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 3 : pub fn is_expired(&self, now: &SystemTime) -> bool {
188 3 : now > &self.valid_until
189 3 : }
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 3 : fn from(range: Range<Lsn>) -> Self {
270 3 : Self {
271 3 : start: range.start,
272 3 : end: range.end,
273 3 : }
274 3 : }
275 : }
276 :
277 : impl From<Range<Key>> for CompactKeyRange {
278 8 : fn from(range: Range<Key>) -> Self {
279 8 : Self {
280 8 : start: range.start,
281 8 : end: range.end,
282 8 : }
283 8 : }
284 : }
285 :
286 : impl From<CompactLsnRange> for Range<Lsn> {
287 5 : fn from(range: CompactLsnRange) -> Self {
288 5 : range.start..range.end
289 5 : }
290 : }
291 :
292 : impl From<CompactKeyRange> for Range<Key> {
293 8 : fn from(range: CompactKeyRange) -> Self {
294 8 : range.start..range.end
295 8 : }
296 : }
297 :
298 : impl CompactLsnRange {
299 2 : pub fn above(lsn: Lsn) -> Self {
300 2 : Self {
301 2 : start: lsn,
302 2 : end: Lsn::MAX,
303 2 : }
304 2 : }
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 : impl TimelineCreateRequest {
324 0 : pub fn mode_tag(&self) -> &'static str {
325 0 : match &self.mode {
326 0 : TimelineCreateRequestMode::Branch { .. } => "branch",
327 0 : TimelineCreateRequestMode::ImportPgdata { .. } => "import",
328 0 : TimelineCreateRequestMode::Bootstrap { .. } => "bootstrap",
329 : }
330 0 : }
331 :
332 0 : pub fn is_import(&self) -> bool {
333 0 : matches!(self.mode, TimelineCreateRequestMode::ImportPgdata { .. })
334 0 : }
335 : }
336 :
337 0 : #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq)]
338 : pub enum ShardImportStatus {
339 : InProgress(Option<ShardImportProgress>),
340 : Done,
341 : Error(String),
342 : }
343 :
344 0 : #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq)]
345 : pub enum ShardImportProgress {
346 : V1(ShardImportProgressV1),
347 : }
348 :
349 0 : #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq)]
350 : pub struct ShardImportProgressV1 {
351 : /// Total number of jobs in the import plan
352 : pub jobs: usize,
353 : /// Number of jobs completed
354 : pub completed: usize,
355 : /// Hash of the plan
356 : pub import_plan_hash: u64,
357 : }
358 :
359 : impl ShardImportStatus {
360 0 : pub fn is_terminal(&self) -> bool {
361 0 : match self {
362 0 : ShardImportStatus::InProgress(_) => false,
363 0 : ShardImportStatus::Done | ShardImportStatus::Error(_) => true,
364 : }
365 0 : }
366 : }
367 :
368 : /// Storage controller specific extensions to [`TimelineInfo`].
369 0 : #[derive(Serialize, Deserialize, Clone)]
370 : pub struct TimelineCreateResponseStorcon {
371 : #[serde(flatten)]
372 : pub timeline_info: TimelineInfo,
373 :
374 : pub safekeepers: Option<SafekeepersInfo>,
375 : }
376 :
377 : /// Safekeepers as returned in timeline creation request to storcon or pushed to
378 : /// cplane in the post migration hook.
379 0 : #[derive(Serialize, Deserialize, Clone)]
380 : pub struct SafekeepersInfo {
381 : pub tenant_id: TenantId,
382 : pub timeline_id: TimelineId,
383 : pub generation: u32,
384 : pub safekeepers: Vec<SafekeeperInfo>,
385 : }
386 :
387 0 : #[derive(Serialize, Deserialize, Clone)]
388 : pub struct SafekeeperInfo {
389 : pub id: NodeId,
390 : pub hostname: String,
391 : }
392 :
393 0 : #[derive(Serialize, Deserialize, Clone)]
394 : #[serde(untagged)]
395 : pub enum TimelineCreateRequestMode {
396 : Branch {
397 : ancestor_timeline_id: TimelineId,
398 : #[serde(default)]
399 : ancestor_start_lsn: Option<Lsn>,
400 : // TODO: cplane sets this, but, the branching code always
401 : // inherits the ancestor's pg_version. Earlier code wasn't
402 : // using a flattened enum, so, it was an accepted field, and
403 : // we continue to accept it by having it here.
404 : pg_version: Option<u32>,
405 : #[serde(default, skip_serializing_if = "std::ops::Not::not")]
406 : read_only: bool,
407 : },
408 : ImportPgdata {
409 : import_pgdata: TimelineCreateRequestModeImportPgdata,
410 : },
411 : // NB: Bootstrap is all-optional, and thus the serde(untagged) will cause serde to stop at Bootstrap.
412 : // (serde picks the first matching enum variant, in declaration order).
413 : Bootstrap {
414 : #[serde(default)]
415 : existing_initdb_timeline_id: Option<TimelineId>,
416 : pg_version: Option<u32>,
417 : },
418 : }
419 :
420 0 : #[derive(Serialize, Deserialize, Clone)]
421 : pub struct TimelineCreateRequestModeImportPgdata {
422 : pub location: ImportPgdataLocation,
423 : pub idempotency_key: ImportPgdataIdempotencyKey,
424 : }
425 :
426 0 : #[derive(Serialize, Deserialize, Clone, Debug)]
427 : pub enum ImportPgdataLocation {
428 : #[cfg(feature = "testing")]
429 : LocalFs { path: Utf8PathBuf },
430 : AwsS3 {
431 : region: String,
432 : bucket: String,
433 : /// A better name for this would be `prefix`; changing requires coordination with cplane.
434 : /// See <https://github.com/neondatabase/cloud/issues/20646>.
435 : key: String,
436 : },
437 : }
438 :
439 0 : #[derive(Serialize, Deserialize, Clone)]
440 : #[serde(transparent)]
441 : pub struct ImportPgdataIdempotencyKey(pub String);
442 :
443 : impl ImportPgdataIdempotencyKey {
444 0 : pub fn random() -> Self {
445 : use rand::Rng;
446 : use rand::distributions::Alphanumeric;
447 0 : Self(
448 0 : rand::thread_rng()
449 0 : .sample_iter(&Alphanumeric)
450 0 : .take(20)
451 0 : .map(char::from)
452 0 : .collect(),
453 0 : )
454 0 : }
455 : }
456 :
457 0 : #[derive(Serialize, Deserialize, Clone)]
458 : pub struct LsnLeaseRequest {
459 : pub lsn: Lsn,
460 : }
461 :
462 0 : #[derive(Serialize, Deserialize)]
463 : pub struct TenantShardSplitRequest {
464 : pub new_shard_count: u8,
465 :
466 : // A tenant's stripe size is only meaningful the first time their shard count goes
467 : // above 1: therefore during a split from 1->N shards, we may modify the stripe size.
468 : //
469 : // If this is set while the stripe count is being increased from an already >1 value,
470 : // then the request will fail with 400.
471 : pub new_stripe_size: Option<ShardStripeSize>,
472 : }
473 :
474 0 : #[derive(Serialize, Deserialize)]
475 : pub struct TenantShardSplitResponse {
476 : pub new_shards: Vec<TenantShardId>,
477 : }
478 :
479 : /// Parameters that apply to all shards in a tenant. Used during tenant creation.
480 0 : #[derive(Serialize, Deserialize, Debug)]
481 : #[serde(deny_unknown_fields)]
482 : pub struct ShardParameters {
483 : pub count: ShardCount,
484 : pub stripe_size: ShardStripeSize,
485 : }
486 :
487 : impl ShardParameters {
488 0 : pub fn is_unsharded(&self) -> bool {
489 0 : self.count.is_unsharded()
490 0 : }
491 : }
492 :
493 : impl Default for ShardParameters {
494 118 : fn default() -> Self {
495 118 : Self {
496 118 : count: ShardCount::new(0),
497 118 : stripe_size: DEFAULT_STRIPE_SIZE,
498 118 : }
499 118 : }
500 : }
501 :
502 : #[derive(Debug, Default, Clone, Eq, PartialEq)]
503 : pub enum FieldPatch<T> {
504 : Upsert(T),
505 : Remove,
506 : #[default]
507 : Noop,
508 : }
509 :
510 : impl<T> FieldPatch<T> {
511 78 : fn is_noop(&self) -> bool {
512 78 : matches!(self, FieldPatch::Noop)
513 78 : }
514 :
515 39 : pub fn apply(self, target: &mut Option<T>) {
516 39 : match self {
517 1 : Self::Upsert(v) => *target = Some(v),
518 1 : Self::Remove => *target = None,
519 37 : Self::Noop => {}
520 : }
521 39 : }
522 :
523 10 : pub fn map<U, E, F: FnOnce(T) -> Result<U, E>>(self, map: F) -> Result<FieldPatch<U>, E> {
524 10 : match self {
525 0 : Self::Upsert(v) => Ok(FieldPatch::<U>::Upsert(map(v)?)),
526 0 : Self::Remove => Ok(FieldPatch::<U>::Remove),
527 10 : Self::Noop => Ok(FieldPatch::<U>::Noop),
528 : }
529 10 : }
530 : }
531 :
532 : impl<'de, T: Deserialize<'de>> Deserialize<'de> for FieldPatch<T> {
533 2 : fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
534 2 : where
535 2 : D: Deserializer<'de>,
536 2 : {
537 2 : Option::deserialize(deserializer).map(|opt| match opt {
538 1 : None => FieldPatch::Remove,
539 1 : Some(val) => FieldPatch::Upsert(val),
540 2 : })
541 2 : }
542 : }
543 :
544 : impl<T: Serialize> Serialize for FieldPatch<T> {
545 2 : fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
546 2 : where
547 2 : S: Serializer,
548 2 : {
549 2 : match self {
550 1 : FieldPatch::Upsert(val) => serializer.serialize_some(val),
551 1 : FieldPatch::Remove => serializer.serialize_none(),
552 0 : FieldPatch::Noop => unreachable!(),
553 : }
554 2 : }
555 : }
556 :
557 2 : #[derive(Serialize, Deserialize, Debug, Default, Clone, Eq, PartialEq)]
558 : #[serde(default)]
559 : pub struct TenantConfigPatch {
560 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
561 : pub checkpoint_distance: FieldPatch<u64>,
562 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
563 : pub checkpoint_timeout: FieldPatch<String>,
564 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
565 : pub compaction_target_size: FieldPatch<u64>,
566 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
567 : pub compaction_period: FieldPatch<String>,
568 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
569 : pub compaction_threshold: FieldPatch<usize>,
570 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
571 : pub compaction_upper_limit: FieldPatch<usize>,
572 : // defer parsing compaction_algorithm, like eviction_policy
573 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
574 : pub compaction_algorithm: FieldPatch<CompactionAlgorithmSettings>,
575 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
576 : pub compaction_shard_ancestor: FieldPatch<bool>,
577 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
578 : pub compaction_l0_first: FieldPatch<bool>,
579 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
580 : pub compaction_l0_semaphore: FieldPatch<bool>,
581 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
582 : pub l0_flush_delay_threshold: FieldPatch<usize>,
583 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
584 : pub l0_flush_stall_threshold: FieldPatch<usize>,
585 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
586 : pub gc_horizon: FieldPatch<u64>,
587 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
588 : pub gc_period: FieldPatch<String>,
589 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
590 : pub image_creation_threshold: FieldPatch<usize>,
591 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
592 : pub pitr_interval: FieldPatch<String>,
593 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
594 : pub walreceiver_connect_timeout: FieldPatch<String>,
595 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
596 : pub lagging_wal_timeout: FieldPatch<String>,
597 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
598 : pub max_lsn_wal_lag: FieldPatch<NonZeroU64>,
599 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
600 : pub eviction_policy: FieldPatch<EvictionPolicy>,
601 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
602 : pub min_resident_size_override: FieldPatch<u64>,
603 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
604 : pub evictions_low_residence_duration_metric_threshold: FieldPatch<String>,
605 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
606 : pub heatmap_period: FieldPatch<String>,
607 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
608 : pub lazy_slru_download: FieldPatch<bool>,
609 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
610 : pub timeline_get_throttle: FieldPatch<ThrottleConfig>,
611 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
612 : pub image_layer_creation_check_threshold: FieldPatch<u8>,
613 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
614 : pub image_creation_preempt_threshold: FieldPatch<usize>,
615 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
616 : pub lsn_lease_length: FieldPatch<String>,
617 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
618 : pub lsn_lease_length_for_ts: FieldPatch<String>,
619 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
620 : pub timeline_offloading: FieldPatch<bool>,
621 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
622 : pub wal_receiver_protocol_override: FieldPatch<PostgresClientProtocol>,
623 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
624 : pub rel_size_v2_enabled: FieldPatch<bool>,
625 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
626 : pub gc_compaction_enabled: FieldPatch<bool>,
627 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
628 : pub gc_compaction_verification: FieldPatch<bool>,
629 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
630 : pub gc_compaction_initial_threshold_kb: FieldPatch<u64>,
631 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
632 : pub gc_compaction_ratio_percent: FieldPatch<u64>,
633 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
634 : pub sampling_ratio: FieldPatch<Option<Ratio>>,
635 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
636 : pub relsize_snapshot_cache_capacity: FieldPatch<usize>,
637 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
638 : pub basebackup_cache_enabled: FieldPatch<bool>,
639 : }
640 :
641 : /// Like [`crate::config::TenantConfigToml`], but preserves the information
642 : /// about which parameters are set and which are not.
643 : ///
644 : /// Used in many places, including durably stored ones.
645 2 : #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Default)]
646 : #[serde(default)] // this maps omitted fields in deserialization to None
647 : pub struct TenantConfig {
648 : #[serde(skip_serializing_if = "Option::is_none")]
649 : pub checkpoint_distance: Option<u64>,
650 :
651 : #[serde(skip_serializing_if = "Option::is_none")]
652 : #[serde(with = "humantime_serde")]
653 : pub checkpoint_timeout: Option<Duration>,
654 :
655 : #[serde(skip_serializing_if = "Option::is_none")]
656 : pub compaction_target_size: Option<u64>,
657 :
658 : #[serde(skip_serializing_if = "Option::is_none")]
659 : #[serde(with = "humantime_serde")]
660 : pub compaction_period: Option<Duration>,
661 :
662 : #[serde(skip_serializing_if = "Option::is_none")]
663 : pub compaction_threshold: Option<usize>,
664 :
665 : #[serde(skip_serializing_if = "Option::is_none")]
666 : pub compaction_upper_limit: Option<usize>,
667 :
668 : #[serde(skip_serializing_if = "Option::is_none")]
669 : pub compaction_algorithm: Option<CompactionAlgorithmSettings>,
670 :
671 : #[serde(skip_serializing_if = "Option::is_none")]
672 : pub compaction_shard_ancestor: Option<bool>,
673 :
674 : #[serde(skip_serializing_if = "Option::is_none")]
675 : pub compaction_l0_first: Option<bool>,
676 :
677 : #[serde(skip_serializing_if = "Option::is_none")]
678 : pub compaction_l0_semaphore: Option<bool>,
679 :
680 : #[serde(skip_serializing_if = "Option::is_none")]
681 : pub l0_flush_delay_threshold: Option<usize>,
682 :
683 : #[serde(skip_serializing_if = "Option::is_none")]
684 : pub l0_flush_stall_threshold: Option<usize>,
685 :
686 : #[serde(skip_serializing_if = "Option::is_none")]
687 : pub gc_horizon: Option<u64>,
688 :
689 : #[serde(skip_serializing_if = "Option::is_none")]
690 : #[serde(with = "humantime_serde")]
691 : pub gc_period: Option<Duration>,
692 :
693 : #[serde(skip_serializing_if = "Option::is_none")]
694 : pub image_creation_threshold: Option<usize>,
695 :
696 : #[serde(skip_serializing_if = "Option::is_none")]
697 : #[serde(with = "humantime_serde")]
698 : pub pitr_interval: Option<Duration>,
699 :
700 : #[serde(skip_serializing_if = "Option::is_none")]
701 : #[serde(with = "humantime_serde")]
702 : pub walreceiver_connect_timeout: Option<Duration>,
703 :
704 : #[serde(skip_serializing_if = "Option::is_none")]
705 : #[serde(with = "humantime_serde")]
706 : pub lagging_wal_timeout: Option<Duration>,
707 :
708 : #[serde(skip_serializing_if = "Option::is_none")]
709 : pub max_lsn_wal_lag: Option<NonZeroU64>,
710 :
711 : #[serde(skip_serializing_if = "Option::is_none")]
712 : pub eviction_policy: Option<EvictionPolicy>,
713 :
714 : #[serde(skip_serializing_if = "Option::is_none")]
715 : pub min_resident_size_override: Option<u64>,
716 :
717 : #[serde(skip_serializing_if = "Option::is_none")]
718 : #[serde(with = "humantime_serde")]
719 : pub evictions_low_residence_duration_metric_threshold: Option<Duration>,
720 :
721 : #[serde(skip_serializing_if = "Option::is_none")]
722 : #[serde(with = "humantime_serde")]
723 : pub heatmap_period: Option<Duration>,
724 :
725 : #[serde(skip_serializing_if = "Option::is_none")]
726 : pub lazy_slru_download: Option<bool>,
727 :
728 : #[serde(skip_serializing_if = "Option::is_none")]
729 : pub timeline_get_throttle: Option<ThrottleConfig>,
730 :
731 : #[serde(skip_serializing_if = "Option::is_none")]
732 : pub image_layer_creation_check_threshold: Option<u8>,
733 :
734 : #[serde(skip_serializing_if = "Option::is_none")]
735 : pub image_creation_preempt_threshold: Option<usize>,
736 :
737 : #[serde(skip_serializing_if = "Option::is_none")]
738 : #[serde(with = "humantime_serde")]
739 : pub lsn_lease_length: Option<Duration>,
740 :
741 : #[serde(skip_serializing_if = "Option::is_none")]
742 : #[serde(with = "humantime_serde")]
743 : pub lsn_lease_length_for_ts: Option<Duration>,
744 :
745 : #[serde(skip_serializing_if = "Option::is_none")]
746 : pub timeline_offloading: Option<bool>,
747 :
748 : #[serde(skip_serializing_if = "Option::is_none")]
749 : pub wal_receiver_protocol_override: Option<PostgresClientProtocol>,
750 :
751 : #[serde(skip_serializing_if = "Option::is_none")]
752 : pub rel_size_v2_enabled: Option<bool>,
753 :
754 : #[serde(skip_serializing_if = "Option::is_none")]
755 : pub gc_compaction_enabled: Option<bool>,
756 :
757 : #[serde(skip_serializing_if = "Option::is_none")]
758 : pub gc_compaction_verification: Option<bool>,
759 :
760 : #[serde(skip_serializing_if = "Option::is_none")]
761 : pub gc_compaction_initial_threshold_kb: Option<u64>,
762 :
763 : #[serde(skip_serializing_if = "Option::is_none")]
764 : pub gc_compaction_ratio_percent: Option<u64>,
765 :
766 : #[serde(skip_serializing_if = "Option::is_none")]
767 : pub sampling_ratio: Option<Option<Ratio>>,
768 :
769 : #[serde(skip_serializing_if = "Option::is_none")]
770 : pub relsize_snapshot_cache_capacity: Option<usize>,
771 :
772 : #[serde(skip_serializing_if = "Option::is_none")]
773 : pub basebackup_cache_enabled: Option<bool>,
774 : }
775 :
776 : impl TenantConfig {
777 1 : pub fn apply_patch(
778 1 : self,
779 1 : patch: TenantConfigPatch,
780 1 : ) -> Result<TenantConfig, humantime::DurationError> {
781 1 : let Self {
782 1 : mut checkpoint_distance,
783 1 : mut checkpoint_timeout,
784 1 : mut compaction_target_size,
785 1 : mut compaction_period,
786 1 : mut compaction_threshold,
787 1 : mut compaction_upper_limit,
788 1 : mut compaction_algorithm,
789 1 : mut compaction_shard_ancestor,
790 1 : mut compaction_l0_first,
791 1 : mut compaction_l0_semaphore,
792 1 : mut l0_flush_delay_threshold,
793 1 : mut l0_flush_stall_threshold,
794 1 : mut gc_horizon,
795 1 : mut gc_period,
796 1 : mut image_creation_threshold,
797 1 : mut pitr_interval,
798 1 : mut walreceiver_connect_timeout,
799 1 : mut lagging_wal_timeout,
800 1 : mut max_lsn_wal_lag,
801 1 : mut eviction_policy,
802 1 : mut min_resident_size_override,
803 1 : mut evictions_low_residence_duration_metric_threshold,
804 1 : mut heatmap_period,
805 1 : mut lazy_slru_download,
806 1 : mut timeline_get_throttle,
807 1 : mut image_layer_creation_check_threshold,
808 1 : mut image_creation_preempt_threshold,
809 1 : mut lsn_lease_length,
810 1 : mut lsn_lease_length_for_ts,
811 1 : mut timeline_offloading,
812 1 : mut wal_receiver_protocol_override,
813 1 : mut rel_size_v2_enabled,
814 1 : mut gc_compaction_enabled,
815 1 : mut gc_compaction_verification,
816 1 : mut gc_compaction_initial_threshold_kb,
817 1 : mut gc_compaction_ratio_percent,
818 1 : mut sampling_ratio,
819 1 : mut relsize_snapshot_cache_capacity,
820 1 : mut basebackup_cache_enabled,
821 1 : } = self;
822 1 :
823 1 : patch.checkpoint_distance.apply(&mut checkpoint_distance);
824 1 : patch
825 1 : .checkpoint_timeout
826 1 : .map(|v| humantime::parse_duration(&v))?
827 1 : .apply(&mut checkpoint_timeout);
828 1 : patch
829 1 : .compaction_target_size
830 1 : .apply(&mut compaction_target_size);
831 1 : patch
832 1 : .compaction_period
833 1 : .map(|v| humantime::parse_duration(&v))?
834 1 : .apply(&mut compaction_period);
835 1 : patch.compaction_threshold.apply(&mut compaction_threshold);
836 1 : patch
837 1 : .compaction_upper_limit
838 1 : .apply(&mut compaction_upper_limit);
839 1 : patch.compaction_algorithm.apply(&mut compaction_algorithm);
840 1 : patch
841 1 : .compaction_shard_ancestor
842 1 : .apply(&mut compaction_shard_ancestor);
843 1 : patch.compaction_l0_first.apply(&mut compaction_l0_first);
844 1 : patch
845 1 : .compaction_l0_semaphore
846 1 : .apply(&mut compaction_l0_semaphore);
847 1 : patch
848 1 : .l0_flush_delay_threshold
849 1 : .apply(&mut l0_flush_delay_threshold);
850 1 : patch
851 1 : .l0_flush_stall_threshold
852 1 : .apply(&mut l0_flush_stall_threshold);
853 1 : patch.gc_horizon.apply(&mut gc_horizon);
854 1 : patch
855 1 : .gc_period
856 1 : .map(|v| humantime::parse_duration(&v))?
857 1 : .apply(&mut gc_period);
858 1 : patch
859 1 : .image_creation_threshold
860 1 : .apply(&mut image_creation_threshold);
861 1 : patch
862 1 : .pitr_interval
863 1 : .map(|v| humantime::parse_duration(&v))?
864 1 : .apply(&mut pitr_interval);
865 1 : patch
866 1 : .walreceiver_connect_timeout
867 1 : .map(|v| humantime::parse_duration(&v))?
868 1 : .apply(&mut walreceiver_connect_timeout);
869 1 : patch
870 1 : .lagging_wal_timeout
871 1 : .map(|v| humantime::parse_duration(&v))?
872 1 : .apply(&mut lagging_wal_timeout);
873 1 : patch.max_lsn_wal_lag.apply(&mut max_lsn_wal_lag);
874 1 : patch.eviction_policy.apply(&mut eviction_policy);
875 1 : patch
876 1 : .min_resident_size_override
877 1 : .apply(&mut min_resident_size_override);
878 1 : patch
879 1 : .evictions_low_residence_duration_metric_threshold
880 1 : .map(|v| humantime::parse_duration(&v))?
881 1 : .apply(&mut evictions_low_residence_duration_metric_threshold);
882 1 : patch
883 1 : .heatmap_period
884 1 : .map(|v| humantime::parse_duration(&v))?
885 1 : .apply(&mut heatmap_period);
886 1 : patch.lazy_slru_download.apply(&mut lazy_slru_download);
887 1 : patch
888 1 : .timeline_get_throttle
889 1 : .apply(&mut timeline_get_throttle);
890 1 : patch
891 1 : .image_layer_creation_check_threshold
892 1 : .apply(&mut image_layer_creation_check_threshold);
893 1 : patch
894 1 : .image_creation_preempt_threshold
895 1 : .apply(&mut image_creation_preempt_threshold);
896 1 : patch
897 1 : .lsn_lease_length
898 1 : .map(|v| humantime::parse_duration(&v))?
899 1 : .apply(&mut lsn_lease_length);
900 1 : patch
901 1 : .lsn_lease_length_for_ts
902 1 : .map(|v| humantime::parse_duration(&v))?
903 1 : .apply(&mut lsn_lease_length_for_ts);
904 1 : patch.timeline_offloading.apply(&mut timeline_offloading);
905 1 : patch
906 1 : .wal_receiver_protocol_override
907 1 : .apply(&mut wal_receiver_protocol_override);
908 1 : patch.rel_size_v2_enabled.apply(&mut rel_size_v2_enabled);
909 1 : patch
910 1 : .gc_compaction_enabled
911 1 : .apply(&mut gc_compaction_enabled);
912 1 : patch
913 1 : .gc_compaction_verification
914 1 : .apply(&mut gc_compaction_verification);
915 1 : patch
916 1 : .gc_compaction_initial_threshold_kb
917 1 : .apply(&mut gc_compaction_initial_threshold_kb);
918 1 : patch
919 1 : .gc_compaction_ratio_percent
920 1 : .apply(&mut gc_compaction_ratio_percent);
921 1 : patch.sampling_ratio.apply(&mut sampling_ratio);
922 1 : patch
923 1 : .relsize_snapshot_cache_capacity
924 1 : .apply(&mut relsize_snapshot_cache_capacity);
925 1 : patch
926 1 : .basebackup_cache_enabled
927 1 : .apply(&mut basebackup_cache_enabled);
928 1 :
929 1 : Ok(Self {
930 1 : checkpoint_distance,
931 1 : checkpoint_timeout,
932 1 : compaction_target_size,
933 1 : compaction_period,
934 1 : compaction_threshold,
935 1 : compaction_upper_limit,
936 1 : compaction_algorithm,
937 1 : compaction_shard_ancestor,
938 1 : compaction_l0_first,
939 1 : compaction_l0_semaphore,
940 1 : l0_flush_delay_threshold,
941 1 : l0_flush_stall_threshold,
942 1 : gc_horizon,
943 1 : gc_period,
944 1 : image_creation_threshold,
945 1 : pitr_interval,
946 1 : walreceiver_connect_timeout,
947 1 : lagging_wal_timeout,
948 1 : max_lsn_wal_lag,
949 1 : eviction_policy,
950 1 : min_resident_size_override,
951 1 : evictions_low_residence_duration_metric_threshold,
952 1 : heatmap_period,
953 1 : lazy_slru_download,
954 1 : timeline_get_throttle,
955 1 : image_layer_creation_check_threshold,
956 1 : image_creation_preempt_threshold,
957 1 : lsn_lease_length,
958 1 : lsn_lease_length_for_ts,
959 1 : timeline_offloading,
960 1 : wal_receiver_protocol_override,
961 1 : rel_size_v2_enabled,
962 1 : gc_compaction_enabled,
963 1 : gc_compaction_verification,
964 1 : gc_compaction_initial_threshold_kb,
965 1 : gc_compaction_ratio_percent,
966 1 : sampling_ratio,
967 1 : relsize_snapshot_cache_capacity,
968 1 : basebackup_cache_enabled,
969 1 : })
970 1 : }
971 :
972 0 : pub fn merge(
973 0 : &self,
974 0 : global_conf: crate::config::TenantConfigToml,
975 0 : ) -> crate::config::TenantConfigToml {
976 0 : crate::config::TenantConfigToml {
977 0 : checkpoint_distance: self
978 0 : .checkpoint_distance
979 0 : .unwrap_or(global_conf.checkpoint_distance),
980 0 : checkpoint_timeout: self
981 0 : .checkpoint_timeout
982 0 : .unwrap_or(global_conf.checkpoint_timeout),
983 0 : compaction_target_size: self
984 0 : .compaction_target_size
985 0 : .unwrap_or(global_conf.compaction_target_size),
986 0 : compaction_period: self
987 0 : .compaction_period
988 0 : .unwrap_or(global_conf.compaction_period),
989 0 : compaction_threshold: self
990 0 : .compaction_threshold
991 0 : .unwrap_or(global_conf.compaction_threshold),
992 0 : compaction_upper_limit: self
993 0 : .compaction_upper_limit
994 0 : .unwrap_or(global_conf.compaction_upper_limit),
995 0 : compaction_algorithm: self
996 0 : .compaction_algorithm
997 0 : .as_ref()
998 0 : .unwrap_or(&global_conf.compaction_algorithm)
999 0 : .clone(),
1000 0 : compaction_shard_ancestor: self
1001 0 : .compaction_shard_ancestor
1002 0 : .unwrap_or(global_conf.compaction_shard_ancestor),
1003 0 : compaction_l0_first: self
1004 0 : .compaction_l0_first
1005 0 : .unwrap_or(global_conf.compaction_l0_first),
1006 0 : compaction_l0_semaphore: self
1007 0 : .compaction_l0_semaphore
1008 0 : .unwrap_or(global_conf.compaction_l0_semaphore),
1009 0 : l0_flush_delay_threshold: self
1010 0 : .l0_flush_delay_threshold
1011 0 : .or(global_conf.l0_flush_delay_threshold),
1012 0 : l0_flush_stall_threshold: self
1013 0 : .l0_flush_stall_threshold
1014 0 : .or(global_conf.l0_flush_stall_threshold),
1015 0 : gc_horizon: self.gc_horizon.unwrap_or(global_conf.gc_horizon),
1016 0 : gc_period: self.gc_period.unwrap_or(global_conf.gc_period),
1017 0 : image_creation_threshold: self
1018 0 : .image_creation_threshold
1019 0 : .unwrap_or(global_conf.image_creation_threshold),
1020 0 : pitr_interval: self.pitr_interval.unwrap_or(global_conf.pitr_interval),
1021 0 : walreceiver_connect_timeout: self
1022 0 : .walreceiver_connect_timeout
1023 0 : .unwrap_or(global_conf.walreceiver_connect_timeout),
1024 0 : lagging_wal_timeout: self
1025 0 : .lagging_wal_timeout
1026 0 : .unwrap_or(global_conf.lagging_wal_timeout),
1027 0 : max_lsn_wal_lag: self.max_lsn_wal_lag.unwrap_or(global_conf.max_lsn_wal_lag),
1028 0 : eviction_policy: self.eviction_policy.unwrap_or(global_conf.eviction_policy),
1029 0 : min_resident_size_override: self
1030 0 : .min_resident_size_override
1031 0 : .or(global_conf.min_resident_size_override),
1032 0 : evictions_low_residence_duration_metric_threshold: self
1033 0 : .evictions_low_residence_duration_metric_threshold
1034 0 : .unwrap_or(global_conf.evictions_low_residence_duration_metric_threshold),
1035 0 : heatmap_period: self.heatmap_period.unwrap_or(global_conf.heatmap_period),
1036 0 : lazy_slru_download: self
1037 0 : .lazy_slru_download
1038 0 : .unwrap_or(global_conf.lazy_slru_download),
1039 0 : timeline_get_throttle: self
1040 0 : .timeline_get_throttle
1041 0 : .clone()
1042 0 : .unwrap_or(global_conf.timeline_get_throttle),
1043 0 : image_layer_creation_check_threshold: self
1044 0 : .image_layer_creation_check_threshold
1045 0 : .unwrap_or(global_conf.image_layer_creation_check_threshold),
1046 0 : image_creation_preempt_threshold: self
1047 0 : .image_creation_preempt_threshold
1048 0 : .unwrap_or(global_conf.image_creation_preempt_threshold),
1049 0 : lsn_lease_length: self
1050 0 : .lsn_lease_length
1051 0 : .unwrap_or(global_conf.lsn_lease_length),
1052 0 : lsn_lease_length_for_ts: self
1053 0 : .lsn_lease_length_for_ts
1054 0 : .unwrap_or(global_conf.lsn_lease_length_for_ts),
1055 0 : timeline_offloading: self
1056 0 : .timeline_offloading
1057 0 : .unwrap_or(global_conf.timeline_offloading),
1058 0 : wal_receiver_protocol_override: self
1059 0 : .wal_receiver_protocol_override
1060 0 : .or(global_conf.wal_receiver_protocol_override),
1061 0 : rel_size_v2_enabled: self
1062 0 : .rel_size_v2_enabled
1063 0 : .unwrap_or(global_conf.rel_size_v2_enabled),
1064 0 : gc_compaction_enabled: self
1065 0 : .gc_compaction_enabled
1066 0 : .unwrap_or(global_conf.gc_compaction_enabled),
1067 0 : gc_compaction_verification: self
1068 0 : .gc_compaction_verification
1069 0 : .unwrap_or(global_conf.gc_compaction_verification),
1070 0 : gc_compaction_initial_threshold_kb: self
1071 0 : .gc_compaction_initial_threshold_kb
1072 0 : .unwrap_or(global_conf.gc_compaction_initial_threshold_kb),
1073 0 : gc_compaction_ratio_percent: self
1074 0 : .gc_compaction_ratio_percent
1075 0 : .unwrap_or(global_conf.gc_compaction_ratio_percent),
1076 0 : sampling_ratio: self.sampling_ratio.unwrap_or(global_conf.sampling_ratio),
1077 0 : relsize_snapshot_cache_capacity: self
1078 0 : .relsize_snapshot_cache_capacity
1079 0 : .unwrap_or(global_conf.relsize_snapshot_cache_capacity),
1080 0 : basebackup_cache_enabled: self
1081 0 : .basebackup_cache_enabled
1082 0 : .unwrap_or(global_conf.basebackup_cache_enabled),
1083 0 : }
1084 0 : }
1085 : }
1086 :
1087 : /// The policy for the aux file storage.
1088 : ///
1089 : /// It can be switched through `switch_aux_file_policy` tenant config.
1090 : /// When the first aux file written, the policy will be persisted in the
1091 : /// `index_part.json` file and has a limited migration path.
1092 : ///
1093 : /// Currently, we only allow the following migration path:
1094 : ///
1095 : /// Unset -> V1
1096 : /// -> V2
1097 : /// -> CrossValidation -> V2
1098 : #[derive(
1099 : Eq,
1100 : PartialEq,
1101 : Debug,
1102 : Copy,
1103 : Clone,
1104 0 : strum_macros::EnumString,
1105 : strum_macros::Display,
1106 1 : serde_with::DeserializeFromStr,
1107 : serde_with::SerializeDisplay,
1108 : )]
1109 : #[strum(serialize_all = "kebab-case")]
1110 : pub enum AuxFilePolicy {
1111 : /// V1 aux file policy: store everything in AUX_FILE_KEY
1112 : #[strum(ascii_case_insensitive)]
1113 : V1,
1114 : /// V2 aux file policy: store in the AUX_FILE keyspace
1115 : #[strum(ascii_case_insensitive)]
1116 : V2,
1117 : /// Cross validation runs both formats on the write path and does validation
1118 : /// on the read path.
1119 : #[strum(ascii_case_insensitive)]
1120 : CrossValidation,
1121 : }
1122 :
1123 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
1124 : #[serde(tag = "kind")]
1125 : pub enum EvictionPolicy {
1126 : NoEviction,
1127 : LayerAccessThreshold(EvictionPolicyLayerAccessThreshold),
1128 : OnlyImitiate(EvictionPolicyLayerAccessThreshold),
1129 : }
1130 :
1131 : impl EvictionPolicy {
1132 0 : pub fn discriminant_str(&self) -> &'static str {
1133 0 : match self {
1134 0 : EvictionPolicy::NoEviction => "NoEviction",
1135 0 : EvictionPolicy::LayerAccessThreshold(_) => "LayerAccessThreshold",
1136 0 : EvictionPolicy::OnlyImitiate(_) => "OnlyImitiate",
1137 : }
1138 0 : }
1139 : }
1140 :
1141 : #[derive(
1142 : Eq,
1143 : PartialEq,
1144 : Debug,
1145 : Copy,
1146 : Clone,
1147 0 : strum_macros::EnumString,
1148 : strum_macros::Display,
1149 0 : serde_with::DeserializeFromStr,
1150 : serde_with::SerializeDisplay,
1151 : )]
1152 : #[strum(serialize_all = "kebab-case")]
1153 : pub enum CompactionAlgorithm {
1154 : Legacy,
1155 : Tiered,
1156 : }
1157 :
1158 : #[derive(
1159 4 : Debug, Clone, Copy, PartialEq, Eq, serde_with::DeserializeFromStr, serde_with::SerializeDisplay,
1160 : )]
1161 : pub enum ImageCompressionAlgorithm {
1162 : // Disabled for writes, support decompressing during read path
1163 : Disabled,
1164 : /// Zstandard compression. Level 0 means and None mean the same (default level). Levels can be negative as well.
1165 : /// For details, see the [manual](http://facebook.github.io/zstd/zstd_manual.html).
1166 : Zstd {
1167 : level: Option<i8>,
1168 : },
1169 : }
1170 :
1171 : impl FromStr for ImageCompressionAlgorithm {
1172 : type Err = anyhow::Error;
1173 8 : fn from_str(s: &str) -> Result<Self, Self::Err> {
1174 8 : let mut components = s.split(['(', ')']);
1175 8 : let first = components
1176 8 : .next()
1177 8 : .ok_or_else(|| anyhow::anyhow!("empty string"))?;
1178 8 : match first {
1179 8 : "disabled" => Ok(ImageCompressionAlgorithm::Disabled),
1180 6 : "zstd" => {
1181 6 : let level = if let Some(v) = components.next() {
1182 4 : let v: i8 = v.parse()?;
1183 4 : Some(v)
1184 : } else {
1185 2 : None
1186 : };
1187 :
1188 6 : Ok(ImageCompressionAlgorithm::Zstd { level })
1189 : }
1190 0 : _ => anyhow::bail!("invalid specifier '{first}'"),
1191 : }
1192 8 : }
1193 : }
1194 :
1195 : impl Display for ImageCompressionAlgorithm {
1196 12 : fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
1197 12 : match self {
1198 3 : ImageCompressionAlgorithm::Disabled => write!(f, "disabled"),
1199 9 : ImageCompressionAlgorithm::Zstd { level } => {
1200 9 : if let Some(level) = level {
1201 6 : write!(f, "zstd({})", level)
1202 : } else {
1203 3 : write!(f, "zstd")
1204 : }
1205 : }
1206 : }
1207 12 : }
1208 : }
1209 :
1210 0 : #[derive(Eq, PartialEq, Debug, Clone, Serialize, Deserialize)]
1211 : pub struct CompactionAlgorithmSettings {
1212 : pub kind: CompactionAlgorithm,
1213 : }
1214 :
1215 0 : #[derive(Debug, PartialEq, Eq, Clone, Deserialize, Serialize)]
1216 : #[serde(tag = "mode", rename_all = "kebab-case")]
1217 : pub enum L0FlushConfig {
1218 : #[serde(rename_all = "snake_case")]
1219 : Direct { max_concurrency: NonZeroUsize },
1220 : }
1221 :
1222 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
1223 : pub struct EvictionPolicyLayerAccessThreshold {
1224 : #[serde(with = "humantime_serde")]
1225 : pub period: Duration,
1226 : #[serde(with = "humantime_serde")]
1227 : pub threshold: Duration,
1228 : }
1229 :
1230 6 : #[derive(Debug, Serialize, Deserialize, Clone, PartialEq, Eq)]
1231 : pub struct ThrottleConfig {
1232 : /// See [`ThrottleConfigTaskKinds`] for why we do the serde `rename`.
1233 : #[serde(rename = "task_kinds")]
1234 : pub enabled: ThrottleConfigTaskKinds,
1235 : pub initial: u32,
1236 : #[serde(with = "humantime_serde")]
1237 : pub refill_interval: Duration,
1238 : pub refill_amount: NonZeroU32,
1239 : pub max: u32,
1240 : }
1241 :
1242 : /// Before <https://github.com/neondatabase/neon/pull/9962>
1243 : /// the throttle was a per `Timeline::get`/`Timeline::get_vectored` call.
1244 : /// The `task_kinds` field controlled which Pageserver "Task Kind"s
1245 : /// were subject to the throttle.
1246 : ///
1247 : /// After that PR, the throttle is applied at pagestream request level
1248 : /// and the `task_kinds` field does not apply since the only task kind
1249 : /// that us subject to the throttle is that of the page service.
1250 : ///
1251 : /// However, we don't want to make a breaking config change right now
1252 : /// because it means we have to migrate all the tenant configs.
1253 : /// This will be done in a future PR.
1254 : ///
1255 : /// In the meantime, we use emptiness / non-emptsiness of the `task_kinds`
1256 : /// field to determine if the throttle is enabled or not.
1257 1 : #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq)]
1258 : #[serde(transparent)]
1259 : pub struct ThrottleConfigTaskKinds(Vec<String>);
1260 :
1261 : impl ThrottleConfigTaskKinds {
1262 128 : pub fn disabled() -> Self {
1263 128 : Self(vec![])
1264 128 : }
1265 119 : pub fn is_enabled(&self) -> bool {
1266 119 : !self.0.is_empty()
1267 119 : }
1268 : }
1269 :
1270 : impl ThrottleConfig {
1271 128 : pub fn disabled() -> Self {
1272 128 : Self {
1273 128 : enabled: ThrottleConfigTaskKinds::disabled(),
1274 128 : // other values don't matter with emtpy `task_kinds`.
1275 128 : initial: 0,
1276 128 : refill_interval: Duration::from_millis(1),
1277 128 : refill_amount: NonZeroU32::new(1).unwrap(),
1278 128 : max: 1,
1279 128 : }
1280 128 : }
1281 : /// The requests per second allowed by the given config.
1282 0 : pub fn steady_rps(&self) -> f64 {
1283 0 : (self.refill_amount.get() as f64) / (self.refill_interval.as_secs_f64())
1284 0 : }
1285 : }
1286 :
1287 : #[cfg(test)]
1288 : mod throttle_config_tests {
1289 : use super::*;
1290 :
1291 : #[test]
1292 1 : fn test_disabled_is_disabled() {
1293 1 : let config = ThrottleConfig::disabled();
1294 1 : assert!(!config.enabled.is_enabled());
1295 1 : }
1296 : #[test]
1297 1 : fn test_enabled_backwards_compat() {
1298 1 : let input = serde_json::json!({
1299 1 : "task_kinds": ["PageRequestHandler"],
1300 1 : "initial": 40000,
1301 1 : "refill_interval": "50ms",
1302 1 : "refill_amount": 1000,
1303 1 : "max": 40000,
1304 1 : "fair": true
1305 1 : });
1306 1 : let config: ThrottleConfig = serde_json::from_value(input).unwrap();
1307 1 : assert!(config.enabled.is_enabled());
1308 1 : }
1309 : }
1310 :
1311 : /// A flattened analog of a `pagesever::tenant::LocationMode`, which
1312 : /// lists out all possible states (and the virtual "Detached" state)
1313 : /// in a flat form rather than using rust-style enums.
1314 0 : #[derive(Serialize, Deserialize, Debug, Clone, Copy, Eq, PartialEq)]
1315 : pub enum LocationConfigMode {
1316 : AttachedSingle,
1317 : AttachedMulti,
1318 : AttachedStale,
1319 : Secondary,
1320 : Detached,
1321 : }
1322 :
1323 0 : #[derive(Serialize, Deserialize, Debug, Clone, Eq, PartialEq)]
1324 : pub struct LocationConfigSecondary {
1325 : pub warm: bool,
1326 : }
1327 :
1328 : /// An alternative representation of `pageserver::tenant::LocationConf`,
1329 : /// for use in external-facing APIs.
1330 0 : #[derive(Serialize, Deserialize, Debug, Clone, Eq, PartialEq)]
1331 : pub struct LocationConfig {
1332 : pub mode: LocationConfigMode,
1333 : /// If attaching, in what generation?
1334 : #[serde(default)]
1335 : pub generation: Option<u32>,
1336 :
1337 : // If requesting mode `Secondary`, configuration for that.
1338 : #[serde(default)]
1339 : pub secondary_conf: Option<LocationConfigSecondary>,
1340 :
1341 : // Shard parameters: if shard_count is nonzero, then other shard_* fields
1342 : // must be set accurately.
1343 : #[serde(default)]
1344 : pub shard_number: u8,
1345 : #[serde(default)]
1346 : pub shard_count: u8,
1347 : #[serde(default)]
1348 : pub shard_stripe_size: u32,
1349 :
1350 : // This configuration only affects attached mode, but should be provided irrespective
1351 : // of the mode, as a secondary location might transition on startup if the response
1352 : // to the `/re-attach` control plane API requests it.
1353 : pub tenant_conf: TenantConfig,
1354 : }
1355 :
1356 0 : #[derive(Serialize, Deserialize)]
1357 : pub struct LocationConfigListResponse {
1358 : pub tenant_shards: Vec<(TenantShardId, Option<LocationConfig>)>,
1359 : }
1360 :
1361 : #[derive(Serialize)]
1362 : pub struct StatusResponse {
1363 : pub id: NodeId,
1364 : }
1365 :
1366 0 : #[derive(Serialize, Deserialize, Debug)]
1367 : #[serde(deny_unknown_fields)]
1368 : pub struct TenantLocationConfigRequest {
1369 : #[serde(flatten)]
1370 : pub config: LocationConfig, // as we have a flattened field, we should reject all unknown fields in it
1371 : }
1372 :
1373 0 : #[derive(Serialize, Deserialize, Debug)]
1374 : #[serde(deny_unknown_fields)]
1375 : pub struct TenantTimeTravelRequest {
1376 : pub shard_counts: Vec<ShardCount>,
1377 : }
1378 :
1379 0 : #[derive(Serialize, Deserialize, Debug)]
1380 : #[serde(deny_unknown_fields)]
1381 : pub struct TenantShardLocation {
1382 : pub shard_id: TenantShardId,
1383 : pub node_id: NodeId,
1384 : }
1385 :
1386 0 : #[derive(Serialize, Deserialize, Debug)]
1387 : #[serde(deny_unknown_fields)]
1388 : pub struct TenantLocationConfigResponse {
1389 : pub shards: Vec<TenantShardLocation>,
1390 : // If the shards' ShardCount count is >1, stripe_size will be set.
1391 : pub stripe_size: Option<ShardStripeSize>,
1392 : }
1393 :
1394 2 : #[derive(Serialize, Deserialize, Debug)]
1395 : #[serde(deny_unknown_fields)]
1396 : pub struct TenantConfigRequest {
1397 : pub tenant_id: TenantId,
1398 : #[serde(flatten)]
1399 : pub config: TenantConfig, // as we have a flattened field, we should reject all unknown fields in it
1400 : }
1401 :
1402 : impl std::ops::Deref for TenantConfigRequest {
1403 : type Target = TenantConfig;
1404 :
1405 0 : fn deref(&self) -> &Self::Target {
1406 0 : &self.config
1407 0 : }
1408 : }
1409 :
1410 : impl TenantConfigRequest {
1411 0 : pub fn new(tenant_id: TenantId) -> TenantConfigRequest {
1412 0 : let config = TenantConfig::default();
1413 0 : TenantConfigRequest { tenant_id, config }
1414 0 : }
1415 : }
1416 :
1417 3 : #[derive(Serialize, Deserialize, Debug)]
1418 : #[serde(deny_unknown_fields)]
1419 : pub struct TenantConfigPatchRequest {
1420 : pub tenant_id: TenantId,
1421 : #[serde(flatten)]
1422 : pub config: TenantConfigPatch, // as we have a flattened field, we should reject all unknown fields in it
1423 : }
1424 :
1425 0 : #[derive(Serialize, Deserialize, Debug)]
1426 : pub struct TenantWaitLsnRequest {
1427 : #[serde(flatten)]
1428 : pub timelines: HashMap<TimelineId, Lsn>,
1429 : pub timeout: Duration,
1430 : }
1431 :
1432 : /// See [`TenantState::attachment_status`] and the OpenAPI docs for context.
1433 0 : #[derive(Serialize, Deserialize, Clone)]
1434 : #[serde(tag = "slug", content = "data", rename_all = "snake_case")]
1435 : pub enum TenantAttachmentStatus {
1436 : Maybe,
1437 : Attached,
1438 : Failed { reason: String },
1439 : }
1440 :
1441 0 : #[derive(Serialize, Deserialize, Clone)]
1442 : pub struct TenantInfo {
1443 : pub id: TenantShardId,
1444 : // NB: intentionally not part of OpenAPI, we don't want to commit to a specific set of TenantState's
1445 : pub state: TenantState,
1446 : /// Sum of the size of all layer files.
1447 : /// If a layer is present in both local FS and S3, it counts only once.
1448 : pub current_physical_size: Option<u64>, // physical size is only included in `tenant_status` endpoint
1449 : pub attachment_status: TenantAttachmentStatus,
1450 : pub generation: u32,
1451 :
1452 : /// Opaque explanation if gc is being blocked.
1453 : ///
1454 : /// Only looked up for the individual tenant detail, not the listing.
1455 : #[serde(skip_serializing_if = "Option::is_none")]
1456 : pub gc_blocking: Option<String>,
1457 : }
1458 :
1459 0 : #[derive(Serialize, Deserialize, Clone)]
1460 : pub struct TenantDetails {
1461 : #[serde(flatten)]
1462 : pub tenant_info: TenantInfo,
1463 :
1464 : pub walredo: Option<WalRedoManagerStatus>,
1465 :
1466 : pub timelines: Vec<TimelineId>,
1467 : }
1468 :
1469 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone, Copy, Debug)]
1470 : pub enum TimelineArchivalState {
1471 : Archived,
1472 : Unarchived,
1473 : }
1474 :
1475 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone)]
1476 : pub enum TimelineVisibilityState {
1477 : Visible,
1478 : Invisible,
1479 : }
1480 :
1481 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone)]
1482 : pub struct TimelineArchivalConfigRequest {
1483 : pub state: TimelineArchivalState,
1484 : }
1485 :
1486 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone)]
1487 : pub struct TimelinePatchIndexPartRequest {
1488 : pub rel_size_migration: Option<RelSizeMigration>,
1489 : pub gc_compaction_last_completed_lsn: Option<Lsn>,
1490 : pub applied_gc_cutoff_lsn: Option<Lsn>,
1491 : #[serde(default)]
1492 : pub force_index_update: bool,
1493 : }
1494 :
1495 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
1496 : pub struct TimelinesInfoAndOffloaded {
1497 : pub timelines: Vec<TimelineInfo>,
1498 : pub offloaded: Vec<OffloadedTimelineInfo>,
1499 : }
1500 :
1501 : /// Analog of [`TimelineInfo`] for offloaded timelines.
1502 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
1503 : pub struct OffloadedTimelineInfo {
1504 : pub tenant_id: TenantShardId,
1505 : pub timeline_id: TimelineId,
1506 : /// Whether the timeline has a parent it has been branched off from or not
1507 : pub ancestor_timeline_id: Option<TimelineId>,
1508 : /// Whether to retain the branch lsn at the ancestor or not
1509 : pub ancestor_retain_lsn: Option<Lsn>,
1510 : /// The time point when the timeline was archived
1511 : pub archived_at: chrono::DateTime<chrono::Utc>,
1512 : }
1513 :
1514 4 : #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
1515 : #[serde(rename_all = "camelCase")]
1516 : pub enum RelSizeMigration {
1517 : /// The tenant is using the old rel_size format.
1518 : /// Note that this enum is persisted as `Option<RelSizeMigration>` in the index part, so
1519 : /// `None` is the same as `Some(RelSizeMigration::Legacy)`.
1520 : Legacy,
1521 : /// The tenant is migrating to the new rel_size format. Both old and new rel_size format are
1522 : /// persisted in the index part. The read path will read both formats and merge them.
1523 : Migrating,
1524 : /// The tenant has migrated to the new rel_size format. Only the new rel_size format is persisted
1525 : /// in the index part, and the read path will not read the old format.
1526 : Migrated,
1527 : }
1528 :
1529 : /// This represents the output of the "timeline_detail" and "timeline_list" API calls.
1530 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
1531 : pub struct TimelineInfo {
1532 : pub tenant_id: TenantShardId,
1533 : pub timeline_id: TimelineId,
1534 :
1535 : pub ancestor_timeline_id: Option<TimelineId>,
1536 : pub ancestor_lsn: Option<Lsn>,
1537 : pub last_record_lsn: Lsn,
1538 : pub prev_record_lsn: Option<Lsn>,
1539 :
1540 : /// The LSN up to which GC has advanced: older data may still exist but it is not available for clients.
1541 : /// This LSN is not suitable for deciding where to create branches etc: use [`TimelineInfo::min_readable_lsn`] instead,
1542 : /// as it is easier to reason about.
1543 : #[serde(default)]
1544 : pub applied_gc_cutoff_lsn: Lsn,
1545 :
1546 : /// The upper bound of data which is either already GC'ed, or elegible to be GC'ed at any time based on PITR interval.
1547 : /// This LSN represents the "end of history" for this timeline, and callers should use it to figure out the oldest
1548 : /// LSN at which it is legal to create a branch or ephemeral endpoint.
1549 : ///
1550 : /// Note that holders of valid LSN leases may be able to create branches and read pages earlier
1551 : /// than this LSN, but new leases may not be taken out earlier than this LSN.
1552 : #[serde(default)]
1553 : pub min_readable_lsn: Lsn,
1554 :
1555 : pub disk_consistent_lsn: Lsn,
1556 :
1557 : /// The LSN that we have succesfully uploaded to remote storage
1558 : pub remote_consistent_lsn: Lsn,
1559 :
1560 : /// The LSN that we are advertizing to safekeepers
1561 : pub remote_consistent_lsn_visible: Lsn,
1562 :
1563 : /// The LSN from the start of the root timeline (never changes)
1564 : pub initdb_lsn: Lsn,
1565 :
1566 : pub current_logical_size: u64,
1567 : pub current_logical_size_is_accurate: bool,
1568 :
1569 : pub directory_entries_counts: Vec<u64>,
1570 :
1571 : /// Sum of the size of all layer files.
1572 : /// If a layer is present in both local FS and S3, it counts only once.
1573 : pub current_physical_size: Option<u64>, // is None when timeline is Unloaded
1574 : pub current_logical_size_non_incremental: Option<u64>,
1575 :
1576 : /// How many bytes of WAL are within this branch's pitr_interval. If the pitr_interval goes
1577 : /// beyond the branch's branch point, we only count up to the branch point.
1578 : pub pitr_history_size: u64,
1579 :
1580 : /// Whether this branch's branch point is within its ancestor's PITR interval (i.e. any
1581 : /// ancestor data used by this branch would have been retained anyway). If this is false, then
1582 : /// this branch may be imposing a cost on the ancestor by causing it to retain layers that it would
1583 : /// otherwise be able to GC.
1584 : pub within_ancestor_pitr: bool,
1585 :
1586 : pub timeline_dir_layer_file_size_sum: Option<u64>,
1587 :
1588 : pub wal_source_connstr: Option<String>,
1589 : pub last_received_msg_lsn: Option<Lsn>,
1590 : /// the timestamp (in microseconds) of the last received message
1591 : pub last_received_msg_ts: Option<u128>,
1592 : pub pg_version: u32,
1593 :
1594 : pub state: TimelineState,
1595 :
1596 : pub walreceiver_status: String,
1597 :
1598 : // ALWAYS add new fields at the end of the struct with `Option` to ensure forward/backward compatibility.
1599 : // Backward compatibility: you will get a JSON not containing the newly-added field.
1600 : // Forward compatibility: a previous version of the pageserver will receive a JSON. serde::Deserialize does
1601 : // not deny unknown fields by default so it's safe to set the field to some value, though it won't be
1602 : // read.
1603 : /// Whether the timeline is archived.
1604 : pub is_archived: Option<bool>,
1605 :
1606 : /// The status of the rel_size migration.
1607 : pub rel_size_migration: Option<RelSizeMigration>,
1608 :
1609 : /// Whether the timeline is invisible in synthetic size calculations.
1610 : pub is_invisible: Option<bool>,
1611 : }
1612 :
1613 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1614 : pub struct LayerMapInfo {
1615 : pub in_memory_layers: Vec<InMemoryLayerInfo>,
1616 : pub historic_layers: Vec<HistoricLayerInfo>,
1617 : }
1618 :
1619 : /// The residence status of a layer
1620 0 : #[derive(Debug, Clone, Copy, Serialize, Deserialize)]
1621 : pub enum LayerResidenceStatus {
1622 : /// Residence status for a layer file that exists locally.
1623 : /// It may also exist on the remote, we don't care here.
1624 : Resident,
1625 : /// Residence status for a layer file that only exists on the remote.
1626 : Evicted,
1627 : }
1628 :
1629 : #[serde_as]
1630 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1631 : pub struct LayerAccessStats {
1632 : #[serde_as(as = "serde_with::TimestampMilliSeconds")]
1633 : pub access_time: SystemTime,
1634 :
1635 : #[serde_as(as = "serde_with::TimestampMilliSeconds")]
1636 : pub residence_time: SystemTime,
1637 :
1638 : pub visible: bool,
1639 : }
1640 :
1641 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1642 : #[serde(tag = "kind")]
1643 : pub enum InMemoryLayerInfo {
1644 : Open { lsn_start: Lsn },
1645 : Frozen { lsn_start: Lsn, lsn_end: Lsn },
1646 : }
1647 :
1648 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1649 : #[serde(tag = "kind")]
1650 : pub enum HistoricLayerInfo {
1651 : Delta {
1652 : layer_file_name: String,
1653 : layer_file_size: u64,
1654 :
1655 : lsn_start: Lsn,
1656 : lsn_end: Lsn,
1657 : remote: bool,
1658 : access_stats: LayerAccessStats,
1659 :
1660 : l0: bool,
1661 : },
1662 : Image {
1663 : layer_file_name: String,
1664 : layer_file_size: u64,
1665 :
1666 : lsn_start: Lsn,
1667 : remote: bool,
1668 : access_stats: LayerAccessStats,
1669 : },
1670 : }
1671 :
1672 : impl HistoricLayerInfo {
1673 0 : pub fn layer_file_name(&self) -> &str {
1674 0 : match self {
1675 : HistoricLayerInfo::Delta {
1676 0 : layer_file_name, ..
1677 0 : } => layer_file_name,
1678 : HistoricLayerInfo::Image {
1679 0 : layer_file_name, ..
1680 0 : } => layer_file_name,
1681 : }
1682 0 : }
1683 0 : pub fn is_remote(&self) -> bool {
1684 0 : match self {
1685 0 : HistoricLayerInfo::Delta { remote, .. } => *remote,
1686 0 : HistoricLayerInfo::Image { remote, .. } => *remote,
1687 : }
1688 0 : }
1689 0 : pub fn set_remote(&mut self, value: bool) {
1690 0 : let field = match self {
1691 0 : HistoricLayerInfo::Delta { remote, .. } => remote,
1692 0 : HistoricLayerInfo::Image { remote, .. } => remote,
1693 : };
1694 0 : *field = value;
1695 0 : }
1696 0 : pub fn layer_file_size(&self) -> u64 {
1697 0 : match self {
1698 : HistoricLayerInfo::Delta {
1699 0 : layer_file_size, ..
1700 0 : } => *layer_file_size,
1701 : HistoricLayerInfo::Image {
1702 0 : layer_file_size, ..
1703 0 : } => *layer_file_size,
1704 : }
1705 0 : }
1706 : }
1707 :
1708 0 : #[derive(Debug, Serialize, Deserialize)]
1709 : pub struct DownloadRemoteLayersTaskSpawnRequest {
1710 : pub max_concurrent_downloads: NonZeroUsize,
1711 : }
1712 :
1713 0 : #[derive(Debug, Serialize, Deserialize)]
1714 : pub struct IngestAuxFilesRequest {
1715 : pub aux_files: HashMap<String, String>,
1716 : }
1717 :
1718 0 : #[derive(Debug, Serialize, Deserialize)]
1719 : pub struct ListAuxFilesRequest {
1720 : pub lsn: Lsn,
1721 : }
1722 :
1723 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
1724 : pub struct DownloadRemoteLayersTaskInfo {
1725 : pub task_id: String,
1726 : pub state: DownloadRemoteLayersTaskState,
1727 : pub total_layer_count: u64, // stable once `completed`
1728 : pub successful_download_count: u64, // stable once `completed`
1729 : pub failed_download_count: u64, // stable once `completed`
1730 : }
1731 :
1732 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
1733 : pub enum DownloadRemoteLayersTaskState {
1734 : Running,
1735 : Completed,
1736 : ShutDown,
1737 : }
1738 :
1739 0 : #[derive(Debug, Serialize, Deserialize)]
1740 : pub struct TimelineGcRequest {
1741 : pub gc_horizon: Option<u64>,
1742 : }
1743 :
1744 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1745 : pub struct WalRedoManagerProcessStatus {
1746 : pub pid: u32,
1747 : }
1748 :
1749 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1750 : pub struct WalRedoManagerStatus {
1751 : pub last_redo_at: Option<chrono::DateTime<chrono::Utc>>,
1752 : pub process: Option<WalRedoManagerProcessStatus>,
1753 : }
1754 :
1755 : /// The progress of a secondary tenant.
1756 : ///
1757 : /// It is mostly useful when doing a long running download: e.g. initiating
1758 : /// a download job, timing out while waiting for it to run, and then inspecting this status to understand
1759 : /// what's happening.
1760 0 : #[derive(Default, Debug, Serialize, Deserialize, Clone)]
1761 : pub struct SecondaryProgress {
1762 : /// The remote storage LastModified time of the heatmap object we last downloaded.
1763 : pub heatmap_mtime: Option<serde_system_time::SystemTime>,
1764 :
1765 : /// The number of layers currently on-disk
1766 : pub layers_downloaded: usize,
1767 : /// The number of layers in the most recently seen heatmap
1768 : pub layers_total: usize,
1769 :
1770 : /// The number of layer bytes currently on-disk
1771 : pub bytes_downloaded: u64,
1772 : /// The number of layer bytes in the most recently seen heatmap
1773 : pub bytes_total: u64,
1774 : }
1775 :
1776 0 : #[derive(Serialize, Deserialize, Debug)]
1777 : pub struct TenantScanRemoteStorageShard {
1778 : pub tenant_shard_id: TenantShardId,
1779 : pub generation: Option<u32>,
1780 : pub stripe_size: Option<ShardStripeSize>,
1781 : }
1782 :
1783 0 : #[derive(Serialize, Deserialize, Debug, Default)]
1784 : pub struct TenantScanRemoteStorageResponse {
1785 : pub shards: Vec<TenantScanRemoteStorageShard>,
1786 : }
1787 :
1788 0 : #[derive(Serialize, Deserialize, Debug, Clone)]
1789 : #[serde(rename_all = "snake_case")]
1790 : pub enum TenantSorting {
1791 : /// Total size of layers on local disk for all timelines in a shard.
1792 : ResidentSize,
1793 : /// The logical size of the largest timeline within a _tenant_ (not shard). Only tracked on
1794 : /// shard 0, contains the sum across all shards.
1795 : MaxLogicalSize,
1796 : /// The logical size of the largest timeline within a _tenant_ (not shard), divided by number of
1797 : /// shards. Only tracked on shard 0, and estimates the per-shard logical size.
1798 : MaxLogicalSizePerShard,
1799 : }
1800 :
1801 : impl Default for TenantSorting {
1802 0 : fn default() -> Self {
1803 0 : Self::ResidentSize
1804 0 : }
1805 : }
1806 :
1807 0 : #[derive(Serialize, Deserialize, Debug, Clone)]
1808 : pub struct TopTenantShardsRequest {
1809 : // How would you like to sort the tenants?
1810 : pub order_by: TenantSorting,
1811 :
1812 : // How many results?
1813 : pub limit: usize,
1814 :
1815 : // Omit tenants with more than this many shards (e.g. if this is the max number of shards
1816 : // that the caller would ever split to)
1817 : pub where_shards_lt: Option<ShardCount>,
1818 :
1819 : // Omit tenants where the ordering metric is less than this (this is an optimization to
1820 : // let us quickly exclude numerous tiny shards)
1821 : pub where_gt: Option<u64>,
1822 : }
1823 :
1824 0 : #[derive(Serialize, Deserialize, Debug, PartialEq, Eq)]
1825 : pub struct TopTenantShardItem {
1826 : pub id: TenantShardId,
1827 :
1828 : /// Total size of layers on local disk for all timelines in this shard.
1829 : pub resident_size: u64,
1830 :
1831 : /// Total size of layers in remote storage for all timelines in this shard.
1832 : pub physical_size: u64,
1833 :
1834 : /// The largest logical size of a timeline within this _tenant_ (not shard). This is only
1835 : /// tracked on shard 0, and contains the sum of the logical size across all shards.
1836 : pub max_logical_size: u64,
1837 :
1838 : /// The largest logical size of a timeline within this _tenant_ (not shard) divided by number of
1839 : /// shards. This is only tracked on shard 0, and is only an estimate as we divide it evenly by
1840 : /// shard count, rounded up.
1841 : pub max_logical_size_per_shard: u64,
1842 : }
1843 :
1844 0 : #[derive(Serialize, Deserialize, Debug, Default)]
1845 : pub struct TopTenantShardsResponse {
1846 : pub shards: Vec<TopTenantShardItem>,
1847 : }
1848 :
1849 : pub mod virtual_file {
1850 :
1851 : #[derive(
1852 : Copy,
1853 : Clone,
1854 : PartialEq,
1855 : Eq,
1856 : Hash,
1857 0 : strum_macros::EnumString,
1858 : strum_macros::Display,
1859 0 : serde_with::DeserializeFromStr,
1860 : serde_with::SerializeDisplay,
1861 : Debug,
1862 : )]
1863 : #[strum(serialize_all = "kebab-case")]
1864 : pub enum IoEngineKind {
1865 : StdFs,
1866 : #[cfg(target_os = "linux")]
1867 : TokioEpollUring,
1868 : }
1869 :
1870 : /// Direct IO modes for a pageserver.
1871 : #[derive(
1872 : Copy,
1873 : Clone,
1874 : PartialEq,
1875 : Eq,
1876 : Hash,
1877 0 : strum_macros::EnumString,
1878 0 : strum_macros::EnumIter,
1879 : strum_macros::Display,
1880 0 : serde_with::DeserializeFromStr,
1881 : serde_with::SerializeDisplay,
1882 : Debug,
1883 : )]
1884 : #[strum(serialize_all = "kebab-case")]
1885 : #[repr(u8)]
1886 : pub enum IoMode {
1887 : /// Uses buffered IO.
1888 : Buffered,
1889 : /// Uses direct IO for reads only.
1890 : Direct,
1891 : /// Use direct IO for reads and writes.
1892 : DirectRw,
1893 : }
1894 :
1895 : impl IoMode {
1896 247 : pub fn preferred() -> Self {
1897 247 : IoMode::DirectRw
1898 247 : }
1899 : }
1900 :
1901 : impl TryFrom<u8> for IoMode {
1902 : type Error = u8;
1903 :
1904 2576 : fn try_from(value: u8) -> Result<Self, Self::Error> {
1905 2576 : Ok(match value {
1906 2576 : v if v == (IoMode::Buffered as u8) => IoMode::Buffered,
1907 2576 : v if v == (IoMode::Direct as u8) => IoMode::Direct,
1908 2576 : v if v == (IoMode::DirectRw as u8) => IoMode::DirectRw,
1909 0 : x => return Err(x),
1910 : })
1911 2576 : }
1912 : }
1913 : }
1914 :
1915 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1916 : pub struct ScanDisposableKeysResponse {
1917 : pub disposable_count: usize,
1918 : pub not_disposable_count: usize,
1919 : }
1920 :
1921 : // Wrapped in libpq CopyData
1922 : #[derive(PartialEq, Eq, Debug)]
1923 : pub enum PagestreamFeMessage {
1924 : Exists(PagestreamExistsRequest),
1925 : Nblocks(PagestreamNblocksRequest),
1926 : GetPage(PagestreamGetPageRequest),
1927 : DbSize(PagestreamDbSizeRequest),
1928 : GetSlruSegment(PagestreamGetSlruSegmentRequest),
1929 : #[cfg(feature = "testing")]
1930 : Test(PagestreamTestRequest),
1931 : }
1932 :
1933 : // Wrapped in libpq CopyData
1934 : #[derive(strum_macros::EnumProperty)]
1935 : pub enum PagestreamBeMessage {
1936 : Exists(PagestreamExistsResponse),
1937 : Nblocks(PagestreamNblocksResponse),
1938 : GetPage(PagestreamGetPageResponse),
1939 : Error(PagestreamErrorResponse),
1940 : DbSize(PagestreamDbSizeResponse),
1941 : GetSlruSegment(PagestreamGetSlruSegmentResponse),
1942 : #[cfg(feature = "testing")]
1943 : Test(PagestreamTestResponse),
1944 : }
1945 :
1946 : // Keep in sync with `pagestore_client.h`
1947 : #[repr(u8)]
1948 : enum PagestreamFeMessageTag {
1949 : Exists = 0,
1950 : Nblocks = 1,
1951 : GetPage = 2,
1952 : DbSize = 3,
1953 : GetSlruSegment = 4,
1954 : /* future tags above this line */
1955 : /// For testing purposes, not available in production.
1956 : #[cfg(feature = "testing")]
1957 : Test = 99,
1958 : }
1959 :
1960 : // Keep in sync with `pagestore_client.h`
1961 : #[repr(u8)]
1962 : enum PagestreamBeMessageTag {
1963 : Exists = 100,
1964 : Nblocks = 101,
1965 : GetPage = 102,
1966 : Error = 103,
1967 : DbSize = 104,
1968 : GetSlruSegment = 105,
1969 : /* future tags above this line */
1970 : /// For testing purposes, not available in production.
1971 : #[cfg(feature = "testing")]
1972 : Test = 199,
1973 : }
1974 :
1975 : impl TryFrom<u8> for PagestreamFeMessageTag {
1976 : type Error = u8;
1977 4 : fn try_from(value: u8) -> Result<Self, u8> {
1978 4 : match value {
1979 1 : 0 => Ok(PagestreamFeMessageTag::Exists),
1980 1 : 1 => Ok(PagestreamFeMessageTag::Nblocks),
1981 1 : 2 => Ok(PagestreamFeMessageTag::GetPage),
1982 1 : 3 => Ok(PagestreamFeMessageTag::DbSize),
1983 0 : 4 => Ok(PagestreamFeMessageTag::GetSlruSegment),
1984 : #[cfg(feature = "testing")]
1985 0 : 99 => Ok(PagestreamFeMessageTag::Test),
1986 0 : _ => Err(value),
1987 : }
1988 4 : }
1989 : }
1990 :
1991 : impl TryFrom<u8> for PagestreamBeMessageTag {
1992 : type Error = u8;
1993 0 : fn try_from(value: u8) -> Result<Self, u8> {
1994 0 : match value {
1995 0 : 100 => Ok(PagestreamBeMessageTag::Exists),
1996 0 : 101 => Ok(PagestreamBeMessageTag::Nblocks),
1997 0 : 102 => Ok(PagestreamBeMessageTag::GetPage),
1998 0 : 103 => Ok(PagestreamBeMessageTag::Error),
1999 0 : 104 => Ok(PagestreamBeMessageTag::DbSize),
2000 0 : 105 => Ok(PagestreamBeMessageTag::GetSlruSegment),
2001 : #[cfg(feature = "testing")]
2002 0 : 199 => Ok(PagestreamBeMessageTag::Test),
2003 0 : _ => Err(value),
2004 : }
2005 0 : }
2006 : }
2007 :
2008 : // A GetPage request contains two LSN values:
2009 : //
2010 : // request_lsn: Get the page version at this point in time. Lsn::Max is a special value that means
2011 : // "get the latest version present". It's used by the primary server, which knows that no one else
2012 : // is writing WAL. 'not_modified_since' must be set to a proper value even if request_lsn is
2013 : // Lsn::Max. Standby servers use the current replay LSN as the request LSN.
2014 : //
2015 : // not_modified_since: Hint to the pageserver that the client knows that the page has not been
2016 : // modified between 'not_modified_since' and the request LSN. It's always correct to set
2017 : // 'not_modified_since equal' to 'request_lsn' (unless Lsn::Max is used as the 'request_lsn'), but
2018 : // passing an earlier LSN can speed up the request, by allowing the pageserver to process the
2019 : // request without waiting for 'request_lsn' to arrive.
2020 : //
2021 : // The now-defunct V1 interface contained only one LSN, and a boolean 'latest' flag. The V1 interface was
2022 : // sufficient for the primary; the 'lsn' was equivalent to the 'not_modified_since' value, and
2023 : // 'latest' was set to true. The V2 interface was added because there was no correct way for a
2024 : // standby to request a page at a particular non-latest LSN, and also include the
2025 : // 'not_modified_since' hint. That led to an awkward choice of either using an old LSN in the
2026 : // request, if the standby knows that the page hasn't been modified since, and risk getting an error
2027 : // if that LSN has fallen behind the GC horizon, or requesting the current replay LSN, which could
2028 : // require the pageserver unnecessarily to wait for the WAL to arrive up to that point. The new V2
2029 : // interface allows sending both LSNs, and let the pageserver do the right thing. There was no
2030 : // difference in the responses between V1 and V2.
2031 : //
2032 : // V3 version of protocol adds request ID to all requests. This request ID is also included in response
2033 : // as well as other fields from requests, which allows to verify that we receive response for our request.
2034 : // We copy fields from request to response to make checking more reliable: request ID is formed from process ID
2035 : // and local counter, so in principle there can be duplicated requests IDs if process PID is reused.
2036 : //
2037 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
2038 : pub enum PagestreamProtocolVersion {
2039 : V2,
2040 : V3,
2041 : }
2042 :
2043 : pub type RequestId = u64;
2044 :
2045 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
2046 : pub struct PagestreamRequest {
2047 : pub reqid: RequestId,
2048 : pub request_lsn: Lsn,
2049 : pub not_modified_since: Lsn,
2050 : }
2051 :
2052 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
2053 : pub struct PagestreamExistsRequest {
2054 : pub hdr: PagestreamRequest,
2055 : pub rel: RelTag,
2056 : }
2057 :
2058 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
2059 : pub struct PagestreamNblocksRequest {
2060 : pub hdr: PagestreamRequest,
2061 : pub rel: RelTag,
2062 : }
2063 :
2064 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
2065 : pub struct PagestreamGetPageRequest {
2066 : pub hdr: PagestreamRequest,
2067 : pub rel: RelTag,
2068 : pub blkno: u32,
2069 : }
2070 :
2071 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
2072 : pub struct PagestreamDbSizeRequest {
2073 : pub hdr: PagestreamRequest,
2074 : pub dbnode: u32,
2075 : }
2076 :
2077 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
2078 : pub struct PagestreamGetSlruSegmentRequest {
2079 : pub hdr: PagestreamRequest,
2080 : pub kind: u8,
2081 : pub segno: u32,
2082 : }
2083 :
2084 : #[derive(Debug)]
2085 : pub struct PagestreamExistsResponse {
2086 : pub req: PagestreamExistsRequest,
2087 : pub exists: bool,
2088 : }
2089 :
2090 : #[derive(Debug)]
2091 : pub struct PagestreamNblocksResponse {
2092 : pub req: PagestreamNblocksRequest,
2093 : pub n_blocks: u32,
2094 : }
2095 :
2096 : #[derive(Debug)]
2097 : pub struct PagestreamGetPageResponse {
2098 : pub req: PagestreamGetPageRequest,
2099 : pub page: Bytes,
2100 : }
2101 :
2102 : #[derive(Debug)]
2103 : pub struct PagestreamGetSlruSegmentResponse {
2104 : pub req: PagestreamGetSlruSegmentRequest,
2105 : pub segment: Bytes,
2106 : }
2107 :
2108 : #[derive(Debug)]
2109 : pub struct PagestreamErrorResponse {
2110 : pub req: PagestreamRequest,
2111 : pub message: String,
2112 : }
2113 :
2114 : #[derive(Debug)]
2115 : pub struct PagestreamDbSizeResponse {
2116 : pub req: PagestreamDbSizeRequest,
2117 : pub db_size: i64,
2118 : }
2119 :
2120 : #[cfg(feature = "testing")]
2121 : #[derive(Debug, PartialEq, Eq, Clone)]
2122 : pub struct PagestreamTestRequest {
2123 : pub hdr: PagestreamRequest,
2124 : pub batch_key: u64,
2125 : pub message: String,
2126 : }
2127 :
2128 : #[cfg(feature = "testing")]
2129 : #[derive(Debug)]
2130 : pub struct PagestreamTestResponse {
2131 : pub req: PagestreamTestRequest,
2132 : }
2133 :
2134 : // This is a cut-down version of TenantHistorySize from the pageserver crate, omitting fields
2135 : // that require pageserver-internal types. It is sufficient to get the total size.
2136 0 : #[derive(Serialize, Deserialize, Debug)]
2137 : pub struct TenantHistorySize {
2138 : pub id: TenantId,
2139 : /// Size is a mixture of WAL and logical size, so the unit is bytes.
2140 : ///
2141 : /// Will be none if `?inputs_only=true` was given.
2142 : pub size: Option<u64>,
2143 : }
2144 :
2145 : impl PagestreamFeMessage {
2146 : /// Serialize a compute -> pageserver message. This is currently only used in testing
2147 : /// tools. Always uses protocol version 3.
2148 4 : pub fn serialize(&self) -> Bytes {
2149 4 : let mut bytes = BytesMut::new();
2150 4 :
2151 4 : match self {
2152 1 : Self::Exists(req) => {
2153 1 : bytes.put_u8(PagestreamFeMessageTag::Exists as u8);
2154 1 : bytes.put_u64(req.hdr.reqid);
2155 1 : bytes.put_u64(req.hdr.request_lsn.0);
2156 1 : bytes.put_u64(req.hdr.not_modified_since.0);
2157 1 : bytes.put_u32(req.rel.spcnode);
2158 1 : bytes.put_u32(req.rel.dbnode);
2159 1 : bytes.put_u32(req.rel.relnode);
2160 1 : bytes.put_u8(req.rel.forknum);
2161 1 : }
2162 :
2163 1 : Self::Nblocks(req) => {
2164 1 : bytes.put_u8(PagestreamFeMessageTag::Nblocks as u8);
2165 1 : bytes.put_u64(req.hdr.reqid);
2166 1 : bytes.put_u64(req.hdr.request_lsn.0);
2167 1 : bytes.put_u64(req.hdr.not_modified_since.0);
2168 1 : bytes.put_u32(req.rel.spcnode);
2169 1 : bytes.put_u32(req.rel.dbnode);
2170 1 : bytes.put_u32(req.rel.relnode);
2171 1 : bytes.put_u8(req.rel.forknum);
2172 1 : }
2173 :
2174 1 : Self::GetPage(req) => {
2175 1 : bytes.put_u8(PagestreamFeMessageTag::GetPage as u8);
2176 1 : bytes.put_u64(req.hdr.reqid);
2177 1 : bytes.put_u64(req.hdr.request_lsn.0);
2178 1 : bytes.put_u64(req.hdr.not_modified_since.0);
2179 1 : bytes.put_u32(req.rel.spcnode);
2180 1 : bytes.put_u32(req.rel.dbnode);
2181 1 : bytes.put_u32(req.rel.relnode);
2182 1 : bytes.put_u8(req.rel.forknum);
2183 1 : bytes.put_u32(req.blkno);
2184 1 : }
2185 :
2186 1 : Self::DbSize(req) => {
2187 1 : bytes.put_u8(PagestreamFeMessageTag::DbSize as u8);
2188 1 : bytes.put_u64(req.hdr.reqid);
2189 1 : bytes.put_u64(req.hdr.request_lsn.0);
2190 1 : bytes.put_u64(req.hdr.not_modified_since.0);
2191 1 : bytes.put_u32(req.dbnode);
2192 1 : }
2193 :
2194 0 : Self::GetSlruSegment(req) => {
2195 0 : bytes.put_u8(PagestreamFeMessageTag::GetSlruSegment as u8);
2196 0 : bytes.put_u64(req.hdr.reqid);
2197 0 : bytes.put_u64(req.hdr.request_lsn.0);
2198 0 : bytes.put_u64(req.hdr.not_modified_since.0);
2199 0 : bytes.put_u8(req.kind);
2200 0 : bytes.put_u32(req.segno);
2201 0 : }
2202 : #[cfg(feature = "testing")]
2203 0 : Self::Test(req) => {
2204 0 : bytes.put_u8(PagestreamFeMessageTag::Test as u8);
2205 0 : bytes.put_u64(req.hdr.reqid);
2206 0 : bytes.put_u64(req.hdr.request_lsn.0);
2207 0 : bytes.put_u64(req.hdr.not_modified_since.0);
2208 0 : bytes.put_u64(req.batch_key);
2209 0 : let message = req.message.as_bytes();
2210 0 : bytes.put_u64(message.len() as u64);
2211 0 : bytes.put_slice(message);
2212 0 : }
2213 : }
2214 :
2215 4 : bytes.into()
2216 4 : }
2217 :
2218 4 : pub fn parse<R: std::io::Read>(
2219 4 : body: &mut R,
2220 4 : protocol_version: PagestreamProtocolVersion,
2221 4 : ) -> anyhow::Result<PagestreamFeMessage> {
2222 : // these correspond to the NeonMessageTag enum in pagestore_client.h
2223 : //
2224 : // TODO: consider using protobuf or serde bincode for less error prone
2225 : // serialization.
2226 4 : let msg_tag = body.read_u8()?;
2227 4 : let (reqid, request_lsn, not_modified_since) = match protocol_version {
2228 : PagestreamProtocolVersion::V2 => (
2229 : 0,
2230 0 : Lsn::from(body.read_u64::<BigEndian>()?),
2231 0 : Lsn::from(body.read_u64::<BigEndian>()?),
2232 : ),
2233 : PagestreamProtocolVersion::V3 => (
2234 4 : body.read_u64::<BigEndian>()?,
2235 4 : Lsn::from(body.read_u64::<BigEndian>()?),
2236 4 : Lsn::from(body.read_u64::<BigEndian>()?),
2237 : ),
2238 : };
2239 :
2240 4 : match PagestreamFeMessageTag::try_from(msg_tag)
2241 4 : .map_err(|tag: u8| anyhow::anyhow!("invalid tag {tag}"))?
2242 : {
2243 : PagestreamFeMessageTag::Exists => {
2244 : Ok(PagestreamFeMessage::Exists(PagestreamExistsRequest {
2245 1 : hdr: PagestreamRequest {
2246 1 : reqid,
2247 1 : request_lsn,
2248 1 : not_modified_since,
2249 1 : },
2250 1 : rel: RelTag {
2251 1 : spcnode: body.read_u32::<BigEndian>()?,
2252 1 : dbnode: body.read_u32::<BigEndian>()?,
2253 1 : relnode: body.read_u32::<BigEndian>()?,
2254 1 : forknum: body.read_u8()?,
2255 : },
2256 : }))
2257 : }
2258 : PagestreamFeMessageTag::Nblocks => {
2259 : Ok(PagestreamFeMessage::Nblocks(PagestreamNblocksRequest {
2260 1 : hdr: PagestreamRequest {
2261 1 : reqid,
2262 1 : request_lsn,
2263 1 : not_modified_since,
2264 1 : },
2265 1 : rel: RelTag {
2266 1 : spcnode: body.read_u32::<BigEndian>()?,
2267 1 : dbnode: body.read_u32::<BigEndian>()?,
2268 1 : relnode: body.read_u32::<BigEndian>()?,
2269 1 : forknum: body.read_u8()?,
2270 : },
2271 : }))
2272 : }
2273 : PagestreamFeMessageTag::GetPage => {
2274 : Ok(PagestreamFeMessage::GetPage(PagestreamGetPageRequest {
2275 1 : hdr: PagestreamRequest {
2276 1 : reqid,
2277 1 : request_lsn,
2278 1 : not_modified_since,
2279 1 : },
2280 1 : rel: RelTag {
2281 1 : spcnode: body.read_u32::<BigEndian>()?,
2282 1 : dbnode: body.read_u32::<BigEndian>()?,
2283 1 : relnode: body.read_u32::<BigEndian>()?,
2284 1 : forknum: body.read_u8()?,
2285 : },
2286 1 : blkno: body.read_u32::<BigEndian>()?,
2287 : }))
2288 : }
2289 : PagestreamFeMessageTag::DbSize => {
2290 : Ok(PagestreamFeMessage::DbSize(PagestreamDbSizeRequest {
2291 1 : hdr: PagestreamRequest {
2292 1 : reqid,
2293 1 : request_lsn,
2294 1 : not_modified_since,
2295 1 : },
2296 1 : dbnode: body.read_u32::<BigEndian>()?,
2297 : }))
2298 : }
2299 : PagestreamFeMessageTag::GetSlruSegment => Ok(PagestreamFeMessage::GetSlruSegment(
2300 : PagestreamGetSlruSegmentRequest {
2301 0 : hdr: PagestreamRequest {
2302 0 : reqid,
2303 0 : request_lsn,
2304 0 : not_modified_since,
2305 0 : },
2306 0 : kind: body.read_u8()?,
2307 0 : segno: body.read_u32::<BigEndian>()?,
2308 : },
2309 : )),
2310 : #[cfg(feature = "testing")]
2311 : PagestreamFeMessageTag::Test => Ok(PagestreamFeMessage::Test(PagestreamTestRequest {
2312 0 : hdr: PagestreamRequest {
2313 0 : reqid,
2314 0 : request_lsn,
2315 0 : not_modified_since,
2316 0 : },
2317 0 : batch_key: body.read_u64::<BigEndian>()?,
2318 : message: {
2319 0 : let len = body.read_u64::<BigEndian>()?;
2320 0 : let mut buf = vec![0; len as usize];
2321 0 : body.read_exact(&mut buf)?;
2322 0 : String::from_utf8(buf)?
2323 : },
2324 : })),
2325 : }
2326 4 : }
2327 : }
2328 :
2329 : impl PagestreamBeMessage {
2330 0 : pub fn serialize(&self, protocol_version: PagestreamProtocolVersion) -> Bytes {
2331 0 : let mut bytes = BytesMut::new();
2332 :
2333 : use PagestreamBeMessageTag as Tag;
2334 0 : match protocol_version {
2335 : PagestreamProtocolVersion::V2 => {
2336 0 : match self {
2337 0 : Self::Exists(resp) => {
2338 0 : bytes.put_u8(Tag::Exists as u8);
2339 0 : bytes.put_u8(resp.exists as u8);
2340 0 : }
2341 :
2342 0 : Self::Nblocks(resp) => {
2343 0 : bytes.put_u8(Tag::Nblocks as u8);
2344 0 : bytes.put_u32(resp.n_blocks);
2345 0 : }
2346 :
2347 0 : Self::GetPage(resp) => {
2348 0 : bytes.put_u8(Tag::GetPage as u8);
2349 0 : bytes.put(&resp.page[..])
2350 : }
2351 :
2352 0 : Self::Error(resp) => {
2353 0 : bytes.put_u8(Tag::Error as u8);
2354 0 : bytes.put(resp.message.as_bytes());
2355 0 : bytes.put_u8(0); // null terminator
2356 0 : }
2357 0 : Self::DbSize(resp) => {
2358 0 : bytes.put_u8(Tag::DbSize as u8);
2359 0 : bytes.put_i64(resp.db_size);
2360 0 : }
2361 :
2362 0 : Self::GetSlruSegment(resp) => {
2363 0 : bytes.put_u8(Tag::GetSlruSegment as u8);
2364 0 : bytes.put_u32((resp.segment.len() / BLCKSZ as usize) as u32);
2365 0 : bytes.put(&resp.segment[..]);
2366 0 : }
2367 :
2368 : #[cfg(feature = "testing")]
2369 0 : Self::Test(resp) => {
2370 0 : bytes.put_u8(Tag::Test as u8);
2371 0 : bytes.put_u64(resp.req.batch_key);
2372 0 : let message = resp.req.message.as_bytes();
2373 0 : bytes.put_u64(message.len() as u64);
2374 0 : bytes.put_slice(message);
2375 0 : }
2376 : }
2377 : }
2378 : PagestreamProtocolVersion::V3 => {
2379 0 : match self {
2380 0 : Self::Exists(resp) => {
2381 0 : bytes.put_u8(Tag::Exists as u8);
2382 0 : bytes.put_u64(resp.req.hdr.reqid);
2383 0 : bytes.put_u64(resp.req.hdr.request_lsn.0);
2384 0 : bytes.put_u64(resp.req.hdr.not_modified_since.0);
2385 0 : bytes.put_u32(resp.req.rel.spcnode);
2386 0 : bytes.put_u32(resp.req.rel.dbnode);
2387 0 : bytes.put_u32(resp.req.rel.relnode);
2388 0 : bytes.put_u8(resp.req.rel.forknum);
2389 0 : bytes.put_u8(resp.exists as u8);
2390 0 : }
2391 :
2392 0 : Self::Nblocks(resp) => {
2393 0 : bytes.put_u8(Tag::Nblocks as u8);
2394 0 : bytes.put_u64(resp.req.hdr.reqid);
2395 0 : bytes.put_u64(resp.req.hdr.request_lsn.0);
2396 0 : bytes.put_u64(resp.req.hdr.not_modified_since.0);
2397 0 : bytes.put_u32(resp.req.rel.spcnode);
2398 0 : bytes.put_u32(resp.req.rel.dbnode);
2399 0 : bytes.put_u32(resp.req.rel.relnode);
2400 0 : bytes.put_u8(resp.req.rel.forknum);
2401 0 : bytes.put_u32(resp.n_blocks);
2402 0 : }
2403 :
2404 0 : Self::GetPage(resp) => {
2405 0 : bytes.put_u8(Tag::GetPage as u8);
2406 0 : bytes.put_u64(resp.req.hdr.reqid);
2407 0 : bytes.put_u64(resp.req.hdr.request_lsn.0);
2408 0 : bytes.put_u64(resp.req.hdr.not_modified_since.0);
2409 0 : bytes.put_u32(resp.req.rel.spcnode);
2410 0 : bytes.put_u32(resp.req.rel.dbnode);
2411 0 : bytes.put_u32(resp.req.rel.relnode);
2412 0 : bytes.put_u8(resp.req.rel.forknum);
2413 0 : bytes.put_u32(resp.req.blkno);
2414 0 : bytes.put(&resp.page[..])
2415 : }
2416 :
2417 0 : Self::Error(resp) => {
2418 0 : bytes.put_u8(Tag::Error as u8);
2419 0 : bytes.put_u64(resp.req.reqid);
2420 0 : bytes.put_u64(resp.req.request_lsn.0);
2421 0 : bytes.put_u64(resp.req.not_modified_since.0);
2422 0 : bytes.put(resp.message.as_bytes());
2423 0 : bytes.put_u8(0); // null terminator
2424 0 : }
2425 0 : Self::DbSize(resp) => {
2426 0 : bytes.put_u8(Tag::DbSize as u8);
2427 0 : bytes.put_u64(resp.req.hdr.reqid);
2428 0 : bytes.put_u64(resp.req.hdr.request_lsn.0);
2429 0 : bytes.put_u64(resp.req.hdr.not_modified_since.0);
2430 0 : bytes.put_u32(resp.req.dbnode);
2431 0 : bytes.put_i64(resp.db_size);
2432 0 : }
2433 :
2434 0 : Self::GetSlruSegment(resp) => {
2435 0 : bytes.put_u8(Tag::GetSlruSegment as u8);
2436 0 : bytes.put_u64(resp.req.hdr.reqid);
2437 0 : bytes.put_u64(resp.req.hdr.request_lsn.0);
2438 0 : bytes.put_u64(resp.req.hdr.not_modified_since.0);
2439 0 : bytes.put_u8(resp.req.kind);
2440 0 : bytes.put_u32(resp.req.segno);
2441 0 : bytes.put_u32((resp.segment.len() / BLCKSZ as usize) as u32);
2442 0 : bytes.put(&resp.segment[..]);
2443 0 : }
2444 :
2445 : #[cfg(feature = "testing")]
2446 0 : Self::Test(resp) => {
2447 0 : bytes.put_u8(Tag::Test as u8);
2448 0 : bytes.put_u64(resp.req.hdr.reqid);
2449 0 : bytes.put_u64(resp.req.hdr.request_lsn.0);
2450 0 : bytes.put_u64(resp.req.hdr.not_modified_since.0);
2451 0 : bytes.put_u64(resp.req.batch_key);
2452 0 : let message = resp.req.message.as_bytes();
2453 0 : bytes.put_u64(message.len() as u64);
2454 0 : bytes.put_slice(message);
2455 0 : }
2456 : }
2457 : }
2458 : }
2459 0 : bytes.into()
2460 0 : }
2461 :
2462 0 : pub fn deserialize(buf: Bytes) -> anyhow::Result<Self> {
2463 0 : let mut buf = buf.reader();
2464 0 : let msg_tag = buf.read_u8()?;
2465 :
2466 : use PagestreamBeMessageTag as Tag;
2467 0 : let ok =
2468 0 : match Tag::try_from(msg_tag).map_err(|tag: u8| anyhow::anyhow!("invalid tag {tag}"))? {
2469 : Tag::Exists => {
2470 0 : let reqid = buf.read_u64::<BigEndian>()?;
2471 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2472 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2473 0 : let rel = RelTag {
2474 0 : spcnode: buf.read_u32::<BigEndian>()?,
2475 0 : dbnode: buf.read_u32::<BigEndian>()?,
2476 0 : relnode: buf.read_u32::<BigEndian>()?,
2477 0 : forknum: buf.read_u8()?,
2478 : };
2479 0 : let exists = buf.read_u8()? != 0;
2480 0 : Self::Exists(PagestreamExistsResponse {
2481 0 : req: PagestreamExistsRequest {
2482 0 : hdr: PagestreamRequest {
2483 0 : reqid,
2484 0 : request_lsn,
2485 0 : not_modified_since,
2486 0 : },
2487 0 : rel,
2488 0 : },
2489 0 : exists,
2490 0 : })
2491 : }
2492 : Tag::Nblocks => {
2493 0 : let reqid = buf.read_u64::<BigEndian>()?;
2494 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2495 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2496 0 : let rel = RelTag {
2497 0 : spcnode: buf.read_u32::<BigEndian>()?,
2498 0 : dbnode: buf.read_u32::<BigEndian>()?,
2499 0 : relnode: buf.read_u32::<BigEndian>()?,
2500 0 : forknum: buf.read_u8()?,
2501 : };
2502 0 : let n_blocks = buf.read_u32::<BigEndian>()?;
2503 0 : Self::Nblocks(PagestreamNblocksResponse {
2504 0 : req: PagestreamNblocksRequest {
2505 0 : hdr: PagestreamRequest {
2506 0 : reqid,
2507 0 : request_lsn,
2508 0 : not_modified_since,
2509 0 : },
2510 0 : rel,
2511 0 : },
2512 0 : n_blocks,
2513 0 : })
2514 : }
2515 : Tag::GetPage => {
2516 0 : let reqid = buf.read_u64::<BigEndian>()?;
2517 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2518 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2519 0 : let rel = RelTag {
2520 0 : spcnode: buf.read_u32::<BigEndian>()?,
2521 0 : dbnode: buf.read_u32::<BigEndian>()?,
2522 0 : relnode: buf.read_u32::<BigEndian>()?,
2523 0 : forknum: buf.read_u8()?,
2524 : };
2525 0 : let blkno = buf.read_u32::<BigEndian>()?;
2526 0 : let mut page = vec![0; 8192]; // TODO: use MaybeUninit
2527 0 : buf.read_exact(&mut page)?;
2528 0 : Self::GetPage(PagestreamGetPageResponse {
2529 0 : req: PagestreamGetPageRequest {
2530 0 : hdr: PagestreamRequest {
2531 0 : reqid,
2532 0 : request_lsn,
2533 0 : not_modified_since,
2534 0 : },
2535 0 : rel,
2536 0 : blkno,
2537 0 : },
2538 0 : page: page.into(),
2539 0 : })
2540 : }
2541 : Tag::Error => {
2542 0 : let reqid = buf.read_u64::<BigEndian>()?;
2543 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2544 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2545 0 : let mut msg = Vec::new();
2546 0 : buf.read_until(0, &mut msg)?;
2547 0 : let cstring = std::ffi::CString::from_vec_with_nul(msg)?;
2548 0 : let rust_str = cstring.to_str()?;
2549 0 : Self::Error(PagestreamErrorResponse {
2550 0 : req: PagestreamRequest {
2551 0 : reqid,
2552 0 : request_lsn,
2553 0 : not_modified_since,
2554 0 : },
2555 0 : message: rust_str.to_owned(),
2556 0 : })
2557 : }
2558 : Tag::DbSize => {
2559 0 : let reqid = buf.read_u64::<BigEndian>()?;
2560 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2561 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2562 0 : let dbnode = buf.read_u32::<BigEndian>()?;
2563 0 : let db_size = buf.read_i64::<BigEndian>()?;
2564 0 : Self::DbSize(PagestreamDbSizeResponse {
2565 0 : req: PagestreamDbSizeRequest {
2566 0 : hdr: PagestreamRequest {
2567 0 : reqid,
2568 0 : request_lsn,
2569 0 : not_modified_since,
2570 0 : },
2571 0 : dbnode,
2572 0 : },
2573 0 : db_size,
2574 0 : })
2575 : }
2576 : Tag::GetSlruSegment => {
2577 0 : let reqid = buf.read_u64::<BigEndian>()?;
2578 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2579 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2580 0 : let kind = buf.read_u8()?;
2581 0 : let segno = buf.read_u32::<BigEndian>()?;
2582 0 : let n_blocks = buf.read_u32::<BigEndian>()?;
2583 0 : let mut segment = vec![0; n_blocks as usize * BLCKSZ as usize];
2584 0 : buf.read_exact(&mut segment)?;
2585 0 : Self::GetSlruSegment(PagestreamGetSlruSegmentResponse {
2586 0 : req: PagestreamGetSlruSegmentRequest {
2587 0 : hdr: PagestreamRequest {
2588 0 : reqid,
2589 0 : request_lsn,
2590 0 : not_modified_since,
2591 0 : },
2592 0 : kind,
2593 0 : segno,
2594 0 : },
2595 0 : segment: segment.into(),
2596 0 : })
2597 : }
2598 : #[cfg(feature = "testing")]
2599 : Tag::Test => {
2600 0 : let reqid = buf.read_u64::<BigEndian>()?;
2601 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2602 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2603 0 : let batch_key = buf.read_u64::<BigEndian>()?;
2604 0 : let len = buf.read_u64::<BigEndian>()?;
2605 0 : let mut msg = vec![0; len as usize];
2606 0 : buf.read_exact(&mut msg)?;
2607 0 : let message = String::from_utf8(msg)?;
2608 0 : Self::Test(PagestreamTestResponse {
2609 0 : req: PagestreamTestRequest {
2610 0 : hdr: PagestreamRequest {
2611 0 : reqid,
2612 0 : request_lsn,
2613 0 : not_modified_since,
2614 0 : },
2615 0 : batch_key,
2616 0 : message,
2617 0 : },
2618 0 : })
2619 : }
2620 : };
2621 0 : let remaining = buf.into_inner();
2622 0 : if !remaining.is_empty() {
2623 0 : anyhow::bail!(
2624 0 : "remaining bytes in msg with tag={msg_tag}: {}",
2625 0 : remaining.len()
2626 0 : );
2627 0 : }
2628 0 : Ok(ok)
2629 0 : }
2630 :
2631 0 : pub fn kind(&self) -> &'static str {
2632 0 : match self {
2633 0 : Self::Exists(_) => "Exists",
2634 0 : Self::Nblocks(_) => "Nblocks",
2635 0 : Self::GetPage(_) => "GetPage",
2636 0 : Self::Error(_) => "Error",
2637 0 : Self::DbSize(_) => "DbSize",
2638 0 : Self::GetSlruSegment(_) => "GetSlruSegment",
2639 : #[cfg(feature = "testing")]
2640 0 : Self::Test(_) => "Test",
2641 : }
2642 0 : }
2643 : }
2644 :
2645 0 : #[derive(Debug, Serialize, Deserialize)]
2646 : pub struct PageTraceEvent {
2647 : pub key: CompactKey,
2648 : pub effective_lsn: Lsn,
2649 : pub time: SystemTime,
2650 : }
2651 :
2652 : impl Default for PageTraceEvent {
2653 0 : fn default() -> Self {
2654 0 : Self {
2655 0 : key: Default::default(),
2656 0 : effective_lsn: Default::default(),
2657 0 : time: std::time::UNIX_EPOCH,
2658 0 : }
2659 0 : }
2660 : }
2661 :
2662 : #[cfg(test)]
2663 : mod tests {
2664 : use std::str::FromStr;
2665 :
2666 : use serde_json::json;
2667 :
2668 : use super::*;
2669 :
2670 : #[test]
2671 1 : fn test_pagestream() {
2672 1 : // Test serialization/deserialization of PagestreamFeMessage
2673 1 : let messages = vec![
2674 1 : PagestreamFeMessage::Exists(PagestreamExistsRequest {
2675 1 : hdr: PagestreamRequest {
2676 1 : reqid: 0,
2677 1 : request_lsn: Lsn(4),
2678 1 : not_modified_since: Lsn(3),
2679 1 : },
2680 1 : rel: RelTag {
2681 1 : forknum: 1,
2682 1 : spcnode: 2,
2683 1 : dbnode: 3,
2684 1 : relnode: 4,
2685 1 : },
2686 1 : }),
2687 1 : PagestreamFeMessage::Nblocks(PagestreamNblocksRequest {
2688 1 : hdr: PagestreamRequest {
2689 1 : reqid: 0,
2690 1 : request_lsn: Lsn(4),
2691 1 : not_modified_since: Lsn(4),
2692 1 : },
2693 1 : rel: RelTag {
2694 1 : forknum: 1,
2695 1 : spcnode: 2,
2696 1 : dbnode: 3,
2697 1 : relnode: 4,
2698 1 : },
2699 1 : }),
2700 1 : PagestreamFeMessage::GetPage(PagestreamGetPageRequest {
2701 1 : hdr: PagestreamRequest {
2702 1 : reqid: 0,
2703 1 : request_lsn: Lsn(4),
2704 1 : not_modified_since: Lsn(3),
2705 1 : },
2706 1 : rel: RelTag {
2707 1 : forknum: 1,
2708 1 : spcnode: 2,
2709 1 : dbnode: 3,
2710 1 : relnode: 4,
2711 1 : },
2712 1 : blkno: 7,
2713 1 : }),
2714 1 : PagestreamFeMessage::DbSize(PagestreamDbSizeRequest {
2715 1 : hdr: PagestreamRequest {
2716 1 : reqid: 0,
2717 1 : request_lsn: Lsn(4),
2718 1 : not_modified_since: Lsn(3),
2719 1 : },
2720 1 : dbnode: 7,
2721 1 : }),
2722 1 : ];
2723 5 : for msg in messages {
2724 4 : let bytes = msg.serialize();
2725 4 : let reconstructed =
2726 4 : PagestreamFeMessage::parse(&mut bytes.reader(), PagestreamProtocolVersion::V3)
2727 4 : .unwrap();
2728 4 : assert!(msg == reconstructed);
2729 : }
2730 1 : }
2731 :
2732 : #[test]
2733 1 : fn test_tenantinfo_serde() {
2734 1 : // Test serialization/deserialization of TenantInfo
2735 1 : let original_active = TenantInfo {
2736 1 : id: TenantShardId::unsharded(TenantId::generate()),
2737 1 : state: TenantState::Active,
2738 1 : current_physical_size: Some(42),
2739 1 : attachment_status: TenantAttachmentStatus::Attached,
2740 1 : generation: 1,
2741 1 : gc_blocking: None,
2742 1 : };
2743 1 : let expected_active = json!({
2744 1 : "id": original_active.id.to_string(),
2745 1 : "state": {
2746 1 : "slug": "Active",
2747 1 : },
2748 1 : "current_physical_size": 42,
2749 1 : "attachment_status": {
2750 1 : "slug":"attached",
2751 1 : },
2752 1 : "generation" : 1
2753 1 : });
2754 1 :
2755 1 : let original_broken = TenantInfo {
2756 1 : id: TenantShardId::unsharded(TenantId::generate()),
2757 1 : state: TenantState::Broken {
2758 1 : reason: "reason".into(),
2759 1 : backtrace: "backtrace info".into(),
2760 1 : },
2761 1 : current_physical_size: Some(42),
2762 1 : attachment_status: TenantAttachmentStatus::Attached,
2763 1 : generation: 1,
2764 1 : gc_blocking: None,
2765 1 : };
2766 1 : let expected_broken = json!({
2767 1 : "id": original_broken.id.to_string(),
2768 1 : "state": {
2769 1 : "slug": "Broken",
2770 1 : "data": {
2771 1 : "backtrace": "backtrace info",
2772 1 : "reason": "reason",
2773 1 : }
2774 1 : },
2775 1 : "current_physical_size": 42,
2776 1 : "attachment_status": {
2777 1 : "slug":"attached",
2778 1 : },
2779 1 : "generation" : 1
2780 1 : });
2781 1 :
2782 1 : assert_eq!(
2783 1 : serde_json::to_value(&original_active).unwrap(),
2784 1 : expected_active
2785 1 : );
2786 :
2787 1 : assert_eq!(
2788 1 : serde_json::to_value(&original_broken).unwrap(),
2789 1 : expected_broken
2790 1 : );
2791 1 : assert!(format!("{:?}", &original_broken.state).contains("reason"));
2792 1 : assert!(format!("{:?}", &original_broken.state).contains("backtrace info"));
2793 1 : }
2794 :
2795 : #[test]
2796 1 : fn test_reject_unknown_field() {
2797 1 : let id = TenantId::generate();
2798 1 : let config_request = json!({
2799 1 : "tenant_id": id.to_string(),
2800 1 : "unknown_field": "unknown_value".to_string(),
2801 1 : });
2802 1 : let err = serde_json::from_value::<TenantConfigRequest>(config_request).unwrap_err();
2803 1 : assert!(
2804 1 : err.to_string().contains("unknown field `unknown_field`"),
2805 0 : "expect unknown field `unknown_field` error, got: {}",
2806 : err
2807 : );
2808 1 : }
2809 :
2810 : #[test]
2811 1 : fn tenantstatus_activating_serde() {
2812 1 : let states = [TenantState::Activating(ActivatingFrom::Attaching)];
2813 1 : let expected = "[{\"slug\":\"Activating\",\"data\":\"Attaching\"}]";
2814 1 :
2815 1 : let actual = serde_json::to_string(&states).unwrap();
2816 1 :
2817 1 : assert_eq!(actual, expected);
2818 :
2819 1 : let parsed = serde_json::from_str::<Vec<TenantState>>(&actual).unwrap();
2820 1 :
2821 1 : assert_eq!(states.as_slice(), &parsed);
2822 1 : }
2823 :
2824 : #[test]
2825 1 : fn tenantstatus_activating_strum() {
2826 1 : // tests added, because we use these for metrics
2827 1 : let examples = [
2828 1 : (line!(), TenantState::Attaching, "Attaching"),
2829 1 : (
2830 1 : line!(),
2831 1 : TenantState::Activating(ActivatingFrom::Attaching),
2832 1 : "Activating",
2833 1 : ),
2834 1 : (line!(), TenantState::Active, "Active"),
2835 1 : (
2836 1 : line!(),
2837 1 : TenantState::Stopping { progress: None },
2838 1 : "Stopping",
2839 1 : ),
2840 1 : (
2841 1 : line!(),
2842 1 : TenantState::Stopping {
2843 1 : progress: Some(completion::Barrier::default()),
2844 1 : },
2845 1 : "Stopping",
2846 1 : ),
2847 1 : (
2848 1 : line!(),
2849 1 : TenantState::Broken {
2850 1 : reason: "Example".into(),
2851 1 : backtrace: "Looooong backtrace".into(),
2852 1 : },
2853 1 : "Broken",
2854 1 : ),
2855 1 : ];
2856 :
2857 7 : for (line, rendered, expected) in examples {
2858 6 : let actual: &'static str = rendered.into();
2859 6 : assert_eq!(actual, expected, "example on {line}");
2860 : }
2861 1 : }
2862 :
2863 : #[test]
2864 1 : fn test_image_compression_algorithm_parsing() {
2865 : use ImageCompressionAlgorithm::*;
2866 1 : let cases = [
2867 1 : ("disabled", Disabled),
2868 1 : ("zstd", Zstd { level: None }),
2869 1 : ("zstd(18)", Zstd { level: Some(18) }),
2870 1 : ("zstd(-3)", Zstd { level: Some(-3) }),
2871 1 : ];
2872 :
2873 5 : for (display, expected) in cases {
2874 4 : assert_eq!(
2875 4 : ImageCompressionAlgorithm::from_str(display).unwrap(),
2876 : expected,
2877 0 : "parsing works"
2878 : );
2879 4 : assert_eq!(format!("{expected}"), display, "Display FromStr roundtrip");
2880 :
2881 4 : let ser = serde_json::to_string(&expected).expect("serialization");
2882 4 : assert_eq!(
2883 4 : serde_json::from_str::<ImageCompressionAlgorithm>(&ser).unwrap(),
2884 : expected,
2885 0 : "serde roundtrip"
2886 : );
2887 :
2888 4 : assert_eq!(
2889 4 : serde_json::Value::String(display.to_string()),
2890 4 : serde_json::to_value(expected).unwrap(),
2891 0 : "Display is the serde serialization"
2892 : );
2893 : }
2894 1 : }
2895 :
2896 : #[test]
2897 1 : fn test_tenant_config_patch_request_serde() {
2898 1 : let patch_request = TenantConfigPatchRequest {
2899 1 : tenant_id: TenantId::from_str("17c6d121946a61e5ab0fe5a2fd4d8215").unwrap(),
2900 1 : config: TenantConfigPatch {
2901 1 : checkpoint_distance: FieldPatch::Upsert(42),
2902 1 : gc_horizon: FieldPatch::Remove,
2903 1 : compaction_threshold: FieldPatch::Noop,
2904 1 : ..TenantConfigPatch::default()
2905 1 : },
2906 1 : };
2907 1 :
2908 1 : let json = serde_json::to_string(&patch_request).unwrap();
2909 1 :
2910 1 : let expected = r#"{"tenant_id":"17c6d121946a61e5ab0fe5a2fd4d8215","checkpoint_distance":42,"gc_horizon":null}"#;
2911 1 : assert_eq!(json, expected);
2912 :
2913 1 : let decoded: TenantConfigPatchRequest = serde_json::from_str(&json).unwrap();
2914 1 : assert_eq!(decoded.tenant_id, patch_request.tenant_id);
2915 1 : assert_eq!(decoded.config, patch_request.config);
2916 :
2917 : // Now apply the patch to a config to demonstrate semantics
2918 :
2919 1 : let base = TenantConfig {
2920 1 : checkpoint_distance: Some(28),
2921 1 : gc_horizon: Some(100),
2922 1 : compaction_target_size: Some(1024),
2923 1 : ..Default::default()
2924 1 : };
2925 1 :
2926 1 : let expected = TenantConfig {
2927 1 : checkpoint_distance: Some(42),
2928 1 : gc_horizon: None,
2929 1 : ..base.clone()
2930 1 : };
2931 1 :
2932 1 : let patched = base.apply_patch(decoded.config).unwrap();
2933 1 :
2934 1 : assert_eq!(patched, expected);
2935 1 : }
2936 : }
|
