Line data Source code
1 : use hyper::{Body, Request, Response, StatusCode};
2 : use safekeeper_api::models;
3 : use safekeeper_api::models::AcceptorStateStatus;
4 : use safekeeper_api::models::SafekeeperStatus;
5 : use safekeeper_api::models::TermSwitchApiEntry;
6 : use safekeeper_api::models::TimelineStatus;
7 : use safekeeper_api::ServerInfo;
8 : use std::collections::HashMap;
9 : use std::fmt;
10 : use std::io::Write as _;
11 : use std::str::FromStr;
12 : use std::sync::Arc;
13 : use storage_broker::proto::SafekeeperTimelineInfo;
14 : use storage_broker::proto::TenantTimelineId as ProtoTenantTimelineId;
15 : use tokio::sync::mpsc;
16 : use tokio::task;
17 : use tokio_stream::wrappers::ReceiverStream;
18 : use tokio_util::sync::CancellationToken;
19 : use tracing::{info_span, Instrument};
20 : use utils::failpoint_support::failpoints_handler;
21 : use utils::http::endpoint::{
22 : profile_cpu_handler, profile_heap_handler, prometheus_metrics_handler, request_span,
23 : ChannelWriter,
24 : };
25 : use utils::http::request::parse_query_param;
26 :
27 : use postgres_ffi::WAL_SEGMENT_SIZE;
28 : use safekeeper_api::models::{SkTimelineInfo, TimelineCopyRequest};
29 : use safekeeper_api::models::{TimelineCreateRequest, TimelineTermBumpRequest};
30 : use utils::{
31 : auth::SwappableJwtAuth,
32 : http::{
33 : endpoint::{self, auth_middleware, check_permission_with},
34 : error::ApiError,
35 : json::{json_request, json_response},
36 : request::{ensure_no_body, parse_request_param},
37 : RequestExt, RouterBuilder,
38 : },
39 : id::{TenantId, TenantTimelineId, TimelineId},
40 : lsn::Lsn,
41 : };
42 :
43 : use crate::debug_dump::TimelineDigestRequest;
44 : use crate::safekeeper::TermLsn;
45 : use crate::timelines_global_map::TimelineDeleteForceResult;
46 : use crate::GlobalTimelines;
47 : use crate::SafeKeeperConf;
48 : use crate::{copy_timeline, debug_dump, patch_control_file, pull_timeline};
49 :
50 : /// Healthcheck handler.
51 0 : async fn status_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
52 0 : check_permission(&request, None)?;
53 0 : let conf = get_conf(&request);
54 0 : let status = SafekeeperStatus { id: conf.my_id };
55 0 : json_response(StatusCode::OK, status)
56 0 : }
57 :
58 0 : fn get_conf(request: &Request<Body>) -> &SafeKeeperConf {
59 0 : request
60 0 : .data::<Arc<SafeKeeperConf>>()
61 0 : .expect("unknown state type")
62 0 : .as_ref()
63 0 : }
64 :
65 0 : fn get_global_timelines(request: &Request<Body>) -> Arc<GlobalTimelines> {
66 0 : request
67 0 : .data::<Arc<GlobalTimelines>>()
68 0 : .expect("unknown state type")
69 0 : .clone()
70 0 : }
71 :
72 0 : fn check_permission(request: &Request<Body>, tenant_id: Option<TenantId>) -> Result<(), ApiError> {
73 0 : check_permission_with(request, |claims| {
74 0 : crate::auth::check_permission(claims, tenant_id)
75 0 : })
76 0 : }
77 :
78 : /// Deactivates all timelines for the tenant and removes its data directory.
79 : /// See `timeline_delete_handler`.
80 0 : async fn tenant_delete_handler(mut request: Request<Body>) -> Result<Response<Body>, ApiError> {
81 0 : let tenant_id = parse_request_param(&request, "tenant_id")?;
82 0 : let only_local = parse_query_param(&request, "only_local")?.unwrap_or(false);
83 0 : check_permission(&request, Some(tenant_id))?;
84 0 : ensure_no_body(&mut request).await?;
85 0 : let global_timelines = get_global_timelines(&request);
86 : // FIXME: `delete_force_all_for_tenant` can return an error for multiple different reasons;
87 : // Using an `InternalServerError` should be fixed when the types support it
88 0 : let delete_info = global_timelines
89 0 : .delete_force_all_for_tenant(&tenant_id, only_local)
90 0 : .await
91 0 : .map_err(ApiError::InternalServerError)?;
92 0 : json_response(
93 0 : StatusCode::OK,
94 0 : delete_info
95 0 : .iter()
96 0 : .map(|(ttid, resp)| (format!("{}", ttid.timeline_id), *resp))
97 0 : .collect::<HashMap<String, TimelineDeleteForceResult>>(),
98 0 : )
99 0 : }
100 :
101 0 : async fn timeline_create_handler(mut request: Request<Body>) -> Result<Response<Body>, ApiError> {
102 0 : let request_data: TimelineCreateRequest = json_request(&mut request).await?;
103 :
104 0 : let ttid = TenantTimelineId {
105 0 : tenant_id: request_data.tenant_id,
106 0 : timeline_id: request_data.timeline_id,
107 0 : };
108 0 : check_permission(&request, Some(ttid.tenant_id))?;
109 :
110 0 : let server_info = ServerInfo {
111 0 : pg_version: request_data.pg_version,
112 0 : system_id: request_data.system_id.unwrap_or(0),
113 0 : wal_seg_size: request_data.wal_seg_size.unwrap_or(WAL_SEGMENT_SIZE as u32),
114 0 : };
115 0 : let global_timelines = get_global_timelines(&request);
116 0 : global_timelines
117 0 : .create(
118 0 : ttid,
119 0 : request_data.mconf,
120 0 : server_info,
121 0 : request_data.start_lsn,
122 0 : request_data.commit_lsn.unwrap_or(request_data.start_lsn),
123 0 : )
124 0 : .await
125 0 : .map_err(ApiError::InternalServerError)?;
126 :
127 0 : json_response(StatusCode::OK, ())
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 : // FIXME: `delete_force` can fail from both internal errors and bad requests. Add better
215 : // error handling here when we're able to.
216 0 : let resp = global_timelines
217 0 : .delete(&ttid, only_local)
218 0 : .await
219 0 : .map_err(ApiError::InternalServerError)?;
220 0 : json_response(StatusCode::OK, resp)
221 0 : }
222 :
223 : /// Pull timeline from peer safekeeper instances.
224 0 : async fn timeline_pull_handler(mut request: Request<Body>) -> Result<Response<Body>, ApiError> {
225 0 : check_permission(&request, None)?;
226 :
227 0 : let data: pull_timeline::Request = json_request(&mut request).await?;
228 0 : let conf = get_conf(&request);
229 0 : let global_timelines = get_global_timelines(&request);
230 :
231 0 : let resp = pull_timeline::handle_request(data, conf.sk_auth_token.clone(), global_timelines)
232 0 : .await
233 0 : .map_err(ApiError::InternalServerError)?;
234 0 : json_response(StatusCode::OK, resp)
235 0 : }
236 :
237 : /// Stream tar archive with all timeline data.
238 0 : async fn timeline_snapshot_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
239 0 : let destination = parse_request_param(&request, "destination_id")?;
240 0 : let ttid = TenantTimelineId::new(
241 0 : parse_request_param(&request, "tenant_id")?,
242 0 : parse_request_param(&request, "timeline_id")?,
243 : );
244 0 : check_permission(&request, Some(ttid.tenant_id))?;
245 :
246 0 : let global_timelines = get_global_timelines(&request);
247 0 : let tli = global_timelines.get(ttid).map_err(ApiError::from)?;
248 :
249 : // To stream the body use wrap_stream which wants Stream of Result<Bytes>,
250 : // so create the chan and write to it in another task.
251 0 : let (tx, rx) = mpsc::channel(1);
252 0 :
253 0 : let conf = get_conf(&request);
254 0 : task::spawn(pull_timeline::stream_snapshot(
255 0 : tli,
256 0 : conf.my_id,
257 0 : destination,
258 0 : tx,
259 0 : ));
260 0 :
261 0 : let rx_stream = ReceiverStream::new(rx);
262 0 : let body = Body::wrap_stream(rx_stream);
263 0 :
264 0 : let response = Response::builder()
265 0 : .status(200)
266 0 : .header(hyper::header::CONTENT_TYPE, "application/octet-stream")
267 0 : .body(body)
268 0 : .unwrap();
269 0 :
270 0 : Ok(response)
271 0 : }
272 :
273 : /// Consider switching timeline membership configuration to the provided one.
274 0 : async fn timeline_membership_handler(
275 0 : mut request: Request<Body>,
276 0 : ) -> Result<Response<Body>, ApiError> {
277 0 : let ttid = TenantTimelineId::new(
278 0 : parse_request_param(&request, "tenant_id")?,
279 0 : parse_request_param(&request, "timeline_id")?,
280 : );
281 0 : check_permission(&request, Some(ttid.tenant_id))?;
282 :
283 0 : let global_timelines = get_global_timelines(&request);
284 0 : let tli = global_timelines.get(ttid).map_err(ApiError::from)?;
285 :
286 0 : let data: models::TimelineMembershipSwitchRequest = json_request(&mut request).await?;
287 0 : let response = tli
288 0 : .membership_switch(data.mconf)
289 0 : .await
290 0 : .map_err(ApiError::InternalServerError)?;
291 :
292 0 : json_response(StatusCode::OK, response)
293 0 : }
294 :
295 0 : async fn timeline_copy_handler(mut request: Request<Body>) -> Result<Response<Body>, ApiError> {
296 0 : check_permission(&request, None)?;
297 :
298 0 : let request_data: TimelineCopyRequest = json_request(&mut request).await?;
299 0 : let source_ttid = TenantTimelineId::new(
300 0 : parse_request_param(&request, "tenant_id")?,
301 0 : parse_request_param(&request, "source_timeline_id")?,
302 : );
303 :
304 0 : let global_timelines = get_global_timelines(&request);
305 0 :
306 0 : copy_timeline::handle_request(copy_timeline::Request{
307 0 : source_ttid,
308 0 : until_lsn: request_data.until_lsn,
309 0 : destination_ttid: TenantTimelineId::new(source_ttid.tenant_id, request_data.target_timeline_id),
310 0 : }, global_timelines)
311 0 : .instrument(info_span!("copy_timeline", from=%source_ttid, to=%request_data.target_timeline_id, until_lsn=%request_data.until_lsn))
312 0 : .await
313 0 : .map_err(ApiError::InternalServerError)?;
314 :
315 0 : json_response(StatusCode::OK, ())
316 0 : }
317 :
318 0 : async fn patch_control_file_handler(
319 0 : mut request: Request<Body>,
320 0 : ) -> Result<Response<Body>, ApiError> {
321 0 : check_permission(&request, None)?;
322 :
323 0 : let ttid = TenantTimelineId::new(
324 0 : parse_request_param(&request, "tenant_id")?,
325 0 : parse_request_param(&request, "timeline_id")?,
326 : );
327 :
328 0 : let global_timelines = get_global_timelines(&request);
329 0 : let tli = global_timelines.get(ttid).map_err(ApiError::from)?;
330 :
331 0 : let patch_request: patch_control_file::Request = json_request(&mut request).await?;
332 0 : let response = patch_control_file::handle_request(tli, patch_request)
333 0 : .await
334 0 : .map_err(ApiError::InternalServerError)?;
335 :
336 0 : json_response(StatusCode::OK, response)
337 0 : }
338 :
339 : /// Force persist control file.
340 0 : async fn timeline_checkpoint_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
341 0 : check_permission(&request, None)?;
342 :
343 0 : let ttid = TenantTimelineId::new(
344 0 : parse_request_param(&request, "tenant_id")?,
345 0 : parse_request_param(&request, "timeline_id")?,
346 : );
347 :
348 0 : let global_timelines = get_global_timelines(&request);
349 0 : let tli = global_timelines.get(ttid)?;
350 0 : tli.write_shared_state()
351 0 : .await
352 : .sk
353 0 : .state_mut()
354 0 : .flush()
355 0 : .await
356 0 : .map_err(ApiError::InternalServerError)?;
357 0 : json_response(StatusCode::OK, ())
358 0 : }
359 :
360 0 : async fn timeline_digest_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
361 0 : let ttid = TenantTimelineId::new(
362 0 : parse_request_param(&request, "tenant_id")?,
363 0 : parse_request_param(&request, "timeline_id")?,
364 : );
365 0 : check_permission(&request, Some(ttid.tenant_id))?;
366 :
367 0 : let global_timelines = get_global_timelines(&request);
368 0 : let from_lsn: Option<Lsn> = parse_query_param(&request, "from_lsn")?;
369 0 : let until_lsn: Option<Lsn> = parse_query_param(&request, "until_lsn")?;
370 :
371 0 : let request = TimelineDigestRequest {
372 0 : from_lsn: from_lsn.ok_or(ApiError::BadRequest(anyhow::anyhow!(
373 0 : "from_lsn is required"
374 0 : )))?,
375 0 : until_lsn: until_lsn.ok_or(ApiError::BadRequest(anyhow::anyhow!(
376 0 : "until_lsn is required"
377 0 : )))?,
378 : };
379 :
380 0 : let tli = global_timelines.get(ttid).map_err(ApiError::from)?;
381 0 : let tli = tli
382 0 : .wal_residence_guard()
383 0 : .await
384 0 : .map_err(ApiError::InternalServerError)?;
385 :
386 0 : let response = debug_dump::calculate_digest(&tli, request)
387 0 : .await
388 0 : .map_err(ApiError::InternalServerError)?;
389 0 : json_response(StatusCode::OK, response)
390 0 : }
391 :
392 : /// Unevict timeline and remove uploaded partial segment(s) from the remote storage.
393 : /// Successfull response returns list of segments existed before the deletion.
394 : /// Aimed for one-off usage not normally needed.
395 0 : async fn timeline_backup_partial_reset(request: Request<Body>) -> Result<Response<Body>, ApiError> {
396 0 : let ttid = TenantTimelineId::new(
397 0 : parse_request_param(&request, "tenant_id")?,
398 0 : parse_request_param(&request, "timeline_id")?,
399 : );
400 0 : check_permission(&request, Some(ttid.tenant_id))?;
401 :
402 0 : let global_timelines = get_global_timelines(&request);
403 0 : let tli = global_timelines.get(ttid).map_err(ApiError::from)?;
404 :
405 0 : let response = tli
406 0 : .backup_partial_reset()
407 0 : .await
408 0 : .map_err(ApiError::InternalServerError)?;
409 0 : json_response(StatusCode::OK, response)
410 0 : }
411 :
412 : /// Make term at least as high as one in request. If one in request is None,
413 : /// increment current one.
414 0 : async fn timeline_term_bump_handler(
415 0 : mut request: Request<Body>,
416 0 : ) -> Result<Response<Body>, ApiError> {
417 0 : let ttid = TenantTimelineId::new(
418 0 : parse_request_param(&request, "tenant_id")?,
419 0 : parse_request_param(&request, "timeline_id")?,
420 : );
421 0 : check_permission(&request, Some(ttid.tenant_id))?;
422 :
423 0 : let request_data: TimelineTermBumpRequest = json_request(&mut request).await?;
424 :
425 0 : let global_timelines = get_global_timelines(&request);
426 0 : let tli = global_timelines.get(ttid).map_err(ApiError::from)?;
427 0 : let response = tli
428 0 : .term_bump(request_data.term)
429 0 : .await
430 0 : .map_err(ApiError::InternalServerError)?;
431 :
432 0 : json_response(StatusCode::OK, response)
433 0 : }
434 :
435 : /// Used only in tests to hand craft required data.
436 0 : async fn record_safekeeper_info(mut request: Request<Body>) -> Result<Response<Body>, ApiError> {
437 0 : let ttid = TenantTimelineId::new(
438 0 : parse_request_param(&request, "tenant_id")?,
439 0 : parse_request_param(&request, "timeline_id")?,
440 : );
441 0 : check_permission(&request, Some(ttid.tenant_id))?;
442 0 : let sk_info: SkTimelineInfo = json_request(&mut request).await?;
443 0 : let proto_sk_info = SafekeeperTimelineInfo {
444 0 : safekeeper_id: 0,
445 0 : tenant_timeline_id: Some(ProtoTenantTimelineId {
446 0 : tenant_id: ttid.tenant_id.as_ref().to_owned(),
447 0 : timeline_id: ttid.timeline_id.as_ref().to_owned(),
448 0 : }),
449 0 : term: sk_info.term.unwrap_or(0),
450 0 : last_log_term: sk_info.last_log_term.unwrap_or(0),
451 0 : flush_lsn: sk_info.flush_lsn.0,
452 0 : commit_lsn: sk_info.commit_lsn.0,
453 0 : remote_consistent_lsn: sk_info.remote_consistent_lsn.0,
454 0 : peer_horizon_lsn: sk_info.peer_horizon_lsn.0,
455 0 : safekeeper_connstr: sk_info.safekeeper_connstr.unwrap_or_else(|| "".to_owned()),
456 0 : http_connstr: sk_info.http_connstr.unwrap_or_else(|| "".to_owned()),
457 0 : backup_lsn: sk_info.backup_lsn.0,
458 0 : local_start_lsn: sk_info.local_start_lsn.0,
459 0 : availability_zone: None,
460 0 : standby_horizon: sk_info.standby_horizon.0,
461 0 : };
462 0 :
463 0 : let global_timelines = get_global_timelines(&request);
464 0 : let tli = global_timelines.get(ttid).map_err(ApiError::from)?;
465 0 : tli.record_safekeeper_info(proto_sk_info)
466 0 : .await
467 0 : .map_err(ApiError::InternalServerError)?;
468 :
469 0 : json_response(StatusCode::OK, ())
470 0 : }
471 :
472 0 : fn parse_kv_str<E: fmt::Display, T: FromStr<Err = E>>(k: &str, v: &str) -> Result<T, ApiError> {
473 0 : v.parse()
474 0 : .map_err(|e| ApiError::BadRequest(anyhow::anyhow!("cannot parse {k}: {e}")))
475 0 : }
476 :
477 : /// Dump debug info about all available safekeeper state.
478 0 : async fn dump_debug_handler(mut request: Request<Body>) -> Result<Response<Body>, ApiError> {
479 0 : check_permission(&request, None)?;
480 0 : ensure_no_body(&mut request).await?;
481 :
482 0 : let mut dump_all: Option<bool> = None;
483 0 : let mut dump_control_file: Option<bool> = None;
484 0 : let mut dump_memory: Option<bool> = None;
485 0 : let mut dump_disk_content: Option<bool> = None;
486 0 : let mut dump_term_history: Option<bool> = None;
487 0 : let mut dump_wal_last_modified: Option<bool> = None;
488 0 : let mut tenant_id: Option<TenantId> = None;
489 0 : let mut timeline_id: Option<TimelineId> = None;
490 0 :
491 0 : let query = request.uri().query().unwrap_or("");
492 0 : let mut values = url::form_urlencoded::parse(query.as_bytes());
493 :
494 0 : for (k, v) in &mut values {
495 0 : match k.as_ref() {
496 0 : "dump_all" => dump_all = Some(parse_kv_str(&k, &v)?),
497 0 : "dump_control_file" => dump_control_file = Some(parse_kv_str(&k, &v)?),
498 0 : "dump_memory" => dump_memory = Some(parse_kv_str(&k, &v)?),
499 0 : "dump_disk_content" => dump_disk_content = Some(parse_kv_str(&k, &v)?),
500 0 : "dump_term_history" => dump_term_history = Some(parse_kv_str(&k, &v)?),
501 0 : "dump_wal_last_modified" => dump_wal_last_modified = Some(parse_kv_str(&k, &v)?),
502 0 : "tenant_id" => tenant_id = Some(parse_kv_str(&k, &v)?),
503 0 : "timeline_id" => timeline_id = Some(parse_kv_str(&k, &v)?),
504 0 : _ => Err(ApiError::BadRequest(anyhow::anyhow!(
505 0 : "Unknown query parameter: {}",
506 0 : k
507 0 : )))?,
508 : }
509 : }
510 :
511 0 : let dump_all = dump_all.unwrap_or(false);
512 0 : let dump_control_file = dump_control_file.unwrap_or(dump_all);
513 0 : let dump_memory = dump_memory.unwrap_or(dump_all);
514 0 : let dump_disk_content = dump_disk_content.unwrap_or(dump_all);
515 0 : let dump_term_history = dump_term_history.unwrap_or(true);
516 0 : let dump_wal_last_modified = dump_wal_last_modified.unwrap_or(dump_all);
517 0 :
518 0 : let global_timelines = get_global_timelines(&request);
519 0 :
520 0 : let args = debug_dump::Args {
521 0 : dump_all,
522 0 : dump_control_file,
523 0 : dump_memory,
524 0 : dump_disk_content,
525 0 : dump_term_history,
526 0 : dump_wal_last_modified,
527 0 : tenant_id,
528 0 : timeline_id,
529 0 : };
530 :
531 0 : let resp = debug_dump::build(args, global_timelines)
532 0 : .await
533 0 : .map_err(ApiError::InternalServerError)?;
534 :
535 0 : let started_at = std::time::Instant::now();
536 0 :
537 0 : let (tx, rx) = mpsc::channel(1);
538 0 :
539 0 : let body = Body::wrap_stream(ReceiverStream::new(rx));
540 0 :
541 0 : let mut writer = ChannelWriter::new(128 * 1024, tx);
542 0 :
543 0 : let response = Response::builder()
544 0 : .status(200)
545 0 : .header(hyper::header::CONTENT_TYPE, "application/octet-stream")
546 0 : .body(body)
547 0 : .unwrap();
548 :
549 0 : let span = info_span!("blocking");
550 0 : tokio::task::spawn_blocking(move || {
551 0 : let _span = span.entered();
552 0 :
553 0 : let res = serde_json::to_writer(&mut writer, &resp)
554 0 : .map_err(std::io::Error::from)
555 0 : .and_then(|_| writer.flush());
556 0 :
557 0 : match res {
558 : Ok(()) => {
559 0 : tracing::info!(
560 0 : bytes = writer.flushed_bytes(),
561 0 : elapsed_ms = started_at.elapsed().as_millis(),
562 0 : "responded /v1/debug_dump"
563 : );
564 : }
565 0 : Err(e) => {
566 0 : tracing::warn!("failed to write out /v1/debug_dump response: {e:#}");
567 : // semantics of this error are quite... unclear. we want to error the stream out to
568 : // abort the response to somehow notify the client that we failed.
569 : //
570 : // though, most likely the reason for failure is that the receiver is already gone.
571 0 : drop(
572 0 : writer
573 0 : .tx
574 0 : .blocking_send(Err(std::io::ErrorKind::BrokenPipe.into())),
575 0 : );
576 : }
577 : }
578 0 : });
579 0 :
580 0 : Ok(response)
581 0 : }
582 :
583 : /// Safekeeper http router.
584 0 : pub fn make_router(
585 0 : conf: Arc<SafeKeeperConf>,
586 0 : global_timelines: Arc<GlobalTimelines>,
587 0 : ) -> RouterBuilder<hyper::Body, ApiError> {
588 0 : let mut router = endpoint::make_router();
589 0 : if conf.http_auth.is_some() {
590 0 : router = router.middleware(auth_middleware(|request| {
591 : const ALLOWLIST_ROUTES: &[&str] =
592 : &["/v1/status", "/metrics", "/profile/cpu", "/profile/heap"];
593 0 : if ALLOWLIST_ROUTES.contains(&request.uri().path()) {
594 0 : None
595 : } else {
596 : // Option<Arc<SwappableJwtAuth>> is always provided as data below, hence unwrap().
597 0 : request
598 0 : .data::<Option<Arc<SwappableJwtAuth>>>()
599 0 : .unwrap()
600 0 : .as_deref()
601 : }
602 0 : }))
603 0 : }
604 :
605 : // NB: on any changes do not forget to update the OpenAPI spec
606 : // located nearby (/safekeeper/src/http/openapi_spec.yaml).
607 0 : let auth = conf.http_auth.clone();
608 0 : router
609 0 : .data(conf)
610 0 : .data(global_timelines)
611 0 : .data(auth)
612 0 : .get("/metrics", |r| request_span(r, prometheus_metrics_handler))
613 0 : .get("/profile/cpu", |r| request_span(r, profile_cpu_handler))
614 0 : .get("/profile/heap", |r| request_span(r, profile_heap_handler))
615 0 : .get("/v1/status", |r| request_span(r, status_handler))
616 0 : .put("/v1/failpoints", |r| {
617 0 : request_span(r, move |r| async {
618 0 : check_permission(&r, None)?;
619 0 : let cancel = CancellationToken::new();
620 0 : failpoints_handler(r, cancel).await
621 0 : })
622 0 : })
623 0 : .delete("/v1/tenant/:tenant_id", |r| {
624 0 : request_span(r, tenant_delete_handler)
625 0 : })
626 0 : // Will be used in the future instead of implicit timeline creation
627 0 : .post("/v1/tenant/timeline", |r| {
628 0 : request_span(r, timeline_create_handler)
629 0 : })
630 0 : .get("/v1/tenant/timeline", |r| {
631 0 : request_span(r, timeline_list_handler)
632 0 : })
633 0 : .get("/v1/tenant/:tenant_id/timeline/:timeline_id", |r| {
634 0 : request_span(r, timeline_status_handler)
635 0 : })
636 0 : .delete("/v1/tenant/:tenant_id/timeline/:timeline_id", |r| {
637 0 : request_span(r, timeline_delete_handler)
638 0 : })
639 0 : .post("/v1/pull_timeline", |r| {
640 0 : request_span(r, timeline_pull_handler)
641 0 : })
642 0 : .get(
643 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/snapshot/:destination_id",
644 0 : |r| request_span(r, timeline_snapshot_handler),
645 0 : )
646 0 : .post(
647 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/membership",
648 0 : |r| request_span(r, timeline_membership_handler),
649 0 : )
650 0 : .post(
651 0 : "/v1/tenant/:tenant_id/timeline/:source_timeline_id/copy",
652 0 : |r| request_span(r, timeline_copy_handler),
653 0 : )
654 0 : .patch(
655 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/control_file",
656 0 : |r| request_span(r, patch_control_file_handler),
657 0 : )
658 0 : .post(
659 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/checkpoint",
660 0 : |r| request_span(r, timeline_checkpoint_handler),
661 0 : )
662 0 : .get("/v1/tenant/:tenant_id/timeline/:timeline_id/digest", |r| {
663 0 : request_span(r, timeline_digest_handler)
664 0 : })
665 0 : .post(
666 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/backup_partial_reset",
667 0 : |r| request_span(r, timeline_backup_partial_reset),
668 0 : )
669 0 : .post(
670 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/term_bump",
671 0 : |r| request_span(r, timeline_term_bump_handler),
672 0 : )
673 0 : .post("/v1/record_safekeeper_info/:tenant_id/:timeline_id", |r| {
674 0 : request_span(r, record_safekeeper_info)
675 0 : })
676 0 : .get("/v1/debug_dump", |r| request_span(r, dump_debug_handler))
677 0 : }
678 :
679 : #[cfg(test)]
680 : mod tests {
681 : use super::*;
682 :
683 : #[test]
684 1 : fn test_term_switch_entry_api_serialize() {
685 1 : let state = AcceptorStateStatus {
686 1 : term: 1,
687 1 : epoch: 1,
688 1 : term_history: vec![TermSwitchApiEntry {
689 1 : term: 1,
690 1 : lsn: Lsn(0x16FFDDDD),
691 1 : }],
692 1 : };
693 1 : let json = serde_json::to_string(&state).unwrap();
694 1 : assert_eq!(
695 1 : json,
696 1 : "{\"term\":1,\"epoch\":1,\"term_history\":[{\"term\":1,\"lsn\":\"0/16FFDDDD\"}]}"
697 1 : );
698 1 : }
699 : }
|
