Line data Source code
1 : use std::collections::{HashMap, HashSet};
2 :
3 : use crate::checks::{
4 : branch_cleanup_and_check_errors, list_timeline_blobs, BlobDataParseResult,
5 : RemoteTimelineBlobData, TenantObjectListing, TimelineAnalysis,
6 : };
7 : use crate::metadata_stream::{stream_tenant_timelines, stream_tenants};
8 : use crate::{init_remote, BucketConfig, NodeKind, RootTarget, TenantShardTimelineId};
9 : use futures_util::{StreamExt, TryStreamExt};
10 : use pageserver::tenant::remote_timeline_client::remote_layer_path;
11 : use pageserver_api::controller_api::MetadataHealthUpdateRequest;
12 : use pageserver_api::shard::TenantShardId;
13 : use remote_storage::GenericRemoteStorage;
14 : use serde::Serialize;
15 : use utils::id::TenantId;
16 : use utils::shard::ShardCount;
17 :
18 : #[derive(Serialize, Default)]
19 : pub struct MetadataSummary {
20 : tenant_count: usize,
21 : timeline_count: usize,
22 : timeline_shard_count: usize,
23 : with_errors: HashSet<TenantShardTimelineId>,
24 : with_warnings: HashSet<TenantShardTimelineId>,
25 : with_orphans: HashSet<TenantShardTimelineId>,
26 : indices_by_version: HashMap<usize, usize>,
27 :
28 : #[serde(skip)]
29 : pub(crate) healthy_tenant_shards: HashSet<TenantShardId>,
30 : #[serde(skip)]
31 : pub(crate) unhealthy_tenant_shards: HashSet<TenantShardId>,
32 : }
33 :
34 : impl MetadataSummary {
35 0 : fn new() -> Self {
36 0 : Self::default()
37 0 : }
38 :
39 0 : fn update_data(&mut self, data: &RemoteTimelineBlobData) {
40 0 : self.timeline_shard_count += 1;
41 : if let BlobDataParseResult::Parsed {
42 0 : index_part,
43 : index_part_generation: _,
44 : s3_layers: _,
45 0 : } = &data.blob_data
46 0 : {
47 0 : *self
48 0 : .indices_by_version
49 0 : .entry(index_part.version())
50 0 : .or_insert(0) += 1;
51 0 : }
52 0 : }
53 :
54 0 : fn update_analysis(&mut self, id: &TenantShardTimelineId, analysis: &TimelineAnalysis) {
55 0 : if analysis.is_healthy() {
56 0 : self.healthy_tenant_shards.insert(id.tenant_shard_id);
57 0 : } else {
58 0 : self.healthy_tenant_shards.remove(&id.tenant_shard_id);
59 0 : self.unhealthy_tenant_shards.insert(id.tenant_shard_id);
60 0 : }
61 :
62 0 : if !analysis.errors.is_empty() {
63 0 : self.with_errors.insert(*id);
64 0 : }
65 :
66 0 : if !analysis.warnings.is_empty() {
67 0 : self.with_warnings.insert(*id);
68 0 : }
69 0 : }
70 :
71 0 : fn notify_timeline_orphan(&mut self, ttid: &TenantShardTimelineId) {
72 0 : self.with_orphans.insert(*ttid);
73 0 : }
74 :
75 : /// Long-form output for printing at end of a scan
76 0 : pub fn summary_string(&self) -> String {
77 0 : let version_summary: String = itertools::join(
78 0 : self.indices_by_version
79 0 : .iter()
80 0 : .map(|(k, v)| format!("{k}: {v}")),
81 0 : ", ",
82 0 : );
83 0 :
84 0 : format!(
85 0 : "Tenants: {}
86 0 : Timelines: {}
87 0 : Timeline-shards: {}
88 0 : With errors: {}
89 0 : With warnings: {}
90 0 : With orphan layers: {}
91 0 : Index versions: {version_summary}
92 0 : ",
93 0 : self.tenant_count,
94 0 : self.timeline_count,
95 0 : self.timeline_shard_count,
96 0 : self.with_errors.len(),
97 0 : self.with_warnings.len(),
98 0 : self.with_orphans.len(),
99 0 : )
100 0 : }
101 :
102 0 : pub fn is_fatal(&self) -> bool {
103 0 : !self.with_errors.is_empty()
104 0 : }
105 :
106 0 : pub fn is_empty(&self) -> bool {
107 0 : self.timeline_shard_count == 0
108 0 : }
109 :
110 0 : pub fn build_health_update_request(&self) -> MetadataHealthUpdateRequest {
111 0 : MetadataHealthUpdateRequest {
112 0 : healthy_tenant_shards: self.healthy_tenant_shards.clone(),
113 0 : unhealthy_tenant_shards: self.unhealthy_tenant_shards.clone(),
114 0 : }
115 0 : }
116 : }
117 :
118 : /// Scan the pageserver metadata in an S3 bucket, reporting errors and statistics.
119 0 : pub async fn scan_pageserver_metadata(
120 0 : bucket_config: BucketConfig,
121 0 : tenant_ids: Vec<TenantShardId>,
122 0 : ) -> anyhow::Result<MetadataSummary> {
123 0 : let (remote_client, target) = init_remote(bucket_config, NodeKind::Pageserver).await?;
124 :
125 0 : let tenants = if tenant_ids.is_empty() {
126 0 : futures::future::Either::Left(stream_tenants(&remote_client, &target))
127 : } else {
128 0 : futures::future::Either::Right(futures::stream::iter(tenant_ids.into_iter().map(Ok)))
129 : };
130 :
131 : // How many tenants to process in parallel. We need to be mindful of pageservers
132 : // accessing the same per tenant prefixes, so use a lower setting than pageservers.
133 : const CONCURRENCY: usize = 32;
134 :
135 : // Generate a stream of TenantTimelineId
136 0 : let timelines = tenants.map_ok(|t| stream_tenant_timelines(&remote_client, &target, t));
137 0 : let timelines = timelines.try_buffered(CONCURRENCY);
138 0 : let timelines = timelines.try_flatten();
139 :
140 : // Generate a stream of S3TimelineBlobData
141 0 : async fn report_on_timeline(
142 0 : remote_client: &GenericRemoteStorage,
143 0 : target: &RootTarget,
144 0 : ttid: TenantShardTimelineId,
145 0 : ) -> anyhow::Result<(TenantShardTimelineId, RemoteTimelineBlobData)> {
146 0 : let data = list_timeline_blobs(remote_client, ttid, target).await?;
147 0 : Ok((ttid, data))
148 0 : }
149 0 : let timelines = timelines.map_ok(|ttid| report_on_timeline(&remote_client, &target, ttid));
150 0 : let mut timelines = std::pin::pin!(timelines.try_buffered(CONCURRENCY));
151 0 :
152 0 : // We must gather all the TenantShardTimelineId->S3TimelineBlobData for each tenant, because different
153 0 : // shards in the same tenant might refer to one anothers' keys if a shard split has happened.
154 0 :
155 0 : let mut tenant_id = None;
156 0 : let mut tenant_objects = TenantObjectListing::default();
157 0 : let mut tenant_timeline_results = Vec::new();
158 :
159 0 : async fn analyze_tenant(
160 0 : remote_client: &GenericRemoteStorage,
161 0 : tenant_id: TenantId,
162 0 : summary: &mut MetadataSummary,
163 0 : mut tenant_objects: TenantObjectListing,
164 0 : timelines: Vec<(TenantShardTimelineId, RemoteTimelineBlobData)>,
165 0 : highest_shard_count: ShardCount,
166 0 : ) {
167 0 : summary.tenant_count += 1;
168 0 :
169 0 : let mut timeline_ids = HashSet::new();
170 0 : let mut timeline_generations = HashMap::new();
171 0 : for (ttid, data) in timelines {
172 0 : if ttid.tenant_shard_id.shard_count == highest_shard_count {
173 : // Only analyze `TenantShardId`s with highest shard count.
174 :
175 : // Stash the generation of each timeline, for later use identifying orphan layers
176 : if let BlobDataParseResult::Parsed {
177 0 : index_part,
178 0 : index_part_generation,
179 0 : s3_layers: _s3_layers,
180 0 : } = &data.blob_data
181 : {
182 0 : if index_part.deleted_at.is_some() {
183 : // skip deleted timeline.
184 0 : tracing::info!("Skip analysis of {} b/c timeline is already deleted", ttid);
185 0 : continue;
186 0 : }
187 0 : timeline_generations.insert(ttid, *index_part_generation);
188 0 : }
189 :
190 : // Apply checks to this timeline shard's metadata, and in the process update `tenant_objects`
191 : // reference counts for layers across the tenant.
192 0 : let analysis = branch_cleanup_and_check_errors(
193 0 : remote_client,
194 0 : &ttid,
195 0 : &mut tenant_objects,
196 0 : None,
197 0 : None,
198 0 : Some(data),
199 0 : )
200 0 : .await;
201 0 : summary.update_analysis(&ttid, &analysis);
202 0 :
203 0 : timeline_ids.insert(ttid.timeline_id);
204 : } else {
205 0 : tracing::info!(
206 0 : "Skip analysis of {} b/c a lower shard count than {}",
207 : ttid,
208 : highest_shard_count.0,
209 : );
210 : }
211 : }
212 :
213 0 : summary.timeline_count += timeline_ids.len();
214 :
215 : // Identifying orphan layers must be done on a tenant-wide basis, because individual
216 : // shards' layers may be referenced by other shards.
217 : //
218 : // Orphan layers are not a corruption, and not an indication of a problem. They are just
219 : // consuming some space in remote storage, and may be cleaned up at leisure.
220 0 : for (shard_index, timeline_id, layer_file, generation) in tenant_objects.get_orphans() {
221 0 : let ttid = TenantShardTimelineId {
222 0 : tenant_shard_id: TenantShardId {
223 0 : tenant_id,
224 0 : shard_count: shard_index.shard_count,
225 0 : shard_number: shard_index.shard_number,
226 0 : },
227 0 : timeline_id,
228 0 : };
229 :
230 0 : if let Some(timeline_generation) = timeline_generations.get(&ttid) {
231 0 : if &generation >= timeline_generation {
232 : // Candidate orphan layer is in the current or future generation relative
233 : // to the index we read for this timeline shard, so its absence from the index
234 : // doesn't make it an orphan: more likely, it is a case where the layer was
235 : // uploaded, but the index referencing the layer wasn't written yet.
236 0 : continue;
237 0 : }
238 0 : }
239 :
240 0 : let orphan_path = remote_layer_path(
241 0 : &tenant_id,
242 0 : &timeline_id,
243 0 : shard_index,
244 0 : &layer_file,
245 0 : generation,
246 0 : );
247 0 :
248 0 : tracing::info!("Orphan layer detected: {orphan_path}");
249 :
250 0 : summary.notify_timeline_orphan(&ttid);
251 : }
252 0 : }
253 :
254 : // Iterate through all the timeline results. These are in key-order, so
255 : // all results for the same tenant will be adjacent. We accumulate these,
256 : // and then call `analyze_tenant` to flush, when we see the next tenant ID.
257 0 : let mut summary = MetadataSummary::new();
258 0 : let mut highest_shard_count = ShardCount::MIN;
259 0 : while let Some(i) = timelines.next().await {
260 0 : let (ttid, data) = i?;
261 0 : summary.update_data(&data);
262 0 :
263 0 : match tenant_id {
264 0 : None => {
265 0 : tenant_id = Some(ttid.tenant_shard_id.tenant_id);
266 0 : highest_shard_count = highest_shard_count.max(ttid.tenant_shard_id.shard_count);
267 0 : }
268 0 : Some(prev_tenant_id) => {
269 0 : if prev_tenant_id != ttid.tenant_shard_id.tenant_id {
270 : // New tenant: analyze this tenant's timelines, clear accumulated tenant_timeline_results
271 0 : let tenant_objects = std::mem::take(&mut tenant_objects);
272 0 : let timelines = std::mem::take(&mut tenant_timeline_results);
273 0 : analyze_tenant(
274 0 : &remote_client,
275 0 : prev_tenant_id,
276 0 : &mut summary,
277 0 : tenant_objects,
278 0 : timelines,
279 0 : highest_shard_count,
280 0 : )
281 0 : .await;
282 0 : tenant_id = Some(ttid.tenant_shard_id.tenant_id);
283 0 : highest_shard_count = ttid.tenant_shard_id.shard_count;
284 0 : } else {
285 0 : highest_shard_count = highest_shard_count.max(ttid.tenant_shard_id.shard_count);
286 0 : }
287 : }
288 : }
289 :
290 0 : match &data.blob_data {
291 : BlobDataParseResult::Parsed {
292 0 : index_part: _index_part,
293 0 : index_part_generation: _index_part_generation,
294 0 : s3_layers,
295 0 : } => {
296 0 : tenant_objects.push(ttid, s3_layers.clone());
297 0 : }
298 0 : BlobDataParseResult::Relic => (),
299 : BlobDataParseResult::Incorrect {
300 : errors: _,
301 0 : s3_layers,
302 0 : } => {
303 0 : tenant_objects.push(ttid, s3_layers.clone());
304 0 : }
305 : }
306 0 : tenant_timeline_results.push((ttid, data));
307 : }
308 :
309 0 : if !tenant_timeline_results.is_empty() {
310 0 : analyze_tenant(
311 0 : &remote_client,
312 0 : tenant_id.expect("Must be set if results are present"),
313 0 : &mut summary,
314 0 : tenant_objects,
315 0 : tenant_timeline_results,
316 0 : highest_shard_count,
317 0 : )
318 0 : .await;
319 0 : }
320 :
321 0 : Ok(summary)
322 0 : }
|
