Line data Source code
1 : pub mod detach_ancestor;
2 : pub mod partitioning;
3 : pub mod utilization;
4 :
5 : use core::ops::Range;
6 : use std::collections::HashMap;
7 : use std::fmt::Display;
8 : use std::io::{BufRead, Read};
9 : use std::num::{NonZeroU32, NonZeroU64, NonZeroUsize};
10 : use std::str::FromStr;
11 : use std::time::{Duration, SystemTime};
12 :
13 : use byteorder::{BigEndian, ReadBytesExt};
14 : use bytes::{Buf, BufMut, Bytes, BytesMut};
15 : #[cfg(feature = "testing")]
16 : use camino::Utf8PathBuf;
17 : use postgres_ffi::BLCKSZ;
18 : use serde::{Deserialize, Deserializer, Serialize, Serializer};
19 : use serde_with::serde_as;
20 : pub use utilization::PageserverUtilization;
21 : use utils::id::{NodeId, TenantId, TimelineId};
22 : use utils::lsn::Lsn;
23 : use utils::postgres_client::PostgresClientProtocol;
24 : use utils::{completion, serde_system_time};
25 :
26 : use crate::config::Ratio;
27 : use crate::key::{CompactKey, Key};
28 : use crate::reltag::RelTag;
29 : use crate::shard::{DEFAULT_STRIPE_SIZE, ShardCount, ShardStripeSize, TenantShardId};
30 :
31 : /// The state of a tenant in this pageserver.
32 : ///
33 : /// ```mermaid
34 : /// stateDiagram-v2
35 : ///
36 : /// [*] --> Attaching: spawn_attach()
37 : ///
38 : /// Attaching --> Activating: activate()
39 : /// Activating --> Active: infallible
40 : ///
41 : /// Attaching --> Broken: attach() failure
42 : ///
43 : /// Active --> Stopping: set_stopping(), part of shutdown & detach
44 : /// Stopping --> Broken: late error in remove_tenant_from_memory
45 : ///
46 : /// Broken --> [*]: ignore / detach / shutdown
47 : /// Stopping --> [*]: remove_from_memory complete
48 : ///
49 : /// Active --> Broken: cfg(testing)-only tenant break point
50 : /// ```
51 : #[derive(
52 : Clone,
53 : PartialEq,
54 : Eq,
55 0 : serde::Serialize,
56 1 : serde::Deserialize,
57 : strum_macros::Display,
58 : strum_macros::VariantNames,
59 : strum_macros::AsRefStr,
60 : strum_macros::IntoStaticStr,
61 : )]
62 : #[serde(tag = "slug", content = "data")]
63 : pub enum TenantState {
64 : /// This tenant is being attached to the pageserver.
65 : ///
66 : /// `set_stopping()` and `set_broken()` do not work in this state and wait for it to pass.
67 : Attaching,
68 : /// The tenant is transitioning from Loading/Attaching to Active.
69 : ///
70 : /// While in this state, the individual timelines are being activated.
71 : ///
72 : /// `set_stopping()` and `set_broken()` do not work in this state and wait for it to pass.
73 : Activating(ActivatingFrom),
74 : /// The tenant has finished activating and is open for business.
75 : ///
76 : /// Transitions out of this state are possible through `set_stopping()` and `set_broken()`.
77 : Active,
78 : /// The tenant is recognized by pageserver, but it is being detached or the
79 : /// system is being shut down.
80 : ///
81 : /// Transitions out of this state are possible through `set_broken()`.
82 : Stopping {
83 : /// The barrier can be used to wait for shutdown to complete. The first caller to set
84 : /// Some(Barrier) is responsible for driving shutdown to completion. Subsequent callers
85 : /// will wait for the first caller's existing barrier.
86 : ///
87 : /// None is set when an attach is cancelled, to signal to shutdown that the attach has in
88 : /// fact cancelled:
89 : ///
90 : /// 1. `shutdown` sees `TenantState::Attaching`, and cancels the tenant.
91 : /// 2. `attach` sets `TenantState::Stopping(None)` and exits.
92 : /// 3. `set_stopping` waits for `TenantState::Stopping(None)` and sets
93 : /// `TenantState::Stopping(Some)` to claim the barrier as the shutdown owner.
94 : //
95 : // Because of https://github.com/serde-rs/serde/issues/2105 this has to be a named field,
96 : // otherwise it will not be skipped during deserialization
97 : #[serde(skip)]
98 : progress: Option<completion::Barrier>,
99 : },
100 : /// The tenant is recognized by the pageserver, but can no longer be used for
101 : /// any operations.
102 : ///
103 : /// If the tenant fails to load or attach, it will transition to this state
104 : /// and it is guaranteed that no background tasks are running in its name.
105 : ///
106 : /// The other way to transition into this state is from `Stopping` state
107 : /// through `set_broken()` called from `remove_tenant_from_memory()`. That happens
108 : /// if the cleanup future executed by `remove_tenant_from_memory()` fails.
109 : Broken { reason: String, backtrace: String },
110 : }
111 :
112 : impl TenantState {
113 0 : pub fn attachment_status(&self) -> TenantAttachmentStatus {
114 : use TenantAttachmentStatus::*;
115 :
116 : // Below TenantState::Activating is used as "transient" or "transparent" state for
117 : // attachment_status determining.
118 0 : match self {
119 : // The attach procedure writes the marker file before adding the Attaching tenant to the tenants map.
120 : // So, technically, we can return Attached here.
121 : // However, as soon as Console observes Attached, it will proceed with the Postgres-level health check.
122 : // But, our attach task might still be fetching the remote timelines, etc.
123 : // So, return `Maybe` while Attaching, making Console wait for the attach task to finish.
124 0 : Self::Attaching | Self::Activating(ActivatingFrom::Attaching) => Maybe,
125 : // We only reach Active after successful load / attach.
126 : // So, call atttachment status Attached.
127 0 : Self::Active => Attached,
128 : // If the (initial or resumed) attach procedure fails, the tenant becomes Broken.
129 : // However, it also becomes Broken if the regular load fails.
130 : // From Console's perspective there's no practical difference
131 : // because attachment_status is polled by console only during attach operation execution.
132 0 : Self::Broken { reason, .. } => Failed {
133 0 : reason: reason.to_owned(),
134 0 : },
135 : // Why is Stopping a Maybe case? Because, during pageserver shutdown,
136 : // we set the Stopping state irrespective of whether the tenant
137 : // has finished attaching or not.
138 0 : Self::Stopping { .. } => Maybe,
139 : }
140 0 : }
141 :
142 0 : pub fn broken_from_reason(reason: String) -> Self {
143 0 : let backtrace_str: String = format!("{}", std::backtrace::Backtrace::force_capture());
144 0 : Self::Broken {
145 0 : reason,
146 0 : backtrace: backtrace_str,
147 0 : }
148 0 : }
149 : }
150 :
151 : impl std::fmt::Debug for TenantState {
152 2 : fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
153 2 : match self {
154 2 : Self::Broken { reason, backtrace } if !reason.is_empty() => {
155 2 : write!(f, "Broken due to: {reason}. Backtrace:\n{backtrace}")
156 : }
157 0 : _ => write!(f, "{self}"),
158 : }
159 2 : }
160 : }
161 :
162 : /// A temporary lease to a specific lsn inside a timeline.
163 : /// Access to the lsn is guaranteed by the pageserver until the expiration indicated by `valid_until`.
164 : #[serde_as]
165 0 : #[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
166 : pub struct LsnLease {
167 : #[serde_as(as = "SystemTimeAsRfc3339Millis")]
168 : pub valid_until: SystemTime,
169 : }
170 :
171 : serde_with::serde_conv!(
172 : SystemTimeAsRfc3339Millis,
173 : SystemTime,
174 0 : |time: &SystemTime| humantime::format_rfc3339_millis(*time).to_string(),
175 0 : |value: String| -> Result<_, humantime::TimestampError> { humantime::parse_rfc3339(&value) }
176 : );
177 :
178 : impl LsnLease {
179 : /// The default length for an explicit LSN lease request (10 minutes).
180 : pub const DEFAULT_LENGTH: Duration = Duration::from_secs(10 * 60);
181 :
182 : /// The default length for an implicit LSN lease granted during
183 : /// `get_lsn_by_timestamp` request (1 minutes).
184 : pub const DEFAULT_LENGTH_FOR_TS: Duration = Duration::from_secs(60);
185 :
186 : /// Checks whether the lease is expired.
187 12 : pub fn is_expired(&self, now: &SystemTime) -> bool {
188 12 : now > &self.valid_until
189 12 : }
190 : }
191 :
192 : /// Controls the detach ancestor behavior.
193 : /// - When set to `NoAncestorAndReparent`, we will only detach a branch if its ancestor is a root branch. It will automatically reparent any children of the ancestor before and at the branch point.
194 : /// - When set to `MultiLevelAndNoReparent`, we will detach a branch from multiple levels of ancestors, and no reparenting will happen at all.
195 : #[derive(Debug, Clone, Copy, Default)]
196 : pub enum DetachBehavior {
197 : #[default]
198 : NoAncestorAndReparent,
199 : MultiLevelAndNoReparent,
200 : }
201 :
202 : impl std::str::FromStr for DetachBehavior {
203 : type Err = &'static str;
204 :
205 0 : fn from_str(s: &str) -> Result<Self, Self::Err> {
206 0 : match s {
207 0 : "no_ancestor_and_reparent" => Ok(DetachBehavior::NoAncestorAndReparent),
208 0 : "multi_level_and_no_reparent" => Ok(DetachBehavior::MultiLevelAndNoReparent),
209 0 : "v1" => Ok(DetachBehavior::NoAncestorAndReparent),
210 0 : "v2" => Ok(DetachBehavior::MultiLevelAndNoReparent),
211 0 : _ => Err("cannot parse detach behavior"),
212 : }
213 0 : }
214 : }
215 :
216 : impl std::fmt::Display for DetachBehavior {
217 0 : fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
218 0 : match self {
219 0 : DetachBehavior::NoAncestorAndReparent => write!(f, "no_ancestor_and_reparent"),
220 0 : DetachBehavior::MultiLevelAndNoReparent => write!(f, "multi_level_and_no_reparent"),
221 : }
222 0 : }
223 : }
224 :
225 : /// The only [`TenantState`] variants we could be `TenantState::Activating` from.
226 : ///
227 : /// XXX: We used to have more variants here, but now it's just one, which makes this rather
228 : /// useless. Remove, once we've checked that there's no client code left that looks at this.
229 1 : #[derive(Clone, Copy, Debug, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
230 : pub enum ActivatingFrom {
231 : /// Arrived to [`TenantState::Activating`] from [`TenantState::Attaching`]
232 : Attaching,
233 : }
234 :
235 : /// A state of a timeline in pageserver's memory.
236 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
237 : pub enum TimelineState {
238 : /// The timeline is recognized by the pageserver but is not yet operational.
239 : /// In particular, the walreceiver connection loop is not running for this timeline.
240 : /// It will eventually transition to state Active or Broken.
241 : Loading,
242 : /// The timeline is fully operational.
243 : /// It can be queried, and the walreceiver connection loop is running.
244 : Active,
245 : /// The timeline was previously Loading or Active but is shutting down.
246 : /// It cannot transition back into any other state.
247 : Stopping,
248 : /// The timeline is broken and not operational (previous states: Loading or Active).
249 : Broken { reason: String, backtrace: String },
250 : }
251 :
252 : #[serde_with::serde_as]
253 0 : #[derive(Debug, Clone, serde::Deserialize, serde::Serialize)]
254 : pub struct CompactLsnRange {
255 : pub start: Lsn,
256 : pub end: Lsn,
257 : }
258 :
259 : #[serde_with::serde_as]
260 0 : #[derive(Debug, Clone, serde::Deserialize, serde::Serialize)]
261 : pub struct CompactKeyRange {
262 : #[serde_as(as = "serde_with::DisplayFromStr")]
263 : pub start: Key,
264 : #[serde_as(as = "serde_with::DisplayFromStr")]
265 : pub end: Key,
266 : }
267 :
268 : impl From<Range<Lsn>> for CompactLsnRange {
269 12 : fn from(range: Range<Lsn>) -> Self {
270 12 : Self {
271 12 : start: range.start,
272 12 : end: range.end,
273 12 : }
274 12 : }
275 : }
276 :
277 : impl From<Range<Key>> for CompactKeyRange {
278 32 : fn from(range: Range<Key>) -> Self {
279 32 : Self {
280 32 : start: range.start,
281 32 : end: range.end,
282 32 : }
283 32 : }
284 : }
285 :
286 : impl From<CompactLsnRange> for Range<Lsn> {
287 20 : fn from(range: CompactLsnRange) -> Self {
288 20 : range.start..range.end
289 20 : }
290 : }
291 :
292 : impl From<CompactKeyRange> for Range<Key> {
293 32 : fn from(range: CompactKeyRange) -> Self {
294 32 : range.start..range.end
295 32 : }
296 : }
297 :
298 : impl CompactLsnRange {
299 8 : pub fn above(lsn: Lsn) -> Self {
300 8 : Self {
301 8 : start: lsn,
302 8 : end: Lsn::MAX,
303 8 : }
304 8 : }
305 : }
306 :
307 : #[derive(Debug, Clone, Serialize)]
308 : pub struct CompactInfoResponse {
309 : pub compact_key_range: Option<CompactKeyRange>,
310 : pub compact_lsn_range: Option<CompactLsnRange>,
311 : pub sub_compaction: bool,
312 : pub running: bool,
313 : pub job_id: usize,
314 : }
315 :
316 0 : #[derive(Serialize, Deserialize, Clone)]
317 : pub struct TimelineCreateRequest {
318 : pub new_timeline_id: TimelineId,
319 : #[serde(flatten)]
320 : pub mode: TimelineCreateRequestMode,
321 : }
322 :
323 : /// Storage controller specific extensions to [`TimelineInfo`].
324 0 : #[derive(Serialize, Deserialize, Clone)]
325 : pub struct TimelineCreateResponseStorcon {
326 : #[serde(flatten)]
327 : pub timeline_info: TimelineInfo,
328 :
329 : pub safekeepers: Option<SafekeepersInfo>,
330 : }
331 :
332 : /// Safekeepers as returned in timeline creation request to storcon or pushed to
333 : /// cplane in the post migration hook.
334 0 : #[derive(Serialize, Deserialize, Clone)]
335 : pub struct SafekeepersInfo {
336 : pub tenant_id: TenantId,
337 : pub timeline_id: TimelineId,
338 : pub generation: u32,
339 : pub safekeepers: Vec<SafekeeperInfo>,
340 : }
341 :
342 0 : #[derive(Serialize, Deserialize, Clone)]
343 : pub struct SafekeeperInfo {
344 : pub id: NodeId,
345 : pub hostname: String,
346 : }
347 :
348 0 : #[derive(Serialize, Deserialize, Clone)]
349 : #[serde(untagged)]
350 : pub enum TimelineCreateRequestMode {
351 : Branch {
352 : ancestor_timeline_id: TimelineId,
353 : #[serde(default)]
354 : ancestor_start_lsn: Option<Lsn>,
355 : // TODO: cplane sets this, but, the branching code always
356 : // inherits the ancestor's pg_version. Earlier code wasn't
357 : // using a flattened enum, so, it was an accepted field, and
358 : // we continue to accept it by having it here.
359 : pg_version: Option<u32>,
360 : },
361 : ImportPgdata {
362 : import_pgdata: TimelineCreateRequestModeImportPgdata,
363 : },
364 : // NB: Bootstrap is all-optional, and thus the serde(untagged) will cause serde to stop at Bootstrap.
365 : // (serde picks the first matching enum variant, in declaration order).
366 : Bootstrap {
367 : #[serde(default)]
368 : existing_initdb_timeline_id: Option<TimelineId>,
369 : pg_version: Option<u32>,
370 : },
371 : }
372 :
373 0 : #[derive(Serialize, Deserialize, Clone)]
374 : pub struct TimelineCreateRequestModeImportPgdata {
375 : pub location: ImportPgdataLocation,
376 : pub idempotency_key: ImportPgdataIdempotencyKey,
377 : }
378 :
379 0 : #[derive(Serialize, Deserialize, Clone, Debug)]
380 : pub enum ImportPgdataLocation {
381 : #[cfg(feature = "testing")]
382 : LocalFs { path: Utf8PathBuf },
383 : AwsS3 {
384 : region: String,
385 : bucket: String,
386 : /// A better name for this would be `prefix`; changing requires coordination with cplane.
387 : /// See <https://github.com/neondatabase/cloud/issues/20646>.
388 : key: String,
389 : },
390 : }
391 :
392 0 : #[derive(Serialize, Deserialize, Clone)]
393 : #[serde(transparent)]
394 : pub struct ImportPgdataIdempotencyKey(pub String);
395 :
396 : impl ImportPgdataIdempotencyKey {
397 0 : pub fn random() -> Self {
398 : use rand::Rng;
399 : use rand::distributions::Alphanumeric;
400 0 : Self(
401 0 : rand::thread_rng()
402 0 : .sample_iter(&Alphanumeric)
403 0 : .take(20)
404 0 : .map(char::from)
405 0 : .collect(),
406 0 : )
407 0 : }
408 : }
409 :
410 0 : #[derive(Serialize, Deserialize, Clone)]
411 : pub struct LsnLeaseRequest {
412 : pub lsn: Lsn,
413 : }
414 :
415 0 : #[derive(Serialize, Deserialize)]
416 : pub struct TenantShardSplitRequest {
417 : pub new_shard_count: u8,
418 :
419 : // A tenant's stripe size is only meaningful the first time their shard count goes
420 : // above 1: therefore during a split from 1->N shards, we may modify the stripe size.
421 : //
422 : // If this is set while the stripe count is being increased from an already >1 value,
423 : // then the request will fail with 400.
424 : pub new_stripe_size: Option<ShardStripeSize>,
425 : }
426 :
427 0 : #[derive(Serialize, Deserialize)]
428 : pub struct TenantShardSplitResponse {
429 : pub new_shards: Vec<TenantShardId>,
430 : }
431 :
432 : /// Parameters that apply to all shards in a tenant. Used during tenant creation.
433 0 : #[derive(Serialize, Deserialize, Debug)]
434 : #[serde(deny_unknown_fields)]
435 : pub struct ShardParameters {
436 : pub count: ShardCount,
437 : pub stripe_size: ShardStripeSize,
438 : }
439 :
440 : impl ShardParameters {
441 0 : pub fn is_unsharded(&self) -> bool {
442 0 : self.count.is_unsharded()
443 0 : }
444 : }
445 :
446 : impl Default for ShardParameters {
447 461 : fn default() -> Self {
448 461 : Self {
449 461 : count: ShardCount::new(0),
450 461 : stripe_size: DEFAULT_STRIPE_SIZE,
451 461 : }
452 461 : }
453 : }
454 :
455 : #[derive(Debug, Default, Clone, Eq, PartialEq)]
456 : pub enum FieldPatch<T> {
457 : Upsert(T),
458 : Remove,
459 : #[default]
460 : Noop,
461 : }
462 :
463 : impl<T> FieldPatch<T> {
464 70 : fn is_noop(&self) -> bool {
465 70 : matches!(self, FieldPatch::Noop)
466 70 : }
467 :
468 35 : pub fn apply(self, target: &mut Option<T>) {
469 35 : match self {
470 1 : Self::Upsert(v) => *target = Some(v),
471 1 : Self::Remove => *target = None,
472 33 : Self::Noop => {}
473 : }
474 35 : }
475 :
476 10 : pub fn map<U, E, F: FnOnce(T) -> Result<U, E>>(self, map: F) -> Result<FieldPatch<U>, E> {
477 10 : match self {
478 0 : Self::Upsert(v) => Ok(FieldPatch::<U>::Upsert(map(v)?)),
479 0 : Self::Remove => Ok(FieldPatch::<U>::Remove),
480 10 : Self::Noop => Ok(FieldPatch::<U>::Noop),
481 : }
482 10 : }
483 : }
484 :
485 : impl<'de, T: Deserialize<'de>> Deserialize<'de> for FieldPatch<T> {
486 2 : fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
487 2 : where
488 2 : D: Deserializer<'de>,
489 2 : {
490 2 : Option::deserialize(deserializer).map(|opt| match opt {
491 1 : None => FieldPatch::Remove,
492 1 : Some(val) => FieldPatch::Upsert(val),
493 2 : })
494 2 : }
495 : }
496 :
497 : impl<T: Serialize> Serialize for FieldPatch<T> {
498 2 : fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
499 2 : where
500 2 : S: Serializer,
501 2 : {
502 2 : match self {
503 1 : FieldPatch::Upsert(val) => serializer.serialize_some(val),
504 1 : FieldPatch::Remove => serializer.serialize_none(),
505 0 : FieldPatch::Noop => unreachable!(),
506 : }
507 2 : }
508 : }
509 :
510 2 : #[derive(Serialize, Deserialize, Debug, Default, Clone, Eq, PartialEq)]
511 : #[serde(default)]
512 : pub struct TenantConfigPatch {
513 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
514 : pub checkpoint_distance: FieldPatch<u64>,
515 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
516 : pub checkpoint_timeout: FieldPatch<String>,
517 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
518 : pub compaction_target_size: FieldPatch<u64>,
519 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
520 : pub compaction_period: FieldPatch<String>,
521 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
522 : pub compaction_threshold: FieldPatch<usize>,
523 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
524 : pub compaction_upper_limit: FieldPatch<usize>,
525 : // defer parsing compaction_algorithm, like eviction_policy
526 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
527 : pub compaction_algorithm: FieldPatch<CompactionAlgorithmSettings>,
528 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
529 : pub compaction_l0_first: FieldPatch<bool>,
530 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
531 : pub compaction_l0_semaphore: FieldPatch<bool>,
532 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
533 : pub l0_flush_delay_threshold: FieldPatch<usize>,
534 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
535 : pub l0_flush_stall_threshold: FieldPatch<usize>,
536 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
537 : pub gc_horizon: FieldPatch<u64>,
538 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
539 : pub gc_period: FieldPatch<String>,
540 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
541 : pub image_creation_threshold: FieldPatch<usize>,
542 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
543 : pub pitr_interval: FieldPatch<String>,
544 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
545 : pub walreceiver_connect_timeout: FieldPatch<String>,
546 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
547 : pub lagging_wal_timeout: FieldPatch<String>,
548 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
549 : pub max_lsn_wal_lag: FieldPatch<NonZeroU64>,
550 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
551 : pub eviction_policy: FieldPatch<EvictionPolicy>,
552 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
553 : pub min_resident_size_override: FieldPatch<u64>,
554 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
555 : pub evictions_low_residence_duration_metric_threshold: FieldPatch<String>,
556 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
557 : pub heatmap_period: FieldPatch<String>,
558 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
559 : pub lazy_slru_download: FieldPatch<bool>,
560 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
561 : pub timeline_get_throttle: FieldPatch<ThrottleConfig>,
562 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
563 : pub image_layer_creation_check_threshold: FieldPatch<u8>,
564 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
565 : pub image_creation_preempt_threshold: FieldPatch<usize>,
566 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
567 : pub lsn_lease_length: FieldPatch<String>,
568 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
569 : pub lsn_lease_length_for_ts: FieldPatch<String>,
570 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
571 : pub timeline_offloading: FieldPatch<bool>,
572 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
573 : pub wal_receiver_protocol_override: FieldPatch<PostgresClientProtocol>,
574 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
575 : pub rel_size_v2_enabled: FieldPatch<bool>,
576 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
577 : pub gc_compaction_enabled: FieldPatch<bool>,
578 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
579 : pub gc_compaction_initial_threshold_kb: FieldPatch<u64>,
580 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
581 : pub gc_compaction_ratio_percent: FieldPatch<u64>,
582 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
583 : pub sampling_ratio: FieldPatch<Option<Ratio>>,
584 : }
585 :
586 : /// Like [`crate::config::TenantConfigToml`], but preserves the information
587 : /// about which parameters are set and which are not.
588 : ///
589 : /// Used in many places, including durably stored ones.
590 8 : #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Default)]
591 : #[serde(default)] // this maps omitted fields in deserialization to None
592 : pub struct TenantConfig {
593 : #[serde(skip_serializing_if = "Option::is_none")]
594 : pub checkpoint_distance: Option<u64>,
595 :
596 : #[serde(skip_serializing_if = "Option::is_none")]
597 : #[serde(with = "humantime_serde")]
598 : pub checkpoint_timeout: Option<Duration>,
599 :
600 : #[serde(skip_serializing_if = "Option::is_none")]
601 : pub compaction_target_size: Option<u64>,
602 :
603 : #[serde(skip_serializing_if = "Option::is_none")]
604 : #[serde(with = "humantime_serde")]
605 : pub compaction_period: Option<Duration>,
606 :
607 : #[serde(skip_serializing_if = "Option::is_none")]
608 : pub compaction_threshold: Option<usize>,
609 :
610 : #[serde(skip_serializing_if = "Option::is_none")]
611 : pub compaction_upper_limit: Option<usize>,
612 :
613 : #[serde(skip_serializing_if = "Option::is_none")]
614 : pub compaction_algorithm: Option<CompactionAlgorithmSettings>,
615 :
616 : #[serde(skip_serializing_if = "Option::is_none")]
617 : pub compaction_l0_first: Option<bool>,
618 :
619 : #[serde(skip_serializing_if = "Option::is_none")]
620 : pub compaction_l0_semaphore: Option<bool>,
621 :
622 : #[serde(skip_serializing_if = "Option::is_none")]
623 : pub l0_flush_delay_threshold: Option<usize>,
624 :
625 : #[serde(skip_serializing_if = "Option::is_none")]
626 : pub l0_flush_stall_threshold: Option<usize>,
627 :
628 : #[serde(skip_serializing_if = "Option::is_none")]
629 : pub gc_horizon: Option<u64>,
630 :
631 : #[serde(skip_serializing_if = "Option::is_none")]
632 : #[serde(with = "humantime_serde")]
633 : pub gc_period: Option<Duration>,
634 :
635 : #[serde(skip_serializing_if = "Option::is_none")]
636 : pub image_creation_threshold: Option<usize>,
637 :
638 : #[serde(skip_serializing_if = "Option::is_none")]
639 : #[serde(with = "humantime_serde")]
640 : pub pitr_interval: Option<Duration>,
641 :
642 : #[serde(skip_serializing_if = "Option::is_none")]
643 : #[serde(with = "humantime_serde")]
644 : pub walreceiver_connect_timeout: Option<Duration>,
645 :
646 : #[serde(skip_serializing_if = "Option::is_none")]
647 : #[serde(with = "humantime_serde")]
648 : pub lagging_wal_timeout: Option<Duration>,
649 :
650 : #[serde(skip_serializing_if = "Option::is_none")]
651 : pub max_lsn_wal_lag: Option<NonZeroU64>,
652 :
653 : #[serde(skip_serializing_if = "Option::is_none")]
654 : pub eviction_policy: Option<EvictionPolicy>,
655 :
656 : #[serde(skip_serializing_if = "Option::is_none")]
657 : pub min_resident_size_override: Option<u64>,
658 :
659 : #[serde(skip_serializing_if = "Option::is_none")]
660 : #[serde(with = "humantime_serde")]
661 : pub evictions_low_residence_duration_metric_threshold: Option<Duration>,
662 :
663 : #[serde(skip_serializing_if = "Option::is_none")]
664 : #[serde(with = "humantime_serde")]
665 : pub heatmap_period: Option<Duration>,
666 :
667 : #[serde(skip_serializing_if = "Option::is_none")]
668 : pub lazy_slru_download: Option<bool>,
669 :
670 : #[serde(skip_serializing_if = "Option::is_none")]
671 : pub timeline_get_throttle: Option<ThrottleConfig>,
672 :
673 : #[serde(skip_serializing_if = "Option::is_none")]
674 : pub image_layer_creation_check_threshold: Option<u8>,
675 :
676 : #[serde(skip_serializing_if = "Option::is_none")]
677 : pub image_creation_preempt_threshold: Option<usize>,
678 :
679 : #[serde(skip_serializing_if = "Option::is_none")]
680 : #[serde(with = "humantime_serde")]
681 : pub lsn_lease_length: Option<Duration>,
682 :
683 : #[serde(skip_serializing_if = "Option::is_none")]
684 : #[serde(with = "humantime_serde")]
685 : pub lsn_lease_length_for_ts: Option<Duration>,
686 :
687 : #[serde(skip_serializing_if = "Option::is_none")]
688 : pub timeline_offloading: Option<bool>,
689 :
690 : #[serde(skip_serializing_if = "Option::is_none")]
691 : pub wal_receiver_protocol_override: Option<PostgresClientProtocol>,
692 :
693 : #[serde(skip_serializing_if = "Option::is_none")]
694 : pub rel_size_v2_enabled: Option<bool>,
695 :
696 : #[serde(skip_serializing_if = "Option::is_none")]
697 : pub gc_compaction_enabled: Option<bool>,
698 :
699 : #[serde(skip_serializing_if = "Option::is_none")]
700 : pub gc_compaction_initial_threshold_kb: Option<u64>,
701 :
702 : #[serde(skip_serializing_if = "Option::is_none")]
703 : pub gc_compaction_ratio_percent: Option<u64>,
704 :
705 : #[serde(skip_serializing_if = "Option::is_none")]
706 : pub sampling_ratio: Option<Option<Ratio>>,
707 : }
708 :
709 : impl TenantConfig {
710 1 : pub fn apply_patch(
711 1 : self,
712 1 : patch: TenantConfigPatch,
713 1 : ) -> Result<TenantConfig, humantime::DurationError> {
714 1 : let Self {
715 1 : mut checkpoint_distance,
716 1 : mut checkpoint_timeout,
717 1 : mut compaction_target_size,
718 1 : mut compaction_period,
719 1 : mut compaction_threshold,
720 1 : mut compaction_upper_limit,
721 1 : mut compaction_algorithm,
722 1 : mut compaction_l0_first,
723 1 : mut compaction_l0_semaphore,
724 1 : mut l0_flush_delay_threshold,
725 1 : mut l0_flush_stall_threshold,
726 1 : mut gc_horizon,
727 1 : mut gc_period,
728 1 : mut image_creation_threshold,
729 1 : mut pitr_interval,
730 1 : mut walreceiver_connect_timeout,
731 1 : mut lagging_wal_timeout,
732 1 : mut max_lsn_wal_lag,
733 1 : mut eviction_policy,
734 1 : mut min_resident_size_override,
735 1 : mut evictions_low_residence_duration_metric_threshold,
736 1 : mut heatmap_period,
737 1 : mut lazy_slru_download,
738 1 : mut timeline_get_throttle,
739 1 : mut image_layer_creation_check_threshold,
740 1 : mut image_creation_preempt_threshold,
741 1 : mut lsn_lease_length,
742 1 : mut lsn_lease_length_for_ts,
743 1 : mut timeline_offloading,
744 1 : mut wal_receiver_protocol_override,
745 1 : mut rel_size_v2_enabled,
746 1 : mut gc_compaction_enabled,
747 1 : mut gc_compaction_initial_threshold_kb,
748 1 : mut gc_compaction_ratio_percent,
749 1 : mut sampling_ratio,
750 1 : } = self;
751 1 :
752 1 : patch.checkpoint_distance.apply(&mut checkpoint_distance);
753 1 : patch
754 1 : .checkpoint_timeout
755 1 : .map(|v| humantime::parse_duration(&v))?
756 1 : .apply(&mut checkpoint_timeout);
757 1 : patch
758 1 : .compaction_target_size
759 1 : .apply(&mut compaction_target_size);
760 1 : patch
761 1 : .compaction_period
762 1 : .map(|v| humantime::parse_duration(&v))?
763 1 : .apply(&mut compaction_period);
764 1 : patch.compaction_threshold.apply(&mut compaction_threshold);
765 1 : patch
766 1 : .compaction_upper_limit
767 1 : .apply(&mut compaction_upper_limit);
768 1 : patch.compaction_algorithm.apply(&mut compaction_algorithm);
769 1 : patch.compaction_l0_first.apply(&mut compaction_l0_first);
770 1 : patch
771 1 : .compaction_l0_semaphore
772 1 : .apply(&mut compaction_l0_semaphore);
773 1 : patch
774 1 : .l0_flush_delay_threshold
775 1 : .apply(&mut l0_flush_delay_threshold);
776 1 : patch
777 1 : .l0_flush_stall_threshold
778 1 : .apply(&mut l0_flush_stall_threshold);
779 1 : patch.gc_horizon.apply(&mut gc_horizon);
780 1 : patch
781 1 : .gc_period
782 1 : .map(|v| humantime::parse_duration(&v))?
783 1 : .apply(&mut gc_period);
784 1 : patch
785 1 : .image_creation_threshold
786 1 : .apply(&mut image_creation_threshold);
787 1 : patch
788 1 : .pitr_interval
789 1 : .map(|v| humantime::parse_duration(&v))?
790 1 : .apply(&mut pitr_interval);
791 1 : patch
792 1 : .walreceiver_connect_timeout
793 1 : .map(|v| humantime::parse_duration(&v))?
794 1 : .apply(&mut walreceiver_connect_timeout);
795 1 : patch
796 1 : .lagging_wal_timeout
797 1 : .map(|v| humantime::parse_duration(&v))?
798 1 : .apply(&mut lagging_wal_timeout);
799 1 : patch.max_lsn_wal_lag.apply(&mut max_lsn_wal_lag);
800 1 : patch.eviction_policy.apply(&mut eviction_policy);
801 1 : patch
802 1 : .min_resident_size_override
803 1 : .apply(&mut min_resident_size_override);
804 1 : patch
805 1 : .evictions_low_residence_duration_metric_threshold
806 1 : .map(|v| humantime::parse_duration(&v))?
807 1 : .apply(&mut evictions_low_residence_duration_metric_threshold);
808 1 : patch
809 1 : .heatmap_period
810 1 : .map(|v| humantime::parse_duration(&v))?
811 1 : .apply(&mut heatmap_period);
812 1 : patch.lazy_slru_download.apply(&mut lazy_slru_download);
813 1 : patch
814 1 : .timeline_get_throttle
815 1 : .apply(&mut timeline_get_throttle);
816 1 : patch
817 1 : .image_layer_creation_check_threshold
818 1 : .apply(&mut image_layer_creation_check_threshold);
819 1 : patch
820 1 : .image_creation_preempt_threshold
821 1 : .apply(&mut image_creation_preempt_threshold);
822 1 : patch
823 1 : .lsn_lease_length
824 1 : .map(|v| humantime::parse_duration(&v))?
825 1 : .apply(&mut lsn_lease_length);
826 1 : patch
827 1 : .lsn_lease_length_for_ts
828 1 : .map(|v| humantime::parse_duration(&v))?
829 1 : .apply(&mut lsn_lease_length_for_ts);
830 1 : patch.timeline_offloading.apply(&mut timeline_offloading);
831 1 : patch
832 1 : .wal_receiver_protocol_override
833 1 : .apply(&mut wal_receiver_protocol_override);
834 1 : patch.rel_size_v2_enabled.apply(&mut rel_size_v2_enabled);
835 1 : patch
836 1 : .gc_compaction_enabled
837 1 : .apply(&mut gc_compaction_enabled);
838 1 : patch
839 1 : .gc_compaction_initial_threshold_kb
840 1 : .apply(&mut gc_compaction_initial_threshold_kb);
841 1 : patch
842 1 : .gc_compaction_ratio_percent
843 1 : .apply(&mut gc_compaction_ratio_percent);
844 1 : patch.sampling_ratio.apply(&mut sampling_ratio);
845 1 :
846 1 : Ok(Self {
847 1 : checkpoint_distance,
848 1 : checkpoint_timeout,
849 1 : compaction_target_size,
850 1 : compaction_period,
851 1 : compaction_threshold,
852 1 : compaction_upper_limit,
853 1 : compaction_algorithm,
854 1 : compaction_l0_first,
855 1 : compaction_l0_semaphore,
856 1 : l0_flush_delay_threshold,
857 1 : l0_flush_stall_threshold,
858 1 : gc_horizon,
859 1 : gc_period,
860 1 : image_creation_threshold,
861 1 : pitr_interval,
862 1 : walreceiver_connect_timeout,
863 1 : lagging_wal_timeout,
864 1 : max_lsn_wal_lag,
865 1 : eviction_policy,
866 1 : min_resident_size_override,
867 1 : evictions_low_residence_duration_metric_threshold,
868 1 : heatmap_period,
869 1 : lazy_slru_download,
870 1 : timeline_get_throttle,
871 1 : image_layer_creation_check_threshold,
872 1 : image_creation_preempt_threshold,
873 1 : lsn_lease_length,
874 1 : lsn_lease_length_for_ts,
875 1 : timeline_offloading,
876 1 : wal_receiver_protocol_override,
877 1 : rel_size_v2_enabled,
878 1 : gc_compaction_enabled,
879 1 : gc_compaction_initial_threshold_kb,
880 1 : gc_compaction_ratio_percent,
881 1 : sampling_ratio,
882 1 : })
883 1 : }
884 :
885 0 : pub fn merge(
886 0 : &self,
887 0 : global_conf: crate::config::TenantConfigToml,
888 0 : ) -> crate::config::TenantConfigToml {
889 0 : crate::config::TenantConfigToml {
890 0 : checkpoint_distance: self
891 0 : .checkpoint_distance
892 0 : .unwrap_or(global_conf.checkpoint_distance),
893 0 : checkpoint_timeout: self
894 0 : .checkpoint_timeout
895 0 : .unwrap_or(global_conf.checkpoint_timeout),
896 0 : compaction_target_size: self
897 0 : .compaction_target_size
898 0 : .unwrap_or(global_conf.compaction_target_size),
899 0 : compaction_period: self
900 0 : .compaction_period
901 0 : .unwrap_or(global_conf.compaction_period),
902 0 : compaction_threshold: self
903 0 : .compaction_threshold
904 0 : .unwrap_or(global_conf.compaction_threshold),
905 0 : compaction_upper_limit: self
906 0 : .compaction_upper_limit
907 0 : .unwrap_or(global_conf.compaction_upper_limit),
908 0 : compaction_algorithm: self
909 0 : .compaction_algorithm
910 0 : .as_ref()
911 0 : .unwrap_or(&global_conf.compaction_algorithm)
912 0 : .clone(),
913 0 : compaction_l0_first: self
914 0 : .compaction_l0_first
915 0 : .unwrap_or(global_conf.compaction_l0_first),
916 0 : compaction_l0_semaphore: self
917 0 : .compaction_l0_semaphore
918 0 : .unwrap_or(global_conf.compaction_l0_semaphore),
919 0 : l0_flush_delay_threshold: self
920 0 : .l0_flush_delay_threshold
921 0 : .or(global_conf.l0_flush_delay_threshold),
922 0 : l0_flush_stall_threshold: self
923 0 : .l0_flush_stall_threshold
924 0 : .or(global_conf.l0_flush_stall_threshold),
925 0 : gc_horizon: self.gc_horizon.unwrap_or(global_conf.gc_horizon),
926 0 : gc_period: self.gc_period.unwrap_or(global_conf.gc_period),
927 0 : image_creation_threshold: self
928 0 : .image_creation_threshold
929 0 : .unwrap_or(global_conf.image_creation_threshold),
930 0 : pitr_interval: self.pitr_interval.unwrap_or(global_conf.pitr_interval),
931 0 : walreceiver_connect_timeout: self
932 0 : .walreceiver_connect_timeout
933 0 : .unwrap_or(global_conf.walreceiver_connect_timeout),
934 0 : lagging_wal_timeout: self
935 0 : .lagging_wal_timeout
936 0 : .unwrap_or(global_conf.lagging_wal_timeout),
937 0 : max_lsn_wal_lag: self.max_lsn_wal_lag.unwrap_or(global_conf.max_lsn_wal_lag),
938 0 : eviction_policy: self.eviction_policy.unwrap_or(global_conf.eviction_policy),
939 0 : min_resident_size_override: self
940 0 : .min_resident_size_override
941 0 : .or(global_conf.min_resident_size_override),
942 0 : evictions_low_residence_duration_metric_threshold: self
943 0 : .evictions_low_residence_duration_metric_threshold
944 0 : .unwrap_or(global_conf.evictions_low_residence_duration_metric_threshold),
945 0 : heatmap_period: self.heatmap_period.unwrap_or(global_conf.heatmap_period),
946 0 : lazy_slru_download: self
947 0 : .lazy_slru_download
948 0 : .unwrap_or(global_conf.lazy_slru_download),
949 0 : timeline_get_throttle: self
950 0 : .timeline_get_throttle
951 0 : .clone()
952 0 : .unwrap_or(global_conf.timeline_get_throttle),
953 0 : image_layer_creation_check_threshold: self
954 0 : .image_layer_creation_check_threshold
955 0 : .unwrap_or(global_conf.image_layer_creation_check_threshold),
956 0 : image_creation_preempt_threshold: self
957 0 : .image_creation_preempt_threshold
958 0 : .unwrap_or(global_conf.image_creation_preempt_threshold),
959 0 : lsn_lease_length: self
960 0 : .lsn_lease_length
961 0 : .unwrap_or(global_conf.lsn_lease_length),
962 0 : lsn_lease_length_for_ts: self
963 0 : .lsn_lease_length_for_ts
964 0 : .unwrap_or(global_conf.lsn_lease_length_for_ts),
965 0 : timeline_offloading: self
966 0 : .timeline_offloading
967 0 : .unwrap_or(global_conf.timeline_offloading),
968 0 : wal_receiver_protocol_override: self
969 0 : .wal_receiver_protocol_override
970 0 : .or(global_conf.wal_receiver_protocol_override),
971 0 : rel_size_v2_enabled: self
972 0 : .rel_size_v2_enabled
973 0 : .unwrap_or(global_conf.rel_size_v2_enabled),
974 0 : gc_compaction_enabled: self
975 0 : .gc_compaction_enabled
976 0 : .unwrap_or(global_conf.gc_compaction_enabled),
977 0 : gc_compaction_initial_threshold_kb: self
978 0 : .gc_compaction_initial_threshold_kb
979 0 : .unwrap_or(global_conf.gc_compaction_initial_threshold_kb),
980 0 : gc_compaction_ratio_percent: self
981 0 : .gc_compaction_ratio_percent
982 0 : .unwrap_or(global_conf.gc_compaction_ratio_percent),
983 0 : sampling_ratio: self.sampling_ratio.unwrap_or(global_conf.sampling_ratio),
984 0 : }
985 0 : }
986 : }
987 :
988 : /// The policy for the aux file storage.
989 : ///
990 : /// It can be switched through `switch_aux_file_policy` tenant config.
991 : /// When the first aux file written, the policy will be persisted in the
992 : /// `index_part.json` file and has a limited migration path.
993 : ///
994 : /// Currently, we only allow the following migration path:
995 : ///
996 : /// Unset -> V1
997 : /// -> V2
998 : /// -> CrossValidation -> V2
999 : #[derive(
1000 : Eq,
1001 : PartialEq,
1002 : Debug,
1003 : Copy,
1004 : Clone,
1005 0 : strum_macros::EnumString,
1006 : strum_macros::Display,
1007 4 : serde_with::DeserializeFromStr,
1008 : serde_with::SerializeDisplay,
1009 : )]
1010 : #[strum(serialize_all = "kebab-case")]
1011 : pub enum AuxFilePolicy {
1012 : /// V1 aux file policy: store everything in AUX_FILE_KEY
1013 : #[strum(ascii_case_insensitive)]
1014 : V1,
1015 : /// V2 aux file policy: store in the AUX_FILE keyspace
1016 : #[strum(ascii_case_insensitive)]
1017 : V2,
1018 : /// Cross validation runs both formats on the write path and does validation
1019 : /// on the read path.
1020 : #[strum(ascii_case_insensitive)]
1021 : CrossValidation,
1022 : }
1023 :
1024 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
1025 : #[serde(tag = "kind")]
1026 : pub enum EvictionPolicy {
1027 : NoEviction,
1028 : LayerAccessThreshold(EvictionPolicyLayerAccessThreshold),
1029 : OnlyImitiate(EvictionPolicyLayerAccessThreshold),
1030 : }
1031 :
1032 : impl EvictionPolicy {
1033 0 : pub fn discriminant_str(&self) -> &'static str {
1034 0 : match self {
1035 0 : EvictionPolicy::NoEviction => "NoEviction",
1036 0 : EvictionPolicy::LayerAccessThreshold(_) => "LayerAccessThreshold",
1037 0 : EvictionPolicy::OnlyImitiate(_) => "OnlyImitiate",
1038 : }
1039 0 : }
1040 : }
1041 :
1042 : #[derive(
1043 : Eq,
1044 : PartialEq,
1045 : Debug,
1046 : Copy,
1047 : Clone,
1048 0 : strum_macros::EnumString,
1049 : strum_macros::Display,
1050 0 : serde_with::DeserializeFromStr,
1051 : serde_with::SerializeDisplay,
1052 : )]
1053 : #[strum(serialize_all = "kebab-case")]
1054 : pub enum CompactionAlgorithm {
1055 : Legacy,
1056 : Tiered,
1057 : }
1058 :
1059 : #[derive(
1060 4 : Debug, Clone, Copy, PartialEq, Eq, serde_with::DeserializeFromStr, serde_with::SerializeDisplay,
1061 : )]
1062 : pub enum ImageCompressionAlgorithm {
1063 : // Disabled for writes, support decompressing during read path
1064 : Disabled,
1065 : /// Zstandard compression. Level 0 means and None mean the same (default level). Levels can be negative as well.
1066 : /// For details, see the [manual](http://facebook.github.io/zstd/zstd_manual.html).
1067 : Zstd {
1068 : level: Option<i8>,
1069 : },
1070 : }
1071 :
1072 : impl FromStr for ImageCompressionAlgorithm {
1073 : type Err = anyhow::Error;
1074 8 : fn from_str(s: &str) -> Result<Self, Self::Err> {
1075 8 : let mut components = s.split(['(', ')']);
1076 8 : let first = components
1077 8 : .next()
1078 8 : .ok_or_else(|| anyhow::anyhow!("empty string"))?;
1079 8 : match first {
1080 8 : "disabled" => Ok(ImageCompressionAlgorithm::Disabled),
1081 6 : "zstd" => {
1082 6 : let level = if let Some(v) = components.next() {
1083 4 : let v: i8 = v.parse()?;
1084 4 : Some(v)
1085 : } else {
1086 2 : None
1087 : };
1088 :
1089 6 : Ok(ImageCompressionAlgorithm::Zstd { level })
1090 : }
1091 0 : _ => anyhow::bail!("invalid specifier '{first}'"),
1092 : }
1093 8 : }
1094 : }
1095 :
1096 : impl Display for ImageCompressionAlgorithm {
1097 12 : fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
1098 12 : match self {
1099 3 : ImageCompressionAlgorithm::Disabled => write!(f, "disabled"),
1100 9 : ImageCompressionAlgorithm::Zstd { level } => {
1101 9 : if let Some(level) = level {
1102 6 : write!(f, "zstd({})", level)
1103 : } else {
1104 3 : write!(f, "zstd")
1105 : }
1106 : }
1107 : }
1108 12 : }
1109 : }
1110 :
1111 0 : #[derive(Eq, PartialEq, Debug, Clone, Serialize, Deserialize)]
1112 : pub struct CompactionAlgorithmSettings {
1113 : pub kind: CompactionAlgorithm,
1114 : }
1115 :
1116 0 : #[derive(Debug, PartialEq, Eq, Clone, Deserialize, Serialize)]
1117 : #[serde(tag = "mode", rename_all = "kebab-case")]
1118 : pub enum L0FlushConfig {
1119 : #[serde(rename_all = "snake_case")]
1120 : Direct { max_concurrency: NonZeroUsize },
1121 : }
1122 :
1123 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
1124 : pub struct EvictionPolicyLayerAccessThreshold {
1125 : #[serde(with = "humantime_serde")]
1126 : pub period: Duration,
1127 : #[serde(with = "humantime_serde")]
1128 : pub threshold: Duration,
1129 : }
1130 :
1131 6 : #[derive(Debug, Serialize, Deserialize, Clone, PartialEq, Eq)]
1132 : pub struct ThrottleConfig {
1133 : /// See [`ThrottleConfigTaskKinds`] for why we do the serde `rename`.
1134 : #[serde(rename = "task_kinds")]
1135 : pub enabled: ThrottleConfigTaskKinds,
1136 : pub initial: u32,
1137 : #[serde(with = "humantime_serde")]
1138 : pub refill_interval: Duration,
1139 : pub refill_amount: NonZeroU32,
1140 : pub max: u32,
1141 : }
1142 :
1143 : /// Before <https://github.com/neondatabase/neon/pull/9962>
1144 : /// the throttle was a per `Timeline::get`/`Timeline::get_vectored` call.
1145 : /// The `task_kinds` field controlled which Pageserver "Task Kind"s
1146 : /// were subject to the throttle.
1147 : ///
1148 : /// After that PR, the throttle is applied at pagestream request level
1149 : /// and the `task_kinds` field does not apply since the only task kind
1150 : /// that us subject to the throttle is that of the page service.
1151 : ///
1152 : /// However, we don't want to make a breaking config change right now
1153 : /// because it means we have to migrate all the tenant configs.
1154 : /// This will be done in a future PR.
1155 : ///
1156 : /// In the meantime, we use emptiness / non-emptsiness of the `task_kinds`
1157 : /// field to determine if the throttle is enabled or not.
1158 1 : #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq)]
1159 : #[serde(transparent)]
1160 : pub struct ThrottleConfigTaskKinds(Vec<String>);
1161 :
1162 : impl ThrottleConfigTaskKinds {
1163 497 : pub fn disabled() -> Self {
1164 497 : Self(vec![])
1165 497 : }
1166 462 : pub fn is_enabled(&self) -> bool {
1167 462 : !self.0.is_empty()
1168 462 : }
1169 : }
1170 :
1171 : impl ThrottleConfig {
1172 497 : pub fn disabled() -> Self {
1173 497 : Self {
1174 497 : enabled: ThrottleConfigTaskKinds::disabled(),
1175 497 : // other values don't matter with emtpy `task_kinds`.
1176 497 : initial: 0,
1177 497 : refill_interval: Duration::from_millis(1),
1178 497 : refill_amount: NonZeroU32::new(1).unwrap(),
1179 497 : max: 1,
1180 497 : }
1181 497 : }
1182 : /// The requests per second allowed by the given config.
1183 0 : pub fn steady_rps(&self) -> f64 {
1184 0 : (self.refill_amount.get() as f64) / (self.refill_interval.as_secs_f64())
1185 0 : }
1186 : }
1187 :
1188 : #[cfg(test)]
1189 : mod throttle_config_tests {
1190 : use super::*;
1191 :
1192 : #[test]
1193 1 : fn test_disabled_is_disabled() {
1194 1 : let config = ThrottleConfig::disabled();
1195 1 : assert!(!config.enabled.is_enabled());
1196 1 : }
1197 : #[test]
1198 1 : fn test_enabled_backwards_compat() {
1199 1 : let input = serde_json::json!({
1200 1 : "task_kinds": ["PageRequestHandler"],
1201 1 : "initial": 40000,
1202 1 : "refill_interval": "50ms",
1203 1 : "refill_amount": 1000,
1204 1 : "max": 40000,
1205 1 : "fair": true
1206 1 : });
1207 1 : let config: ThrottleConfig = serde_json::from_value(input).unwrap();
1208 1 : assert!(config.enabled.is_enabled());
1209 1 : }
1210 : }
1211 :
1212 : /// A flattened analog of a `pagesever::tenant::LocationMode`, which
1213 : /// lists out all possible states (and the virtual "Detached" state)
1214 : /// in a flat form rather than using rust-style enums.
1215 0 : #[derive(Serialize, Deserialize, Debug, Clone, Copy, Eq, PartialEq)]
1216 : pub enum LocationConfigMode {
1217 : AttachedSingle,
1218 : AttachedMulti,
1219 : AttachedStale,
1220 : Secondary,
1221 : Detached,
1222 : }
1223 :
1224 0 : #[derive(Serialize, Deserialize, Debug, Clone, Eq, PartialEq)]
1225 : pub struct LocationConfigSecondary {
1226 : pub warm: bool,
1227 : }
1228 :
1229 : /// An alternative representation of `pageserver::tenant::LocationConf`,
1230 : /// for use in external-facing APIs.
1231 0 : #[derive(Serialize, Deserialize, Debug, Clone, Eq, PartialEq)]
1232 : pub struct LocationConfig {
1233 : pub mode: LocationConfigMode,
1234 : /// If attaching, in what generation?
1235 : #[serde(default)]
1236 : pub generation: Option<u32>,
1237 :
1238 : // If requesting mode `Secondary`, configuration for that.
1239 : #[serde(default)]
1240 : pub secondary_conf: Option<LocationConfigSecondary>,
1241 :
1242 : // Shard parameters: if shard_count is nonzero, then other shard_* fields
1243 : // must be set accurately.
1244 : #[serde(default)]
1245 : pub shard_number: u8,
1246 : #[serde(default)]
1247 : pub shard_count: u8,
1248 : #[serde(default)]
1249 : pub shard_stripe_size: u32,
1250 :
1251 : // This configuration only affects attached mode, but should be provided irrespective
1252 : // of the mode, as a secondary location might transition on startup if the response
1253 : // to the `/re-attach` control plane API requests it.
1254 : pub tenant_conf: TenantConfig,
1255 : }
1256 :
1257 0 : #[derive(Serialize, Deserialize)]
1258 : pub struct LocationConfigListResponse {
1259 : pub tenant_shards: Vec<(TenantShardId, Option<LocationConfig>)>,
1260 : }
1261 :
1262 : #[derive(Serialize)]
1263 : pub struct StatusResponse {
1264 : pub id: NodeId,
1265 : }
1266 :
1267 0 : #[derive(Serialize, Deserialize, Debug)]
1268 : #[serde(deny_unknown_fields)]
1269 : pub struct TenantLocationConfigRequest {
1270 : #[serde(flatten)]
1271 : pub config: LocationConfig, // as we have a flattened field, we should reject all unknown fields in it
1272 : }
1273 :
1274 0 : #[derive(Serialize, Deserialize, Debug)]
1275 : #[serde(deny_unknown_fields)]
1276 : pub struct TenantTimeTravelRequest {
1277 : pub shard_counts: Vec<ShardCount>,
1278 : }
1279 :
1280 0 : #[derive(Serialize, Deserialize, Debug)]
1281 : #[serde(deny_unknown_fields)]
1282 : pub struct TenantShardLocation {
1283 : pub shard_id: TenantShardId,
1284 : pub node_id: NodeId,
1285 : }
1286 :
1287 0 : #[derive(Serialize, Deserialize, Debug)]
1288 : #[serde(deny_unknown_fields)]
1289 : pub struct TenantLocationConfigResponse {
1290 : pub shards: Vec<TenantShardLocation>,
1291 : // If the shards' ShardCount count is >1, stripe_size will be set.
1292 : pub stripe_size: Option<ShardStripeSize>,
1293 : }
1294 :
1295 2 : #[derive(Serialize, Deserialize, Debug)]
1296 : #[serde(deny_unknown_fields)]
1297 : pub struct TenantConfigRequest {
1298 : pub tenant_id: TenantId,
1299 : #[serde(flatten)]
1300 : pub config: TenantConfig, // as we have a flattened field, we should reject all unknown fields in it
1301 : }
1302 :
1303 : impl std::ops::Deref for TenantConfigRequest {
1304 : type Target = TenantConfig;
1305 :
1306 0 : fn deref(&self) -> &Self::Target {
1307 0 : &self.config
1308 0 : }
1309 : }
1310 :
1311 : impl TenantConfigRequest {
1312 0 : pub fn new(tenant_id: TenantId) -> TenantConfigRequest {
1313 0 : let config = TenantConfig::default();
1314 0 : TenantConfigRequest { tenant_id, config }
1315 0 : }
1316 : }
1317 :
1318 3 : #[derive(Serialize, Deserialize, Debug)]
1319 : #[serde(deny_unknown_fields)]
1320 : pub struct TenantConfigPatchRequest {
1321 : pub tenant_id: TenantId,
1322 : #[serde(flatten)]
1323 : pub config: TenantConfigPatch, // as we have a flattened field, we should reject all unknown fields in it
1324 : }
1325 :
1326 0 : #[derive(Serialize, Deserialize, Debug)]
1327 : pub struct TenantWaitLsnRequest {
1328 : #[serde(flatten)]
1329 : pub timelines: HashMap<TimelineId, Lsn>,
1330 : pub timeout: Duration,
1331 : }
1332 :
1333 : /// See [`TenantState::attachment_status`] and the OpenAPI docs for context.
1334 0 : #[derive(Serialize, Deserialize, Clone)]
1335 : #[serde(tag = "slug", content = "data", rename_all = "snake_case")]
1336 : pub enum TenantAttachmentStatus {
1337 : Maybe,
1338 : Attached,
1339 : Failed { reason: String },
1340 : }
1341 :
1342 0 : #[derive(Serialize, Deserialize, Clone)]
1343 : pub struct TenantInfo {
1344 : pub id: TenantShardId,
1345 : // NB: intentionally not part of OpenAPI, we don't want to commit to a specific set of TenantState's
1346 : pub state: TenantState,
1347 : /// Sum of the size of all layer files.
1348 : /// If a layer is present in both local FS and S3, it counts only once.
1349 : pub current_physical_size: Option<u64>, // physical size is only included in `tenant_status` endpoint
1350 : pub attachment_status: TenantAttachmentStatus,
1351 : pub generation: u32,
1352 :
1353 : /// Opaque explanation if gc is being blocked.
1354 : ///
1355 : /// Only looked up for the individual tenant detail, not the listing.
1356 : #[serde(skip_serializing_if = "Option::is_none")]
1357 : pub gc_blocking: Option<String>,
1358 : }
1359 :
1360 0 : #[derive(Serialize, Deserialize, Clone)]
1361 : pub struct TenantDetails {
1362 : #[serde(flatten)]
1363 : pub tenant_info: TenantInfo,
1364 :
1365 : pub walredo: Option<WalRedoManagerStatus>,
1366 :
1367 : pub timelines: Vec<TimelineId>,
1368 : }
1369 :
1370 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone, Copy, Debug)]
1371 : pub enum TimelineArchivalState {
1372 : Archived,
1373 : Unarchived,
1374 : }
1375 :
1376 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone)]
1377 : pub enum TimelineVisibilityState {
1378 : Visible,
1379 : Invisible,
1380 : }
1381 :
1382 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone)]
1383 : pub struct TimelineArchivalConfigRequest {
1384 : pub state: TimelineArchivalState,
1385 : }
1386 :
1387 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone)]
1388 : pub struct TimelinePatchIndexPartRequest {
1389 : pub rel_size_migration: Option<RelSizeMigration>,
1390 : pub gc_compaction_last_completed_lsn: Option<Lsn>,
1391 : pub applied_gc_cutoff_lsn: Option<Lsn>,
1392 : #[serde(default)]
1393 : pub force_index_update: bool,
1394 : }
1395 :
1396 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
1397 : pub struct TimelinesInfoAndOffloaded {
1398 : pub timelines: Vec<TimelineInfo>,
1399 : pub offloaded: Vec<OffloadedTimelineInfo>,
1400 : }
1401 :
1402 : /// Analog of [`TimelineInfo`] for offloaded timelines.
1403 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
1404 : pub struct OffloadedTimelineInfo {
1405 : pub tenant_id: TenantShardId,
1406 : pub timeline_id: TimelineId,
1407 : /// Whether the timeline has a parent it has been branched off from or not
1408 : pub ancestor_timeline_id: Option<TimelineId>,
1409 : /// Whether to retain the branch lsn at the ancestor or not
1410 : pub ancestor_retain_lsn: Option<Lsn>,
1411 : /// The time point when the timeline was archived
1412 : pub archived_at: chrono::DateTime<chrono::Utc>,
1413 : }
1414 :
1415 16 : #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
1416 : #[serde(rename_all = "camelCase")]
1417 : pub enum RelSizeMigration {
1418 : /// The tenant is using the old rel_size format.
1419 : /// Note that this enum is persisted as `Option<RelSizeMigration>` in the index part, so
1420 : /// `None` is the same as `Some(RelSizeMigration::Legacy)`.
1421 : Legacy,
1422 : /// The tenant is migrating to the new rel_size format. Both old and new rel_size format are
1423 : /// persisted in the index part. The read path will read both formats and merge them.
1424 : Migrating,
1425 : /// The tenant has migrated to the new rel_size format. Only the new rel_size format is persisted
1426 : /// in the index part, and the read path will not read the old format.
1427 : Migrated,
1428 : }
1429 :
1430 : /// This represents the output of the "timeline_detail" and "timeline_list" API calls.
1431 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
1432 : pub struct TimelineInfo {
1433 : pub tenant_id: TenantShardId,
1434 : pub timeline_id: TimelineId,
1435 :
1436 : pub ancestor_timeline_id: Option<TimelineId>,
1437 : pub ancestor_lsn: Option<Lsn>,
1438 : pub last_record_lsn: Lsn,
1439 : pub prev_record_lsn: Option<Lsn>,
1440 :
1441 : /// The LSN up to which GC has advanced: older data may still exist but it is not available for clients.
1442 : /// This LSN is not suitable for deciding where to create branches etc: use [`TimelineInfo::min_readable_lsn`] instead,
1443 : /// as it is easier to reason about.
1444 : #[serde(default)]
1445 : pub applied_gc_cutoff_lsn: Lsn,
1446 :
1447 : /// The upper bound of data which is either already GC'ed, or elegible to be GC'ed at any time based on PITR interval.
1448 : /// This LSN represents the "end of history" for this timeline, and callers should use it to figure out the oldest
1449 : /// LSN at which it is legal to create a branch or ephemeral endpoint.
1450 : ///
1451 : /// Note that holders of valid LSN leases may be able to create branches and read pages earlier
1452 : /// than this LSN, but new leases may not be taken out earlier than this LSN.
1453 : #[serde(default)]
1454 : pub min_readable_lsn: Lsn,
1455 :
1456 : pub disk_consistent_lsn: Lsn,
1457 :
1458 : /// The LSN that we have succesfully uploaded to remote storage
1459 : pub remote_consistent_lsn: Lsn,
1460 :
1461 : /// The LSN that we are advertizing to safekeepers
1462 : pub remote_consistent_lsn_visible: Lsn,
1463 :
1464 : /// The LSN from the start of the root timeline (never changes)
1465 : pub initdb_lsn: Lsn,
1466 :
1467 : pub current_logical_size: u64,
1468 : pub current_logical_size_is_accurate: bool,
1469 :
1470 : pub directory_entries_counts: Vec<u64>,
1471 :
1472 : /// Sum of the size of all layer files.
1473 : /// If a layer is present in both local FS and S3, it counts only once.
1474 : pub current_physical_size: Option<u64>, // is None when timeline is Unloaded
1475 : pub current_logical_size_non_incremental: Option<u64>,
1476 :
1477 : /// How many bytes of WAL are within this branch's pitr_interval. If the pitr_interval goes
1478 : /// beyond the branch's branch point, we only count up to the branch point.
1479 : pub pitr_history_size: u64,
1480 :
1481 : /// Whether this branch's branch point is within its ancestor's PITR interval (i.e. any
1482 : /// ancestor data used by this branch would have been retained anyway). If this is false, then
1483 : /// this branch may be imposing a cost on the ancestor by causing it to retain layers that it would
1484 : /// otherwise be able to GC.
1485 : pub within_ancestor_pitr: bool,
1486 :
1487 : pub timeline_dir_layer_file_size_sum: Option<u64>,
1488 :
1489 : pub wal_source_connstr: Option<String>,
1490 : pub last_received_msg_lsn: Option<Lsn>,
1491 : /// the timestamp (in microseconds) of the last received message
1492 : pub last_received_msg_ts: Option<u128>,
1493 : pub pg_version: u32,
1494 :
1495 : pub state: TimelineState,
1496 :
1497 : pub walreceiver_status: String,
1498 :
1499 : // ALWAYS add new fields at the end of the struct with `Option` to ensure forward/backward compatibility.
1500 : // Backward compatibility: you will get a JSON not containing the newly-added field.
1501 : // Forward compatibility: a previous version of the pageserver will receive a JSON. serde::Deserialize does
1502 : // not deny unknown fields by default so it's safe to set the field to some value, though it won't be
1503 : // read.
1504 : /// Whether the timeline is archived.
1505 : pub is_archived: Option<bool>,
1506 :
1507 : /// The status of the rel_size migration.
1508 : pub rel_size_migration: Option<RelSizeMigration>,
1509 :
1510 : /// Whether the timeline is invisible in synthetic size calculations.
1511 : pub is_invisible: Option<bool>,
1512 : }
1513 :
1514 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1515 : pub struct LayerMapInfo {
1516 : pub in_memory_layers: Vec<InMemoryLayerInfo>,
1517 : pub historic_layers: Vec<HistoricLayerInfo>,
1518 : }
1519 :
1520 : /// The residence status of a layer
1521 0 : #[derive(Debug, Clone, Copy, Serialize, Deserialize)]
1522 : pub enum LayerResidenceStatus {
1523 : /// Residence status for a layer file that exists locally.
1524 : /// It may also exist on the remote, we don't care here.
1525 : Resident,
1526 : /// Residence status for a layer file that only exists on the remote.
1527 : Evicted,
1528 : }
1529 :
1530 : #[serde_as]
1531 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1532 : pub struct LayerAccessStats {
1533 : #[serde_as(as = "serde_with::TimestampMilliSeconds")]
1534 : pub access_time: SystemTime,
1535 :
1536 : #[serde_as(as = "serde_with::TimestampMilliSeconds")]
1537 : pub residence_time: SystemTime,
1538 :
1539 : pub visible: bool,
1540 : }
1541 :
1542 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1543 : #[serde(tag = "kind")]
1544 : pub enum InMemoryLayerInfo {
1545 : Open { lsn_start: Lsn },
1546 : Frozen { lsn_start: Lsn, lsn_end: Lsn },
1547 : }
1548 :
1549 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1550 : #[serde(tag = "kind")]
1551 : pub enum HistoricLayerInfo {
1552 : Delta {
1553 : layer_file_name: String,
1554 : layer_file_size: u64,
1555 :
1556 : lsn_start: Lsn,
1557 : lsn_end: Lsn,
1558 : remote: bool,
1559 : access_stats: LayerAccessStats,
1560 :
1561 : l0: bool,
1562 : },
1563 : Image {
1564 : layer_file_name: String,
1565 : layer_file_size: u64,
1566 :
1567 : lsn_start: Lsn,
1568 : remote: bool,
1569 : access_stats: LayerAccessStats,
1570 : },
1571 : }
1572 :
1573 : impl HistoricLayerInfo {
1574 0 : pub fn layer_file_name(&self) -> &str {
1575 0 : match self {
1576 : HistoricLayerInfo::Delta {
1577 0 : layer_file_name, ..
1578 0 : } => layer_file_name,
1579 : HistoricLayerInfo::Image {
1580 0 : layer_file_name, ..
1581 0 : } => layer_file_name,
1582 : }
1583 0 : }
1584 0 : pub fn is_remote(&self) -> bool {
1585 0 : match self {
1586 0 : HistoricLayerInfo::Delta { remote, .. } => *remote,
1587 0 : HistoricLayerInfo::Image { remote, .. } => *remote,
1588 : }
1589 0 : }
1590 0 : pub fn set_remote(&mut self, value: bool) {
1591 0 : let field = match self {
1592 0 : HistoricLayerInfo::Delta { remote, .. } => remote,
1593 0 : HistoricLayerInfo::Image { remote, .. } => remote,
1594 : };
1595 0 : *field = value;
1596 0 : }
1597 0 : pub fn layer_file_size(&self) -> u64 {
1598 0 : match self {
1599 : HistoricLayerInfo::Delta {
1600 0 : layer_file_size, ..
1601 0 : } => *layer_file_size,
1602 : HistoricLayerInfo::Image {
1603 0 : layer_file_size, ..
1604 0 : } => *layer_file_size,
1605 : }
1606 0 : }
1607 : }
1608 :
1609 0 : #[derive(Debug, Serialize, Deserialize)]
1610 : pub struct DownloadRemoteLayersTaskSpawnRequest {
1611 : pub max_concurrent_downloads: NonZeroUsize,
1612 : }
1613 :
1614 0 : #[derive(Debug, Serialize, Deserialize)]
1615 : pub struct IngestAuxFilesRequest {
1616 : pub aux_files: HashMap<String, String>,
1617 : }
1618 :
1619 0 : #[derive(Debug, Serialize, Deserialize)]
1620 : pub struct ListAuxFilesRequest {
1621 : pub lsn: Lsn,
1622 : }
1623 :
1624 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
1625 : pub struct DownloadRemoteLayersTaskInfo {
1626 : pub task_id: String,
1627 : pub state: DownloadRemoteLayersTaskState,
1628 : pub total_layer_count: u64, // stable once `completed`
1629 : pub successful_download_count: u64, // stable once `completed`
1630 : pub failed_download_count: u64, // stable once `completed`
1631 : }
1632 :
1633 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
1634 : pub enum DownloadRemoteLayersTaskState {
1635 : Running,
1636 : Completed,
1637 : ShutDown,
1638 : }
1639 :
1640 0 : #[derive(Debug, Serialize, Deserialize)]
1641 : pub struct TimelineGcRequest {
1642 : pub gc_horizon: Option<u64>,
1643 : }
1644 :
1645 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1646 : pub struct WalRedoManagerProcessStatus {
1647 : pub pid: u32,
1648 : }
1649 :
1650 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1651 : pub struct WalRedoManagerStatus {
1652 : pub last_redo_at: Option<chrono::DateTime<chrono::Utc>>,
1653 : pub process: Option<WalRedoManagerProcessStatus>,
1654 : }
1655 :
1656 : /// The progress of a secondary tenant.
1657 : ///
1658 : /// It is mostly useful when doing a long running download: e.g. initiating
1659 : /// a download job, timing out while waiting for it to run, and then inspecting this status to understand
1660 : /// what's happening.
1661 0 : #[derive(Default, Debug, Serialize, Deserialize, Clone)]
1662 : pub struct SecondaryProgress {
1663 : /// The remote storage LastModified time of the heatmap object we last downloaded.
1664 : pub heatmap_mtime: Option<serde_system_time::SystemTime>,
1665 :
1666 : /// The number of layers currently on-disk
1667 : pub layers_downloaded: usize,
1668 : /// The number of layers in the most recently seen heatmap
1669 : pub layers_total: usize,
1670 :
1671 : /// The number of layer bytes currently on-disk
1672 : pub bytes_downloaded: u64,
1673 : /// The number of layer bytes in the most recently seen heatmap
1674 : pub bytes_total: u64,
1675 : }
1676 :
1677 0 : #[derive(Serialize, Deserialize, Debug)]
1678 : pub struct TenantScanRemoteStorageShard {
1679 : pub tenant_shard_id: TenantShardId,
1680 : pub generation: Option<u32>,
1681 : pub stripe_size: Option<ShardStripeSize>,
1682 : }
1683 :
1684 0 : #[derive(Serialize, Deserialize, Debug, Default)]
1685 : pub struct TenantScanRemoteStorageResponse {
1686 : pub shards: Vec<TenantScanRemoteStorageShard>,
1687 : }
1688 :
1689 0 : #[derive(Serialize, Deserialize, Debug, Clone)]
1690 : #[serde(rename_all = "snake_case")]
1691 : pub enum TenantSorting {
1692 : /// Total size of layers on local disk for all timelines in a shard.
1693 : ResidentSize,
1694 : /// The logical size of the largest timeline within a _tenant_ (not shard). Only tracked on
1695 : /// shard 0, contains the sum across all shards.
1696 : MaxLogicalSize,
1697 : /// The logical size of the largest timeline within a _tenant_ (not shard), divided by number of
1698 : /// shards. Only tracked on shard 0, and estimates the per-shard logical size.
1699 : MaxLogicalSizePerShard,
1700 : }
1701 :
1702 : impl Default for TenantSorting {
1703 0 : fn default() -> Self {
1704 0 : Self::ResidentSize
1705 0 : }
1706 : }
1707 :
1708 0 : #[derive(Serialize, Deserialize, Debug, Clone)]
1709 : pub struct TopTenantShardsRequest {
1710 : // How would you like to sort the tenants?
1711 : pub order_by: TenantSorting,
1712 :
1713 : // How many results?
1714 : pub limit: usize,
1715 :
1716 : // Omit tenants with more than this many shards (e.g. if this is the max number of shards
1717 : // that the caller would ever split to)
1718 : pub where_shards_lt: Option<ShardCount>,
1719 :
1720 : // Omit tenants where the ordering metric is less than this (this is an optimization to
1721 : // let us quickly exclude numerous tiny shards)
1722 : pub where_gt: Option<u64>,
1723 : }
1724 :
1725 0 : #[derive(Serialize, Deserialize, Debug, PartialEq, Eq)]
1726 : pub struct TopTenantShardItem {
1727 : pub id: TenantShardId,
1728 :
1729 : /// Total size of layers on local disk for all timelines in this shard.
1730 : pub resident_size: u64,
1731 :
1732 : /// Total size of layers in remote storage for all timelines in this shard.
1733 : pub physical_size: u64,
1734 :
1735 : /// The largest logical size of a timeline within this _tenant_ (not shard). This is only
1736 : /// tracked on shard 0, and contains the sum of the logical size across all shards.
1737 : pub max_logical_size: u64,
1738 :
1739 : /// The largest logical size of a timeline within this _tenant_ (not shard) divided by number of
1740 : /// shards. This is only tracked on shard 0, and is only an estimate as we divide it evenly by
1741 : /// shard count, rounded up.
1742 : pub max_logical_size_per_shard: u64,
1743 : }
1744 :
1745 0 : #[derive(Serialize, Deserialize, Debug, Default)]
1746 : pub struct TopTenantShardsResponse {
1747 : pub shards: Vec<TopTenantShardItem>,
1748 : }
1749 :
1750 : pub mod virtual_file {
1751 : #[derive(
1752 : Copy,
1753 : Clone,
1754 : PartialEq,
1755 : Eq,
1756 : Hash,
1757 0 : strum_macros::EnumString,
1758 : strum_macros::Display,
1759 0 : serde_with::DeserializeFromStr,
1760 : serde_with::SerializeDisplay,
1761 : Debug,
1762 : )]
1763 : #[strum(serialize_all = "kebab-case")]
1764 : pub enum IoEngineKind {
1765 : StdFs,
1766 : #[cfg(target_os = "linux")]
1767 : TokioEpollUring,
1768 : }
1769 :
1770 : /// Direct IO modes for a pageserver.
1771 : #[derive(
1772 : Copy,
1773 : Clone,
1774 : PartialEq,
1775 : Eq,
1776 : Hash,
1777 0 : strum_macros::EnumString,
1778 : strum_macros::Display,
1779 0 : serde_with::DeserializeFromStr,
1780 : serde_with::SerializeDisplay,
1781 : Debug,
1782 : )]
1783 : #[strum(serialize_all = "kebab-case")]
1784 : #[repr(u8)]
1785 : pub enum IoMode {
1786 : /// Uses buffered IO.
1787 : Buffered,
1788 : /// Uses direct IO, error out if the operation fails.
1789 : #[cfg(target_os = "linux")]
1790 : Direct,
1791 : }
1792 :
1793 : impl IoMode {
1794 496 : pub const fn preferred() -> Self {
1795 496 : Self::Buffered
1796 496 : }
1797 : }
1798 :
1799 : impl TryFrom<u8> for IoMode {
1800 : type Error = u8;
1801 :
1802 5120 : fn try_from(value: u8) -> Result<Self, Self::Error> {
1803 5120 : Ok(match value {
1804 5120 : v if v == (IoMode::Buffered as u8) => IoMode::Buffered,
1805 : #[cfg(target_os = "linux")]
1806 0 : v if v == (IoMode::Direct as u8) => IoMode::Direct,
1807 0 : x => return Err(x),
1808 : })
1809 5120 : }
1810 : }
1811 : }
1812 :
1813 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1814 : pub struct ScanDisposableKeysResponse {
1815 : pub disposable_count: usize,
1816 : pub not_disposable_count: usize,
1817 : }
1818 :
1819 : // Wrapped in libpq CopyData
1820 : #[derive(PartialEq, Eq, Debug)]
1821 : pub enum PagestreamFeMessage {
1822 : Exists(PagestreamExistsRequest),
1823 : Nblocks(PagestreamNblocksRequest),
1824 : GetPage(PagestreamGetPageRequest),
1825 : DbSize(PagestreamDbSizeRequest),
1826 : GetSlruSegment(PagestreamGetSlruSegmentRequest),
1827 : #[cfg(feature = "testing")]
1828 : Test(PagestreamTestRequest),
1829 : }
1830 :
1831 : // Wrapped in libpq CopyData
1832 : #[derive(strum_macros::EnumProperty)]
1833 : pub enum PagestreamBeMessage {
1834 : Exists(PagestreamExistsResponse),
1835 : Nblocks(PagestreamNblocksResponse),
1836 : GetPage(PagestreamGetPageResponse),
1837 : Error(PagestreamErrorResponse),
1838 : DbSize(PagestreamDbSizeResponse),
1839 : GetSlruSegment(PagestreamGetSlruSegmentResponse),
1840 : #[cfg(feature = "testing")]
1841 : Test(PagestreamTestResponse),
1842 : }
1843 :
1844 : // Keep in sync with `pagestore_client.h`
1845 : #[repr(u8)]
1846 : enum PagestreamFeMessageTag {
1847 : Exists = 0,
1848 : Nblocks = 1,
1849 : GetPage = 2,
1850 : DbSize = 3,
1851 : GetSlruSegment = 4,
1852 : /* future tags above this line */
1853 : /// For testing purposes, not available in production.
1854 : #[cfg(feature = "testing")]
1855 : Test = 99,
1856 : }
1857 :
1858 : // Keep in sync with `pagestore_client.h`
1859 : #[repr(u8)]
1860 : enum PagestreamBeMessageTag {
1861 : Exists = 100,
1862 : Nblocks = 101,
1863 : GetPage = 102,
1864 : Error = 103,
1865 : DbSize = 104,
1866 : GetSlruSegment = 105,
1867 : /* future tags above this line */
1868 : /// For testing purposes, not available in production.
1869 : #[cfg(feature = "testing")]
1870 : Test = 199,
1871 : }
1872 :
1873 : impl TryFrom<u8> for PagestreamFeMessageTag {
1874 : type Error = u8;
1875 4 : fn try_from(value: u8) -> Result<Self, u8> {
1876 4 : match value {
1877 1 : 0 => Ok(PagestreamFeMessageTag::Exists),
1878 1 : 1 => Ok(PagestreamFeMessageTag::Nblocks),
1879 1 : 2 => Ok(PagestreamFeMessageTag::GetPage),
1880 1 : 3 => Ok(PagestreamFeMessageTag::DbSize),
1881 0 : 4 => Ok(PagestreamFeMessageTag::GetSlruSegment),
1882 : #[cfg(feature = "testing")]
1883 0 : 99 => Ok(PagestreamFeMessageTag::Test),
1884 0 : _ => Err(value),
1885 : }
1886 4 : }
1887 : }
1888 :
1889 : impl TryFrom<u8> for PagestreamBeMessageTag {
1890 : type Error = u8;
1891 0 : fn try_from(value: u8) -> Result<Self, u8> {
1892 0 : match value {
1893 0 : 100 => Ok(PagestreamBeMessageTag::Exists),
1894 0 : 101 => Ok(PagestreamBeMessageTag::Nblocks),
1895 0 : 102 => Ok(PagestreamBeMessageTag::GetPage),
1896 0 : 103 => Ok(PagestreamBeMessageTag::Error),
1897 0 : 104 => Ok(PagestreamBeMessageTag::DbSize),
1898 0 : 105 => Ok(PagestreamBeMessageTag::GetSlruSegment),
1899 : #[cfg(feature = "testing")]
1900 0 : 199 => Ok(PagestreamBeMessageTag::Test),
1901 0 : _ => Err(value),
1902 : }
1903 0 : }
1904 : }
1905 :
1906 : // A GetPage request contains two LSN values:
1907 : //
1908 : // request_lsn: Get the page version at this point in time. Lsn::Max is a special value that means
1909 : // "get the latest version present". It's used by the primary server, which knows that no one else
1910 : // is writing WAL. 'not_modified_since' must be set to a proper value even if request_lsn is
1911 : // Lsn::Max. Standby servers use the current replay LSN as the request LSN.
1912 : //
1913 : // not_modified_since: Hint to the pageserver that the client knows that the page has not been
1914 : // modified between 'not_modified_since' and the request LSN. It's always correct to set
1915 : // 'not_modified_since equal' to 'request_lsn' (unless Lsn::Max is used as the 'request_lsn'), but
1916 : // passing an earlier LSN can speed up the request, by allowing the pageserver to process the
1917 : // request without waiting for 'request_lsn' to arrive.
1918 : //
1919 : // The now-defunct V1 interface contained only one LSN, and a boolean 'latest' flag. The V1 interface was
1920 : // sufficient for the primary; the 'lsn' was equivalent to the 'not_modified_since' value, and
1921 : // 'latest' was set to true. The V2 interface was added because there was no correct way for a
1922 : // standby to request a page at a particular non-latest LSN, and also include the
1923 : // 'not_modified_since' hint. That led to an awkward choice of either using an old LSN in the
1924 : // request, if the standby knows that the page hasn't been modified since, and risk getting an error
1925 : // if that LSN has fallen behind the GC horizon, or requesting the current replay LSN, which could
1926 : // require the pageserver unnecessarily to wait for the WAL to arrive up to that point. The new V2
1927 : // interface allows sending both LSNs, and let the pageserver do the right thing. There was no
1928 : // difference in the responses between V1 and V2.
1929 : //
1930 : // V3 version of protocol adds request ID to all requests. This request ID is also included in response
1931 : // as well as other fields from requests, which allows to verify that we receive response for our request.
1932 : // We copy fields from request to response to make checking more reliable: request ID is formed from process ID
1933 : // and local counter, so in principle there can be duplicated requests IDs if process PID is reused.
1934 : //
1935 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
1936 : pub enum PagestreamProtocolVersion {
1937 : V2,
1938 : V3,
1939 : }
1940 :
1941 : pub type RequestId = u64;
1942 :
1943 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
1944 : pub struct PagestreamRequest {
1945 : pub reqid: RequestId,
1946 : pub request_lsn: Lsn,
1947 : pub not_modified_since: Lsn,
1948 : }
1949 :
1950 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
1951 : pub struct PagestreamExistsRequest {
1952 : pub hdr: PagestreamRequest,
1953 : pub rel: RelTag,
1954 : }
1955 :
1956 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
1957 : pub struct PagestreamNblocksRequest {
1958 : pub hdr: PagestreamRequest,
1959 : pub rel: RelTag,
1960 : }
1961 :
1962 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
1963 : pub struct PagestreamGetPageRequest {
1964 : pub hdr: PagestreamRequest,
1965 : pub rel: RelTag,
1966 : pub blkno: u32,
1967 : }
1968 :
1969 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
1970 : pub struct PagestreamDbSizeRequest {
1971 : pub hdr: PagestreamRequest,
1972 : pub dbnode: u32,
1973 : }
1974 :
1975 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
1976 : pub struct PagestreamGetSlruSegmentRequest {
1977 : pub hdr: PagestreamRequest,
1978 : pub kind: u8,
1979 : pub segno: u32,
1980 : }
1981 :
1982 : #[derive(Debug)]
1983 : pub struct PagestreamExistsResponse {
1984 : pub req: PagestreamExistsRequest,
1985 : pub exists: bool,
1986 : }
1987 :
1988 : #[derive(Debug)]
1989 : pub struct PagestreamNblocksResponse {
1990 : pub req: PagestreamNblocksRequest,
1991 : pub n_blocks: u32,
1992 : }
1993 :
1994 : #[derive(Debug)]
1995 : pub struct PagestreamGetPageResponse {
1996 : pub req: PagestreamGetPageRequest,
1997 : pub page: Bytes,
1998 : }
1999 :
2000 : #[derive(Debug)]
2001 : pub struct PagestreamGetSlruSegmentResponse {
2002 : pub req: PagestreamGetSlruSegmentRequest,
2003 : pub segment: Bytes,
2004 : }
2005 :
2006 : #[derive(Debug)]
2007 : pub struct PagestreamErrorResponse {
2008 : pub req: PagestreamRequest,
2009 : pub message: String,
2010 : }
2011 :
2012 : #[derive(Debug)]
2013 : pub struct PagestreamDbSizeResponse {
2014 : pub req: PagestreamDbSizeRequest,
2015 : pub db_size: i64,
2016 : }
2017 :
2018 : #[cfg(feature = "testing")]
2019 : #[derive(Debug, PartialEq, Eq, Clone)]
2020 : pub struct PagestreamTestRequest {
2021 : pub hdr: PagestreamRequest,
2022 : pub batch_key: u64,
2023 : pub message: String,
2024 : }
2025 :
2026 : #[cfg(feature = "testing")]
2027 : #[derive(Debug)]
2028 : pub struct PagestreamTestResponse {
2029 : pub req: PagestreamTestRequest,
2030 : }
2031 :
2032 : // This is a cut-down version of TenantHistorySize from the pageserver crate, omitting fields
2033 : // that require pageserver-internal types. It is sufficient to get the total size.
2034 0 : #[derive(Serialize, Deserialize, Debug)]
2035 : pub struct TenantHistorySize {
2036 : pub id: TenantId,
2037 : /// Size is a mixture of WAL and logical size, so the unit is bytes.
2038 : ///
2039 : /// Will be none if `?inputs_only=true` was given.
2040 : pub size: Option<u64>,
2041 : }
2042 :
2043 : impl PagestreamFeMessage {
2044 : /// Serialize a compute -> pageserver message. This is currently only used in testing
2045 : /// tools. Always uses protocol version 3.
2046 4 : pub fn serialize(&self) -> Bytes {
2047 4 : let mut bytes = BytesMut::new();
2048 4 :
2049 4 : match self {
2050 1 : Self::Exists(req) => {
2051 1 : bytes.put_u8(PagestreamFeMessageTag::Exists as u8);
2052 1 : bytes.put_u64(req.hdr.reqid);
2053 1 : bytes.put_u64(req.hdr.request_lsn.0);
2054 1 : bytes.put_u64(req.hdr.not_modified_since.0);
2055 1 : bytes.put_u32(req.rel.spcnode);
2056 1 : bytes.put_u32(req.rel.dbnode);
2057 1 : bytes.put_u32(req.rel.relnode);
2058 1 : bytes.put_u8(req.rel.forknum);
2059 1 : }
2060 :
2061 1 : Self::Nblocks(req) => {
2062 1 : bytes.put_u8(PagestreamFeMessageTag::Nblocks as u8);
2063 1 : bytes.put_u64(req.hdr.reqid);
2064 1 : bytes.put_u64(req.hdr.request_lsn.0);
2065 1 : bytes.put_u64(req.hdr.not_modified_since.0);
2066 1 : bytes.put_u32(req.rel.spcnode);
2067 1 : bytes.put_u32(req.rel.dbnode);
2068 1 : bytes.put_u32(req.rel.relnode);
2069 1 : bytes.put_u8(req.rel.forknum);
2070 1 : }
2071 :
2072 1 : Self::GetPage(req) => {
2073 1 : bytes.put_u8(PagestreamFeMessageTag::GetPage as u8);
2074 1 : bytes.put_u64(req.hdr.reqid);
2075 1 : bytes.put_u64(req.hdr.request_lsn.0);
2076 1 : bytes.put_u64(req.hdr.not_modified_since.0);
2077 1 : bytes.put_u32(req.rel.spcnode);
2078 1 : bytes.put_u32(req.rel.dbnode);
2079 1 : bytes.put_u32(req.rel.relnode);
2080 1 : bytes.put_u8(req.rel.forknum);
2081 1 : bytes.put_u32(req.blkno);
2082 1 : }
2083 :
2084 1 : Self::DbSize(req) => {
2085 1 : bytes.put_u8(PagestreamFeMessageTag::DbSize as u8);
2086 1 : bytes.put_u64(req.hdr.reqid);
2087 1 : bytes.put_u64(req.hdr.request_lsn.0);
2088 1 : bytes.put_u64(req.hdr.not_modified_since.0);
2089 1 : bytes.put_u32(req.dbnode);
2090 1 : }
2091 :
2092 0 : Self::GetSlruSegment(req) => {
2093 0 : bytes.put_u8(PagestreamFeMessageTag::GetSlruSegment as u8);
2094 0 : bytes.put_u64(req.hdr.reqid);
2095 0 : bytes.put_u64(req.hdr.request_lsn.0);
2096 0 : bytes.put_u64(req.hdr.not_modified_since.0);
2097 0 : bytes.put_u8(req.kind);
2098 0 : bytes.put_u32(req.segno);
2099 0 : }
2100 : #[cfg(feature = "testing")]
2101 0 : Self::Test(req) => {
2102 0 : bytes.put_u8(PagestreamFeMessageTag::Test as u8);
2103 0 : bytes.put_u64(req.hdr.reqid);
2104 0 : bytes.put_u64(req.hdr.request_lsn.0);
2105 0 : bytes.put_u64(req.hdr.not_modified_since.0);
2106 0 : bytes.put_u64(req.batch_key);
2107 0 : let message = req.message.as_bytes();
2108 0 : bytes.put_u64(message.len() as u64);
2109 0 : bytes.put_slice(message);
2110 0 : }
2111 : }
2112 :
2113 4 : bytes.into()
2114 4 : }
2115 :
2116 4 : pub fn parse<R: std::io::Read>(
2117 4 : body: &mut R,
2118 4 : protocol_version: PagestreamProtocolVersion,
2119 4 : ) -> anyhow::Result<PagestreamFeMessage> {
2120 : // these correspond to the NeonMessageTag enum in pagestore_client.h
2121 : //
2122 : // TODO: consider using protobuf or serde bincode for less error prone
2123 : // serialization.
2124 4 : let msg_tag = body.read_u8()?;
2125 4 : let (reqid, request_lsn, not_modified_since) = match protocol_version {
2126 : PagestreamProtocolVersion::V2 => (
2127 : 0,
2128 0 : Lsn::from(body.read_u64::<BigEndian>()?),
2129 0 : Lsn::from(body.read_u64::<BigEndian>()?),
2130 : ),
2131 : PagestreamProtocolVersion::V3 => (
2132 4 : body.read_u64::<BigEndian>()?,
2133 4 : Lsn::from(body.read_u64::<BigEndian>()?),
2134 4 : Lsn::from(body.read_u64::<BigEndian>()?),
2135 : ),
2136 : };
2137 :
2138 4 : match PagestreamFeMessageTag::try_from(msg_tag)
2139 4 : .map_err(|tag: u8| anyhow::anyhow!("invalid tag {tag}"))?
2140 : {
2141 : PagestreamFeMessageTag::Exists => {
2142 : Ok(PagestreamFeMessage::Exists(PagestreamExistsRequest {
2143 1 : hdr: PagestreamRequest {
2144 1 : reqid,
2145 1 : request_lsn,
2146 1 : not_modified_since,
2147 1 : },
2148 1 : rel: RelTag {
2149 1 : spcnode: body.read_u32::<BigEndian>()?,
2150 1 : dbnode: body.read_u32::<BigEndian>()?,
2151 1 : relnode: body.read_u32::<BigEndian>()?,
2152 1 : forknum: body.read_u8()?,
2153 : },
2154 : }))
2155 : }
2156 : PagestreamFeMessageTag::Nblocks => {
2157 : Ok(PagestreamFeMessage::Nblocks(PagestreamNblocksRequest {
2158 1 : hdr: PagestreamRequest {
2159 1 : reqid,
2160 1 : request_lsn,
2161 1 : not_modified_since,
2162 1 : },
2163 1 : rel: RelTag {
2164 1 : spcnode: body.read_u32::<BigEndian>()?,
2165 1 : dbnode: body.read_u32::<BigEndian>()?,
2166 1 : relnode: body.read_u32::<BigEndian>()?,
2167 1 : forknum: body.read_u8()?,
2168 : },
2169 : }))
2170 : }
2171 : PagestreamFeMessageTag::GetPage => {
2172 : Ok(PagestreamFeMessage::GetPage(PagestreamGetPageRequest {
2173 1 : hdr: PagestreamRequest {
2174 1 : reqid,
2175 1 : request_lsn,
2176 1 : not_modified_since,
2177 1 : },
2178 1 : rel: RelTag {
2179 1 : spcnode: body.read_u32::<BigEndian>()?,
2180 1 : dbnode: body.read_u32::<BigEndian>()?,
2181 1 : relnode: body.read_u32::<BigEndian>()?,
2182 1 : forknum: body.read_u8()?,
2183 : },
2184 1 : blkno: body.read_u32::<BigEndian>()?,
2185 : }))
2186 : }
2187 : PagestreamFeMessageTag::DbSize => {
2188 : Ok(PagestreamFeMessage::DbSize(PagestreamDbSizeRequest {
2189 1 : hdr: PagestreamRequest {
2190 1 : reqid,
2191 1 : request_lsn,
2192 1 : not_modified_since,
2193 1 : },
2194 1 : dbnode: body.read_u32::<BigEndian>()?,
2195 : }))
2196 : }
2197 : PagestreamFeMessageTag::GetSlruSegment => Ok(PagestreamFeMessage::GetSlruSegment(
2198 : PagestreamGetSlruSegmentRequest {
2199 0 : hdr: PagestreamRequest {
2200 0 : reqid,
2201 0 : request_lsn,
2202 0 : not_modified_since,
2203 0 : },
2204 0 : kind: body.read_u8()?,
2205 0 : segno: body.read_u32::<BigEndian>()?,
2206 : },
2207 : )),
2208 : #[cfg(feature = "testing")]
2209 : PagestreamFeMessageTag::Test => Ok(PagestreamFeMessage::Test(PagestreamTestRequest {
2210 0 : hdr: PagestreamRequest {
2211 0 : reqid,
2212 0 : request_lsn,
2213 0 : not_modified_since,
2214 0 : },
2215 0 : batch_key: body.read_u64::<BigEndian>()?,
2216 : message: {
2217 0 : let len = body.read_u64::<BigEndian>()?;
2218 0 : let mut buf = vec![0; len as usize];
2219 0 : body.read_exact(&mut buf)?;
2220 0 : String::from_utf8(buf)?
2221 : },
2222 : })),
2223 : }
2224 4 : }
2225 : }
2226 :
2227 : impl PagestreamBeMessage {
2228 0 : pub fn serialize(&self, protocol_version: PagestreamProtocolVersion) -> Bytes {
2229 0 : let mut bytes = BytesMut::new();
2230 :
2231 : use PagestreamBeMessageTag as Tag;
2232 0 : match protocol_version {
2233 : PagestreamProtocolVersion::V2 => {
2234 0 : match self {
2235 0 : Self::Exists(resp) => {
2236 0 : bytes.put_u8(Tag::Exists as u8);
2237 0 : bytes.put_u8(resp.exists as u8);
2238 0 : }
2239 :
2240 0 : Self::Nblocks(resp) => {
2241 0 : bytes.put_u8(Tag::Nblocks as u8);
2242 0 : bytes.put_u32(resp.n_blocks);
2243 0 : }
2244 :
2245 0 : Self::GetPage(resp) => {
2246 0 : bytes.put_u8(Tag::GetPage as u8);
2247 0 : bytes.put(&resp.page[..])
2248 : }
2249 :
2250 0 : Self::Error(resp) => {
2251 0 : bytes.put_u8(Tag::Error as u8);
2252 0 : bytes.put(resp.message.as_bytes());
2253 0 : bytes.put_u8(0); // null terminator
2254 0 : }
2255 0 : Self::DbSize(resp) => {
2256 0 : bytes.put_u8(Tag::DbSize as u8);
2257 0 : bytes.put_i64(resp.db_size);
2258 0 : }
2259 :
2260 0 : Self::GetSlruSegment(resp) => {
2261 0 : bytes.put_u8(Tag::GetSlruSegment as u8);
2262 0 : bytes.put_u32((resp.segment.len() / BLCKSZ as usize) as u32);
2263 0 : bytes.put(&resp.segment[..]);
2264 0 : }
2265 :
2266 : #[cfg(feature = "testing")]
2267 0 : Self::Test(resp) => {
2268 0 : bytes.put_u8(Tag::Test as u8);
2269 0 : bytes.put_u64(resp.req.batch_key);
2270 0 : let message = resp.req.message.as_bytes();
2271 0 : bytes.put_u64(message.len() as u64);
2272 0 : bytes.put_slice(message);
2273 0 : }
2274 : }
2275 : }
2276 : PagestreamProtocolVersion::V3 => {
2277 0 : match self {
2278 0 : Self::Exists(resp) => {
2279 0 : bytes.put_u8(Tag::Exists as u8);
2280 0 : bytes.put_u64(resp.req.hdr.reqid);
2281 0 : bytes.put_u64(resp.req.hdr.request_lsn.0);
2282 0 : bytes.put_u64(resp.req.hdr.not_modified_since.0);
2283 0 : bytes.put_u32(resp.req.rel.spcnode);
2284 0 : bytes.put_u32(resp.req.rel.dbnode);
2285 0 : bytes.put_u32(resp.req.rel.relnode);
2286 0 : bytes.put_u8(resp.req.rel.forknum);
2287 0 : bytes.put_u8(resp.exists as u8);
2288 0 : }
2289 :
2290 0 : Self::Nblocks(resp) => {
2291 0 : bytes.put_u8(Tag::Nblocks as u8);
2292 0 : bytes.put_u64(resp.req.hdr.reqid);
2293 0 : bytes.put_u64(resp.req.hdr.request_lsn.0);
2294 0 : bytes.put_u64(resp.req.hdr.not_modified_since.0);
2295 0 : bytes.put_u32(resp.req.rel.spcnode);
2296 0 : bytes.put_u32(resp.req.rel.dbnode);
2297 0 : bytes.put_u32(resp.req.rel.relnode);
2298 0 : bytes.put_u8(resp.req.rel.forknum);
2299 0 : bytes.put_u32(resp.n_blocks);
2300 0 : }
2301 :
2302 0 : Self::GetPage(resp) => {
2303 0 : bytes.put_u8(Tag::GetPage as u8);
2304 0 : bytes.put_u64(resp.req.hdr.reqid);
2305 0 : bytes.put_u64(resp.req.hdr.request_lsn.0);
2306 0 : bytes.put_u64(resp.req.hdr.not_modified_since.0);
2307 0 : bytes.put_u32(resp.req.rel.spcnode);
2308 0 : bytes.put_u32(resp.req.rel.dbnode);
2309 0 : bytes.put_u32(resp.req.rel.relnode);
2310 0 : bytes.put_u8(resp.req.rel.forknum);
2311 0 : bytes.put_u32(resp.req.blkno);
2312 0 : bytes.put(&resp.page[..])
2313 : }
2314 :
2315 0 : Self::Error(resp) => {
2316 0 : bytes.put_u8(Tag::Error as u8);
2317 0 : bytes.put_u64(resp.req.reqid);
2318 0 : bytes.put_u64(resp.req.request_lsn.0);
2319 0 : bytes.put_u64(resp.req.not_modified_since.0);
2320 0 : bytes.put(resp.message.as_bytes());
2321 0 : bytes.put_u8(0); // null terminator
2322 0 : }
2323 0 : Self::DbSize(resp) => {
2324 0 : bytes.put_u8(Tag::DbSize as u8);
2325 0 : bytes.put_u64(resp.req.hdr.reqid);
2326 0 : bytes.put_u64(resp.req.hdr.request_lsn.0);
2327 0 : bytes.put_u64(resp.req.hdr.not_modified_since.0);
2328 0 : bytes.put_u32(resp.req.dbnode);
2329 0 : bytes.put_i64(resp.db_size);
2330 0 : }
2331 :
2332 0 : Self::GetSlruSegment(resp) => {
2333 0 : bytes.put_u8(Tag::GetSlruSegment as u8);
2334 0 : bytes.put_u64(resp.req.hdr.reqid);
2335 0 : bytes.put_u64(resp.req.hdr.request_lsn.0);
2336 0 : bytes.put_u64(resp.req.hdr.not_modified_since.0);
2337 0 : bytes.put_u8(resp.req.kind);
2338 0 : bytes.put_u32(resp.req.segno);
2339 0 : bytes.put_u32((resp.segment.len() / BLCKSZ as usize) as u32);
2340 0 : bytes.put(&resp.segment[..]);
2341 0 : }
2342 :
2343 : #[cfg(feature = "testing")]
2344 0 : Self::Test(resp) => {
2345 0 : bytes.put_u8(Tag::Test as u8);
2346 0 : bytes.put_u64(resp.req.hdr.reqid);
2347 0 : bytes.put_u64(resp.req.hdr.request_lsn.0);
2348 0 : bytes.put_u64(resp.req.hdr.not_modified_since.0);
2349 0 : bytes.put_u64(resp.req.batch_key);
2350 0 : let message = resp.req.message.as_bytes();
2351 0 : bytes.put_u64(message.len() as u64);
2352 0 : bytes.put_slice(message);
2353 0 : }
2354 : }
2355 : }
2356 : }
2357 0 : bytes.into()
2358 0 : }
2359 :
2360 0 : pub fn deserialize(buf: Bytes) -> anyhow::Result<Self> {
2361 0 : let mut buf = buf.reader();
2362 0 : let msg_tag = buf.read_u8()?;
2363 :
2364 : use PagestreamBeMessageTag as Tag;
2365 0 : let ok =
2366 0 : match Tag::try_from(msg_tag).map_err(|tag: u8| anyhow::anyhow!("invalid tag {tag}"))? {
2367 : Tag::Exists => {
2368 0 : let reqid = buf.read_u64::<BigEndian>()?;
2369 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2370 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2371 0 : let rel = RelTag {
2372 0 : spcnode: buf.read_u32::<BigEndian>()?,
2373 0 : dbnode: buf.read_u32::<BigEndian>()?,
2374 0 : relnode: buf.read_u32::<BigEndian>()?,
2375 0 : forknum: buf.read_u8()?,
2376 : };
2377 0 : let exists = buf.read_u8()? != 0;
2378 0 : Self::Exists(PagestreamExistsResponse {
2379 0 : req: PagestreamExistsRequest {
2380 0 : hdr: PagestreamRequest {
2381 0 : reqid,
2382 0 : request_lsn,
2383 0 : not_modified_since,
2384 0 : },
2385 0 : rel,
2386 0 : },
2387 0 : exists,
2388 0 : })
2389 : }
2390 : Tag::Nblocks => {
2391 0 : let reqid = buf.read_u64::<BigEndian>()?;
2392 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2393 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2394 0 : let rel = RelTag {
2395 0 : spcnode: buf.read_u32::<BigEndian>()?,
2396 0 : dbnode: buf.read_u32::<BigEndian>()?,
2397 0 : relnode: buf.read_u32::<BigEndian>()?,
2398 0 : forknum: buf.read_u8()?,
2399 : };
2400 0 : let n_blocks = buf.read_u32::<BigEndian>()?;
2401 0 : Self::Nblocks(PagestreamNblocksResponse {
2402 0 : req: PagestreamNblocksRequest {
2403 0 : hdr: PagestreamRequest {
2404 0 : reqid,
2405 0 : request_lsn,
2406 0 : not_modified_since,
2407 0 : },
2408 0 : rel,
2409 0 : },
2410 0 : n_blocks,
2411 0 : })
2412 : }
2413 : Tag::GetPage => {
2414 0 : let reqid = buf.read_u64::<BigEndian>()?;
2415 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2416 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2417 0 : let rel = RelTag {
2418 0 : spcnode: buf.read_u32::<BigEndian>()?,
2419 0 : dbnode: buf.read_u32::<BigEndian>()?,
2420 0 : relnode: buf.read_u32::<BigEndian>()?,
2421 0 : forknum: buf.read_u8()?,
2422 : };
2423 0 : let blkno = buf.read_u32::<BigEndian>()?;
2424 0 : let mut page = vec![0; 8192]; // TODO: use MaybeUninit
2425 0 : buf.read_exact(&mut page)?;
2426 0 : Self::GetPage(PagestreamGetPageResponse {
2427 0 : req: PagestreamGetPageRequest {
2428 0 : hdr: PagestreamRequest {
2429 0 : reqid,
2430 0 : request_lsn,
2431 0 : not_modified_since,
2432 0 : },
2433 0 : rel,
2434 0 : blkno,
2435 0 : },
2436 0 : page: page.into(),
2437 0 : })
2438 : }
2439 : Tag::Error => {
2440 0 : let reqid = buf.read_u64::<BigEndian>()?;
2441 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2442 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2443 0 : let mut msg = Vec::new();
2444 0 : buf.read_until(0, &mut msg)?;
2445 0 : let cstring = std::ffi::CString::from_vec_with_nul(msg)?;
2446 0 : let rust_str = cstring.to_str()?;
2447 0 : Self::Error(PagestreamErrorResponse {
2448 0 : req: PagestreamRequest {
2449 0 : reqid,
2450 0 : request_lsn,
2451 0 : not_modified_since,
2452 0 : },
2453 0 : message: rust_str.to_owned(),
2454 0 : })
2455 : }
2456 : Tag::DbSize => {
2457 0 : let reqid = buf.read_u64::<BigEndian>()?;
2458 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2459 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2460 0 : let dbnode = buf.read_u32::<BigEndian>()?;
2461 0 : let db_size = buf.read_i64::<BigEndian>()?;
2462 0 : Self::DbSize(PagestreamDbSizeResponse {
2463 0 : req: PagestreamDbSizeRequest {
2464 0 : hdr: PagestreamRequest {
2465 0 : reqid,
2466 0 : request_lsn,
2467 0 : not_modified_since,
2468 0 : },
2469 0 : dbnode,
2470 0 : },
2471 0 : db_size,
2472 0 : })
2473 : }
2474 : Tag::GetSlruSegment => {
2475 0 : let reqid = buf.read_u64::<BigEndian>()?;
2476 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2477 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2478 0 : let kind = buf.read_u8()?;
2479 0 : let segno = buf.read_u32::<BigEndian>()?;
2480 0 : let n_blocks = buf.read_u32::<BigEndian>()?;
2481 0 : let mut segment = vec![0; n_blocks as usize * BLCKSZ as usize];
2482 0 : buf.read_exact(&mut segment)?;
2483 0 : Self::GetSlruSegment(PagestreamGetSlruSegmentResponse {
2484 0 : req: PagestreamGetSlruSegmentRequest {
2485 0 : hdr: PagestreamRequest {
2486 0 : reqid,
2487 0 : request_lsn,
2488 0 : not_modified_since,
2489 0 : },
2490 0 : kind,
2491 0 : segno,
2492 0 : },
2493 0 : segment: segment.into(),
2494 0 : })
2495 : }
2496 : #[cfg(feature = "testing")]
2497 : Tag::Test => {
2498 0 : let reqid = buf.read_u64::<BigEndian>()?;
2499 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2500 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2501 0 : let batch_key = buf.read_u64::<BigEndian>()?;
2502 0 : let len = buf.read_u64::<BigEndian>()?;
2503 0 : let mut msg = vec![0; len as usize];
2504 0 : buf.read_exact(&mut msg)?;
2505 0 : let message = String::from_utf8(msg)?;
2506 0 : Self::Test(PagestreamTestResponse {
2507 0 : req: PagestreamTestRequest {
2508 0 : hdr: PagestreamRequest {
2509 0 : reqid,
2510 0 : request_lsn,
2511 0 : not_modified_since,
2512 0 : },
2513 0 : batch_key,
2514 0 : message,
2515 0 : },
2516 0 : })
2517 : }
2518 : };
2519 0 : let remaining = buf.into_inner();
2520 0 : if !remaining.is_empty() {
2521 0 : anyhow::bail!(
2522 0 : "remaining bytes in msg with tag={msg_tag}: {}",
2523 0 : remaining.len()
2524 0 : );
2525 0 : }
2526 0 : Ok(ok)
2527 0 : }
2528 :
2529 0 : pub fn kind(&self) -> &'static str {
2530 0 : match self {
2531 0 : Self::Exists(_) => "Exists",
2532 0 : Self::Nblocks(_) => "Nblocks",
2533 0 : Self::GetPage(_) => "GetPage",
2534 0 : Self::Error(_) => "Error",
2535 0 : Self::DbSize(_) => "DbSize",
2536 0 : Self::GetSlruSegment(_) => "GetSlruSegment",
2537 : #[cfg(feature = "testing")]
2538 0 : Self::Test(_) => "Test",
2539 : }
2540 0 : }
2541 : }
2542 :
2543 0 : #[derive(Debug, Serialize, Deserialize)]
2544 : pub struct PageTraceEvent {
2545 : pub key: CompactKey,
2546 : pub effective_lsn: Lsn,
2547 : pub time: SystemTime,
2548 : }
2549 :
2550 : impl Default for PageTraceEvent {
2551 0 : fn default() -> Self {
2552 0 : Self {
2553 0 : key: Default::default(),
2554 0 : effective_lsn: Default::default(),
2555 0 : time: std::time::UNIX_EPOCH,
2556 0 : }
2557 0 : }
2558 : }
2559 :
2560 : #[cfg(test)]
2561 : mod tests {
2562 : use std::str::FromStr;
2563 :
2564 : use serde_json::json;
2565 :
2566 : use super::*;
2567 :
2568 : #[test]
2569 1 : fn test_pagestream() {
2570 1 : // Test serialization/deserialization of PagestreamFeMessage
2571 1 : let messages = vec![
2572 1 : PagestreamFeMessage::Exists(PagestreamExistsRequest {
2573 1 : hdr: PagestreamRequest {
2574 1 : reqid: 0,
2575 1 : request_lsn: Lsn(4),
2576 1 : not_modified_since: Lsn(3),
2577 1 : },
2578 1 : rel: RelTag {
2579 1 : forknum: 1,
2580 1 : spcnode: 2,
2581 1 : dbnode: 3,
2582 1 : relnode: 4,
2583 1 : },
2584 1 : }),
2585 1 : PagestreamFeMessage::Nblocks(PagestreamNblocksRequest {
2586 1 : hdr: PagestreamRequest {
2587 1 : reqid: 0,
2588 1 : request_lsn: Lsn(4),
2589 1 : not_modified_since: Lsn(4),
2590 1 : },
2591 1 : rel: RelTag {
2592 1 : forknum: 1,
2593 1 : spcnode: 2,
2594 1 : dbnode: 3,
2595 1 : relnode: 4,
2596 1 : },
2597 1 : }),
2598 1 : PagestreamFeMessage::GetPage(PagestreamGetPageRequest {
2599 1 : hdr: PagestreamRequest {
2600 1 : reqid: 0,
2601 1 : request_lsn: Lsn(4),
2602 1 : not_modified_since: Lsn(3),
2603 1 : },
2604 1 : rel: RelTag {
2605 1 : forknum: 1,
2606 1 : spcnode: 2,
2607 1 : dbnode: 3,
2608 1 : relnode: 4,
2609 1 : },
2610 1 : blkno: 7,
2611 1 : }),
2612 1 : PagestreamFeMessage::DbSize(PagestreamDbSizeRequest {
2613 1 : hdr: PagestreamRequest {
2614 1 : reqid: 0,
2615 1 : request_lsn: Lsn(4),
2616 1 : not_modified_since: Lsn(3),
2617 1 : },
2618 1 : dbnode: 7,
2619 1 : }),
2620 1 : ];
2621 5 : for msg in messages {
2622 4 : let bytes = msg.serialize();
2623 4 : let reconstructed =
2624 4 : PagestreamFeMessage::parse(&mut bytes.reader(), PagestreamProtocolVersion::V3)
2625 4 : .unwrap();
2626 4 : assert!(msg == reconstructed);
2627 : }
2628 1 : }
2629 :
2630 : #[test]
2631 1 : fn test_tenantinfo_serde() {
2632 1 : // Test serialization/deserialization of TenantInfo
2633 1 : let original_active = TenantInfo {
2634 1 : id: TenantShardId::unsharded(TenantId::generate()),
2635 1 : state: TenantState::Active,
2636 1 : current_physical_size: Some(42),
2637 1 : attachment_status: TenantAttachmentStatus::Attached,
2638 1 : generation: 1,
2639 1 : gc_blocking: None,
2640 1 : };
2641 1 : let expected_active = json!({
2642 1 : "id": original_active.id.to_string(),
2643 1 : "state": {
2644 1 : "slug": "Active",
2645 1 : },
2646 1 : "current_physical_size": 42,
2647 1 : "attachment_status": {
2648 1 : "slug":"attached",
2649 1 : },
2650 1 : "generation" : 1
2651 1 : });
2652 1 :
2653 1 : let original_broken = TenantInfo {
2654 1 : id: TenantShardId::unsharded(TenantId::generate()),
2655 1 : state: TenantState::Broken {
2656 1 : reason: "reason".into(),
2657 1 : backtrace: "backtrace info".into(),
2658 1 : },
2659 1 : current_physical_size: Some(42),
2660 1 : attachment_status: TenantAttachmentStatus::Attached,
2661 1 : generation: 1,
2662 1 : gc_blocking: None,
2663 1 : };
2664 1 : let expected_broken = json!({
2665 1 : "id": original_broken.id.to_string(),
2666 1 : "state": {
2667 1 : "slug": "Broken",
2668 1 : "data": {
2669 1 : "backtrace": "backtrace info",
2670 1 : "reason": "reason",
2671 1 : }
2672 1 : },
2673 1 : "current_physical_size": 42,
2674 1 : "attachment_status": {
2675 1 : "slug":"attached",
2676 1 : },
2677 1 : "generation" : 1
2678 1 : });
2679 1 :
2680 1 : assert_eq!(
2681 1 : serde_json::to_value(&original_active).unwrap(),
2682 1 : expected_active
2683 1 : );
2684 :
2685 1 : assert_eq!(
2686 1 : serde_json::to_value(&original_broken).unwrap(),
2687 1 : expected_broken
2688 1 : );
2689 1 : assert!(format!("{:?}", &original_broken.state).contains("reason"));
2690 1 : assert!(format!("{:?}", &original_broken.state).contains("backtrace info"));
2691 1 : }
2692 :
2693 : #[test]
2694 1 : fn test_reject_unknown_field() {
2695 1 : let id = TenantId::generate();
2696 1 : let config_request = json!({
2697 1 : "tenant_id": id.to_string(),
2698 1 : "unknown_field": "unknown_value".to_string(),
2699 1 : });
2700 1 : let err = serde_json::from_value::<TenantConfigRequest>(config_request).unwrap_err();
2701 1 : assert!(
2702 1 : err.to_string().contains("unknown field `unknown_field`"),
2703 0 : "expect unknown field `unknown_field` error, got: {}",
2704 : err
2705 : );
2706 1 : }
2707 :
2708 : #[test]
2709 1 : fn tenantstatus_activating_serde() {
2710 1 : let states = [TenantState::Activating(ActivatingFrom::Attaching)];
2711 1 : let expected = "[{\"slug\":\"Activating\",\"data\":\"Attaching\"}]";
2712 1 :
2713 1 : let actual = serde_json::to_string(&states).unwrap();
2714 1 :
2715 1 : assert_eq!(actual, expected);
2716 :
2717 1 : let parsed = serde_json::from_str::<Vec<TenantState>>(&actual).unwrap();
2718 1 :
2719 1 : assert_eq!(states.as_slice(), &parsed);
2720 1 : }
2721 :
2722 : #[test]
2723 1 : fn tenantstatus_activating_strum() {
2724 1 : // tests added, because we use these for metrics
2725 1 : let examples = [
2726 1 : (line!(), TenantState::Attaching, "Attaching"),
2727 1 : (
2728 1 : line!(),
2729 1 : TenantState::Activating(ActivatingFrom::Attaching),
2730 1 : "Activating",
2731 1 : ),
2732 1 : (line!(), TenantState::Active, "Active"),
2733 1 : (
2734 1 : line!(),
2735 1 : TenantState::Stopping { progress: None },
2736 1 : "Stopping",
2737 1 : ),
2738 1 : (
2739 1 : line!(),
2740 1 : TenantState::Stopping {
2741 1 : progress: Some(completion::Barrier::default()),
2742 1 : },
2743 1 : "Stopping",
2744 1 : ),
2745 1 : (
2746 1 : line!(),
2747 1 : TenantState::Broken {
2748 1 : reason: "Example".into(),
2749 1 : backtrace: "Looooong backtrace".into(),
2750 1 : },
2751 1 : "Broken",
2752 1 : ),
2753 1 : ];
2754 :
2755 7 : for (line, rendered, expected) in examples {
2756 6 : let actual: &'static str = rendered.into();
2757 6 : assert_eq!(actual, expected, "example on {line}");
2758 : }
2759 1 : }
2760 :
2761 : #[test]
2762 1 : fn test_image_compression_algorithm_parsing() {
2763 : use ImageCompressionAlgorithm::*;
2764 1 : let cases = [
2765 1 : ("disabled", Disabled),
2766 1 : ("zstd", Zstd { level: None }),
2767 1 : ("zstd(18)", Zstd { level: Some(18) }),
2768 1 : ("zstd(-3)", Zstd { level: Some(-3) }),
2769 1 : ];
2770 :
2771 5 : for (display, expected) in cases {
2772 4 : assert_eq!(
2773 4 : ImageCompressionAlgorithm::from_str(display).unwrap(),
2774 : expected,
2775 0 : "parsing works"
2776 : );
2777 4 : assert_eq!(format!("{expected}"), display, "Display FromStr roundtrip");
2778 :
2779 4 : let ser = serde_json::to_string(&expected).expect("serialization");
2780 4 : assert_eq!(
2781 4 : serde_json::from_str::<ImageCompressionAlgorithm>(&ser).unwrap(),
2782 : expected,
2783 0 : "serde roundtrip"
2784 : );
2785 :
2786 4 : assert_eq!(
2787 4 : serde_json::Value::String(display.to_string()),
2788 4 : serde_json::to_value(expected).unwrap(),
2789 0 : "Display is the serde serialization"
2790 : );
2791 : }
2792 1 : }
2793 :
2794 : #[test]
2795 1 : fn test_tenant_config_patch_request_serde() {
2796 1 : let patch_request = TenantConfigPatchRequest {
2797 1 : tenant_id: TenantId::from_str("17c6d121946a61e5ab0fe5a2fd4d8215").unwrap(),
2798 1 : config: TenantConfigPatch {
2799 1 : checkpoint_distance: FieldPatch::Upsert(42),
2800 1 : gc_horizon: FieldPatch::Remove,
2801 1 : compaction_threshold: FieldPatch::Noop,
2802 1 : ..TenantConfigPatch::default()
2803 1 : },
2804 1 : };
2805 1 :
2806 1 : let json = serde_json::to_string(&patch_request).unwrap();
2807 1 :
2808 1 : let expected = r#"{"tenant_id":"17c6d121946a61e5ab0fe5a2fd4d8215","checkpoint_distance":42,"gc_horizon":null}"#;
2809 1 : assert_eq!(json, expected);
2810 :
2811 1 : let decoded: TenantConfigPatchRequest = serde_json::from_str(&json).unwrap();
2812 1 : assert_eq!(decoded.tenant_id, patch_request.tenant_id);
2813 1 : assert_eq!(decoded.config, patch_request.config);
2814 :
2815 : // Now apply the patch to a config to demonstrate semantics
2816 :
2817 1 : let base = TenantConfig {
2818 1 : checkpoint_distance: Some(28),
2819 1 : gc_horizon: Some(100),
2820 1 : compaction_target_size: Some(1024),
2821 1 : ..Default::default()
2822 1 : };
2823 1 :
2824 1 : let expected = TenantConfig {
2825 1 : checkpoint_distance: Some(42),
2826 1 : gc_horizon: None,
2827 1 : ..base.clone()
2828 1 : };
2829 1 :
2830 1 : let patched = base.apply_patch(decoded.config).unwrap();
2831 1 :
2832 1 : assert_eq!(patched, expected);
2833 1 : }
2834 : }
|
