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 bytes::Bytes;
11 : use measured::{label::LabelValue, metric::histogram, FixedCardinalityLabel, MetricGroup};
12 : use metrics::NeonMetrics;
13 : use once_cell::sync::Lazy;
14 : use std::sync::Mutex;
15 : use strum::IntoEnumIterator;
16 :
17 : use crate::{
18 : persistence::{DatabaseError, DatabaseOperation},
19 : service::LeadershipStatus,
20 : };
21 :
22 : pub(crate) static METRICS_REGISTRY: Lazy<StorageControllerMetrics> =
23 : Lazy::new(StorageControllerMetrics::default);
24 :
25 0 : pub fn preinitialize_metrics() {
26 0 : Lazy::force(&METRICS_REGISTRY);
27 0 : }
28 :
29 : pub(crate) struct StorageControllerMetrics {
30 : pub(crate) metrics_group: StorageControllerMetricGroup,
31 : encoder: Mutex<measured::text::BufferedTextEncoder>,
32 : }
33 :
34 3 : #[derive(measured::MetricGroup)]
35 : #[metric(new())]
36 : pub(crate) struct StorageControllerMetricGroup {
37 : /// Count of how many times we spawn a reconcile task
38 : pub(crate) storage_controller_reconcile_spawn: measured::Counter,
39 :
40 : /// Reconciler tasks completed, broken down by success/failure/cancelled
41 : pub(crate) storage_controller_reconcile_complete:
42 : measured::CounterVec<ReconcileCompleteLabelGroupSet>,
43 :
44 : /// Count of how many times we make an optimization change to a tenant's scheduling
45 : pub(crate) storage_controller_schedule_optimization: measured::Counter,
46 :
47 : /// HTTP request status counters for handled requests
48 : pub(crate) storage_controller_http_request_status:
49 : measured::CounterVec<HttpRequestStatusLabelGroupSet>,
50 :
51 : /// HTTP request handler latency across all status codes
52 : #[metric(metadata = histogram::Thresholds::exponential_buckets(0.1, 2.0))]
53 : pub(crate) storage_controller_http_request_latency:
54 : measured::HistogramVec<HttpRequestLatencyLabelGroupSet, 5>,
55 :
56 : /// Count of HTTP requests to the pageserver that resulted in an error,
57 : /// broken down by the pageserver node id, request name and method
58 : pub(crate) storage_controller_pageserver_request_error:
59 : measured::CounterVec<PageserverRequestLabelGroupSet>,
60 :
61 : /// Latency of HTTP requests to the pageserver, broken down by pageserver
62 : /// node id, request name and method. This include both successful and unsuccessful
63 : /// requests.
64 : #[metric(metadata = histogram::Thresholds::exponential_buckets(0.1, 2.0))]
65 : pub(crate) storage_controller_pageserver_request_latency:
66 : measured::HistogramVec<PageserverRequestLabelGroupSet, 5>,
67 :
68 : /// Count of pass-through HTTP requests to the pageserver that resulted in an error,
69 : /// broken down by the pageserver node id, request name and method
70 : pub(crate) storage_controller_passthrough_request_error:
71 : measured::CounterVec<PageserverRequestLabelGroupSet>,
72 :
73 : /// Latency of pass-through HTTP requests to the pageserver, broken down by pageserver
74 : /// node id, request name and method. This include both successful and unsuccessful
75 : /// requests.
76 : #[metric(metadata = histogram::Thresholds::exponential_buckets(0.1, 2.0))]
77 : pub(crate) storage_controller_passthrough_request_latency:
78 : measured::HistogramVec<PageserverRequestLabelGroupSet, 5>,
79 :
80 : /// Count of errors in database queries, broken down by error type and operation.
81 : pub(crate) storage_controller_database_query_error:
82 : measured::CounterVec<DatabaseQueryErrorLabelGroupSet>,
83 :
84 : /// Latency of database queries, broken down by operation.
85 : #[metric(metadata = histogram::Thresholds::exponential_buckets(0.1, 2.0))]
86 : pub(crate) storage_controller_database_query_latency:
87 : measured::HistogramVec<DatabaseQueryLatencyLabelGroupSet, 5>,
88 :
89 : pub(crate) storage_controller_leadership_status: measured::GaugeVec<LeadershipStatusGroupSet>,
90 :
91 : /// HTTP request status counters for handled requests
92 : pub(crate) storage_controller_reconcile_long_running:
93 : measured::CounterVec<ReconcileLongRunningLabelGroupSet>,
94 : }
95 :
96 : impl StorageControllerMetrics {
97 0 : pub(crate) fn encode(&self, neon_metrics: &NeonMetrics) -> Bytes {
98 0 : let mut encoder = self.encoder.lock().unwrap();
99 0 : neon_metrics
100 0 : .collect_group_into(&mut *encoder)
101 0 : .unwrap_or_else(|infallible| match infallible {});
102 0 : self.metrics_group
103 0 : .collect_group_into(&mut *encoder)
104 0 : .unwrap_or_else(|infallible| match infallible {});
105 0 : encoder.finish()
106 0 : }
107 : }
108 :
109 : impl Default for StorageControllerMetrics {
110 3 : fn default() -> Self {
111 3 : let mut metrics_group = StorageControllerMetricGroup::new();
112 3 : metrics_group
113 3 : .storage_controller_reconcile_complete
114 3 : .init_all_dense();
115 3 :
116 3 : Self {
117 3 : metrics_group,
118 3 : encoder: Mutex::new(measured::text::BufferedTextEncoder::new()),
119 3 : }
120 3 : }
121 : }
122 :
123 9 : #[derive(measured::LabelGroup)]
124 : #[label(set = ReconcileCompleteLabelGroupSet)]
125 : pub(crate) struct ReconcileCompleteLabelGroup {
126 : pub(crate) status: ReconcileOutcome,
127 : }
128 :
129 6 : #[derive(measured::LabelGroup)]
130 : #[label(set = HttpRequestStatusLabelGroupSet)]
131 : pub(crate) struct HttpRequestStatusLabelGroup<'a> {
132 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
133 : pub(crate) path: &'a str,
134 : pub(crate) method: Method,
135 : pub(crate) status: StatusCode,
136 : }
137 :
138 6 : #[derive(measured::LabelGroup)]
139 : #[label(set = HttpRequestLatencyLabelGroupSet)]
140 : pub(crate) struct HttpRequestLatencyLabelGroup<'a> {
141 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
142 : pub(crate) path: &'a str,
143 : pub(crate) method: Method,
144 : }
145 :
146 24 : #[derive(measured::LabelGroup, Clone)]
147 : #[label(set = PageserverRequestLabelGroupSet)]
148 : pub(crate) struct PageserverRequestLabelGroup<'a> {
149 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
150 : pub(crate) pageserver_id: &'a str,
151 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
152 : pub(crate) path: &'a str,
153 : pub(crate) method: Method,
154 : }
155 :
156 12 : #[derive(measured::LabelGroup)]
157 : #[label(set = DatabaseQueryErrorLabelGroupSet)]
158 : pub(crate) struct DatabaseQueryErrorLabelGroup {
159 : pub(crate) error_type: DatabaseErrorLabel,
160 : pub(crate) operation: DatabaseOperation,
161 : }
162 :
163 9 : #[derive(measured::LabelGroup)]
164 : #[label(set = DatabaseQueryLatencyLabelGroupSet)]
165 : pub(crate) struct DatabaseQueryLatencyLabelGroup {
166 : pub(crate) operation: DatabaseOperation,
167 : }
168 :
169 9 : #[derive(measured::LabelGroup)]
170 : #[label(set = LeadershipStatusGroupSet)]
171 : pub(crate) struct LeadershipStatusGroup {
172 : pub(crate) status: LeadershipStatus,
173 : }
174 :
175 6 : #[derive(measured::LabelGroup, Clone)]
176 : #[label(set = ReconcileLongRunningLabelGroupSet)]
177 : pub(crate) struct ReconcileLongRunningLabelGroup<'a> {
178 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
179 : pub(crate) tenant_id: &'a str,
180 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
181 : pub(crate) shard_number: &'a str,
182 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
183 : pub(crate) sequence: &'a str,
184 : }
185 :
186 : #[derive(FixedCardinalityLabel, Clone, Copy)]
187 : pub(crate) enum ReconcileOutcome {
188 : #[label(rename = "ok")]
189 : Success,
190 : Error,
191 : Cancel,
192 : }
193 :
194 : #[derive(FixedCardinalityLabel, Copy, Clone)]
195 : pub(crate) enum Method {
196 : Get,
197 : Put,
198 : Post,
199 : Delete,
200 : Other,
201 : }
202 :
203 : impl From<hyper::Method> for Method {
204 0 : fn from(value: hyper::Method) -> Self {
205 0 : if value == hyper::Method::GET {
206 0 : Method::Get
207 0 : } else if value == hyper::Method::PUT {
208 0 : Method::Put
209 0 : } else if value == hyper::Method::POST {
210 0 : Method::Post
211 0 : } else if value == hyper::Method::DELETE {
212 0 : Method::Delete
213 : } else {
214 0 : Method::Other
215 : }
216 0 : }
217 : }
218 :
219 : #[derive(Clone, Copy)]
220 : pub(crate) struct StatusCode(pub(crate) hyper::http::StatusCode);
221 :
222 : impl LabelValue for StatusCode {
223 0 : fn visit<V: measured::label::LabelVisitor>(&self, v: V) -> V::Output {
224 0 : v.write_int(self.0.as_u16() as i64)
225 0 : }
226 : }
227 :
228 : impl FixedCardinalityLabel for StatusCode {
229 0 : fn cardinality() -> usize {
230 0 : (100..1000).len()
231 0 : }
232 :
233 0 : fn encode(&self) -> usize {
234 0 : self.0.as_u16() as usize
235 0 : }
236 :
237 0 : fn decode(value: usize) -> Self {
238 0 : Self(hyper::http::StatusCode::from_u16(u16::try_from(value).unwrap()).unwrap())
239 0 : }
240 : }
241 :
242 : #[derive(FixedCardinalityLabel, Clone, Copy)]
243 : pub(crate) enum DatabaseErrorLabel {
244 : Query,
245 : Connection,
246 : ConnectionPool,
247 : Logical,
248 : Migration,
249 : }
250 :
251 : impl DatabaseError {
252 0 : pub(crate) fn error_label(&self) -> DatabaseErrorLabel {
253 0 : match self {
254 0 : Self::Query(_) => DatabaseErrorLabel::Query,
255 0 : Self::Connection(_) => DatabaseErrorLabel::Connection,
256 0 : Self::ConnectionPool(_) => DatabaseErrorLabel::ConnectionPool,
257 0 : Self::Logical(_) => DatabaseErrorLabel::Logical,
258 0 : Self::Migration(_) => DatabaseErrorLabel::Migration,
259 : }
260 0 : }
261 : }
262 :
263 : /// Update the leadership status metric gauges to reflect the requested status
264 0 : pub(crate) fn update_leadership_status(status: LeadershipStatus) {
265 0 : let status_metric = &METRICS_REGISTRY
266 0 : .metrics_group
267 0 : .storage_controller_leadership_status;
268 :
269 0 : for s in LeadershipStatus::iter() {
270 0 : if s == status {
271 0 : status_metric.set(LeadershipStatusGroup { status: s }, 1);
272 0 : } else {
273 0 : status_metric.set(LeadershipStatusGroup { status: s }, 0);
274 0 : }
275 : }
276 0 : }
|
