Line data Source code
1 : use pageserver_api::models::detach_ancestor::AncestorDetached;
2 : use pageserver_api::models::{
3 : LocationConfig, LocationConfigListResponse, PageserverUtilization, SecondaryProgress,
4 : TenantScanRemoteStorageResponse, TenantShardSplitRequest, TenantShardSplitResponse,
5 : TenantWaitLsnRequest, TimelineArchivalConfigRequest, TimelineCreateRequest, TimelineInfo,
6 : TopTenantShardsRequest, TopTenantShardsResponse,
7 : };
8 : use pageserver_api::shard::TenantShardId;
9 : use pageserver_client::BlockUnblock;
10 : use pageserver_client::mgmt_api::{Client, Result};
11 : use reqwest::{Certificate, StatusCode};
12 : use utils::id::{NodeId, TenantId, TimelineId};
13 :
14 : /// Thin wrapper around [`pageserver_client::mgmt_api::Client`]. It allows the storage
15 : /// controller to collect metrics in a non-intrusive manner.
16 : #[derive(Debug, Clone)]
17 : pub(crate) struct PageserverClient {
18 : inner: Client,
19 : node_id_label: String,
20 : }
21 :
22 : macro_rules! measured_request {
23 : ($name:literal, $method:expr, $node_id: expr, $invoke:expr) => {{
24 : let labels = crate::metrics::PageserverRequestLabelGroup {
25 : pageserver_id: $node_id,
26 : path: $name,
27 : method: $method,
28 : };
29 :
30 : let latency = &crate::metrics::METRICS_REGISTRY
31 : .metrics_group
32 : .storage_controller_pageserver_request_latency;
33 : let _timer_guard = latency.start_timer(labels.clone());
34 :
35 : let res = $invoke;
36 :
37 : if res.is_err() {
38 : let error_counters = &crate::metrics::METRICS_REGISTRY
39 : .metrics_group
40 : .storage_controller_pageserver_request_error;
41 : error_counters.inc(labels)
42 : }
43 :
44 : res
45 : }};
46 : }
47 :
48 : impl PageserverClient {
49 0 : pub(crate) fn new(
50 0 : node_id: NodeId,
51 0 : mgmt_api_endpoint: String,
52 0 : jwt: Option<&str>,
53 0 : ssl_ca_cert: Option<Certificate>,
54 0 : ) -> Result<Self> {
55 0 : Ok(Self {
56 0 : inner: Client::new(mgmt_api_endpoint, jwt, ssl_ca_cert)?,
57 0 : node_id_label: node_id.0.to_string(),
58 : })
59 0 : }
60 :
61 0 : pub(crate) fn from_client(
62 0 : node_id: NodeId,
63 0 : raw_client: reqwest::Client,
64 0 : mgmt_api_endpoint: String,
65 0 : jwt: Option<&str>,
66 0 : ) -> Self {
67 0 : Self {
68 0 : inner: Client::from_client(raw_client, mgmt_api_endpoint, jwt),
69 0 : node_id_label: node_id.0.to_string(),
70 0 : }
71 0 : }
72 :
73 0 : pub(crate) async fn tenant_delete(&self, tenant_shard_id: TenantShardId) -> Result<StatusCode> {
74 0 : measured_request!(
75 0 : "tenant",
76 0 : crate::metrics::Method::Delete,
77 0 : &self.node_id_label,
78 0 : self.inner.tenant_delete(tenant_shard_id).await
79 0 : )
80 0 : }
81 :
82 0 : pub(crate) async fn tenant_time_travel_remote_storage(
83 0 : &self,
84 0 : tenant_shard_id: TenantShardId,
85 0 : timestamp: &str,
86 0 : done_if_after: &str,
87 0 : ) -> Result<()> {
88 0 : measured_request!(
89 0 : "tenant_time_travel_remote_storage",
90 0 : crate::metrics::Method::Put,
91 0 : &self.node_id_label,
92 0 : self.inner
93 0 : .tenant_time_travel_remote_storage(tenant_shard_id, timestamp, done_if_after)
94 0 : .await
95 0 : )
96 0 : }
97 :
98 0 : pub(crate) async fn tenant_scan_remote_storage(
99 0 : &self,
100 0 : tenant_id: TenantId,
101 0 : ) -> Result<TenantScanRemoteStorageResponse> {
102 0 : measured_request!(
103 0 : "tenant_scan_remote_storage",
104 0 : crate::metrics::Method::Get,
105 0 : &self.node_id_label,
106 0 : self.inner.tenant_scan_remote_storage(tenant_id).await
107 0 : )
108 0 : }
109 :
110 0 : pub(crate) async fn tenant_secondary_download(
111 0 : &self,
112 0 : tenant_id: TenantShardId,
113 0 : wait: Option<std::time::Duration>,
114 0 : ) -> Result<(StatusCode, SecondaryProgress)> {
115 0 : measured_request!(
116 0 : "tenant_secondary_download",
117 0 : crate::metrics::Method::Post,
118 0 : &self.node_id_label,
119 0 : self.inner.tenant_secondary_download(tenant_id, wait).await
120 0 : )
121 0 : }
122 :
123 0 : pub(crate) async fn tenant_secondary_status(
124 0 : &self,
125 0 : tenant_shard_id: TenantShardId,
126 0 : ) -> Result<SecondaryProgress> {
127 0 : measured_request!(
128 0 : "tenant_secondary_status",
129 0 : crate::metrics::Method::Get,
130 0 : &self.node_id_label,
131 0 : self.inner.tenant_secondary_status(tenant_shard_id).await
132 0 : )
133 0 : }
134 :
135 0 : pub(crate) async fn tenant_heatmap_upload(&self, tenant_id: TenantShardId) -> Result<()> {
136 0 : measured_request!(
137 0 : "tenant_heatmap_upload",
138 0 : crate::metrics::Method::Post,
139 0 : &self.node_id_label,
140 0 : self.inner.tenant_heatmap_upload(tenant_id).await
141 0 : )
142 0 : }
143 :
144 0 : pub(crate) async fn location_config(
145 0 : &self,
146 0 : tenant_shard_id: TenantShardId,
147 0 : config: LocationConfig,
148 0 : flush_ms: Option<std::time::Duration>,
149 0 : lazy: bool,
150 0 : ) -> Result<()> {
151 0 : measured_request!(
152 0 : "location_config",
153 0 : crate::metrics::Method::Put,
154 0 : &self.node_id_label,
155 0 : self.inner
156 0 : .location_config(tenant_shard_id, config, flush_ms, lazy)
157 0 : .await
158 0 : )
159 0 : }
160 :
161 0 : pub(crate) async fn list_location_config(&self) -> Result<LocationConfigListResponse> {
162 0 : measured_request!(
163 0 : "location_configs",
164 0 : crate::metrics::Method::Get,
165 0 : &self.node_id_label,
166 0 : self.inner.list_location_config().await
167 0 : )
168 0 : }
169 :
170 0 : pub(crate) async fn get_location_config(
171 0 : &self,
172 0 : tenant_shard_id: TenantShardId,
173 0 : ) -> Result<Option<LocationConfig>> {
174 0 : measured_request!(
175 0 : "location_config",
176 0 : crate::metrics::Method::Get,
177 0 : &self.node_id_label,
178 0 : self.inner.get_location_config(tenant_shard_id).await
179 0 : )
180 0 : }
181 :
182 0 : pub(crate) async fn timeline_create(
183 0 : &self,
184 0 : tenant_shard_id: TenantShardId,
185 0 : req: &TimelineCreateRequest,
186 0 : ) -> Result<TimelineInfo> {
187 0 : measured_request!(
188 0 : "timeline",
189 0 : crate::metrics::Method::Post,
190 0 : &self.node_id_label,
191 0 : self.inner.timeline_create(tenant_shard_id, req).await
192 0 : )
193 0 : }
194 :
195 0 : pub(crate) async fn timeline_delete(
196 0 : &self,
197 0 : tenant_shard_id: TenantShardId,
198 0 : timeline_id: TimelineId,
199 0 : ) -> Result<StatusCode> {
200 0 : measured_request!(
201 0 : "timeline",
202 0 : crate::metrics::Method::Delete,
203 0 : &self.node_id_label,
204 0 : self.inner
205 0 : .timeline_delete(tenant_shard_id, timeline_id)
206 0 : .await
207 0 : )
208 0 : }
209 :
210 0 : pub(crate) async fn tenant_shard_split(
211 0 : &self,
212 0 : tenant_shard_id: TenantShardId,
213 0 : req: TenantShardSplitRequest,
214 0 : ) -> Result<TenantShardSplitResponse> {
215 0 : measured_request!(
216 0 : "tenant_shard_split",
217 0 : crate::metrics::Method::Put,
218 0 : &self.node_id_label,
219 0 : self.inner.tenant_shard_split(tenant_shard_id, req).await
220 0 : )
221 0 : }
222 :
223 0 : pub(crate) async fn timeline_list(
224 0 : &self,
225 0 : tenant_shard_id: &TenantShardId,
226 0 : ) -> Result<Vec<TimelineInfo>> {
227 0 : measured_request!(
228 0 : "timelines",
229 0 : crate::metrics::Method::Get,
230 0 : &self.node_id_label,
231 0 : self.inner.timeline_list(tenant_shard_id).await
232 0 : )
233 0 : }
234 :
235 0 : pub(crate) async fn timeline_archival_config(
236 0 : &self,
237 0 : tenant_shard_id: TenantShardId,
238 0 : timeline_id: TimelineId,
239 0 : req: &TimelineArchivalConfigRequest,
240 0 : ) -> Result<()> {
241 0 : measured_request!(
242 0 : "timeline_archival_config",
243 0 : crate::metrics::Method::Put,
244 0 : &self.node_id_label,
245 0 : self.inner
246 0 : .timeline_archival_config(tenant_shard_id, timeline_id, req)
247 0 : .await
248 0 : )
249 0 : }
250 :
251 0 : pub(crate) async fn timeline_detach_ancestor(
252 0 : &self,
253 0 : tenant_shard_id: TenantShardId,
254 0 : timeline_id: TimelineId,
255 0 : ) -> Result<AncestorDetached> {
256 0 : measured_request!(
257 0 : "timeline_detach_ancestor",
258 0 : crate::metrics::Method::Put,
259 0 : &self.node_id_label,
260 0 : self.inner
261 0 : .timeline_detach_ancestor(tenant_shard_id, timeline_id)
262 0 : .await
263 0 : )
264 0 : }
265 :
266 0 : pub(crate) async fn timeline_block_unblock_gc(
267 0 : &self,
268 0 : tenant_shard_id: TenantShardId,
269 0 : timeline_id: TimelineId,
270 0 : dir: BlockUnblock,
271 0 : ) -> Result<()> {
272 0 : // measuring these makes no sense because we synchronize with the gc loop and remote
273 0 : // storage on block_gc so there should be huge outliers
274 0 : measured_request!(
275 0 : "timeline_block_unblock_gc",
276 0 : crate::metrics::Method::Post,
277 0 : &self.node_id_label,
278 0 : self.inner
279 0 : .timeline_block_unblock_gc(tenant_shard_id, timeline_id, dir)
280 0 : .await
281 0 : )
282 0 : }
283 :
284 0 : pub(crate) async fn timeline_download_heatmap_layers(
285 0 : &self,
286 0 : tenant_shard_id: TenantShardId,
287 0 : timeline_id: TimelineId,
288 0 : concurrency: Option<usize>,
289 0 : recurse: bool,
290 0 : ) -> Result<()> {
291 0 : measured_request!(
292 0 : "download_heatmap_layers",
293 0 : crate::metrics::Method::Post,
294 0 : &self.node_id_label,
295 0 : self.inner
296 0 : .timeline_download_heatmap_layers(
297 0 : tenant_shard_id,
298 0 : timeline_id,
299 0 : concurrency,
300 0 : recurse
301 0 : )
302 0 : .await
303 0 : )
304 0 : }
305 :
306 0 : pub(crate) async fn get_utilization(&self) -> Result<PageserverUtilization> {
307 0 : measured_request!(
308 0 : "utilization",
309 0 : crate::metrics::Method::Get,
310 0 : &self.node_id_label,
311 0 : self.inner.get_utilization().await
312 0 : )
313 0 : }
314 :
315 0 : pub(crate) async fn top_tenant_shards(
316 0 : &self,
317 0 : request: TopTenantShardsRequest,
318 0 : ) -> Result<TopTenantShardsResponse> {
319 0 : measured_request!(
320 0 : "top_tenants",
321 0 : crate::metrics::Method::Post,
322 0 : &self.node_id_label,
323 0 : self.inner.top_tenant_shards(request).await
324 0 : )
325 0 : }
326 :
327 0 : pub(crate) async fn wait_lsn(
328 0 : &self,
329 0 : tenant_shard_id: TenantShardId,
330 0 : request: TenantWaitLsnRequest,
331 0 : ) -> Result<StatusCode> {
332 0 : measured_request!(
333 0 : "wait_lsn",
334 0 : crate::metrics::Method::Post,
335 0 : &self.node_id_label,
336 0 : self.inner.wait_lsn(tenant_shard_id, request).await
337 0 : )
338 0 : }
339 : }
|
