 0 : "refused to switch into {}, node {} is not a member of it",
364 0 : data.mconf, my_id
365 0 : )));
366 0 : }
367 0 : let req_gen = data.mconf.generation;
368 0 : let response = tli
369 0 : .membership_switch(data.mconf)
370 0 : .await
371 0 : .map_err(ApiError::InternalServerError)?;
372 :
373 : // Return 409 if request was ignored.
374 0 : if req_gen == response.current_conf.generation {
375 0 : json_response(StatusCode::OK, response)
376 : } else {
377 0 : Err(ApiError::Conflict(format!(
378 0 : "request to switch into {} ignored, current generation {}",
379 0 : req_gen, response.current_conf.generation
380 0 : )))
381 : }
382 0 : }
383 :
384 0 : async fn timeline_copy_handler(mut request: Request<Body>) -> Result<Response<Body>, ApiError> {
385 0 : check_permission(&request, None)?;
386 :
387 0 : let request_data: TimelineCopyRequest = json_request(&mut request).await?;
388 0 : let source_ttid = TenantTimelineId::new(
389 0 : parse_request_param(&request, "tenant_id")?,
390 0 : parse_request_param(&request, "source_timeline_id")?,
391 : );
392 :
393 0 : let global_timelines = get_global_timelines(&request);
394 0 :
395 0 : copy_timeline::handle_request(copy_timeline::Request{
396 0 : source_ttid,
397 0 : until_lsn: request_data.until_lsn,
398 0 : destination_ttid: TenantTimelineId::new(source_ttid.tenant_id, request_data.target_timeline_id),
399 0 : }, global_timelines)
400 0 : .instrument(info_span!("copy_timeline", from=%source_ttid, to=%request_data.target_timeline_id, until_lsn=%request_data.until_lsn))
401 0 : .await
402 0 : .map_err(ApiError::InternalServerError)?;
403 :
404 0 : json_response(StatusCode::OK, ())
405 0 : }
406 :
407 0 : async fn patch_control_file_handler(
408 0 : mut request: Request<Body>,
409 0 : ) -> Result<Response<Body>, ApiError> {
410 0 : check_permission(&request, None)?;
411 :
412 0 : let ttid = TenantTimelineId::new(
413 0 : parse_request_param(&request, "tenant_id")?,
414 0 : parse_request_param(&request, "timeline_id")?,
415 : );
416 :
417 0 : let global_timelines = get_global_timelines(&request);
418 0 : let tli = global_timelines.get(ttid).map_err(ApiError::from)?;
419 :
420 0 : let patch_request: patch_control_file::Request = json_request(&mut request).await?;
421 0 : let response = patch_control_file::handle_request(tli, patch_request)
422 0 : .await
423 0 : .map_err(ApiError::InternalServerError)?;
424 :
425 0 : json_response(StatusCode::OK, response)
426 0 : }
427 :
428 : /// Force persist control file.
429 0 : async fn timeline_checkpoint_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
430 0 : check_permission(&request, None)?;
431 :
432 0 : let ttid = TenantTimelineId::new(
433 0 : parse_request_param(&request, "tenant_id")?,
434 0 : parse_request_param(&request, "timeline_id")?,
435 : );
436 :
437 0 : let global_timelines = get_global_timelines(&request);
438 0 : let tli = global_timelines.get(ttid)?;
439 0 : tli.write_shared_state()
440 0 : .await
441 : .sk
442 0 : .state_mut()
443 0 : .flush()
444 0 : .await
445 0 : .map_err(ApiError::InternalServerError)?;
446 0 : json_response(StatusCode::OK, ())
447 0 : }
448 :
449 0 : async fn timeline_digest_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
450 0 : let ttid = TenantTimelineId::new(
451 0 : parse_request_param(&request, "tenant_id")?,
452 0 : parse_request_param(&request, "timeline_id")?,
453 : );
454 0 : check_permission(&request, Some(ttid.tenant_id))?;
455 :
456 0 : let global_timelines = get_global_timelines(&request);
457 0 : let from_lsn: Option<Lsn> = parse_query_param(&request, "from_lsn")?;
458 0 : let until_lsn: Option<Lsn> = parse_query_param(&request, "until_lsn")?;
459 :
460 0 : let request = TimelineDigestRequest {
461 0 : from_lsn: from_lsn.ok_or(ApiError::BadRequest(anyhow::anyhow!(
462 0 : "from_lsn is required"
463 0 : )))?,
464 0 : until_lsn: until_lsn.ok_or(ApiError::BadRequest(anyhow::anyhow!(
465 0 : "until_lsn is required"
466 0 : )))?,
467 : };
468 :
469 0 : let tli = global_timelines.get(ttid).map_err(ApiError::from)?;
470 0 : let tli = tli
471 0 : .wal_residence_guard()
472 0 : .await
473 0 : .map_err(ApiError::InternalServerError)?;
474 :
475 0 : let response = debug_dump::calculate_digest(&tli, request)
476 0 : .await
477 0 : .map_err(ApiError::InternalServerError)?;
478 0 : json_response(StatusCode::OK, response)
479 0 : }
480 :
481 : /// Unevict timeline and remove uploaded partial segment(s) from the remote storage.
482 : /// Successfull response returns list of segments existed before the deletion.
483 : /// Aimed for one-off usage not normally needed.
484 0 : async fn timeline_backup_partial_reset(request: Request<Body>) -> Result<Response<Body>, ApiError> {
485 0 : let ttid = TenantTimelineId::new(
486 0 : parse_request_param(&request, "tenant_id")?,
487 0 : parse_request_param(&request, "timeline_id")?,
488 : );
489 0 : check_permission(&request, Some(ttid.tenant_id))?;
490 :
491 0 : let global_timelines = get_global_timelines(&request);
492 0 : let tli = global_timelines.get(ttid).map_err(ApiError::from)?;
493 :
494 0 : let response = tli
495 0 : .backup_partial_reset()
496 0 : .await
497 0 : .map_err(ApiError::InternalServerError)?;
498 0 : json_response(StatusCode::OK, response)
499 0 : }
500 :
501 : /// Make term at least as high as one in request. If one in request is None,
502 : /// increment current one.
503 0 : async fn timeline_term_bump_handler(
504 0 : mut request: Request<Body>,
505 0 : ) -> Result<Response<Body>, ApiError> {
506 0 : let ttid = TenantTimelineId::new(
507 0 : parse_request_param(&request, "tenant_id")?,
508 0 : parse_request_param(&request, "timeline_id")?,
509 : );
510 0 : check_permission(&request, Some(ttid.tenant_id))?;
511 :
512 0 : let request_data: TimelineTermBumpRequest = json_request(&mut request).await?;
513 :
514 0 : let global_timelines = get_global_timelines(&request);
515 0 : let tli = global_timelines.get(ttid).map_err(ApiError::from)?;
516 0 : let response = tli
517 0 : .term_bump(request_data.term)
518 0 : .await
519 0 : .map_err(ApiError::InternalServerError)?;
520 :
521 0 : json_response(StatusCode::OK, response)
522 0 : }
523 :
524 : /// Used only in tests to hand craft required data.
525 0 : async fn record_safekeeper_info(mut request: Request<Body>) -> Result<Response<Body>, ApiError> {
526 0 : let ttid = TenantTimelineId::new(
527 0 : parse_request_param(&request, "tenant_id")?,
528 0 : parse_request_param(&request, "timeline_id")?,
529 : );
530 0 : check_permission(&request, Some(ttid.tenant_id))?;
531 0 : let sk_info: SkTimelineInfo = json_request(&mut request).await?;
532 0 : let proto_sk_info = SafekeeperTimelineInfo {
533 0 : safekeeper_id: 0,
534 0 : tenant_timeline_id: Some(ProtoTenantTimelineId {
535 0 : tenant_id: ttid.tenant_id.as_ref().to_owned(),
536 0 : timeline_id: ttid.timeline_id.as_ref().to_owned(),
537 0 : }),
538 0 : term: sk_info.term.unwrap_or(0),
539 0 : last_log_term: sk_info.last_log_term.unwrap_or(0),
540 0 : flush_lsn: sk_info.flush_lsn.0,
541 0 : commit_lsn: sk_info.commit_lsn.0,
542 0 : remote_consistent_lsn: sk_info.remote_consistent_lsn.0,
543 0 : peer_horizon_lsn: sk_info.peer_horizon_lsn.0,
544 0 : safekeeper_connstr: sk_info.safekeeper_connstr.unwrap_or_else(|| "".to_owned()),
545 0 : http_connstr: sk_info.http_connstr.unwrap_or_else(|| "".to_owned()),
546 0 : https_connstr: sk_info.https_connstr,
547 0 : backup_lsn: sk_info.backup_lsn.0,
548 0 : local_start_lsn: sk_info.local_start_lsn.0,
549 0 : availability_zone: None,
550 0 : standby_horizon: sk_info.standby_horizon.0,
551 0 : };
552 0 :
553 0 : let global_timelines = get_global_timelines(&request);
554 0 : let tli = global_timelines.get(ttid).map_err(ApiError::from)?;
555 0 : tli.record_safekeeper_info(proto_sk_info)
556 0 : .await
557 0 : .map_err(ApiError::InternalServerError)?;
558 :
559 0 : json_response(StatusCode::OK, ())
560 0 : }
561 :
562 0 : fn parse_kv_str<E: fmt::Display, T: FromStr<Err = E>>(k: &str, v: &str) -> Result<T, ApiError> {
563 0 : v.parse()
564 0 : .map_err(|e| ApiError::BadRequest(anyhow::anyhow!("cannot parse {k}: {e}")))
565 0 : }
566 :
567 : /// Dump debug info about all available safekeeper state.
568 0 : async fn dump_debug_handler(mut request: Request<Body>) -> Result<Response<Body>, ApiError> {
569 0 : check_permission(&request, None)?;
570 0 : ensure_no_body(&mut request).await?;
571 :
572 0 : let mut dump_all: Option<bool> = None;
573 0 : let mut dump_control_file: Option<bool> = None;
574 0 : let mut dump_memory: Option<bool> = None;
575 0 : let mut dump_disk_content: Option<bool> = None;
576 0 : let mut dump_term_history: Option<bool> = None;
577 0 : let mut dump_wal_last_modified: Option<bool> = None;
578 0 : let mut tenant_id: Option<TenantId> = None;
579 0 : let mut timeline_id: Option<TimelineId> = None;
580 0 :
581 0 : let query = request.uri().query().unwrap_or("");
582 0 : let mut values = url::form_urlencoded::parse(query.as_bytes());
583 :
584 0 : for (k, v) in &mut values {
585 0 : match k.as_ref() {
586 0 : "dump_all" => dump_all = Some(parse_kv_str(&k, &v)?),
587 0 : "dump_control_file" => dump_control_file = Some(parse_kv_str(&k, &v)?),
588 0 : "dump_memory" => dump_memory = Some(parse_kv_str(&k, &v)?),
589 0 : "dump_disk_content" => dump_disk_content = Some(parse_kv_str(&k, &v)?),
590 0 : "dump_term_history" => dump_term_history = Some(parse_kv_str(&k, &v)?),
591 0 : "dump_wal_last_modified" => dump_wal_last_modified = Some(parse_kv_str(&k, &v)?),
592 0 : "tenant_id" => tenant_id = Some(parse_kv_str(&k, &v)?),
593 0 : "timeline_id" => timeline_id = Some(parse_kv_str(&k, &v)?),
594 0 : _ => Err(ApiError::BadRequest(anyhow::anyhow!(
595 0 : "Unknown query parameter: {}",
596 0 : k
597 0 : )))?,
598 : }
599 : }
600 :
601 0 : let dump_all = dump_all.unwrap_or(false);
602 0 : let dump_control_file = dump_control_file.unwrap_or(dump_all);
603 0 : let dump_memory = dump_memory.unwrap_or(dump_all);
604 0 : let dump_disk_content = dump_disk_content.unwrap_or(dump_all);
605 0 : let dump_term_history = dump_term_history.unwrap_or(true);
606 0 : let dump_wal_last_modified = dump_wal_last_modified.unwrap_or(dump_all);
607 0 :
608 0 : let global_timelines = get_global_timelines(&request);
609 0 :
610 0 : let args = debug_dump::Args {
611 0 : dump_all,
612 0 : dump_control_file,
613 0 : dump_memory,
614 0 : dump_disk_content,
615 0 : dump_term_history,
616 0 : dump_wal_last_modified,
617 0 : tenant_id,
618 0 : timeline_id,
619 0 : };
620 :
621 0 : let resp = debug_dump::build(args, global_timelines)
622 0 : .await
623 0 : .map_err(ApiError::InternalServerError)?;
624 :
625 0 : let started_at = std::time::Instant::now();
626 0 :
627 0 : let (tx, rx) = mpsc::channel(1);
628 0 :
629 0 : let body = Body::wrap_stream(ReceiverStream::new(rx));
630 0 :
631 0 : let mut writer = ChannelWriter::new(128 * 1024, tx);
632 0 :
633 0 : let response = Response::builder()
634 0 : .status(200)
635 0 : .header(hyper::header::CONTENT_TYPE, "application/octet-stream")
636 0 : .body(body)
637 0 : .unwrap();
638 :
639 0 : let span = info_span!("blocking");
640 0 : tokio::task::spawn_blocking(move || {
641 0 : let _span = span.entered();
642 0 :
643 0 : let res = serde_json::to_writer(&mut writer, &resp)
644 0 : .map_err(std::io::Error::from)
645 0 : .and_then(|_| writer.flush());
646 0 :
647 0 : match res {
648 : Ok(()) => {
649 0 : tracing::info!(
650 0 : bytes = writer.flushed_bytes(),
651 0 : elapsed_ms = started_at.elapsed().as_millis(),
652 0 : "responded /v1/debug_dump"
653 : );
654 : }
655 0 : Err(e) => {
656 0 : tracing::warn!("failed to write out /v1/debug_dump response: {e:#}");
657 : // semantics of this error are quite... unclear. we want to error the stream out to
658 : // abort the response to somehow notify the client that we failed.
659 : //
660 : // though, most likely the reason for failure is that the receiver is already gone.
661 0 : drop(
662 0 : writer
663 0 : .tx
664 0 : .blocking_send(Err(std::io::ErrorKind::BrokenPipe.into())),
665 0 : );
666 : }
667 : }
668 0 : });
669 0 :
670 0 : Ok(response)
671 0 : }
672 :
673 : /// Safekeeper http router.
674 0 : pub fn make_router(
675 0 : conf: Arc<SafeKeeperConf>,
676 0 : global_timelines: Arc<GlobalTimelines>,
677 0 : ) -> RouterBuilder<hyper::Body, ApiError> {
678 0 : let mut router = endpoint::make_router();
679 0 : if conf.http_auth.is_some() {
680 0 : router = router.middleware(auth_middleware(|request| {
681 : const ALLOWLIST_ROUTES: &[&str] =
682 : &["/v1/status", "/metrics", "/profile/cpu", "/profile/heap"];
683 0 : if ALLOWLIST_ROUTES.contains(&request.uri().path()) {
684 0 : None
685 : } else {
686 : // Option<Arc<SwappableJwtAuth>> is always provided as data below, hence unwrap().
687 0 : request
688 0 : .data::<Option<Arc<SwappableJwtAuth>>>()
689 0 : .unwrap()
690 0 : .as_deref()
691 : }
692 0 : }))
693 0 : }
694 :
695 : // NB: on any changes do not forget to update the OpenAPI spec
696 : // located nearby (/safekeeper/src/http/openapi_spec.yaml).
697 0 : let auth = conf.http_auth.clone();
698 0 : router
699 0 : .data(conf)
700 0 : .data(global_timelines)
701 0 : .data(auth)
702 0 : .get("/metrics", |r| request_span(r, prometheus_metrics_handler))
703 0 : .get("/profile/cpu", |r| request_span(r, profile_cpu_handler))
704 0 : .get("/profile/heap", |r| request_span(r, profile_heap_handler))
705 0 : .get("/v1/status", |r| request_span(r, status_handler))
706 0 : .put("/v1/failpoints", |r| {
707 0 : request_span(r, move |r| async {
708 0 : check_permission(&r, None)?;
709 0 : let cancel = CancellationToken::new();
710 0 : failpoints_handler(r, cancel).await
711 0 : })
712 0 : })
713 0 : .get("/v1/utilization", |r| request_span(r, utilization_handler))
714 0 : .delete("/v1/tenant/:tenant_id", |r| {
715 0 : request_span(r, tenant_delete_handler)
716 0 : })
717 0 : // Will be used in the future instead of implicit timeline creation
718 0 : .post("/v1/tenant/timeline", |r| {
719 0 : request_span(r, timeline_create_handler)
720 0 : })
721 0 : .get("/v1/tenant/timeline", |r| {
722 0 : request_span(r, timeline_list_handler)
723 0 : })
724 0 : .get("/v1/tenant/:tenant_id/timeline/:timeline_id", |r| {
725 0 : request_span(r, timeline_status_handler)
726 0 : })
727 0 : .delete("/v1/tenant/:tenant_id/timeline/:timeline_id", |r| {
728 0 : request_span(r, timeline_delete_handler)
729 0 : })
730 0 : .post("/v1/pull_timeline", |r| {
731 0 : request_span(r, timeline_pull_handler)
732 0 : })
733 0 : .put("/v1/tenant/:tenant_id/timeline/:timeline_id/exclude", |r| {
734 0 : request_span(r, timeline_exclude_handler)
735 0 : })
736 0 : .get(
737 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/snapshot/:destination_id",
738 0 : |r| request_span(r, timeline_snapshot_handler),
739 0 : )
740 0 : .put(
741 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/membership",
742 0 : |r| request_span(r, timeline_membership_handler),
743 0 : )
744 0 : .post(
745 0 : "/v1/tenant/:tenant_id/timeline/:source_timeline_id/copy",
746 0 : |r| request_span(r, timeline_copy_handler),
747 0 : )
748 0 : .patch(
749 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/control_file",
750 0 : |r| request_span(r, patch_control_file_handler),
751 0 : )
752 0 : .post(
753 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/checkpoint",
754 0 : |r| request_span(r, timeline_checkpoint_handler),
755 0 : )
756 0 : .get("/v1/tenant/:tenant_id/timeline/:timeline_id/digest", |r| {
757 0 : request_span(r, timeline_digest_handler)
758 0 : })
759 0 : .post(
760 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/backup_partial_reset",
761 0 : |r| request_span(r, timeline_backup_partial_reset),
762 0 : )
763 0 : .post(
764 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/term_bump",
765 0 : |r| request_span(r, timeline_term_bump_handler),
766 0 : )
767 0 : .post("/v1/record_safekeeper_info/:tenant_id/:timeline_id", |r| {
768 0 : request_span(r, record_safekeeper_info)
769 0 : })
770 0 : .get("/v1/debug_dump", |r| request_span(r, dump_debug_handler))
771 0 : }
772 :
773 : #[cfg(test)]
774 : mod tests {
775 : use super::*;
776 :
777 : #[test]
778 1 : fn test_term_switch_entry_api_serialize() {
779 1 : let state = AcceptorStateStatus {
780 1 : term: 1,
781 1 : epoch: 1,
782 1 : term_history: vec![TermSwitchApiEntry {
783 1 : term: 1,
784 1 : lsn: Lsn(0x16FFDDDD),
785 1 : }],
786 1 : };
787 1 : let json = serde_json::to_string(&state).unwrap();
788 1 : assert_eq!(
789 1 : json,
790 1 : "{\"term\":1,\"epoch\":1,\"term_history\":[{\"term\":1,\"lsn\":\"0/16FFDDDD\"}]}"
791 1 : );
792 1 : }
793 : }
|
