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