Line data Source code
1 : use std::collections::{HashMap, HashSet};
2 : use std::fmt::Display;
3 : use std::str::FromStr;
4 : use std::time::{Duration, Instant};
5 :
6 : /// Request/response types for the storage controller
7 : /// API (`/control/v1` prefix). Implemented by the server
8 : /// in [`storage_controller::http`]
9 : use serde::{Deserialize, Serialize};
10 : use utils::id::{NodeId, TenantId};
11 :
12 : use crate::models::PageserverUtilization;
13 : use crate::{
14 : models::{ShardParameters, TenantConfig},
15 : shard::{ShardStripeSize, TenantShardId},
16 : };
17 :
18 3 : #[derive(Serialize, Deserialize, Debug)]
19 : #[serde(deny_unknown_fields)]
20 : pub struct TenantCreateRequest {
21 : pub new_tenant_id: TenantShardId,
22 : #[serde(default)]
23 : #[serde(skip_serializing_if = "Option::is_none")]
24 : pub generation: Option<u32>,
25 :
26 : // If omitted, create a single shard with TenantShardId::unsharded()
27 : #[serde(default)]
28 : #[serde(skip_serializing_if = "ShardParameters::is_unsharded")]
29 : pub shard_parameters: ShardParameters,
30 :
31 : #[serde(default)]
32 : #[serde(skip_serializing_if = "Option::is_none")]
33 : pub placement_policy: Option<PlacementPolicy>,
34 :
35 : #[serde(flatten)]
36 : pub config: TenantConfig, // as we have a flattened field, we should reject all unknown fields in it
37 : }
38 :
39 0 : #[derive(Serialize, Deserialize)]
40 : pub struct TenantCreateResponseShard {
41 : pub shard_id: TenantShardId,
42 : pub node_id: NodeId,
43 : pub generation: u32,
44 : }
45 :
46 0 : #[derive(Serialize, Deserialize)]
47 : pub struct TenantCreateResponse {
48 : pub shards: Vec<TenantCreateResponseShard>,
49 : }
50 :
51 0 : #[derive(Serialize, Deserialize, Debug, Clone)]
52 : pub struct NodeRegisterRequest {
53 : pub node_id: NodeId,
54 :
55 : pub listen_pg_addr: String,
56 : pub listen_pg_port: u16,
57 :
58 : pub listen_http_addr: String,
59 : pub listen_http_port: u16,
60 :
61 : pub availability_zone_id: AvailabilityZone,
62 : }
63 :
64 0 : #[derive(Serialize, Deserialize)]
65 : pub struct NodeConfigureRequest {
66 : pub node_id: NodeId,
67 :
68 : pub availability: Option<NodeAvailabilityWrapper>,
69 : pub scheduling: Option<NodeSchedulingPolicy>,
70 : }
71 :
72 0 : #[derive(Serialize, Deserialize)]
73 : pub struct TenantPolicyRequest {
74 : pub placement: Option<PlacementPolicy>,
75 : pub scheduling: Option<ShardSchedulingPolicy>,
76 : }
77 :
78 0 : #[derive(Clone, Serialize, Deserialize, PartialEq, Eq, Hash, Debug, PartialOrd, Ord)]
79 : pub struct AvailabilityZone(pub String);
80 :
81 : impl Display for AvailabilityZone {
82 0 : fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
83 0 : write!(f, "{}", self.0)
84 0 : }
85 : }
86 :
87 0 : #[derive(Serialize, Deserialize)]
88 : pub struct ShardsPreferredAzsRequest {
89 : #[serde(flatten)]
90 : pub preferred_az_ids: HashMap<TenantShardId, AvailabilityZone>,
91 : }
92 :
93 0 : #[derive(Serialize, Deserialize)]
94 : pub struct ShardsPreferredAzsResponse {
95 : pub updated: Vec<TenantShardId>,
96 : }
97 :
98 0 : #[derive(Serialize, Deserialize, Debug)]
99 : pub struct TenantLocateResponseShard {
100 : pub shard_id: TenantShardId,
101 : pub node_id: NodeId,
102 :
103 : pub listen_pg_addr: String,
104 : pub listen_pg_port: u16,
105 :
106 : pub listen_http_addr: String,
107 : pub listen_http_port: u16,
108 : }
109 :
110 0 : #[derive(Serialize, Deserialize)]
111 : pub struct TenantLocateResponse {
112 : pub shards: Vec<TenantLocateResponseShard>,
113 : pub shard_params: ShardParameters,
114 : }
115 :
116 0 : #[derive(Serialize, Deserialize, Debug)]
117 : pub struct TenantDescribeResponse {
118 : pub tenant_id: TenantId,
119 : pub shards: Vec<TenantDescribeResponseShard>,
120 : pub stripe_size: ShardStripeSize,
121 : pub policy: PlacementPolicy,
122 : pub config: TenantConfig,
123 : }
124 :
125 0 : #[derive(Serialize, Deserialize, Debug)]
126 : pub struct NodeShardResponse {
127 : pub node_id: NodeId,
128 : pub shards: Vec<NodeShard>,
129 : }
130 :
131 0 : #[derive(Serialize, Deserialize, Debug)]
132 : pub struct NodeShard {
133 : pub tenant_shard_id: TenantShardId,
134 : /// Whether the shard is observed secondary on a specific node. True = yes, False = no, None = not on this node.
135 : pub is_observed_secondary: Option<bool>,
136 : /// Whether the shard is intended to be a secondary on a specific node. True = yes, False = no, None = not on this node.
137 : pub is_intended_secondary: Option<bool>,
138 : }
139 :
140 0 : #[derive(Serialize, Deserialize)]
141 : pub struct NodeDescribeResponse {
142 : pub id: NodeId,
143 :
144 : pub availability: NodeAvailabilityWrapper,
145 : pub scheduling: NodeSchedulingPolicy,
146 :
147 : pub listen_http_addr: String,
148 : pub listen_http_port: u16,
149 :
150 : pub listen_pg_addr: String,
151 : pub listen_pg_port: u16,
152 : }
153 :
154 0 : #[derive(Serialize, Deserialize, Debug)]
155 : pub struct TenantDescribeResponseShard {
156 : pub tenant_shard_id: TenantShardId,
157 :
158 : pub node_attached: Option<NodeId>,
159 : pub node_secondary: Vec<NodeId>,
160 :
161 : pub last_error: String,
162 :
163 : /// A task is currently running to reconcile this tenant's intent state with the state on pageservers
164 : pub is_reconciling: bool,
165 : /// This shard failed in sending a compute notification to the cloud control plane, and a retry is pending.
166 : pub is_pending_compute_notification: bool,
167 : /// A shard split is currently underway
168 : pub is_splitting: bool,
169 :
170 : pub scheduling_policy: ShardSchedulingPolicy,
171 :
172 : pub preferred_az_id: Option<String>,
173 : }
174 :
175 : /// Migration request for a given tenant shard to a given node.
176 : ///
177 : /// Explicitly migrating a particular shard is a low level operation
178 : /// TODO: higher level "Reschedule tenant" operation where the request
179 : /// specifies some constraints, e.g. asking it to get off particular node(s)
180 0 : #[derive(Serialize, Deserialize, Debug)]
181 : pub struct TenantShardMigrateRequest {
182 : pub tenant_shard_id: TenantShardId,
183 : pub node_id: NodeId,
184 : }
185 :
186 : #[derive(Serialize, Clone, Debug)]
187 : #[serde(into = "NodeAvailabilityWrapper")]
188 : pub enum NodeAvailability {
189 : // Normal, happy state
190 : Active(PageserverUtilization),
191 : // Node is warming up, but we expect it to become available soon. Covers
192 : // the time span between the re-attach response being composed on the storage controller
193 : // and the first successful heartbeat after the processing of the re-attach response
194 : // finishes on the pageserver.
195 : WarmingUp(Instant),
196 : // Offline: Tenants shouldn't try to attach here, but they may assume that their
197 : // secondary locations on this node still exist. Newly added nodes are in this
198 : // state until we successfully contact them.
199 : Offline,
200 : }
201 :
202 : impl PartialEq for NodeAvailability {
203 0 : fn eq(&self, other: &Self) -> bool {
204 : use NodeAvailability::*;
205 0 : matches!(
206 0 : (self, other),
207 : (Active(_), Active(_)) | (Offline, Offline) | (WarmingUp(_), WarmingUp(_))
208 : )
209 0 : }
210 : }
211 :
212 : impl Eq for NodeAvailability {}
213 :
214 : // This wrapper provides serde functionality and it should only be used to
215 : // communicate with external callers which don't know or care about the
216 : // utilisation score of the pageserver it is targeting.
217 0 : #[derive(Serialize, Deserialize, Clone, Copy, Debug)]
218 : pub enum NodeAvailabilityWrapper {
219 : Active,
220 : WarmingUp,
221 : Offline,
222 : }
223 :
224 : impl From<NodeAvailabilityWrapper> for NodeAvailability {
225 0 : fn from(val: NodeAvailabilityWrapper) -> Self {
226 0 : match val {
227 : // Assume the worst utilisation score to begin with. It will later be updated by
228 : // the heartbeats.
229 : NodeAvailabilityWrapper::Active => {
230 0 : NodeAvailability::Active(PageserverUtilization::full())
231 : }
232 0 : NodeAvailabilityWrapper::WarmingUp => NodeAvailability::WarmingUp(Instant::now()),
233 0 : NodeAvailabilityWrapper::Offline => NodeAvailability::Offline,
234 : }
235 0 : }
236 : }
237 :
238 : impl From<NodeAvailability> for NodeAvailabilityWrapper {
239 0 : fn from(val: NodeAvailability) -> Self {
240 0 : match val {
241 0 : NodeAvailability::Active(_) => NodeAvailabilityWrapper::Active,
242 0 : NodeAvailability::WarmingUp(_) => NodeAvailabilityWrapper::WarmingUp,
243 0 : NodeAvailability::Offline => NodeAvailabilityWrapper::Offline,
244 : }
245 0 : }
246 : }
247 :
248 : /// Scheduling policy enables us to selectively disable some automatic actions that the
249 : /// controller performs on a tenant shard. This is only set to a non-default value by
250 : /// human intervention, and it is reset to the default value (Active) when the tenant's
251 : /// placement policy is modified away from Attached.
252 : ///
253 : /// The typical use of a non-Active scheduling policy is one of:
254 : /// - Pinnning a shard to a node (i.e. migrating it there & setting a non-Active scheduling policy)
255 : /// - Working around a bug (e.g. if something is flapping and we need to stop it until the bug is fixed)
256 : ///
257 : /// If you're not sure which policy to use to pin a shard to its current location, you probably
258 : /// want Pause.
259 0 : #[derive(Serialize, Deserialize, Clone, Copy, Eq, PartialEq, Debug)]
260 : pub enum ShardSchedulingPolicy {
261 : // Normal mode: the tenant's scheduled locations may be updated at will, including
262 : // for non-essential optimization.
263 : Active,
264 :
265 : // Disable optimizations, but permit scheduling when necessary to fulfil the PlacementPolicy.
266 : // For example, this still permits a node's attachment location to change to a secondary in
267 : // response to a node failure, or to assign a new secondary if a node was removed.
268 : Essential,
269 :
270 : // No scheduling: leave the shard running wherever it currently is. Even if the shard is
271 : // unavailable, it will not be rescheduled to another node.
272 : Pause,
273 :
274 : // No reconciling: we will make no location_conf API calls to pageservers at all. If the
275 : // shard is unavailable, it stays that way. If a node fails, this shard doesn't get failed over.
276 : Stop,
277 : }
278 :
279 : impl Default for ShardSchedulingPolicy {
280 12525 : fn default() -> Self {
281 12525 : Self::Active
282 12525 : }
283 : }
284 :
285 0 : #[derive(Serialize, Deserialize, Clone, Copy, Eq, PartialEq, Debug)]
286 : pub enum NodeSchedulingPolicy {
287 : Active,
288 : Filling,
289 : Pause,
290 : PauseForRestart,
291 : Draining,
292 : }
293 :
294 : impl FromStr for NodeSchedulingPolicy {
295 : type Err = anyhow::Error;
296 :
297 0 : fn from_str(s: &str) -> Result<Self, Self::Err> {
298 0 : match s {
299 0 : "active" => Ok(Self::Active),
300 0 : "filling" => Ok(Self::Filling),
301 0 : "pause" => Ok(Self::Pause),
302 0 : "pause_for_restart" => Ok(Self::PauseForRestart),
303 0 : "draining" => Ok(Self::Draining),
304 0 : _ => Err(anyhow::anyhow!("Unknown scheduling state '{s}'")),
305 : }
306 0 : }
307 : }
308 :
309 : impl From<NodeSchedulingPolicy> for String {
310 0 : fn from(value: NodeSchedulingPolicy) -> String {
311 : use NodeSchedulingPolicy::*;
312 0 : match value {
313 0 : Active => "active",
314 0 : Filling => "filling",
315 0 : Pause => "pause",
316 0 : PauseForRestart => "pause_for_restart",
317 0 : Draining => "draining",
318 : }
319 0 : .to_string()
320 0 : }
321 : }
322 :
323 : /// Controls how tenant shards are mapped to locations on pageservers, e.g. whether
324 : /// to create secondary locations.
325 4 : #[derive(Clone, Serialize, Deserialize, Debug, PartialEq, Eq)]
326 : pub enum PlacementPolicy {
327 : /// Normal live state: one attached pageserver and zero or more secondaries.
328 : Attached(usize),
329 : /// Create one secondary mode locations. This is useful when onboarding
330 : /// a tenant, or for an idle tenant that we might want to bring online quickly.
331 : Secondary,
332 :
333 : /// Do not attach to any pageservers. This is appropriate for tenants that
334 : /// have been idle for a long time, where we do not mind some delay in making
335 : /// them available in future.
336 : Detached,
337 : }
338 :
339 0 : #[derive(Serialize, Deserialize, Debug)]
340 : pub struct TenantShardMigrateResponse {}
341 :
342 : /// Metadata health record posted from scrubber.
343 0 : #[derive(Serialize, Deserialize, Debug)]
344 : pub struct MetadataHealthRecord {
345 : pub tenant_shard_id: TenantShardId,
346 : pub healthy: bool,
347 : pub last_scrubbed_at: chrono::DateTime<chrono::Utc>,
348 : }
349 :
350 0 : #[derive(Serialize, Deserialize, Debug)]
351 : pub struct MetadataHealthUpdateRequest {
352 : pub healthy_tenant_shards: HashSet<TenantShardId>,
353 : pub unhealthy_tenant_shards: HashSet<TenantShardId>,
354 : }
355 :
356 0 : #[derive(Serialize, Deserialize, Debug)]
357 : pub struct MetadataHealthUpdateResponse {}
358 :
359 0 : #[derive(Serialize, Deserialize, Debug)]
360 : pub struct MetadataHealthListUnhealthyResponse {
361 : pub unhealthy_tenant_shards: Vec<TenantShardId>,
362 : }
363 :
364 0 : #[derive(Serialize, Deserialize, Debug)]
365 : pub struct MetadataHealthListOutdatedRequest {
366 : #[serde(with = "humantime_serde")]
367 : pub not_scrubbed_for: Duration,
368 : }
369 :
370 0 : #[derive(Serialize, Deserialize, Debug)]
371 : pub struct MetadataHealthListOutdatedResponse {
372 : pub health_records: Vec<MetadataHealthRecord>,
373 : }
374 :
375 : /// Publicly exposed safekeeper description
376 : ///
377 : /// The `active` flag which we have in the DB is not included on purpose: it is deprecated.
378 0 : #[derive(Serialize, Deserialize, Clone)]
379 : pub struct SafekeeperDescribeResponse {
380 : pub id: NodeId,
381 : pub region_id: String,
382 : /// 1 is special, it means just created (not currently posted to storcon).
383 : /// Zero or negative is not really expected.
384 : /// Otherwise the number from `release-$(number_of_commits_on_branch)` tag.
385 : pub version: i64,
386 : pub host: String,
387 : pub port: i32,
388 : pub http_port: i32,
389 : pub availability_zone_id: String,
390 : }
391 :
392 : #[cfg(test)]
393 : mod test {
394 : use super::*;
395 : use serde_json;
396 :
397 : /// Check stability of PlacementPolicy's serialization
398 : #[test]
399 1 : fn placement_policy_encoding() -> anyhow::Result<()> {
400 1 : let v = PlacementPolicy::Attached(1);
401 1 : let encoded = serde_json::to_string(&v)?;
402 1 : assert_eq!(encoded, "{\"Attached\":1}");
403 1 : assert_eq!(serde_json::from_str::<PlacementPolicy>(&encoded)?, v);
404 :
405 1 : let v = PlacementPolicy::Detached;
406 1 : let encoded = serde_json::to_string(&v)?;
407 1 : assert_eq!(encoded, "\"Detached\"");
408 1 : assert_eq!(serde_json::from_str::<PlacementPolicy>(&encoded)?, v);
409 1 : Ok(())
410 1 : }
411 :
412 : #[test]
413 1 : fn test_reject_unknown_field() {
414 1 : let id = TenantId::generate();
415 1 : let create_request = serde_json::json!({
416 1 : "new_tenant_id": id.to_string(),
417 1 : "unknown_field": "unknown_value".to_string(),
418 1 : });
419 1 : let err = serde_json::from_value::<TenantCreateRequest>(create_request).unwrap_err();
420 1 : assert!(
421 1 : err.to_string().contains("unknown field `unknown_field`"),
422 0 : "expect unknown field `unknown_field` error, got: {}",
423 : err
424 : );
425 1 : }
426 : }
|
