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 13 : #[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 : /// Size of the in-memory map of tenant shards
41 : pub(crate) storage_controller_tenant_shards: measured::Gauge,
42 :
43 : /// Size of the in-memory map of pageserver_nodes
44 : pub(crate) storage_controller_pageserver_nodes: measured::Gauge,
45 :
46 : /// Reconciler tasks completed, broken down by success/failure/cancelled
47 : pub(crate) storage_controller_reconcile_complete:
48 : measured::CounterVec<ReconcileCompleteLabelGroupSet>,
49 :
50 : /// Count of how many times we make an optimization change to a tenant's scheduling
51 : pub(crate) storage_controller_schedule_optimization: measured::Counter,
52 :
53 : /// How many shards are not scheduled into their preferred AZ
54 : pub(crate) storage_controller_schedule_az_violation: measured::Gauge,
55 :
56 : /// How many shard locations (secondary or attached) on each node
57 : pub(crate) storage_controller_node_shards: measured::GaugeVec<NodeLabelGroupSet>,
58 :
59 : /// How many _attached_ shard locations on each node
60 : pub(crate) storage_controller_node_attached_shards: measured::GaugeVec<NodeLabelGroupSet>,
61 :
62 : /// How many _home_ shard locations on each node (i.e. the node's AZ matches the shard's
63 : /// preferred AZ)
64 : pub(crate) storage_controller_node_home_shards: measured::GaugeVec<NodeLabelGroupSet>,
65 :
66 : /// How many shards would like to reconcile but were blocked by concurrency limits
67 : pub(crate) storage_controller_pending_reconciles: measured::Gauge,
68 :
69 : /// HTTP request status counters for handled requests
70 : pub(crate) storage_controller_http_request_status:
71 : measured::CounterVec<HttpRequestStatusLabelGroupSet>,
72 :
73 : /// HTTP request handler latency across all status codes
74 : #[metric(metadata = histogram::Thresholds::exponential_buckets(0.1, 2.0))]
75 : pub(crate) storage_controller_http_request_latency:
76 : measured::HistogramVec<HttpRequestLatencyLabelGroupSet, 5>,
77 :
78 : /// Count of HTTP requests to the pageserver that resulted in an error,
79 : /// broken down by the pageserver node id, request name and method
80 : pub(crate) storage_controller_pageserver_request_error:
81 : measured::CounterVec<PageserverRequestLabelGroupSet>,
82 :
83 : /// Latency of HTTP requests to the pageserver, broken down by pageserver
84 : /// node id, request name and method. This include both successful and unsuccessful
85 : /// requests.
86 : #[metric(metadata = histogram::Thresholds::exponential_buckets(0.1, 2.0))]
87 : pub(crate) storage_controller_pageserver_request_latency:
88 : measured::HistogramVec<PageserverRequestLabelGroupSet, 5>,
89 :
90 : /// Count of pass-through HTTP requests to the pageserver that resulted in an error,
91 : /// broken down by the pageserver node id, request name and method
92 : pub(crate) storage_controller_passthrough_request_error:
93 : measured::CounterVec<PageserverRequestLabelGroupSet>,
94 :
95 : /// Latency of pass-through HTTP requests to the pageserver, broken down by pageserver
96 : /// node id, request name and method. This include both successful and unsuccessful
97 : /// requests.
98 : #[metric(metadata = histogram::Thresholds::exponential_buckets(0.1, 2.0))]
99 : pub(crate) storage_controller_passthrough_request_latency:
100 : measured::HistogramVec<PageserverRequestLabelGroupSet, 5>,
101 :
102 : /// Count of errors in database queries, broken down by error type and operation.
103 : pub(crate) storage_controller_database_query_error:
104 : measured::CounterVec<DatabaseQueryErrorLabelGroupSet>,
105 :
106 : /// Latency of database queries, broken down by operation.
107 : #[metric(metadata = histogram::Thresholds::exponential_buckets(0.1, 2.0))]
108 : pub(crate) storage_controller_database_query_latency:
109 : measured::HistogramVec<DatabaseQueryLatencyLabelGroupSet, 5>,
110 :
111 : pub(crate) storage_controller_leadership_status: measured::GaugeVec<LeadershipStatusGroupSet>,
112 :
113 : /// HTTP request status counters for handled requests
114 : pub(crate) storage_controller_reconcile_long_running:
115 : measured::CounterVec<ReconcileLongRunningLabelGroupSet>,
116 : }
117 :
118 : impl StorageControllerMetrics {
119 0 : pub(crate) fn encode(&self, neon_metrics: &NeonMetrics) -> Bytes {
120 0 : let mut encoder = self.encoder.lock().unwrap();
121 0 : neon_metrics
122 0 : .collect_group_into(&mut *encoder)
123 0 : .unwrap_or_else(|infallible| match infallible {});
124 0 : self.metrics_group
125 0 : .collect_group_into(&mut *encoder)
126 0 : .unwrap_or_else(|infallible| match infallible {});
127 0 : encoder.finish()
128 0 : }
129 : }
130 :
131 : impl Default for StorageControllerMetrics {
132 13 : fn default() -> Self {
133 13 : let mut metrics_group = StorageControllerMetricGroup::new();
134 13 : metrics_group
135 13 : .storage_controller_reconcile_complete
136 13 : .init_all_dense();
137 13 :
138 13 : Self {
139 13 : metrics_group,
140 13 : encoder: Mutex::new(measured::text::BufferedTextEncoder::new()),
141 13 : }
142 13 : }
143 : }
144 :
145 78 : #[derive(measured::LabelGroup, Clone)]
146 : #[label(set = NodeLabelGroupSet)]
147 : pub(crate) struct NodeLabelGroup<'a> {
148 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
149 : pub(crate) az: &'a str,
150 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
151 : pub(crate) node_id: &'a str,
152 : }
153 :
154 39 : #[derive(measured::LabelGroup)]
155 : #[label(set = ReconcileCompleteLabelGroupSet)]
156 : pub(crate) struct ReconcileCompleteLabelGroup {
157 : pub(crate) status: ReconcileOutcome,
158 : }
159 :
160 26 : #[derive(measured::LabelGroup)]
161 : #[label(set = HttpRequestStatusLabelGroupSet)]
162 : pub(crate) struct HttpRequestStatusLabelGroup<'a> {
163 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
164 : pub(crate) path: &'a str,
165 : pub(crate) method: Method,
166 : pub(crate) status: StatusCode,
167 : }
168 :
169 26 : #[derive(measured::LabelGroup)]
170 : #[label(set = HttpRequestLatencyLabelGroupSet)]
171 : pub(crate) struct HttpRequestLatencyLabelGroup<'a> {
172 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
173 : pub(crate) path: &'a str,
174 : pub(crate) method: Method,
175 : }
176 :
177 104 : #[derive(measured::LabelGroup, Clone)]
178 : #[label(set = PageserverRequestLabelGroupSet)]
179 : pub(crate) struct PageserverRequestLabelGroup<'a> {
180 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
181 : pub(crate) pageserver_id: &'a str,
182 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
183 : pub(crate) path: &'a str,
184 : pub(crate) method: Method,
185 : }
186 :
187 52 : #[derive(measured::LabelGroup)]
188 : #[label(set = DatabaseQueryErrorLabelGroupSet)]
189 : pub(crate) struct DatabaseQueryErrorLabelGroup {
190 : pub(crate) error_type: DatabaseErrorLabel,
191 : pub(crate) operation: DatabaseOperation,
192 : }
193 :
194 39 : #[derive(measured::LabelGroup)]
195 : #[label(set = DatabaseQueryLatencyLabelGroupSet)]
196 : pub(crate) struct DatabaseQueryLatencyLabelGroup {
197 : pub(crate) operation: DatabaseOperation,
198 : }
199 :
200 39 : #[derive(measured::LabelGroup)]
201 : #[label(set = LeadershipStatusGroupSet)]
202 : pub(crate) struct LeadershipStatusGroup {
203 : pub(crate) status: LeadershipStatus,
204 : }
205 :
206 26 : #[derive(measured::LabelGroup, Clone)]
207 : #[label(set = ReconcileLongRunningLabelGroupSet)]
208 : pub(crate) struct ReconcileLongRunningLabelGroup<'a> {
209 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
210 : pub(crate) tenant_id: &'a str,
211 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
212 : pub(crate) shard_number: &'a str,
213 : #[label(dynamic_with = lasso::ThreadedRodeo, default)]
214 : pub(crate) sequence: &'a str,
215 : }
216 :
217 : #[derive(FixedCardinalityLabel, Clone, Copy)]
218 : pub(crate) enum ReconcileOutcome {
219 : #[label(rename = "ok")]
220 : Success,
221 : Error,
222 : Cancel,
223 : }
224 :
225 : #[derive(FixedCardinalityLabel, Copy, Clone)]
226 : pub(crate) enum Method {
227 : Get,
228 : Put,
229 : Post,
230 : Delete,
231 : Other,
232 : }
233 :
234 : impl From<hyper::Method> for Method {
235 0 : fn from(value: hyper::Method) -> Self {
236 0 : if value == hyper::Method::GET {
237 0 : Method::Get
238 0 : } else if value == hyper::Method::PUT {
239 0 : Method::Put
240 0 : } else if value == hyper::Method::POST {
241 0 : Method::Post
242 0 : } else if value == hyper::Method::DELETE {
243 0 : Method::Delete
244 : } else {
245 0 : Method::Other
246 : }
247 0 : }
248 : }
249 :
250 : #[derive(Clone, Copy)]
251 : pub(crate) struct StatusCode(pub(crate) hyper::http::StatusCode);
252 :
253 : impl LabelValue for StatusCode {
254 0 : fn visit<V: measured::label::LabelVisitor>(&self, v: V) -> V::Output {
255 0 : v.write_int(self.0.as_u16() as i64)
256 0 : }
257 : }
258 :
259 : impl FixedCardinalityLabel for StatusCode {
260 0 : fn cardinality() -> usize {
261 0 : (100..1000).len()
262 0 : }
263 :
264 0 : fn encode(&self) -> usize {
265 0 : self.0.as_u16() as usize
266 0 : }
267 :
268 0 : fn decode(value: usize) -> Self {
269 0 : Self(hyper::http::StatusCode::from_u16(u16::try_from(value).unwrap()).unwrap())
270 0 : }
271 : }
272 :
273 : #[derive(FixedCardinalityLabel, Clone, Copy)]
274 : pub(crate) enum DatabaseErrorLabel {
275 : Query,
276 : Connection,
277 : ConnectionPool,
278 : Logical,
279 : Migration,
280 : }
281 :
282 : impl DatabaseError {
283 0 : pub(crate) fn error_label(&self) -> DatabaseErrorLabel {
284 0 : match self {
285 0 : Self::Query(_) => DatabaseErrorLabel::Query,
286 0 : Self::Connection(_) => DatabaseErrorLabel::Connection,
287 0 : Self::ConnectionPool(_) => DatabaseErrorLabel::ConnectionPool,
288 0 : Self::Logical(_) => DatabaseErrorLabel::Logical,
289 0 : Self::Migration(_) => DatabaseErrorLabel::Migration,
290 : }
291 0 : }
292 : }
293 :
294 : /// Update the leadership status metric gauges to reflect the requested status
295 0 : pub(crate) fn update_leadership_status(status: LeadershipStatus) {
296 0 : let status_metric = &METRICS_REGISTRY
297 0 : .metrics_group
298 0 : .storage_controller_leadership_status;
299 :
300 0 : for s in LeadershipStatus::iter() {
301 0 : if s == status {
302 0 : status_metric.set(LeadershipStatusGroup { status: s }, 1);
303 0 : } else {
304 0 : status_metric.set(LeadershipStatusGroup { status: s }, 0);
305 0 : }
306 : }
307 0 : }
|
