Line data Source code
1 : use std::collections::HashSet;
2 : use std::str::FromStr;
3 : use std::sync::Arc;
4 : use std::time::Duration;
5 :
6 : use super::safekeeper_reconciler::ScheduleRequest;
7 : use crate::heartbeater::SafekeeperState;
8 : use crate::metrics;
9 : use crate::persistence::{
10 : DatabaseError, SafekeeperTimelineOpKind, TimelinePendingOpPersistence, TimelinePersistence,
11 : };
12 : use crate::safekeeper::Safekeeper;
13 : use anyhow::Context;
14 : use http_utils::error::ApiError;
15 : use pageserver_api::controller_api::{
16 : SafekeeperDescribeResponse, SkSchedulingPolicy, TimelineImportRequest,
17 : };
18 : use pageserver_api::models::{self, SafekeeperInfo, SafekeepersInfo, TimelineInfo};
19 : use safekeeper_api::membership::{MemberSet, SafekeeperId};
20 : use tokio::task::JoinSet;
21 : use tokio_util::sync::CancellationToken;
22 : use utils::id::{NodeId, TenantId, TimelineId};
23 : use utils::logging::SecretString;
24 : use utils::lsn::Lsn;
25 :
26 : use super::Service;
27 :
28 : impl Service {
29 : /// Timeline creation on safekeepers
30 : ///
31 : /// Returns `Ok(left)` if the timeline has been created on a quorum of safekeepers,
32 : /// where `left` contains the list of safekeepers that didn't have a successful response.
33 : /// Assumes tenant lock is held while calling this function.
34 0 : pub(super) async fn tenant_timeline_create_safekeepers_quorum(
35 0 : &self,
36 0 : tenant_id: TenantId,
37 0 : timeline_id: TimelineId,
38 0 : pg_version: u32,
39 0 : timeline_persistence: &TimelinePersistence,
40 0 : ) -> Result<Vec<NodeId>, ApiError> {
41 0 : // If quorum is reached, return if we are outside of a specified timeout
42 0 : let jwt = self
43 0 : .config
44 0 : .safekeeper_jwt_token
45 0 : .clone()
46 0 : .map(SecretString::from);
47 0 : let mut joinset = JoinSet::new();
48 0 :
49 0 : let safekeepers = {
50 0 : let locked = self.inner.read().unwrap();
51 0 : locked.safekeepers.clone()
52 0 : };
53 0 :
54 0 : let mut members = Vec::new();
55 0 : for sk_id in timeline_persistence.sk_set.iter() {
56 0 : let sk_id = NodeId(*sk_id as u64);
57 0 : let Some(safekeeper) = safekeepers.get(&sk_id) else {
58 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(
59 0 : "couldn't find entry for safekeeper with id {sk_id}"
60 0 : )))?;
61 : };
62 0 : members.push(SafekeeperId {
63 0 : id: sk_id,
64 0 : host: safekeeper.skp.host.clone(),
65 0 : pg_port: safekeeper.skp.port as u16,
66 0 : });
67 : }
68 0 : let mset = MemberSet::new(members).map_err(ApiError::InternalServerError)?;
69 0 : let mconf = safekeeper_api::membership::Configuration::new(mset);
70 0 :
71 0 : let req = safekeeper_api::models::TimelineCreateRequest {
72 0 : commit_lsn: None,
73 0 : mconf,
74 0 : pg_version,
75 0 : start_lsn: timeline_persistence.start_lsn.0,
76 0 : system_id: None,
77 0 : tenant_id,
78 0 : timeline_id,
79 0 : wal_seg_size: None,
80 0 : };
81 : const SK_CREATE_TIMELINE_RECONCILE_TIMEOUT: Duration = Duration::from_secs(30);
82 0 : for sk in timeline_persistence.sk_set.iter() {
83 0 : let sk_id = NodeId(*sk as u64);
84 0 : let safekeepers = safekeepers.clone();
85 0 : let http_client = self.http_client.clone();
86 0 : let jwt = jwt.clone();
87 0 : let req = req.clone();
88 0 : joinset.spawn(async move {
89 0 : // Unwrap is fine as we already would have returned error above
90 0 : let sk_p = safekeepers.get(&sk_id).unwrap();
91 0 : let res = sk_p
92 0 : .with_client_retries(
93 0 : |client| {
94 0 : let req = req.clone();
95 0 : async move { client.create_timeline(&req).await }
96 0 : },
97 0 : &http_client,
98 0 : &jwt,
99 0 : 3,
100 0 : 3,
101 0 : SK_CREATE_TIMELINE_RECONCILE_TIMEOUT,
102 0 : &CancellationToken::new(),
103 0 : )
104 0 : .await;
105 0 : (sk_id, sk_p.skp.host.clone(), res)
106 0 : });
107 0 : }
108 : // After we have built the joinset, we now wait for the tasks to complete,
109 : // but with a specified timeout to make sure we return swiftly, either with
110 : // a failure or success.
111 0 : let reconcile_deadline = tokio::time::Instant::now() + SK_CREATE_TIMELINE_RECONCILE_TIMEOUT;
112 0 :
113 0 : // Wait until all tasks finish or timeout is hit, whichever occurs
114 0 : // first.
115 0 : let mut reconcile_results = Vec::new();
116 : loop {
117 0 : if let Ok(res) = tokio::time::timeout_at(reconcile_deadline, joinset.join_next()).await
118 : {
119 0 : let Some(res) = res else { break };
120 0 : match res {
121 0 : Ok(res) => {
122 0 : tracing::info!(
123 0 : "response from safekeeper id:{} at {}: {:?}",
124 : res.0,
125 : res.1,
126 : res.2
127 : );
128 0 : reconcile_results.push(res);
129 : }
130 0 : Err(join_err) => {
131 0 : tracing::info!("join_err for task in joinset: {join_err}");
132 : }
133 : }
134 : } else {
135 0 : tracing::info!(
136 0 : "timeout for creation call after {} responses",
137 0 : reconcile_results.len()
138 : );
139 0 : break;
140 : }
141 : }
142 :
143 : // Now check now if quorum was reached in reconcile_results.
144 0 : let total_result_count = reconcile_results.len();
145 0 : let remaining = reconcile_results
146 0 : .into_iter()
147 0 : .filter_map(|res| res.2.is_err().then_some(res.0))
148 0 : .collect::<Vec<_>>();
149 0 : tracing::info!(
150 0 : "Got {} non-successful responses from initial creation request of total {total_result_count} responses",
151 0 : remaining.len()
152 : );
153 0 : if remaining.len() >= 2 {
154 : // Failure
155 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(
156 0 : "not enough successful reconciliations to reach quorum, please retry: {} errored",
157 0 : remaining.len()
158 0 : )));
159 0 : }
160 0 :
161 0 : Ok(remaining)
162 0 : }
163 :
164 : /// Create timeline in controller database and on safekeepers.
165 : /// `timeline_info` is result of timeline creation on pageserver.
166 : ///
167 : /// All actions must be idempotent as the call is retried until success. It
168 : /// tries to create timeline in the db and on at least majority of
169 : /// safekeepers + queue creation for safekeepers which missed it in the db
170 : /// for infinite retries; after that, call returns Ok.
171 : ///
172 : /// The idea is that once this is reached as long as we have alive majority
173 : /// of safekeepers it is expected to get eventually operational as storcon
174 : /// will be able to seed timeline on nodes which missed creation by making
175 : /// pull_timeline from peers. On the other hand we don't want to fail
176 : /// timeline creation if one safekeeper is down.
177 0 : pub(super) async fn tenant_timeline_create_safekeepers(
178 0 : self: &Arc<Self>,
179 0 : tenant_id: TenantId,
180 0 : timeline_info: &TimelineInfo,
181 0 : create_mode: models::TimelineCreateRequestMode,
182 0 : ) -> Result<SafekeepersInfo, ApiError> {
183 0 : let timeline_id = timeline_info.timeline_id;
184 0 : let pg_version = timeline_info.pg_version * 10000;
185 : // Initially start_lsn is determined by last_record_lsn in pageserver
186 : // response as it does initdb. However, later we persist it and in sk
187 : // creation calls replace with the value from the timeline row if it
188 : // previously existed as on retries in theory endpoint might have
189 : // already written some data and advanced last_record_lsn, while we want
190 : // safekeepers to have consistent start_lsn.
191 0 : let start_lsn = match create_mode {
192 0 : models::TimelineCreateRequestMode::Bootstrap { .. } => timeline_info.last_record_lsn,
193 0 : models::TimelineCreateRequestMode::Branch { .. } => timeline_info.last_record_lsn,
194 : models::TimelineCreateRequestMode::ImportPgdata { .. } => {
195 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(
196 0 : "import pgdata doesn't specify the start lsn, aborting creation on safekeepers"
197 0 : )))?;
198 : }
199 : };
200 : // Choose initial set of safekeepers respecting affinity
201 0 : let sks = self.safekeepers_for_new_timeline().await?;
202 0 : let sks_persistence = sks.iter().map(|sk| sk.id.0 as i64).collect::<Vec<_>>();
203 0 : // Add timeline to db
204 0 : let mut timeline_persist = TimelinePersistence {
205 0 : tenant_id: tenant_id.to_string(),
206 0 : timeline_id: timeline_id.to_string(),
207 0 : start_lsn: start_lsn.into(),
208 0 : generation: 0,
209 0 : sk_set: sks_persistence.clone(),
210 0 : new_sk_set: None,
211 0 : cplane_notified_generation: 0,
212 0 : deleted_at: None,
213 0 : };
214 0 : let inserted = self
215 0 : .persistence
216 0 : .insert_timeline(timeline_persist.clone())
217 0 : .await?;
218 0 : if !inserted {
219 0 : if let Some(existent_persist) = self
220 0 : .persistence
221 0 : .get_timeline(tenant_id, timeline_id)
222 0 : .await?
223 0 : {
224 0 : // Replace with what we have in the db, to get stuff like the generation right.
225 0 : // We do still repeat the http calls to the safekeepers. After all, we could have
226 0 : // crashed right after the wrote to the DB.
227 0 : timeline_persist = existent_persist;
228 0 : } else {
229 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(
230 0 : "insertion said timeline already in db, but looking it up, it was gone"
231 0 : )));
232 : }
233 0 : }
234 : // Create the timeline on a quorum of safekeepers
235 0 : let remaining = self
236 0 : .tenant_timeline_create_safekeepers_quorum(
237 0 : tenant_id,
238 0 : timeline_id,
239 0 : pg_version,
240 0 : &timeline_persist,
241 0 : )
242 0 : .await?;
243 :
244 : // For the remaining safekeepers, take care of their reconciliation asynchronously
245 0 : for &remaining_id in remaining.iter() {
246 0 : let pending_op = TimelinePendingOpPersistence {
247 0 : tenant_id: tenant_id.to_string(),
248 0 : timeline_id: timeline_id.to_string(),
249 0 : generation: timeline_persist.generation,
250 0 : op_kind: crate::persistence::SafekeeperTimelineOpKind::Pull,
251 0 : sk_id: remaining_id.0 as i64,
252 0 : };
253 0 : tracing::info!("writing pending op for sk id {remaining_id}");
254 0 : self.persistence.insert_pending_op(pending_op).await?;
255 : }
256 0 : if !remaining.is_empty() {
257 0 : let mut locked = self.inner.write().unwrap();
258 0 : for remaining_id in remaining {
259 0 : let Some(sk) = locked.safekeepers.get(&remaining_id) else {
260 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(
261 0 : "Couldn't find safekeeper with id {remaining_id}"
262 0 : )));
263 : };
264 0 : let Ok(host_list) = sks
265 0 : .iter()
266 0 : .map(|sk| {
267 0 : Ok((
268 0 : sk.id,
269 0 : locked
270 0 : .safekeepers
271 0 : .get(&sk.id)
272 0 : .ok_or_else(|| {
273 0 : ApiError::InternalServerError(anyhow::anyhow!(
274 0 : "Couldn't find safekeeper with id {} to pull from",
275 0 : sk.id
276 0 : ))
277 0 : })?
278 0 : .base_url(),
279 : ))
280 0 : })
281 0 : .collect::<Result<_, ApiError>>()
282 : else {
283 0 : continue;
284 : };
285 0 : let req = ScheduleRequest {
286 0 : safekeeper: Box::new(sk.clone()),
287 0 : host_list,
288 0 : tenant_id,
289 0 : timeline_id: Some(timeline_id),
290 0 : generation: timeline_persist.generation as u32,
291 0 : kind: crate::persistence::SafekeeperTimelineOpKind::Pull,
292 0 : };
293 0 : locked.safekeeper_reconcilers.schedule_request(self, req);
294 : }
295 0 : }
296 :
297 0 : Ok(SafekeepersInfo {
298 0 : generation: timeline_persist.generation as u32,
299 0 : safekeepers: sks,
300 0 : tenant_id,
301 0 : timeline_id,
302 0 : })
303 0 : }
304 :
305 : /// Directly insert the timeline into the database without reconciling it with safekeepers.
306 : ///
307 : /// Useful if the timeline already exists on the specified safekeepers,
308 : /// but we want to make it storage controller managed.
309 0 : pub(crate) async fn timeline_import(&self, req: TimelineImportRequest) -> Result<(), ApiError> {
310 0 : let persistence = TimelinePersistence {
311 0 : tenant_id: req.tenant_id.to_string(),
312 0 : timeline_id: req.timeline_id.to_string(),
313 0 : start_lsn: Lsn::INVALID.into(),
314 0 : generation: 1,
315 0 : sk_set: req.sk_set.iter().map(|sk_id| sk_id.0 as i64).collect(),
316 0 : new_sk_set: None,
317 0 : cplane_notified_generation: 1,
318 0 : deleted_at: None,
319 0 : };
320 0 : let inserted = self.persistence.insert_timeline(persistence).await?;
321 0 : if inserted {
322 0 : tracing::info!("imported timeline into db");
323 : } else {
324 0 : tracing::info!("didn't import timeline into db, as it is already present in db");
325 : }
326 0 : Ok(())
327 0 : }
328 :
329 : /// Perform timeline deletion on safekeepers. Will return success: we persist the deletion into the reconciler.
330 0 : pub(super) async fn tenant_timeline_delete_safekeepers(
331 0 : self: &Arc<Self>,
332 0 : tenant_id: TenantId,
333 0 : timeline_id: TimelineId,
334 0 : ) -> Result<(), ApiError> {
335 0 : let tl = self
336 0 : .persistence
337 0 : .get_timeline(tenant_id, timeline_id)
338 0 : .await?;
339 0 : let Some(tl) = tl else {
340 0 : tracing::info!(
341 0 : "timeline {tenant_id}/{timeline_id} doesn't exist in timelines table, no deletions on safekeepers needed"
342 : );
343 0 : return Ok(());
344 : };
345 0 : self.persistence
346 0 : .timeline_set_deleted_at(tenant_id, timeline_id)
347 0 : .await?;
348 0 : let all_sks = tl
349 0 : .new_sk_set
350 0 : .iter()
351 0 : .flatten()
352 0 : .chain(tl.sk_set.iter())
353 0 : .collect::<HashSet<_>>();
354 :
355 : // Schedule reconciliations
356 0 : for &sk_id in all_sks.iter() {
357 0 : let pending_op = TimelinePendingOpPersistence {
358 0 : tenant_id: tenant_id.to_string(),
359 0 : timeline_id: timeline_id.to_string(),
360 0 : generation: tl.generation,
361 0 : op_kind: SafekeeperTimelineOpKind::Delete,
362 0 : sk_id: *sk_id,
363 0 : };
364 0 : tracing::info!("writing pending op for sk id {sk_id}");
365 0 : self.persistence.insert_pending_op(pending_op).await?;
366 : }
367 : {
368 0 : let mut locked = self.inner.write().unwrap();
369 0 : for sk_id in all_sks {
370 0 : let sk_id = NodeId(*sk_id as u64);
371 0 : let Some(sk) = locked.safekeepers.get(&sk_id) else {
372 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(
373 0 : "Couldn't find safekeeper with id {sk_id}"
374 0 : )));
375 : };
376 :
377 0 : let req = ScheduleRequest {
378 0 : safekeeper: Box::new(sk.clone()),
379 0 : // we don't use this for this kind, put a dummy value
380 0 : host_list: Vec::new(),
381 0 : tenant_id,
382 0 : timeline_id: Some(timeline_id),
383 0 : generation: tl.generation as u32,
384 0 : kind: SafekeeperTimelineOpKind::Delete,
385 0 : };
386 0 : locked.safekeeper_reconcilers.schedule_request(self, req);
387 : }
388 : }
389 0 : Ok(())
390 0 : }
391 :
392 : /// Perform tenant deletion on safekeepers.
393 0 : pub(super) async fn tenant_delete_safekeepers(
394 0 : self: &Arc<Self>,
395 0 : tenant_id: TenantId,
396 0 : ) -> Result<(), ApiError> {
397 0 : let timeline_list = self
398 0 : .persistence
399 0 : .list_timelines_for_tenant(tenant_id)
400 0 : .await?;
401 :
402 0 : if timeline_list.is_empty() {
403 : // Early exit: the tenant is either empty or not migrated to the storcon yet
404 0 : tracing::info!("Skipping tenant delete as the timeline doesn't exist in db");
405 0 : return Ok(());
406 0 : }
407 :
408 0 : let timeline_list = timeline_list
409 0 : .into_iter()
410 0 : .map(|timeline| {
411 0 : let timeline_id = TimelineId::from_str(&timeline.timeline_id)
412 0 : .context("timeline id loaded from db")
413 0 : .map_err(ApiError::InternalServerError)?;
414 0 : Ok((timeline_id, timeline))
415 0 : })
416 0 : .collect::<Result<Vec<_>, ApiError>>()?;
417 :
418 : // Remove pending ops from db, and set `deleted_at`.
419 : // We cancel them in a later iteration once we hold the state lock.
420 0 : for (timeline_id, _timeline) in timeline_list.iter() {
421 0 : self.persistence
422 0 : .remove_pending_ops_for_timeline(tenant_id, Some(*timeline_id))
423 0 : .await?;
424 0 : self.persistence
425 0 : .timeline_set_deleted_at(tenant_id, *timeline_id)
426 0 : .await?;
427 : }
428 :
429 : // The list of safekeepers that have any of the timelines
430 0 : let mut sk_list = HashSet::new();
431 :
432 : // List all pending ops for all timelines, cancel them
433 0 : for (_timeline_id, timeline) in timeline_list.iter() {
434 0 : let sk_iter = timeline
435 0 : .sk_set
436 0 : .iter()
437 0 : .chain(timeline.new_sk_set.iter().flatten())
438 0 : .map(|id| NodeId(*id as u64));
439 0 : sk_list.extend(sk_iter);
440 0 : }
441 :
442 0 : for &sk_id in sk_list.iter() {
443 0 : let pending_op = TimelinePendingOpPersistence {
444 0 : tenant_id: tenant_id.to_string(),
445 0 : timeline_id: String::new(),
446 0 : generation: i32::MAX,
447 0 : op_kind: SafekeeperTimelineOpKind::Delete,
448 0 : sk_id: sk_id.0 as i64,
449 0 : };
450 0 : tracing::info!("writing pending op for sk id {sk_id}");
451 0 : self.persistence.insert_pending_op(pending_op).await?;
452 : }
453 :
454 0 : let mut locked = self.inner.write().unwrap();
455 :
456 0 : for (timeline_id, _timeline) in timeline_list.iter() {
457 0 : for sk_id in sk_list.iter() {
458 0 : locked
459 0 : .safekeeper_reconcilers
460 0 : .cancel_reconciles_for_timeline(*sk_id, tenant_id, Some(*timeline_id));
461 0 : }
462 : }
463 :
464 : // unwrap is safe: we return above for an empty timeline list
465 0 : let max_generation = timeline_list
466 0 : .iter()
467 0 : .map(|(_tl_id, tl)| tl.generation as u32)
468 0 : .max()
469 0 : .unwrap();
470 :
471 0 : for sk_id in sk_list {
472 0 : let Some(safekeeper) = locked.safekeepers.get(&sk_id) else {
473 0 : tracing::warn!("Couldn't find safekeeper with id {sk_id}");
474 0 : continue;
475 : };
476 : // Add pending op for tenant deletion
477 0 : let req = ScheduleRequest {
478 0 : generation: max_generation,
479 0 : host_list: Vec::new(),
480 0 : kind: SafekeeperTimelineOpKind::Delete,
481 0 : safekeeper: Box::new(safekeeper.clone()),
482 0 : tenant_id,
483 0 : timeline_id: None,
484 0 : };
485 0 : locked.safekeeper_reconcilers.schedule_request(self, req);
486 : }
487 0 : Ok(())
488 0 : }
489 :
490 : /// Choose safekeepers for the new timeline: 3 in different azs.
491 0 : pub(crate) async fn safekeepers_for_new_timeline(
492 0 : &self,
493 0 : ) -> Result<Vec<SafekeeperInfo>, ApiError> {
494 0 : // Number of safekeepers in different AZs we are looking for
495 0 : let wanted_count = 3;
496 0 : let mut all_safekeepers = {
497 0 : let locked = self.inner.read().unwrap();
498 0 : locked
499 0 : .safekeepers
500 0 : .iter()
501 0 : .filter_map(|sk| {
502 0 : if sk.1.scheduling_policy() != SkSchedulingPolicy::Active {
503 : // If we don't want to schedule stuff onto the safekeeper, respect that.
504 0 : return None;
505 0 : }
506 0 : let utilization_opt = if let SafekeeperState::Available {
507 : last_seen_at: _,
508 0 : utilization,
509 0 : } = sk.1.availability()
510 : {
511 0 : Some(utilization)
512 : } else {
513 : // non-available safekeepers still get a chance for new timelines,
514 : // but put them last in the list.
515 0 : None
516 : };
517 0 : let info = SafekeeperInfo {
518 0 : hostname: sk.1.skp.host.clone(),
519 0 : id: NodeId(sk.1.skp.id as u64),
520 0 : };
521 0 : Some((utilization_opt, info, sk.1.skp.availability_zone_id.clone()))
522 0 : })
523 0 : .collect::<Vec<_>>()
524 0 : };
525 0 : all_safekeepers.sort_by_key(|sk| {
526 0 : (
527 0 : sk.0.as_ref()
528 0 : .map(|ut| ut.timeline_count)
529 0 : .unwrap_or(u64::MAX),
530 0 : // Use the id to decide on equal scores for reliability
531 0 : sk.1.id.0,
532 0 : )
533 0 : });
534 0 : let mut sks = Vec::new();
535 0 : let mut azs = HashSet::new();
536 0 : for (_sk_util, sk_info, az_id) in all_safekeepers.iter() {
537 0 : if !azs.insert(az_id) {
538 0 : continue;
539 0 : }
540 0 : sks.push(sk_info.clone());
541 0 : if sks.len() == wanted_count {
542 0 : break;
543 0 : }
544 : }
545 0 : if sks.len() == wanted_count {
546 0 : Ok(sks)
547 : } else {
548 0 : Err(ApiError::InternalServerError(anyhow::anyhow!(
549 0 : "couldn't find {wanted_count} safekeepers in different AZs for new timeline (found: {}, total active: {})",
550 0 : sks.len(),
551 0 : all_safekeepers.len(),
552 0 : )))
553 : }
554 0 : }
555 :
556 0 : pub(crate) async fn safekeepers_list(
557 0 : &self,
558 0 : ) -> Result<Vec<SafekeeperDescribeResponse>, DatabaseError> {
559 0 : let locked = self.inner.read().unwrap();
560 0 : let mut list = locked
561 0 : .safekeepers
562 0 : .iter()
563 0 : .map(|sk| sk.1.describe_response())
564 0 : .collect::<Result<Vec<_>, _>>()?;
565 0 : list.sort_by_key(|v| v.id);
566 0 : Ok(list)
567 0 : }
568 :
569 0 : pub(crate) async fn get_safekeeper(
570 0 : &self,
571 0 : id: i64,
572 0 : ) -> Result<SafekeeperDescribeResponse, DatabaseError> {
573 0 : let locked = self.inner.read().unwrap();
574 0 : let sk = locked
575 0 : .safekeepers
576 0 : .get(&NodeId(id as u64))
577 0 : .ok_or(diesel::result::Error::NotFound)?;
578 0 : sk.describe_response()
579 0 : }
580 :
581 0 : pub(crate) async fn upsert_safekeeper(
582 0 : &self,
583 0 : record: crate::persistence::SafekeeperUpsert,
584 0 : ) -> Result<(), ApiError> {
585 0 : let node_id = NodeId(record.id as u64);
586 0 : let use_https = self.config.use_https_safekeeper_api;
587 0 :
588 0 : if use_https && record.https_port.is_none() {
589 0 : return Err(ApiError::PreconditionFailed(
590 0 : format!(
591 0 : "cannot upsert safekeeper {node_id}: \
592 0 : https is enabled, but https port is not specified"
593 0 : )
594 0 : .into(),
595 0 : ));
596 0 : }
597 0 :
598 0 : self.persistence.safekeeper_upsert(record.clone()).await?;
599 : {
600 0 : let mut locked = self.inner.write().unwrap();
601 0 : let mut safekeepers = (*locked.safekeepers).clone();
602 0 : match safekeepers.entry(node_id) {
603 0 : std::collections::hash_map::Entry::Occupied(mut entry) => entry
604 0 : .get_mut()
605 0 : .update_from_record(record)
606 0 : .expect("all preconditions should be checked before upsert to database"),
607 0 : std::collections::hash_map::Entry::Vacant(entry) => {
608 0 : entry.insert(
609 0 : Safekeeper::from_persistence(
610 0 : crate::persistence::SafekeeperPersistence::from_upsert(
611 0 : record,
612 0 : SkSchedulingPolicy::Pause,
613 0 : ),
614 0 : CancellationToken::new(),
615 0 : use_https,
616 0 : )
617 0 : .expect("all preconditions should be checked before upsert to database"),
618 0 : );
619 0 : }
620 : }
621 0 : locked.safekeepers = Arc::new(safekeepers);
622 0 : metrics::METRICS_REGISTRY
623 0 : .metrics_group
624 0 : .storage_controller_safekeeper_nodes
625 0 : .set(locked.safekeepers.len() as i64);
626 0 : metrics::METRICS_REGISTRY
627 0 : .metrics_group
628 0 : .storage_controller_https_safekeeper_nodes
629 0 : .set(
630 0 : locked
631 0 : .safekeepers
632 0 : .values()
633 0 : .filter(|s| s.has_https_port())
634 0 : .count() as i64,
635 0 : );
636 0 : }
637 0 : Ok(())
638 0 : }
639 :
640 0 : pub(crate) async fn set_safekeeper_scheduling_policy(
641 0 : &self,
642 0 : id: i64,
643 0 : scheduling_policy: SkSchedulingPolicy,
644 0 : ) -> Result<(), DatabaseError> {
645 0 : self.persistence
646 0 : .set_safekeeper_scheduling_policy(id, scheduling_policy)
647 0 : .await?;
648 0 : let node_id = NodeId(id as u64);
649 0 : // After the change has been persisted successfully, update the in-memory state
650 0 : {
651 0 : let mut locked = self.inner.write().unwrap();
652 0 : let mut safekeepers = (*locked.safekeepers).clone();
653 0 : let sk = safekeepers
654 0 : .get_mut(&node_id)
655 0 : .ok_or(DatabaseError::Logical("Not found".to_string()))?;
656 0 : sk.set_scheduling_policy(scheduling_policy);
657 0 :
658 0 : match scheduling_policy {
659 0 : SkSchedulingPolicy::Active => (),
660 0 : SkSchedulingPolicy::Decomissioned | SkSchedulingPolicy::Pause => {
661 0 : locked.safekeeper_reconcilers.cancel_safekeeper(node_id);
662 0 : }
663 : }
664 :
665 0 : locked.safekeepers = Arc::new(safekeepers);
666 0 : }
667 0 : Ok(())
668 0 : }
669 : }
|
