Line data Source code
1 : //!
2 : //! This module provides metric definitions for the storage controller.
3 : //!
4 : //! All metrics are grouped in [`StorageControllerMetricGroup`]. [`StorageControllerMetrics`] holds
5 : //! the mentioned metrics and their encoder. It's globally available via the [`METRICS_REGISTRY`]
6 : //! constant.
7 : //!
8 : //! The rest of the code defines label group types and deals with converting outer types to labels.
9 : //!
10 : use std::sync::Mutex;
11 :
12 : use bytes::Bytes;
13 : use measured::label::LabelValue;
14 : use measured::metric::histogram;
15 : use measured::{FixedCardinalityLabel, MetricGroup};
16 : use metrics::NeonMetrics;
17 : use once_cell::sync::Lazy;
18 : use strum::IntoEnumIterator;
19 :
20 : use crate::persistence::{DatabaseError, DatabaseOperation};
21 : use crate::service::LeadershipStatus;
22 :
23 : pub(crate) static METRICS_REGISTRY: Lazy<StorageControllerMetrics> =
24 : Lazy::new(StorageControllerMetrics::default);
25 :
26 0 : pub fn preinitialize_metrics() {
27 0 : Lazy::force(&METRICS_REGISTRY);
28 0 : }
29 :
30 : pub(crate) struct StorageControllerMetrics {
31 : pub(crate) metrics_group: StorageControllerMetricGroup,
32 : encoder: Mutex<measured::text::BufferedTextEncoder>,
33 : }
34 :
35 : #[derive(measured::MetricGroup)]
36 : #[metric(new())]
37 : pub(crate) struct StorageControllerMetricGroup {
38 : /// Count of how many times we spawn a reconcile task
39 : pub(crate) storage_controller_reconcile_spawn: measured::Counter,
40 :
41 : /// Size of the in-memory map of tenant shards
42 : pub(crate) storage_controller_tenant_shards: measured::Gauge,
43 :
44 : /// Size of the in-memory map of pageserver_nodes
45 : pub(crate) storage_controller_pageserver_nodes: measured::Gauge,
46 :
47 : /// Count of how many pageserver nodes from in-memory map have https configured
48 : pub(crate) storage_controller_https_pageserver_nodes: measured::Gauge,
49 :
50 : /// Size of the in-memory map of safekeeper_nodes
51 : pub(crate) storage_controller_safekeeper_nodes: measured::Gauge,
52 :
53 : /// Count of how many safekeeper nodes from in-memory map have https configured
54 : pub(crate) storage_controller_https_safekeeper_nodes: measured::Gauge,
55 :
56 : /// Reconciler tasks completed, broken down by success/failure/cancelled
57 : pub(crate) storage_controller_reconcile_complete:
58 : measured::CounterVec<ReconcileCompleteLabelGroupSet>,
59 :
60 : /// Count of how many times we make an optimization change to a tenant's scheduling
61 : pub(crate) storage_controller_schedule_optimization: measured::Counter,
62 :
63 : /// How many shards are not scheduled into their preferred AZ
64 : pub(crate) storage_controller_schedule_az_violation: measured::Gauge,
65 :
66 : /// How many shard locations (secondary or attached) on each node
67 : pub(crate) storage_controller_node_shards: measured::GaugeVec<NodeLabelGroupSet>,
68 :
69 : /// How many _attached_ shard locations on each node
70 : pub(crate) storage_controller_node_attached_shards: measured::GaugeVec<NodeLabelGroupSet>,
71 :
72 : /// How many _home_ shard locations on each node (i.e. the node's AZ matches the shard's
73 : /// preferred AZ)
74 : pub(crate) storage_controller_node_home_shards: measured::GaugeVec<NodeLabelGroupSet>,
75 :
76 : /// How many shards would like to reconcile but were blocked by concurrency limits
77 : pub(crate) storage_controller_pending_reconciles: measured::Gauge,
78 :
79 : /// How many shards are keep-failing and will be ignored when considering to run optimizations
80 : pub(crate) storage_controller_keep_failing_reconciles: measured::Gauge,
81 :
82 : /// HTTP request status counters for handled requests
83 : pub(crate) storage_controller_http_request_status:
84 : measured::CounterVec<HttpRequestStatusLabelGroupSet>,
85 :
86 : /// HTTP request handler latency across all status codes
87 : #[metric(metadata = histogram::Thresholds::exponential_buckets(0.1, 2.0))]
88 : pub(crate) storage_controller_http_request_latency:
89 : measured::HistogramVec<HttpRequestLatencyLabelGroupSet, 5>,
90 :
91 : /// HTTP rate limiting latency across all tenants and endpoints
92 : #[metric(metadata = histogram::Thresholds::exponential_buckets(0.1, 10.0))]
93 : pub(crate) storage_controller_http_request_rate_limited: measured::Histogram<10>,
94 :
95 : /// Count of HTTP requests to the pageserver that resulted in an error,
96 : /// broken down by the pageserver node id, request name and method
97 : pub(crate) storage_controller_pageserver_request_error:
98 : measured::CounterVec<PageserverRequestLabelGroupSet>,
99 :
100 : /// Count of HTTP requests to the safekeeper that resulted in an error,
101 : /// broken down by the safekeeper node id, request name and method
102 : pub(crate) storage_controller_safekeeper_request_error:
103 : measured::CounterVec<SafekeeperRequestLabelGroupSet>,
104 :
105 : /// Latency of HTTP requests to the pageserver, broken down by pageserver
106 : /// node id, request name and method. This include both successful and unsuccessful
107 : /// requests.
108 : #[metric(metadata = histogram::Thresholds::exponential_buckets(0.1, 2.0))]
109 : pub(crate) storage_controller_pageserver_request_latency:
110 : measured::HistogramVec<PageserverRequestLabelGroupSet, 5>,
111 :
112 : /// Latency of HTTP requests to the safekeeper, broken down by safekeeper
113 : /// node id, request name and method. This include both successful and unsuccessful
114 : /// requests.
115 : #[metric(metadata = histogram::Thresholds::exponential_buckets(0.1, 2.0))]
116 : pub(crate) storage_controller_safekeeper_request_latency:
117 : measured::HistogramVec<SafekeeperRequestLabelGroupSet, 5>,
118 :
119 : /// Count of pass-through HTTP requests to the pageserver that resulted in an error,
120 : /// broken down by the pageserver node id, request name and method
121 : pub(crate) storage_controller_passthrough_request_error:
122 : measured::CounterVec<PageserverRequestLabelGroupSet>,
123 :
124 : /// Latency of pass-through HTTP requests to the pageserver, broken down by pageserver
125 : /// node id, request name and method. This include both successful and unsuccessful
126 : /// requests.
127 : #[metric(metadata = histogram::Thresholds::exponential_buckets(0.1, 2.0))]
128 : pub(crate) storage_controller_passthrough_request_latency:
129 : measured::HistogramVec<PageserverRequestLabelGroupSet, 5>,
130 :
131 : /// Count of errors in database queries, broken down by error type and operation.
132 : pub(crate) storage_controller_database_query_error:
133 : measured::CounterVec<DatabaseQueryErrorLabelGroupSet>,
134 :
135 : /// Latency of database queries, broken down by operation.
136 : #[metric(metadata = histogram::Thresholds::exponential_buckets(0.1, 2.0))]
137 : pub(crate) storage_controller_database_query_latency:
138 : measured::HistogramVec<DatabaseQueryLatencyLabelGroupSet, 5>,
139 :
140 : pub(crate) storage_controller_leadership_status: measured::GaugeVec<LeadershipStatusGroupSet>,
141 :
142 : /// Indicator of stucked (long-running) reconciles, broken down by tenant, shard and sequence.
143 : /// The metric is automatically removed once the reconciliation completes.
144 : pub(crate) storage_controller_reconcile_long_running:
145 : measured::CounterVec<ReconcileLongRunningLabelGroupSet>,
146 :
147 : /// Indicator of safekeeper reconciler queue depth, broken down by safekeeper, excluding ongoing reconciles.
148 : pub(crate) storage_controller_safekeeper_reconciles_queued:
149 : measured::GaugeVec<SafekeeperReconcilerLabelGroupSet>,
150 :
151 : /// Indicator of completed safekeeper reconciles, broken down by safekeeper.
152 : pub(crate) storage_controller_safekeeper_reconciles_complete:
153 : measured::CounterVec<SafekeeperReconcilerLabelGroupSet>,
154 : }
155 :
156 : impl StorageControllerMetrics {
157 0 : pub(crate) fn encode(&self, neon_metrics: &NeonMetrics) -> Bytes {
158 0 : let mut encoder = self.encoder.lock().unwrap();
159 0 : neon_metrics
160 0 : .collect_group_into(&mut *encoder)
161 0 : .unwrap_or_else(|infallible| match infallible {});
162 0 : self.metrics_group
163 0 : .collect_group_into(&mut *encoder)
164 0 : .unwrap_or_else(|infallible| match infallible {});
165 0 : encoder.finish()
166 0 : }
167 : }
168 :
169 : impl Default for StorageControllerMetrics {
170 17 : fn default() -> Self {
171 17 : let mut metrics_group = StorageControllerMetricGroup::new();
172 17 : metrics_group
173 17 : .storage_controller_reconcile_complete
174 17 : .init_all_dense();
175 :
176 17 : Self {
177 17 : metrics_group,
178 17 : encoder: Mutex::new(measured::text::BufferedTextEncoder::new()),
179 17 : }
180 17 : }
181 : }
182 :
183 : #[derive(measured::LabelGroup, Clone)]
184 : #[label(set = NodeLabelGroupSet)]
185 : pub(crate) struct NodeLabelGroup<'a> {
186 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
187 : pub(crate) az: &'a str,
188 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
189 : pub(crate) node_id: &'a str,
190 : }
191 :
192 : #[derive(measured::LabelGroup)]
193 : #[label(set = ReconcileCompleteLabelGroupSet)]
194 : pub(crate) struct ReconcileCompleteLabelGroup {
195 : pub(crate) status: ReconcileOutcome,
196 : }
197 :
198 : #[derive(measured::LabelGroup)]
199 : #[label(set = HttpRequestStatusLabelGroupSet)]
200 : pub(crate) struct HttpRequestStatusLabelGroup<'a> {
201 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
202 : pub(crate) path: &'a str,
203 : pub(crate) method: Method,
204 : pub(crate) status: StatusCode,
205 : }
206 :
207 : #[derive(measured::LabelGroup)]
208 : #[label(set = HttpRequestLatencyLabelGroupSet)]
209 : pub(crate) struct HttpRequestLatencyLabelGroup<'a> {
210 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
211 : pub(crate) path: &'a str,
212 : pub(crate) method: Method,
213 : }
214 :
215 : #[derive(measured::LabelGroup, Clone)]
216 : #[label(set = PageserverRequestLabelGroupSet)]
217 : pub(crate) struct PageserverRequestLabelGroup<'a> {
218 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
219 : pub(crate) pageserver_id: &'a str,
220 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
221 : pub(crate) path: &'a str,
222 : pub(crate) method: Method,
223 : }
224 :
225 : #[derive(measured::LabelGroup, Clone)]
226 : #[label(set = SafekeeperRequestLabelGroupSet)]
227 : pub(crate) struct SafekeeperRequestLabelGroup<'a> {
228 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
229 : pub(crate) safekeeper_id: &'a str,
230 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
231 : pub(crate) path: &'a str,
232 : pub(crate) method: Method,
233 : }
234 :
235 : #[derive(measured::LabelGroup)]
236 : #[label(set = DatabaseQueryErrorLabelGroupSet)]
237 : pub(crate) struct DatabaseQueryErrorLabelGroup {
238 : pub(crate) error_type: DatabaseErrorLabel,
239 : pub(crate) operation: DatabaseOperation,
240 : }
241 :
242 : #[derive(measured::LabelGroup)]
243 : #[label(set = DatabaseQueryLatencyLabelGroupSet)]
244 : pub(crate) struct DatabaseQueryLatencyLabelGroup {
245 : pub(crate) operation: DatabaseOperation,
246 : }
247 :
248 : #[derive(measured::LabelGroup)]
249 : #[label(set = LeadershipStatusGroupSet)]
250 : pub(crate) struct LeadershipStatusGroup {
251 : pub(crate) status: LeadershipStatus,
252 : }
253 :
254 : #[derive(measured::LabelGroup, Clone)]
255 : #[label(set = ReconcileLongRunningLabelGroupSet)]
256 : pub(crate) struct ReconcileLongRunningLabelGroup<'a> {
257 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
258 : pub(crate) tenant_id: &'a str,
259 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
260 : pub(crate) shard_number: &'a str,
261 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
262 : pub(crate) sequence: &'a str,
263 : }
264 :
265 : #[derive(FixedCardinalityLabel, Clone, Copy)]
266 : pub(crate) enum ReconcileOutcome {
267 : #[label(rename = "ok")]
268 : Success,
269 : Error,
270 : Cancel,
271 : }
272 :
273 : #[derive(FixedCardinalityLabel, Copy, Clone)]
274 : pub(crate) enum Method {
275 : Get,
276 : Put,
277 : Post,
278 : Delete,
279 : Other,
280 : }
281 :
282 : #[derive(measured::LabelGroup, Clone)]
283 : #[label(set = SafekeeperReconcilerLabelGroupSet)]
284 : pub(crate) struct SafekeeperReconcilerLabelGroup<'a> {
285 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
286 : pub(crate) sk_az: &'a str,
287 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
288 : pub(crate) sk_node_id: &'a str,
289 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
290 : pub(crate) sk_hostname: &'a str,
291 : }
292 :
293 : impl From<hyper::Method> for Method {
294 0 : fn from(value: hyper::Method) -> Self {
295 0 : if value == hyper::Method::GET {
296 0 : Method::Get
297 0 : } else if value == hyper::Method::PUT {
298 0 : Method::Put
299 0 : } else if value == hyper::Method::POST {
300 0 : Method::Post
301 0 : } else if value == hyper::Method::DELETE {
302 0 : Method::Delete
303 : } else {
304 0 : Method::Other
305 : }
306 0 : }
307 : }
308 :
309 : #[derive(Clone, Copy)]
310 : pub(crate) struct StatusCode(pub(crate) hyper::http::StatusCode);
311 :
312 : impl LabelValue for StatusCode {
313 0 : fn visit<V: measured::label::LabelVisitor>(&self, v: V) -> V::Output {
314 0 : v.write_int(self.0.as_u16() as i64)
315 0 : }
316 : }
317 :
318 : impl FixedCardinalityLabel for StatusCode {
319 0 : fn cardinality() -> usize {
320 0 : (100..1000).len()
321 0 : }
322 :
323 0 : fn encode(&self) -> usize {
324 0 : self.0.as_u16() as usize
325 0 : }
326 :
327 0 : fn decode(value: usize) -> Self {
328 0 : Self(hyper::http::StatusCode::from_u16(u16::try_from(value).unwrap()).unwrap())
329 0 : }
330 : }
331 :
332 : #[derive(FixedCardinalityLabel, Clone, Copy)]
333 : pub(crate) enum DatabaseErrorLabel {
334 : Query,
335 : Connection,
336 : ConnectionPool,
337 : Logical,
338 : Migration,
339 : Cas,
340 : }
341 :
342 : impl DatabaseError {
343 0 : pub(crate) fn error_label(&self) -> DatabaseErrorLabel {
344 0 : match self {
345 0 : Self::Query(_) => DatabaseErrorLabel::Query,
346 0 : Self::Connection(_) => DatabaseErrorLabel::Connection,
347 0 : Self::ConnectionPool(_) => DatabaseErrorLabel::ConnectionPool,
348 0 : Self::Logical(_) => DatabaseErrorLabel::Logical,
349 0 : Self::Migration(_) => DatabaseErrorLabel::Migration,
350 0 : Self::Cas(_) => DatabaseErrorLabel::Cas,
351 : }
352 0 : }
353 : }
354 :
355 : /// Update the leadership status metric gauges to reflect the requested status
356 0 : pub(crate) fn update_leadership_status(status: LeadershipStatus) {
357 0 : let status_metric = &METRICS_REGISTRY
358 0 : .metrics_group
359 0 : .storage_controller_leadership_status;
360 :
361 0 : for s in LeadershipStatus::iter() {
362 0 : if s == status {
363 0 : status_metric.set(LeadershipStatusGroup { status: s }, 1);
364 0 : } else {
365 0 : status_metric.set(LeadershipStatusGroup { status: s }, 0);
366 0 : }
367 : }
368 0 : }
|
