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 17 : #[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 : /// Reconciler tasks completed, broken down by success/failure/cancelled
48 : pub(crate) storage_controller_reconcile_complete:
49 : measured::CounterVec<ReconcileCompleteLabelGroupSet>,
50 :
51 : /// Count of how many times we make an optimization change to a tenant's scheduling
52 : pub(crate) storage_controller_schedule_optimization: measured::Counter,
53 :
54 : /// How many shards are not scheduled into their preferred AZ
55 : pub(crate) storage_controller_schedule_az_violation: measured::Gauge,
56 :
57 : /// How many shard locations (secondary or attached) on each node
58 : pub(crate) storage_controller_node_shards: measured::GaugeVec<NodeLabelGroupSet>,
59 :
60 : /// How many _attached_ shard locations on each node
61 : pub(crate) storage_controller_node_attached_shards: measured::GaugeVec<NodeLabelGroupSet>,
62 :
63 : /// How many _home_ shard locations on each node (i.e. the node's AZ matches the shard's
64 : /// preferred AZ)
65 : pub(crate) storage_controller_node_home_shards: measured::GaugeVec<NodeLabelGroupSet>,
66 :
67 : /// How many shards would like to reconcile but were blocked by concurrency limits
68 : pub(crate) storage_controller_pending_reconciles: measured::Gauge,
69 :
70 : /// HTTP request status counters for handled requests
71 : pub(crate) storage_controller_http_request_status:
72 : measured::CounterVec<HttpRequestStatusLabelGroupSet>,
73 :
74 : /// HTTP request handler latency across all status codes
75 : #[metric(metadata = histogram::Thresholds::exponential_buckets(0.1, 2.0))]
76 : pub(crate) storage_controller_http_request_latency:
77 : measured::HistogramVec<HttpRequestLatencyLabelGroupSet, 5>,
78 :
79 : /// HTTP rate limiting latency across all tenants and endpoints
80 : #[metric(metadata = histogram::Thresholds::exponential_buckets(0.1, 10.0))]
81 : pub(crate) storage_controller_http_request_rate_limited: measured::Histogram<10>,
82 :
83 : /// Count of HTTP requests to the pageserver that resulted in an error,
84 : /// broken down by the pageserver node id, request name and method
85 : pub(crate) storage_controller_pageserver_request_error:
86 : measured::CounterVec<PageserverRequestLabelGroupSet>,
87 :
88 : /// Count of HTTP requests to the safekeeper that resulted in an error,
89 : /// broken down by the safekeeper node id, request name and method
90 : pub(crate) storage_controller_safekeeper_request_error:
91 : measured::CounterVec<PageserverRequestLabelGroupSet>,
92 :
93 : /// Latency of HTTP requests to the pageserver, broken down by pageserver
94 : /// node id, request name and method. This include both successful and unsuccessful
95 : /// requests.
96 : #[metric(metadata = histogram::Thresholds::exponential_buckets(0.1, 2.0))]
97 : pub(crate) storage_controller_pageserver_request_latency:
98 : measured::HistogramVec<PageserverRequestLabelGroupSet, 5>,
99 :
100 : /// Latency of HTTP requests to the safekeeper, broken down by safekeeper
101 : /// node id, request name and method. This include both successful and unsuccessful
102 : /// requests.
103 : #[metric(metadata = histogram::Thresholds::exponential_buckets(0.1, 2.0))]
104 : pub(crate) storage_controller_safekeeper_request_latency:
105 : measured::HistogramVec<PageserverRequestLabelGroupSet, 5>,
106 :
107 : /// Count of pass-through HTTP requests to the pageserver that resulted in an error,
108 : /// broken down by the pageserver node id, request name and method
109 : pub(crate) storage_controller_passthrough_request_error:
110 : measured::CounterVec<PageserverRequestLabelGroupSet>,
111 :
112 : /// Latency of pass-through HTTP requests to the pageserver, broken down by pageserver
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_passthrough_request_latency:
117 : measured::HistogramVec<PageserverRequestLabelGroupSet, 5>,
118 :
119 : /// Count of errors in database queries, broken down by error type and operation.
120 : pub(crate) storage_controller_database_query_error:
121 : measured::CounterVec<DatabaseQueryErrorLabelGroupSet>,
122 :
123 : /// Latency of database queries, broken down by operation.
124 : #[metric(metadata = histogram::Thresholds::exponential_buckets(0.1, 2.0))]
125 : pub(crate) storage_controller_database_query_latency:
126 : measured::HistogramVec<DatabaseQueryLatencyLabelGroupSet, 5>,
127 :
128 : pub(crate) storage_controller_leadership_status: measured::GaugeVec<LeadershipStatusGroupSet>,
129 :
130 : /// HTTP request status counters for handled requests
131 : pub(crate) storage_controller_reconcile_long_running:
132 : measured::CounterVec<ReconcileLongRunningLabelGroupSet>,
133 : }
134 :
135 : impl StorageControllerMetrics {
136 0 : pub(crate) fn encode(&self, neon_metrics: &NeonMetrics) -> Bytes {
137 0 : let mut encoder = self.encoder.lock().unwrap();
138 0 : neon_metrics
139 0 : .collect_group_into(&mut *encoder)
140 0 : .unwrap_or_else(|infallible| match infallible {});
141 0 : self.metrics_group
142 0 : .collect_group_into(&mut *encoder)
143 0 : .unwrap_or_else(|infallible| match infallible {});
144 0 : encoder.finish()
145 0 : }
146 : }
147 :
148 : impl Default for StorageControllerMetrics {
149 17 : fn default() -> Self {
150 17 : let mut metrics_group = StorageControllerMetricGroup::new();
151 17 : metrics_group
152 17 : .storage_controller_reconcile_complete
153 17 : .init_all_dense();
154 17 :
155 17 : Self {
156 17 : metrics_group,
157 17 : encoder: Mutex::new(measured::text::BufferedTextEncoder::new()),
158 17 : }
159 17 : }
160 : }
161 :
162 102 : #[derive(measured::LabelGroup, Clone)]
163 : #[label(set = NodeLabelGroupSet)]
164 : pub(crate) struct NodeLabelGroup<'a> {
165 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
166 : pub(crate) az: &'a str,
167 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
168 : pub(crate) node_id: &'a str,
169 : }
170 :
171 51 : #[derive(measured::LabelGroup)]
172 : #[label(set = ReconcileCompleteLabelGroupSet)]
173 : pub(crate) struct ReconcileCompleteLabelGroup {
174 : pub(crate) status: ReconcileOutcome,
175 : }
176 :
177 34 : #[derive(measured::LabelGroup)]
178 : #[label(set = HttpRequestStatusLabelGroupSet)]
179 : pub(crate) struct HttpRequestStatusLabelGroup<'a> {
180 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
181 : pub(crate) path: &'a str,
182 : pub(crate) method: Method,
183 : pub(crate) status: StatusCode,
184 : }
185 :
186 34 : #[derive(measured::LabelGroup)]
187 : #[label(set = HttpRequestLatencyLabelGroupSet)]
188 : pub(crate) struct HttpRequestLatencyLabelGroup<'a> {
189 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
190 : pub(crate) path: &'a str,
191 : pub(crate) method: Method,
192 : }
193 :
194 204 : #[derive(measured::LabelGroup, Clone)]
195 : #[label(set = PageserverRequestLabelGroupSet)]
196 : pub(crate) struct PageserverRequestLabelGroup<'a> {
197 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
198 : pub(crate) pageserver_id: &'a str,
199 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
200 : pub(crate) path: &'a str,
201 : pub(crate) method: Method,
202 : }
203 :
204 68 : #[derive(measured::LabelGroup)]
205 : #[label(set = DatabaseQueryErrorLabelGroupSet)]
206 : pub(crate) struct DatabaseQueryErrorLabelGroup {
207 : pub(crate) error_type: DatabaseErrorLabel,
208 : pub(crate) operation: DatabaseOperation,
209 : }
210 :
211 51 : #[derive(measured::LabelGroup)]
212 : #[label(set = DatabaseQueryLatencyLabelGroupSet)]
213 : pub(crate) struct DatabaseQueryLatencyLabelGroup {
214 : pub(crate) operation: DatabaseOperation,
215 : }
216 :
217 51 : #[derive(measured::LabelGroup)]
218 : #[label(set = LeadershipStatusGroupSet)]
219 : pub(crate) struct LeadershipStatusGroup {
220 : pub(crate) status: LeadershipStatus,
221 : }
222 :
223 34 : #[derive(measured::LabelGroup, Clone)]
224 : #[label(set = ReconcileLongRunningLabelGroupSet)]
225 : pub(crate) struct ReconcileLongRunningLabelGroup<'a> {
226 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
227 : pub(crate) tenant_id: &'a str,
228 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
229 : pub(crate) shard_number: &'a str,
230 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
231 : pub(crate) sequence: &'a str,
232 : }
233 :
234 : #[derive(FixedCardinalityLabel, Clone, Copy)]
235 : pub(crate) enum ReconcileOutcome {
236 : #[label(rename = "ok")]
237 : Success,
238 : Error,
239 : Cancel,
240 : }
241 :
242 : #[derive(FixedCardinalityLabel, Copy, Clone)]
243 : pub(crate) enum Method {
244 : Get,
245 : Put,
246 : Post,
247 : Delete,
248 : Other,
249 : }
250 :
251 : impl From<hyper::Method> for Method {
252 0 : fn from(value: hyper::Method) -> Self {
253 0 : if value == hyper::Method::GET {
254 0 : Method::Get
255 0 : } else if value == hyper::Method::PUT {
256 0 : Method::Put
257 0 : } else if value == hyper::Method::POST {
258 0 : Method::Post
259 0 : } else if value == hyper::Method::DELETE {
260 0 : Method::Delete
261 : } else {
262 0 : Method::Other
263 : }
264 0 : }
265 : }
266 :
267 : #[derive(Clone, Copy)]
268 : pub(crate) struct StatusCode(pub(crate) hyper::http::StatusCode);
269 :
270 : impl LabelValue for StatusCode {
271 0 : fn visit<V: measured::label::LabelVisitor>(&self, v: V) -> V::Output {
272 0 : v.write_int(self.0.as_u16() as i64)
273 0 : }
274 : }
275 :
276 : impl FixedCardinalityLabel for StatusCode {
277 0 : fn cardinality() -> usize {
278 0 : (100..1000).len()
279 0 : }
280 :
281 0 : fn encode(&self) -> usize {
282 0 : self.0.as_u16() as usize
283 0 : }
284 :
285 0 : fn decode(value: usize) -> Self {
286 0 : Self(hyper::http::StatusCode::from_u16(u16::try_from(value).unwrap()).unwrap())
287 0 : }
288 : }
289 :
290 : #[derive(FixedCardinalityLabel, Clone, Copy)]
291 : pub(crate) enum DatabaseErrorLabel {
292 : Query,
293 : Connection,
294 : ConnectionPool,
295 : Logical,
296 : Migration,
297 : }
298 :
299 : impl DatabaseError {
300 0 : pub(crate) fn error_label(&self) -> DatabaseErrorLabel {
301 0 : match self {
302 0 : Self::Query(_) => DatabaseErrorLabel::Query,
303 0 : Self::Connection(_) => DatabaseErrorLabel::Connection,
304 0 : Self::ConnectionPool(_) => DatabaseErrorLabel::ConnectionPool,
305 0 : Self::Logical(_) => DatabaseErrorLabel::Logical,
306 0 : Self::Migration(_) => DatabaseErrorLabel::Migration,
307 : }
308 0 : }
309 : }
310 :
311 : /// Update the leadership status metric gauges to reflect the requested status
312 0 : pub(crate) fn update_leadership_status(status: LeadershipStatus) {
313 0 : let status_metric = &METRICS_REGISTRY
314 0 : .metrics_group
315 0 : .storage_controller_leadership_status;
316 :
317 0 : for s in LeadershipStatus::iter() {
318 0 : if s == status {
319 0 : status_metric.set(LeadershipStatusGroup { status: s }, 1);
320 0 : } else {
321 0 : status_metric.set(LeadershipStatusGroup { status: s }, 0);
322 0 : }
323 : }
324 0 : }
|