Line data Source code
1 : pub mod detach_ancestor;
2 : pub mod partitioning;
3 : pub mod utilization;
4 :
5 : #[cfg(feature = "testing")]
6 : use camino::Utf8PathBuf;
7 : pub use utilization::PageserverUtilization;
8 :
9 : use core::ops::Range;
10 : use std::{
11 : collections::HashMap,
12 : fmt::Display,
13 : io::{BufRead, Read},
14 : num::{NonZeroU32, NonZeroU64, NonZeroUsize},
15 : str::FromStr,
16 : time::{Duration, SystemTime},
17 : };
18 :
19 : use byteorder::{BigEndian, ReadBytesExt};
20 : use postgres_ffi::BLCKSZ;
21 : use serde::{Deserialize, Deserializer, Serialize, Serializer};
22 : use serde_with::serde_as;
23 : use utils::{
24 : completion,
25 : id::{NodeId, TenantId, TimelineId},
26 : lsn::Lsn,
27 : postgres_client::PostgresClientProtocol,
28 : serde_system_time,
29 : };
30 :
31 : use crate::{
32 : key::{CompactKey, Key},
33 : reltag::RelTag,
34 : shard::{ShardCount, ShardStripeSize, TenantShardId},
35 : };
36 : use bytes::{Buf, BufMut, Bytes, BytesMut};
37 :
38 : /// The state of a tenant in this pageserver.
39 : ///
40 : /// ```mermaid
41 : /// stateDiagram-v2
42 : ///
43 : /// [*] --> Attaching: spawn_attach()
44 : ///
45 : /// Attaching --> Activating: activate()
46 : /// Activating --> Active: infallible
47 : ///
48 : /// Attaching --> Broken: attach() failure
49 : ///
50 : /// Active --> Stopping: set_stopping(), part of shutdown & detach
51 : /// Stopping --> Broken: late error in remove_tenant_from_memory
52 : ///
53 : /// Broken --> [*]: ignore / detach / shutdown
54 : /// Stopping --> [*]: remove_from_memory complete
55 : ///
56 : /// Active --> Broken: cfg(testing)-only tenant break point
57 : /// ```
58 : #[derive(
59 : Clone,
60 : PartialEq,
61 : Eq,
62 0 : serde::Serialize,
63 1 : serde::Deserialize,
64 : strum_macros::Display,
65 : strum_macros::VariantNames,
66 : strum_macros::AsRefStr,
67 : strum_macros::IntoStaticStr,
68 : )]
69 : #[serde(tag = "slug", content = "data")]
70 : pub enum TenantState {
71 : /// This tenant is being attached to the pageserver.
72 : ///
73 : /// `set_stopping()` and `set_broken()` do not work in this state and wait for it to pass.
74 : Attaching,
75 : /// The tenant is transitioning from Loading/Attaching to Active.
76 : ///
77 : /// While in this state, the individual timelines are being activated.
78 : ///
79 : /// `set_stopping()` and `set_broken()` do not work in this state and wait for it to pass.
80 : Activating(ActivatingFrom),
81 : /// The tenant has finished activating and is open for business.
82 : ///
83 : /// Transitions out of this state are possible through `set_stopping()` and `set_broken()`.
84 : Active,
85 : /// The tenant is recognized by pageserver, but it is being detached or the
86 : /// system is being shut down.
87 : ///
88 : /// Transitions out of this state are possible through `set_broken()`.
89 : Stopping {
90 : // Because of https://github.com/serde-rs/serde/issues/2105 this has to be a named field,
91 : // otherwise it will not be skipped during deserialization
92 : #[serde(skip)]
93 : progress: completion::Barrier,
94 : },
95 : /// The tenant is recognized by the pageserver, but can no longer be used for
96 : /// any operations.
97 : ///
98 : /// If the tenant fails to load or attach, it will transition to this state
99 : /// and it is guaranteed that no background tasks are running in its name.
100 : ///
101 : /// The other way to transition into this state is from `Stopping` state
102 : /// through `set_broken()` called from `remove_tenant_from_memory()`. That happens
103 : /// if the cleanup future executed by `remove_tenant_from_memory()` fails.
104 : Broken { reason: String, backtrace: String },
105 : }
106 :
107 : impl TenantState {
108 0 : pub fn attachment_status(&self) -> TenantAttachmentStatus {
109 : use TenantAttachmentStatus::*;
110 :
111 : // Below TenantState::Activating is used as "transient" or "transparent" state for
112 : // attachment_status determining.
113 0 : match self {
114 : // The attach procedure writes the marker file before adding the Attaching tenant to the tenants map.
115 : // So, technically, we can return Attached here.
116 : // However, as soon as Console observes Attached, it will proceed with the Postgres-level health check.
117 : // But, our attach task might still be fetching the remote timelines, etc.
118 : // So, return `Maybe` while Attaching, making Console wait for the attach task to finish.
119 0 : Self::Attaching | Self::Activating(ActivatingFrom::Attaching) => Maybe,
120 : // We only reach Active after successful load / attach.
121 : // So, call atttachment status Attached.
122 0 : Self::Active => Attached,
123 : // If the (initial or resumed) attach procedure fails, the tenant becomes Broken.
124 : // However, it also becomes Broken if the regular load fails.
125 : // From Console's perspective there's no practical difference
126 : // because attachment_status is polled by console only during attach operation execution.
127 0 : Self::Broken { reason, .. } => Failed {
128 0 : reason: reason.to_owned(),
129 0 : },
130 : // Why is Stopping a Maybe case? Because, during pageserver shutdown,
131 : // we set the Stopping state irrespective of whether the tenant
132 : // has finished attaching or not.
133 0 : Self::Stopping { .. } => Maybe,
134 : }
135 0 : }
136 :
137 0 : pub fn broken_from_reason(reason: String) -> Self {
138 0 : let backtrace_str: String = format!("{}", std::backtrace::Backtrace::force_capture());
139 0 : Self::Broken {
140 0 : reason,
141 0 : backtrace: backtrace_str,
142 0 : }
143 0 : }
144 : }
145 :
146 : impl std::fmt::Debug for TenantState {
147 2 : fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
148 2 : match self {
149 2 : Self::Broken { reason, backtrace } if !reason.is_empty() => {
150 2 : write!(f, "Broken due to: {reason}. Backtrace:\n{backtrace}")
151 : }
152 0 : _ => write!(f, "{self}"),
153 : }
154 2 : }
155 : }
156 :
157 : /// A temporary lease to a specific lsn inside a timeline.
158 : /// Access to the lsn is guaranteed by the pageserver until the expiration indicated by `valid_until`.
159 : #[serde_as]
160 0 : #[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
161 : pub struct LsnLease {
162 : #[serde_as(as = "SystemTimeAsRfc3339Millis")]
163 : pub valid_until: SystemTime,
164 : }
165 :
166 : serde_with::serde_conv!(
167 : SystemTimeAsRfc3339Millis,
168 : SystemTime,
169 0 : |time: &SystemTime| humantime::format_rfc3339_millis(*time).to_string(),
170 0 : |value: String| -> Result<_, humantime::TimestampError> { humantime::parse_rfc3339(&value) }
171 : );
172 :
173 : impl LsnLease {
174 : /// The default length for an explicit LSN lease request (10 minutes).
175 : pub const DEFAULT_LENGTH: Duration = Duration::from_secs(10 * 60);
176 :
177 : /// The default length for an implicit LSN lease granted during
178 : /// `get_lsn_by_timestamp` request (1 minutes).
179 : pub const DEFAULT_LENGTH_FOR_TS: Duration = Duration::from_secs(60);
180 :
181 : /// Checks whether the lease is expired.
182 12 : pub fn is_expired(&self, now: &SystemTime) -> bool {
183 12 : now > &self.valid_until
184 12 : }
185 : }
186 :
187 : /// The only [`TenantState`] variants we could be `TenantState::Activating` from.
188 : ///
189 : /// XXX: We used to have more variants here, but now it's just one, which makes this rather
190 : /// useless. Remove, once we've checked that there's no client code left that looks at this.
191 1 : #[derive(Clone, Copy, Debug, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
192 : pub enum ActivatingFrom {
193 : /// Arrived to [`TenantState::Activating`] from [`TenantState::Attaching`]
194 : Attaching,
195 : }
196 :
197 : /// A state of a timeline in pageserver's memory.
198 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
199 : pub enum TimelineState {
200 : /// The timeline is recognized by the pageserver but is not yet operational.
201 : /// In particular, the walreceiver connection loop is not running for this timeline.
202 : /// It will eventually transition to state Active or Broken.
203 : Loading,
204 : /// The timeline is fully operational.
205 : /// It can be queried, and the walreceiver connection loop is running.
206 : Active,
207 : /// The timeline was previously Loading or Active but is shutting down.
208 : /// It cannot transition back into any other state.
209 : Stopping,
210 : /// The timeline is broken and not operational (previous states: Loading or Active).
211 : Broken { reason: String, backtrace: String },
212 : }
213 :
214 : #[serde_with::serde_as]
215 0 : #[derive(Debug, Clone, serde::Deserialize, serde::Serialize)]
216 : pub struct CompactLsnRange {
217 : pub start: Lsn,
218 : pub end: Lsn,
219 : }
220 :
221 : #[serde_with::serde_as]
222 0 : #[derive(Debug, Clone, serde::Deserialize, serde::Serialize)]
223 : pub struct CompactKeyRange {
224 : #[serde_as(as = "serde_with::DisplayFromStr")]
225 : pub start: Key,
226 : #[serde_as(as = "serde_with::DisplayFromStr")]
227 : pub end: Key,
228 : }
229 :
230 : impl From<Range<Lsn>> for CompactLsnRange {
231 12 : fn from(range: Range<Lsn>) -> Self {
232 12 : Self {
233 12 : start: range.start,
234 12 : end: range.end,
235 12 : }
236 12 : }
237 : }
238 :
239 : impl From<Range<Key>> for CompactKeyRange {
240 32 : fn from(range: Range<Key>) -> Self {
241 32 : Self {
242 32 : start: range.start,
243 32 : end: range.end,
244 32 : }
245 32 : }
246 : }
247 :
248 : impl From<CompactLsnRange> for Range<Lsn> {
249 20 : fn from(range: CompactLsnRange) -> Self {
250 20 : range.start..range.end
251 20 : }
252 : }
253 :
254 : impl From<CompactKeyRange> for Range<Key> {
255 32 : fn from(range: CompactKeyRange) -> Self {
256 32 : range.start..range.end
257 32 : }
258 : }
259 :
260 : impl CompactLsnRange {
261 8 : pub fn above(lsn: Lsn) -> Self {
262 8 : Self {
263 8 : start: lsn,
264 8 : end: Lsn::MAX,
265 8 : }
266 8 : }
267 : }
268 :
269 : #[derive(Debug, Clone, Serialize)]
270 : pub struct CompactInfoResponse {
271 : pub compact_key_range: Option<CompactKeyRange>,
272 : pub compact_lsn_range: Option<CompactLsnRange>,
273 : pub sub_compaction: bool,
274 : pub running: bool,
275 : pub job_id: usize,
276 : }
277 :
278 0 : #[derive(Serialize, Deserialize, Clone)]
279 : pub struct TimelineCreateRequest {
280 : pub new_timeline_id: TimelineId,
281 : #[serde(flatten)]
282 : pub mode: TimelineCreateRequestMode,
283 : }
284 :
285 0 : #[derive(Serialize, Deserialize, Clone)]
286 : #[serde(untagged)]
287 : pub enum TimelineCreateRequestMode {
288 : Branch {
289 : ancestor_timeline_id: TimelineId,
290 : #[serde(default)]
291 : ancestor_start_lsn: Option<Lsn>,
292 : // TODO: cplane sets this, but, the branching code always
293 : // inherits the ancestor's pg_version. Earlier code wasn't
294 : // using a flattened enum, so, it was an accepted field, and
295 : // we continue to accept it by having it here.
296 : pg_version: Option<u32>,
297 : },
298 : ImportPgdata {
299 : import_pgdata: TimelineCreateRequestModeImportPgdata,
300 : },
301 : // NB: Bootstrap is all-optional, and thus the serde(untagged) will cause serde to stop at Bootstrap.
302 : // (serde picks the first matching enum variant, in declaration order).
303 : Bootstrap {
304 : #[serde(default)]
305 : existing_initdb_timeline_id: Option<TimelineId>,
306 : pg_version: Option<u32>,
307 : },
308 : }
309 :
310 0 : #[derive(Serialize, Deserialize, Clone)]
311 : pub struct TimelineCreateRequestModeImportPgdata {
312 : pub location: ImportPgdataLocation,
313 : pub idempotency_key: ImportPgdataIdempotencyKey,
314 : }
315 :
316 0 : #[derive(Serialize, Deserialize, Clone, Debug)]
317 : pub enum ImportPgdataLocation {
318 : #[cfg(feature = "testing")]
319 : LocalFs { path: Utf8PathBuf },
320 : AwsS3 {
321 : region: String,
322 : bucket: String,
323 : /// A better name for this would be `prefix`; changing requires coordination with cplane.
324 : /// See <https://github.com/neondatabase/cloud/issues/20646>.
325 : key: String,
326 : },
327 : }
328 :
329 0 : #[derive(Serialize, Deserialize, Clone)]
330 : #[serde(transparent)]
331 : pub struct ImportPgdataIdempotencyKey(pub String);
332 :
333 : impl ImportPgdataIdempotencyKey {
334 0 : pub fn random() -> Self {
335 : use rand::{distributions::Alphanumeric, Rng};
336 0 : Self(
337 0 : rand::thread_rng()
338 0 : .sample_iter(&Alphanumeric)
339 0 : .take(20)
340 0 : .map(char::from)
341 0 : .collect(),
342 0 : )
343 0 : }
344 : }
345 :
346 0 : #[derive(Serialize, Deserialize, Clone)]
347 : pub struct LsnLeaseRequest {
348 : pub lsn: Lsn,
349 : }
350 :
351 0 : #[derive(Serialize, Deserialize)]
352 : pub struct TenantShardSplitRequest {
353 : pub new_shard_count: u8,
354 :
355 : // A tenant's stripe size is only meaningful the first time their shard count goes
356 : // above 1: therefore during a split from 1->N shards, we may modify the stripe size.
357 : //
358 : // If this is set while the stripe count is being increased from an already >1 value,
359 : // then the request will fail with 400.
360 : pub new_stripe_size: Option<ShardStripeSize>,
361 : }
362 :
363 0 : #[derive(Serialize, Deserialize)]
364 : pub struct TenantShardSplitResponse {
365 : pub new_shards: Vec<TenantShardId>,
366 : }
367 :
368 : /// Parameters that apply to all shards in a tenant. Used during tenant creation.
369 0 : #[derive(Serialize, Deserialize, Debug)]
370 : #[serde(deny_unknown_fields)]
371 : pub struct ShardParameters {
372 : pub count: ShardCount,
373 : pub stripe_size: ShardStripeSize,
374 : }
375 :
376 : impl ShardParameters {
377 : pub const DEFAULT_STRIPE_SIZE: ShardStripeSize = ShardStripeSize(256 * 1024 / 8);
378 :
379 0 : pub fn is_unsharded(&self) -> bool {
380 0 : self.count.is_unsharded()
381 0 : }
382 : }
383 :
384 : impl Default for ShardParameters {
385 441 : fn default() -> Self {
386 441 : Self {
387 441 : count: ShardCount::new(0),
388 441 : stripe_size: Self::DEFAULT_STRIPE_SIZE,
389 441 : }
390 441 : }
391 : }
392 :
393 : #[derive(Debug, Default, Clone, Eq, PartialEq)]
394 : pub enum FieldPatch<T> {
395 : Upsert(T),
396 : Remove,
397 : #[default]
398 : Noop,
399 : }
400 :
401 : impl<T> FieldPatch<T> {
402 64 : fn is_noop(&self) -> bool {
403 64 : matches!(self, FieldPatch::Noop)
404 64 : }
405 :
406 32 : pub fn apply(self, target: &mut Option<T>) {
407 32 : match self {
408 1 : Self::Upsert(v) => *target = Some(v),
409 1 : Self::Remove => *target = None,
410 30 : Self::Noop => {}
411 : }
412 32 : }
413 :
414 0 : pub fn map<U, E, F: FnOnce(T) -> Result<U, E>>(self, map: F) -> Result<FieldPatch<U>, E> {
415 0 : match self {
416 0 : Self::Upsert(v) => Ok(FieldPatch::<U>::Upsert(map(v)?)),
417 0 : Self::Remove => Ok(FieldPatch::<U>::Remove),
418 0 : Self::Noop => Ok(FieldPatch::<U>::Noop),
419 : }
420 0 : }
421 : }
422 :
423 : impl<'de, T: Deserialize<'de>> Deserialize<'de> for FieldPatch<T> {
424 2 : fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
425 2 : where
426 2 : D: Deserializer<'de>,
427 2 : {
428 2 : Option::deserialize(deserializer).map(|opt| match opt {
429 1 : None => FieldPatch::Remove,
430 1 : Some(val) => FieldPatch::Upsert(val),
431 2 : })
432 2 : }
433 : }
434 :
435 : impl<T: Serialize> Serialize for FieldPatch<T> {
436 2 : fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
437 2 : where
438 2 : S: Serializer,
439 2 : {
440 2 : match self {
441 1 : FieldPatch::Upsert(val) => serializer.serialize_some(val),
442 1 : FieldPatch::Remove => serializer.serialize_none(),
443 0 : FieldPatch::Noop => unreachable!(),
444 : }
445 2 : }
446 : }
447 :
448 2 : #[derive(Serialize, Deserialize, Debug, Default, Clone, Eq, PartialEq)]
449 : #[serde(default)]
450 : pub struct TenantConfigPatch {
451 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
452 : pub checkpoint_distance: FieldPatch<u64>,
453 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
454 : pub checkpoint_timeout: FieldPatch<String>,
455 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
456 : pub compaction_target_size: FieldPatch<u64>,
457 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
458 : pub compaction_period: FieldPatch<String>,
459 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
460 : pub compaction_threshold: FieldPatch<usize>,
461 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
462 : pub compaction_upper_limit: FieldPatch<usize>,
463 : // defer parsing compaction_algorithm, like eviction_policy
464 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
465 : pub compaction_algorithm: FieldPatch<CompactionAlgorithmSettings>,
466 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
467 : pub l0_flush_delay_threshold: FieldPatch<usize>,
468 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
469 : pub l0_flush_stall_threshold: FieldPatch<usize>,
470 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
471 : pub l0_flush_wait_upload: FieldPatch<bool>,
472 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
473 : pub gc_horizon: FieldPatch<u64>,
474 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
475 : pub gc_period: FieldPatch<String>,
476 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
477 : pub image_creation_threshold: FieldPatch<usize>,
478 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
479 : pub pitr_interval: FieldPatch<String>,
480 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
481 : pub walreceiver_connect_timeout: FieldPatch<String>,
482 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
483 : pub lagging_wal_timeout: FieldPatch<String>,
484 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
485 : pub max_lsn_wal_lag: FieldPatch<NonZeroU64>,
486 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
487 : pub eviction_policy: FieldPatch<EvictionPolicy>,
488 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
489 : pub min_resident_size_override: FieldPatch<u64>,
490 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
491 : pub evictions_low_residence_duration_metric_threshold: FieldPatch<String>,
492 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
493 : pub heatmap_period: FieldPatch<String>,
494 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
495 : pub lazy_slru_download: FieldPatch<bool>,
496 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
497 : pub timeline_get_throttle: FieldPatch<ThrottleConfig>,
498 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
499 : pub image_layer_creation_check_threshold: FieldPatch<u8>,
500 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
501 : pub lsn_lease_length: FieldPatch<String>,
502 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
503 : pub lsn_lease_length_for_ts: FieldPatch<String>,
504 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
505 : pub timeline_offloading: FieldPatch<bool>,
506 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
507 : pub wal_receiver_protocol_override: FieldPatch<PostgresClientProtocol>,
508 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
509 : pub rel_size_v2_enabled: FieldPatch<bool>,
510 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
511 : pub gc_compaction_enabled: FieldPatch<bool>,
512 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
513 : pub gc_compaction_initial_threshold_kb: FieldPatch<u64>,
514 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
515 : pub gc_compaction_ratio_percent: FieldPatch<u64>,
516 : }
517 :
518 : /// An alternative representation of `pageserver::tenant::TenantConf` with
519 : /// simpler types.
520 0 : #[derive(Serialize, Deserialize, Debug, Default, Clone, Eq, PartialEq)]
521 : pub struct TenantConfig {
522 : pub checkpoint_distance: Option<u64>,
523 : pub checkpoint_timeout: Option<String>,
524 : pub compaction_target_size: Option<u64>,
525 : pub compaction_period: Option<String>,
526 : pub compaction_threshold: Option<usize>,
527 : pub compaction_upper_limit: Option<usize>,
528 : // defer parsing compaction_algorithm, like eviction_policy
529 : pub compaction_algorithm: Option<CompactionAlgorithmSettings>,
530 : pub l0_flush_delay_threshold: Option<usize>,
531 : pub l0_flush_stall_threshold: Option<usize>,
532 : pub l0_flush_wait_upload: Option<bool>,
533 : pub gc_horizon: Option<u64>,
534 : pub gc_period: Option<String>,
535 : pub image_creation_threshold: Option<usize>,
536 : pub pitr_interval: Option<String>,
537 : pub walreceiver_connect_timeout: Option<String>,
538 : pub lagging_wal_timeout: Option<String>,
539 : pub max_lsn_wal_lag: Option<NonZeroU64>,
540 : pub eviction_policy: Option<EvictionPolicy>,
541 : pub min_resident_size_override: Option<u64>,
542 : pub evictions_low_residence_duration_metric_threshold: Option<String>,
543 : pub heatmap_period: Option<String>,
544 : pub lazy_slru_download: Option<bool>,
545 : pub timeline_get_throttle: Option<ThrottleConfig>,
546 : pub image_layer_creation_check_threshold: Option<u8>,
547 : pub lsn_lease_length: Option<String>,
548 : pub lsn_lease_length_for_ts: Option<String>,
549 : pub timeline_offloading: Option<bool>,
550 : pub wal_receiver_protocol_override: Option<PostgresClientProtocol>,
551 : pub rel_size_v2_enabled: Option<bool>,
552 : pub gc_compaction_enabled: Option<bool>,
553 : pub gc_compaction_initial_threshold_kb: Option<u64>,
554 : pub gc_compaction_ratio_percent: Option<u64>,
555 : }
556 :
557 : impl TenantConfig {
558 1 : pub fn apply_patch(self, patch: TenantConfigPatch) -> TenantConfig {
559 1 : let Self {
560 1 : mut checkpoint_distance,
561 1 : mut checkpoint_timeout,
562 1 : mut compaction_target_size,
563 1 : mut compaction_period,
564 1 : mut compaction_threshold,
565 1 : mut compaction_upper_limit,
566 1 : mut compaction_algorithm,
567 1 : mut l0_flush_delay_threshold,
568 1 : mut l0_flush_stall_threshold,
569 1 : mut l0_flush_wait_upload,
570 1 : mut gc_horizon,
571 1 : mut gc_period,
572 1 : mut image_creation_threshold,
573 1 : mut pitr_interval,
574 1 : mut walreceiver_connect_timeout,
575 1 : mut lagging_wal_timeout,
576 1 : mut max_lsn_wal_lag,
577 1 : mut eviction_policy,
578 1 : mut min_resident_size_override,
579 1 : mut evictions_low_residence_duration_metric_threshold,
580 1 : mut heatmap_period,
581 1 : mut lazy_slru_download,
582 1 : mut timeline_get_throttle,
583 1 : mut image_layer_creation_check_threshold,
584 1 : mut lsn_lease_length,
585 1 : mut lsn_lease_length_for_ts,
586 1 : mut timeline_offloading,
587 1 : mut wal_receiver_protocol_override,
588 1 : mut rel_size_v2_enabled,
589 1 : mut gc_compaction_enabled,
590 1 : mut gc_compaction_initial_threshold_kb,
591 1 : mut gc_compaction_ratio_percent,
592 1 : } = self;
593 1 :
594 1 : patch.checkpoint_distance.apply(&mut checkpoint_distance);
595 1 : patch.checkpoint_timeout.apply(&mut checkpoint_timeout);
596 1 : patch
597 1 : .compaction_target_size
598 1 : .apply(&mut compaction_target_size);
599 1 : patch.compaction_period.apply(&mut compaction_period);
600 1 : patch.compaction_threshold.apply(&mut compaction_threshold);
601 1 : patch
602 1 : .compaction_upper_limit
603 1 : .apply(&mut compaction_upper_limit);
604 1 : patch.compaction_algorithm.apply(&mut compaction_algorithm);
605 1 : patch
606 1 : .l0_flush_delay_threshold
607 1 : .apply(&mut l0_flush_delay_threshold);
608 1 : patch
609 1 : .l0_flush_stall_threshold
610 1 : .apply(&mut l0_flush_stall_threshold);
611 1 : patch.l0_flush_wait_upload.apply(&mut l0_flush_wait_upload);
612 1 : patch.gc_horizon.apply(&mut gc_horizon);
613 1 : patch.gc_period.apply(&mut gc_period);
614 1 : patch
615 1 : .image_creation_threshold
616 1 : .apply(&mut image_creation_threshold);
617 1 : patch.pitr_interval.apply(&mut pitr_interval);
618 1 : patch
619 1 : .walreceiver_connect_timeout
620 1 : .apply(&mut walreceiver_connect_timeout);
621 1 : patch.lagging_wal_timeout.apply(&mut lagging_wal_timeout);
622 1 : patch.max_lsn_wal_lag.apply(&mut max_lsn_wal_lag);
623 1 : patch.eviction_policy.apply(&mut eviction_policy);
624 1 : patch
625 1 : .min_resident_size_override
626 1 : .apply(&mut min_resident_size_override);
627 1 : patch
628 1 : .evictions_low_residence_duration_metric_threshold
629 1 : .apply(&mut evictions_low_residence_duration_metric_threshold);
630 1 : patch.heatmap_period.apply(&mut heatmap_period);
631 1 : patch.lazy_slru_download.apply(&mut lazy_slru_download);
632 1 : patch
633 1 : .timeline_get_throttle
634 1 : .apply(&mut timeline_get_throttle);
635 1 : patch
636 1 : .image_layer_creation_check_threshold
637 1 : .apply(&mut image_layer_creation_check_threshold);
638 1 : patch.lsn_lease_length.apply(&mut lsn_lease_length);
639 1 : patch
640 1 : .lsn_lease_length_for_ts
641 1 : .apply(&mut lsn_lease_length_for_ts);
642 1 : patch.timeline_offloading.apply(&mut timeline_offloading);
643 1 : patch
644 1 : .wal_receiver_protocol_override
645 1 : .apply(&mut wal_receiver_protocol_override);
646 1 : patch.rel_size_v2_enabled.apply(&mut rel_size_v2_enabled);
647 1 : patch
648 1 : .gc_compaction_enabled
649 1 : .apply(&mut gc_compaction_enabled);
650 1 : patch
651 1 : .gc_compaction_initial_threshold_kb
652 1 : .apply(&mut gc_compaction_initial_threshold_kb);
653 1 : patch
654 1 : .gc_compaction_ratio_percent
655 1 : .apply(&mut gc_compaction_ratio_percent);
656 1 :
657 1 : Self {
658 1 : checkpoint_distance,
659 1 : checkpoint_timeout,
660 1 : compaction_target_size,
661 1 : compaction_period,
662 1 : compaction_threshold,
663 1 : compaction_upper_limit,
664 1 : compaction_algorithm,
665 1 : l0_flush_delay_threshold,
666 1 : l0_flush_stall_threshold,
667 1 : l0_flush_wait_upload,
668 1 : gc_horizon,
669 1 : gc_period,
670 1 : image_creation_threshold,
671 1 : pitr_interval,
672 1 : walreceiver_connect_timeout,
673 1 : lagging_wal_timeout,
674 1 : max_lsn_wal_lag,
675 1 : eviction_policy,
676 1 : min_resident_size_override,
677 1 : evictions_low_residence_duration_metric_threshold,
678 1 : heatmap_period,
679 1 : lazy_slru_download,
680 1 : timeline_get_throttle,
681 1 : image_layer_creation_check_threshold,
682 1 : lsn_lease_length,
683 1 : lsn_lease_length_for_ts,
684 1 : timeline_offloading,
685 1 : wal_receiver_protocol_override,
686 1 : rel_size_v2_enabled,
687 1 : gc_compaction_enabled,
688 1 : gc_compaction_initial_threshold_kb,
689 1 : gc_compaction_ratio_percent,
690 1 : }
691 1 : }
692 : }
693 :
694 : /// The policy for the aux file storage.
695 : ///
696 : /// It can be switched through `switch_aux_file_policy` tenant config.
697 : /// When the first aux file written, the policy will be persisted in the
698 : /// `index_part.json` file and has a limited migration path.
699 : ///
700 : /// Currently, we only allow the following migration path:
701 : ///
702 : /// Unset -> V1
703 : /// -> V2
704 : /// -> CrossValidation -> V2
705 : #[derive(
706 : Eq,
707 : PartialEq,
708 : Debug,
709 : Copy,
710 : Clone,
711 0 : strum_macros::EnumString,
712 : strum_macros::Display,
713 4 : serde_with::DeserializeFromStr,
714 : serde_with::SerializeDisplay,
715 : )]
716 : #[strum(serialize_all = "kebab-case")]
717 : pub enum AuxFilePolicy {
718 : /// V1 aux file policy: store everything in AUX_FILE_KEY
719 : #[strum(ascii_case_insensitive)]
720 : V1,
721 : /// V2 aux file policy: store in the AUX_FILE keyspace
722 : #[strum(ascii_case_insensitive)]
723 : V2,
724 : /// Cross validation runs both formats on the write path and does validation
725 : /// on the read path.
726 : #[strum(ascii_case_insensitive)]
727 : CrossValidation,
728 : }
729 :
730 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
731 : #[serde(tag = "kind")]
732 : pub enum EvictionPolicy {
733 : NoEviction,
734 : LayerAccessThreshold(EvictionPolicyLayerAccessThreshold),
735 : OnlyImitiate(EvictionPolicyLayerAccessThreshold),
736 : }
737 :
738 : impl EvictionPolicy {
739 0 : pub fn discriminant_str(&self) -> &'static str {
740 0 : match self {
741 0 : EvictionPolicy::NoEviction => "NoEviction",
742 0 : EvictionPolicy::LayerAccessThreshold(_) => "LayerAccessThreshold",
743 0 : EvictionPolicy::OnlyImitiate(_) => "OnlyImitiate",
744 : }
745 0 : }
746 : }
747 :
748 : #[derive(
749 : Eq,
750 : PartialEq,
751 : Debug,
752 : Copy,
753 : Clone,
754 0 : strum_macros::EnumString,
755 : strum_macros::Display,
756 0 : serde_with::DeserializeFromStr,
757 : serde_with::SerializeDisplay,
758 : )]
759 : #[strum(serialize_all = "kebab-case")]
760 : pub enum CompactionAlgorithm {
761 : Legacy,
762 : Tiered,
763 : }
764 :
765 : #[derive(
766 4 : Debug, Clone, Copy, PartialEq, Eq, serde_with::DeserializeFromStr, serde_with::SerializeDisplay,
767 : )]
768 : pub enum ImageCompressionAlgorithm {
769 : // Disabled for writes, support decompressing during read path
770 : Disabled,
771 : /// Zstandard compression. Level 0 means and None mean the same (default level). Levels can be negative as well.
772 : /// For details, see the [manual](http://facebook.github.io/zstd/zstd_manual.html).
773 : Zstd {
774 : level: Option<i8>,
775 : },
776 : }
777 :
778 : impl FromStr for ImageCompressionAlgorithm {
779 : type Err = anyhow::Error;
780 8 : fn from_str(s: &str) -> Result<Self, Self::Err> {
781 8 : let mut components = s.split(['(', ')']);
782 8 : let first = components
783 8 : .next()
784 8 : .ok_or_else(|| anyhow::anyhow!("empty string"))?;
785 8 : match first {
786 8 : "disabled" => Ok(ImageCompressionAlgorithm::Disabled),
787 6 : "zstd" => {
788 6 : let level = if let Some(v) = components.next() {
789 4 : let v: i8 = v.parse()?;
790 4 : Some(v)
791 : } else {
792 2 : None
793 : };
794 :
795 6 : Ok(ImageCompressionAlgorithm::Zstd { level })
796 : }
797 0 : _ => anyhow::bail!("invalid specifier '{first}'"),
798 : }
799 8 : }
800 : }
801 :
802 : impl Display for ImageCompressionAlgorithm {
803 12 : fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
804 12 : match self {
805 3 : ImageCompressionAlgorithm::Disabled => write!(f, "disabled"),
806 9 : ImageCompressionAlgorithm::Zstd { level } => {
807 9 : if let Some(level) = level {
808 6 : write!(f, "zstd({})", level)
809 : } else {
810 3 : write!(f, "zstd")
811 : }
812 : }
813 : }
814 12 : }
815 : }
816 :
817 0 : #[derive(Eq, PartialEq, Debug, Clone, Serialize, Deserialize)]
818 : pub struct CompactionAlgorithmSettings {
819 : pub kind: CompactionAlgorithm,
820 : }
821 :
822 8 : #[derive(Debug, PartialEq, Eq, Clone, Deserialize, Serialize)]
823 : #[serde(tag = "mode", rename_all = "kebab-case", deny_unknown_fields)]
824 : pub enum L0FlushConfig {
825 : #[serde(rename_all = "snake_case")]
826 : Direct { max_concurrency: NonZeroUsize },
827 : }
828 :
829 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
830 : pub struct EvictionPolicyLayerAccessThreshold {
831 : #[serde(with = "humantime_serde")]
832 : pub period: Duration,
833 : #[serde(with = "humantime_serde")]
834 : pub threshold: Duration,
835 : }
836 :
837 6 : #[derive(Debug, Serialize, Deserialize, Clone, PartialEq, Eq)]
838 : pub struct ThrottleConfig {
839 : /// See [`ThrottleConfigTaskKinds`] for why we do the serde `rename`.
840 : #[serde(rename = "task_kinds")]
841 : pub enabled: ThrottleConfigTaskKinds,
842 : pub initial: u32,
843 : #[serde(with = "humantime_serde")]
844 : pub refill_interval: Duration,
845 : pub refill_amount: NonZeroU32,
846 : pub max: u32,
847 : }
848 :
849 : /// Before <https://github.com/neondatabase/neon/pull/9962>
850 : /// the throttle was a per `Timeline::get`/`Timeline::get_vectored` call.
851 : /// The `task_kinds` field controlled which Pageserver "Task Kind"s
852 : /// were subject to the throttle.
853 : ///
854 : /// After that PR, the throttle is applied at pagestream request level
855 : /// and the `task_kinds` field does not apply since the only task kind
856 : /// that us subject to the throttle is that of the page service.
857 : ///
858 : /// However, we don't want to make a breaking config change right now
859 : /// because it means we have to migrate all the tenant configs.
860 : /// This will be done in a future PR.
861 : ///
862 : /// In the meantime, we use emptiness / non-emptsiness of the `task_kinds`
863 : /// field to determine if the throttle is enabled or not.
864 1 : #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq)]
865 : #[serde(transparent)]
866 : pub struct ThrottleConfigTaskKinds(Vec<String>);
867 :
868 : impl ThrottleConfigTaskKinds {
869 901 : pub fn disabled() -> Self {
870 901 : Self(vec![])
871 901 : }
872 442 : pub fn is_enabled(&self) -> bool {
873 442 : !self.0.is_empty()
874 442 : }
875 : }
876 :
877 : impl ThrottleConfig {
878 901 : pub fn disabled() -> Self {
879 901 : Self {
880 901 : enabled: ThrottleConfigTaskKinds::disabled(),
881 901 : // other values don't matter with emtpy `task_kinds`.
882 901 : initial: 0,
883 901 : refill_interval: Duration::from_millis(1),
884 901 : refill_amount: NonZeroU32::new(1).unwrap(),
885 901 : max: 1,
886 901 : }
887 901 : }
888 : /// The requests per second allowed by the given config.
889 0 : pub fn steady_rps(&self) -> f64 {
890 0 : (self.refill_amount.get() as f64) / (self.refill_interval.as_secs_f64())
891 0 : }
892 : }
893 :
894 : #[cfg(test)]
895 : mod throttle_config_tests {
896 : use super::*;
897 :
898 : #[test]
899 1 : fn test_disabled_is_disabled() {
900 1 : let config = ThrottleConfig::disabled();
901 1 : assert!(!config.enabled.is_enabled());
902 1 : }
903 : #[test]
904 1 : fn test_enabled_backwards_compat() {
905 1 : let input = serde_json::json!({
906 1 : "task_kinds": ["PageRequestHandler"],
907 1 : "initial": 40000,
908 1 : "refill_interval": "50ms",
909 1 : "refill_amount": 1000,
910 1 : "max": 40000,
911 1 : "fair": true
912 1 : });
913 1 : let config: ThrottleConfig = serde_json::from_value(input).unwrap();
914 1 : assert!(config.enabled.is_enabled());
915 1 : }
916 : }
917 :
918 : /// A flattened analog of a `pagesever::tenant::LocationMode`, which
919 : /// lists out all possible states (and the virtual "Detached" state)
920 : /// in a flat form rather than using rust-style enums.
921 0 : #[derive(Serialize, Deserialize, Debug, Clone, Copy, Eq, PartialEq)]
922 : pub enum LocationConfigMode {
923 : AttachedSingle,
924 : AttachedMulti,
925 : AttachedStale,
926 : Secondary,
927 : Detached,
928 : }
929 :
930 0 : #[derive(Serialize, Deserialize, Debug, Clone, Eq, PartialEq)]
931 : pub struct LocationConfigSecondary {
932 : pub warm: bool,
933 : }
934 :
935 : /// An alternative representation of `pageserver::tenant::LocationConf`,
936 : /// for use in external-facing APIs.
937 0 : #[derive(Serialize, Deserialize, Debug, Clone, Eq, PartialEq)]
938 : pub struct LocationConfig {
939 : pub mode: LocationConfigMode,
940 : /// If attaching, in what generation?
941 : #[serde(default)]
942 : pub generation: Option<u32>,
943 :
944 : // If requesting mode `Secondary`, configuration for that.
945 : #[serde(default)]
946 : pub secondary_conf: Option<LocationConfigSecondary>,
947 :
948 : // Shard parameters: if shard_count is nonzero, then other shard_* fields
949 : // must be set accurately.
950 : #[serde(default)]
951 : pub shard_number: u8,
952 : #[serde(default)]
953 : pub shard_count: u8,
954 : #[serde(default)]
955 : pub shard_stripe_size: u32,
956 :
957 : // This configuration only affects attached mode, but should be provided irrespective
958 : // of the mode, as a secondary location might transition on startup if the response
959 : // to the `/re-attach` control plane API requests it.
960 : pub tenant_conf: TenantConfig,
961 : }
962 :
963 0 : #[derive(Serialize, Deserialize)]
964 : pub struct LocationConfigListResponse {
965 : pub tenant_shards: Vec<(TenantShardId, Option<LocationConfig>)>,
966 : }
967 :
968 : #[derive(Serialize)]
969 : pub struct StatusResponse {
970 : pub id: NodeId,
971 : }
972 :
973 0 : #[derive(Serialize, Deserialize, Debug)]
974 : #[serde(deny_unknown_fields)]
975 : pub struct TenantLocationConfigRequest {
976 : #[serde(flatten)]
977 : pub config: LocationConfig, // as we have a flattened field, we should reject all unknown fields in it
978 : }
979 :
980 0 : #[derive(Serialize, Deserialize, Debug)]
981 : #[serde(deny_unknown_fields)]
982 : pub struct TenantTimeTravelRequest {
983 : pub shard_counts: Vec<ShardCount>,
984 : }
985 :
986 0 : #[derive(Serialize, Deserialize, Debug)]
987 : #[serde(deny_unknown_fields)]
988 : pub struct TenantShardLocation {
989 : pub shard_id: TenantShardId,
990 : pub node_id: NodeId,
991 : }
992 :
993 0 : #[derive(Serialize, Deserialize, Debug)]
994 : #[serde(deny_unknown_fields)]
995 : pub struct TenantLocationConfigResponse {
996 : pub shards: Vec<TenantShardLocation>,
997 : // If the shards' ShardCount count is >1, stripe_size will be set.
998 : pub stripe_size: Option<ShardStripeSize>,
999 : }
1000 :
1001 2 : #[derive(Serialize, Deserialize, Debug)]
1002 : #[serde(deny_unknown_fields)]
1003 : pub struct TenantConfigRequest {
1004 : pub tenant_id: TenantId,
1005 : #[serde(flatten)]
1006 : pub config: TenantConfig, // as we have a flattened field, we should reject all unknown fields in it
1007 : }
1008 :
1009 : impl std::ops::Deref for TenantConfigRequest {
1010 : type Target = TenantConfig;
1011 :
1012 0 : fn deref(&self) -> &Self::Target {
1013 0 : &self.config
1014 0 : }
1015 : }
1016 :
1017 : impl TenantConfigRequest {
1018 0 : pub fn new(tenant_id: TenantId) -> TenantConfigRequest {
1019 0 : let config = TenantConfig::default();
1020 0 : TenantConfigRequest { tenant_id, config }
1021 0 : }
1022 : }
1023 :
1024 3 : #[derive(Serialize, Deserialize, Debug)]
1025 : #[serde(deny_unknown_fields)]
1026 : pub struct TenantConfigPatchRequest {
1027 : pub tenant_id: TenantId,
1028 : #[serde(flatten)]
1029 : pub config: TenantConfigPatch, // as we have a flattened field, we should reject all unknown fields in it
1030 : }
1031 :
1032 0 : #[derive(Serialize, Deserialize, Debug)]
1033 : pub struct TenantWaitLsnRequest {
1034 : #[serde(flatten)]
1035 : pub timelines: HashMap<TimelineId, Lsn>,
1036 : pub timeout: Duration,
1037 : }
1038 :
1039 : /// See [`TenantState::attachment_status`] and the OpenAPI docs for context.
1040 0 : #[derive(Serialize, Deserialize, Clone)]
1041 : #[serde(tag = "slug", content = "data", rename_all = "snake_case")]
1042 : pub enum TenantAttachmentStatus {
1043 : Maybe,
1044 : Attached,
1045 : Failed { reason: String },
1046 : }
1047 :
1048 0 : #[derive(Serialize, Deserialize, Clone)]
1049 : pub struct TenantInfo {
1050 : pub id: TenantShardId,
1051 : // NB: intentionally not part of OpenAPI, we don't want to commit to a specific set of TenantState's
1052 : pub state: TenantState,
1053 : /// Sum of the size of all layer files.
1054 : /// If a layer is present in both local FS and S3, it counts only once.
1055 : pub current_physical_size: Option<u64>, // physical size is only included in `tenant_status` endpoint
1056 : pub attachment_status: TenantAttachmentStatus,
1057 : pub generation: u32,
1058 :
1059 : /// Opaque explanation if gc is being blocked.
1060 : ///
1061 : /// Only looked up for the individual tenant detail, not the listing. This is purely for
1062 : /// debugging, not included in openapi.
1063 : #[serde(skip_serializing_if = "Option::is_none")]
1064 : pub gc_blocking: Option<String>,
1065 : }
1066 :
1067 0 : #[derive(Serialize, Deserialize, Clone)]
1068 : pub struct TenantDetails {
1069 : #[serde(flatten)]
1070 : pub tenant_info: TenantInfo,
1071 :
1072 : pub walredo: Option<WalRedoManagerStatus>,
1073 :
1074 : pub timelines: Vec<TimelineId>,
1075 : }
1076 :
1077 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone, Copy, Debug)]
1078 : pub enum TimelineArchivalState {
1079 : Archived,
1080 : Unarchived,
1081 : }
1082 :
1083 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone)]
1084 : pub struct TimelineArchivalConfigRequest {
1085 : pub state: TimelineArchivalState,
1086 : }
1087 :
1088 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
1089 : pub struct TimelinesInfoAndOffloaded {
1090 : pub timelines: Vec<TimelineInfo>,
1091 : pub offloaded: Vec<OffloadedTimelineInfo>,
1092 : }
1093 :
1094 : /// Analog of [`TimelineInfo`] for offloaded timelines.
1095 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
1096 : pub struct OffloadedTimelineInfo {
1097 : pub tenant_id: TenantShardId,
1098 : pub timeline_id: TimelineId,
1099 : /// Whether the timeline has a parent it has been branched off from or not
1100 : pub ancestor_timeline_id: Option<TimelineId>,
1101 : /// Whether to retain the branch lsn at the ancestor or not
1102 : pub ancestor_retain_lsn: Option<Lsn>,
1103 : /// The time point when the timeline was archived
1104 : pub archived_at: chrono::DateTime<chrono::Utc>,
1105 : }
1106 :
1107 : /// This represents the output of the "timeline_detail" and "timeline_list" API calls.
1108 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
1109 : pub struct TimelineInfo {
1110 : pub tenant_id: TenantShardId,
1111 : pub timeline_id: TimelineId,
1112 :
1113 : pub ancestor_timeline_id: Option<TimelineId>,
1114 : pub ancestor_lsn: Option<Lsn>,
1115 : pub last_record_lsn: Lsn,
1116 : pub prev_record_lsn: Option<Lsn>,
1117 : pub latest_gc_cutoff_lsn: Lsn,
1118 : pub disk_consistent_lsn: Lsn,
1119 :
1120 : /// The LSN that we have succesfully uploaded to remote storage
1121 : pub remote_consistent_lsn: Lsn,
1122 :
1123 : /// The LSN that we are advertizing to safekeepers
1124 : pub remote_consistent_lsn_visible: Lsn,
1125 :
1126 : /// The LSN from the start of the root timeline (never changes)
1127 : pub initdb_lsn: Lsn,
1128 :
1129 : pub current_logical_size: u64,
1130 : pub current_logical_size_is_accurate: bool,
1131 :
1132 : pub directory_entries_counts: Vec<u64>,
1133 :
1134 : /// Sum of the size of all layer files.
1135 : /// If a layer is present in both local FS and S3, it counts only once.
1136 : pub current_physical_size: Option<u64>, // is None when timeline is Unloaded
1137 : pub current_logical_size_non_incremental: Option<u64>,
1138 :
1139 : /// How many bytes of WAL are within this branch's pitr_interval. If the pitr_interval goes
1140 : /// beyond the branch's branch point, we only count up to the branch point.
1141 : pub pitr_history_size: u64,
1142 :
1143 : /// Whether this branch's branch point is within its ancestor's PITR interval (i.e. any
1144 : /// ancestor data used by this branch would have been retained anyway). If this is false, then
1145 : /// this branch may be imposing a cost on the ancestor by causing it to retain layers that it would
1146 : /// otherwise be able to GC.
1147 : pub within_ancestor_pitr: bool,
1148 :
1149 : pub timeline_dir_layer_file_size_sum: Option<u64>,
1150 :
1151 : pub wal_source_connstr: Option<String>,
1152 : pub last_received_msg_lsn: Option<Lsn>,
1153 : /// the timestamp (in microseconds) of the last received message
1154 : pub last_received_msg_ts: Option<u128>,
1155 : pub pg_version: u32,
1156 :
1157 : pub state: TimelineState,
1158 :
1159 : pub walreceiver_status: String,
1160 :
1161 : // ALWAYS add new fields at the end of the struct with `Option` to ensure forward/backward compatibility.
1162 : // Backward compatibility: you will get a JSON not containing the newly-added field.
1163 : // Forward compatibility: a previous version of the pageserver will receive a JSON. serde::Deserialize does
1164 : // not deny unknown fields by default so it's safe to set the field to some value, though it won't be
1165 : // read.
1166 : pub is_archived: Option<bool>,
1167 : }
1168 :
1169 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1170 : pub struct LayerMapInfo {
1171 : pub in_memory_layers: Vec<InMemoryLayerInfo>,
1172 : pub historic_layers: Vec<HistoricLayerInfo>,
1173 : }
1174 :
1175 : /// The residence status of a layer
1176 0 : #[derive(Debug, Clone, Copy, Serialize, Deserialize)]
1177 : pub enum LayerResidenceStatus {
1178 : /// Residence status for a layer file that exists locally.
1179 : /// It may also exist on the remote, we don't care here.
1180 : Resident,
1181 : /// Residence status for a layer file that only exists on the remote.
1182 : Evicted,
1183 : }
1184 :
1185 : #[serde_as]
1186 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1187 : pub struct LayerAccessStats {
1188 : #[serde_as(as = "serde_with::TimestampMilliSeconds")]
1189 : pub access_time: SystemTime,
1190 :
1191 : #[serde_as(as = "serde_with::TimestampMilliSeconds")]
1192 : pub residence_time: SystemTime,
1193 :
1194 : pub visible: bool,
1195 : }
1196 :
1197 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1198 : #[serde(tag = "kind")]
1199 : pub enum InMemoryLayerInfo {
1200 : Open { lsn_start: Lsn },
1201 : Frozen { lsn_start: Lsn, lsn_end: Lsn },
1202 : }
1203 :
1204 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1205 : #[serde(tag = "kind")]
1206 : pub enum HistoricLayerInfo {
1207 : Delta {
1208 : layer_file_name: String,
1209 : layer_file_size: u64,
1210 :
1211 : lsn_start: Lsn,
1212 : lsn_end: Lsn,
1213 : remote: bool,
1214 : access_stats: LayerAccessStats,
1215 :
1216 : l0: bool,
1217 : },
1218 : Image {
1219 : layer_file_name: String,
1220 : layer_file_size: u64,
1221 :
1222 : lsn_start: Lsn,
1223 : remote: bool,
1224 : access_stats: LayerAccessStats,
1225 : },
1226 : }
1227 :
1228 : impl HistoricLayerInfo {
1229 0 : pub fn layer_file_name(&self) -> &str {
1230 0 : match self {
1231 : HistoricLayerInfo::Delta {
1232 0 : layer_file_name, ..
1233 0 : } => layer_file_name,
1234 : HistoricLayerInfo::Image {
1235 0 : layer_file_name, ..
1236 0 : } => layer_file_name,
1237 : }
1238 0 : }
1239 0 : pub fn is_remote(&self) -> bool {
1240 0 : match self {
1241 0 : HistoricLayerInfo::Delta { remote, .. } => *remote,
1242 0 : HistoricLayerInfo::Image { remote, .. } => *remote,
1243 : }
1244 0 : }
1245 0 : pub fn set_remote(&mut self, value: bool) {
1246 0 : let field = match self {
1247 0 : HistoricLayerInfo::Delta { remote, .. } => remote,
1248 0 : HistoricLayerInfo::Image { remote, .. } => remote,
1249 : };
1250 0 : *field = value;
1251 0 : }
1252 0 : pub fn layer_file_size(&self) -> u64 {
1253 0 : match self {
1254 : HistoricLayerInfo::Delta {
1255 0 : layer_file_size, ..
1256 0 : } => *layer_file_size,
1257 : HistoricLayerInfo::Image {
1258 0 : layer_file_size, ..
1259 0 : } => *layer_file_size,
1260 : }
1261 0 : }
1262 : }
1263 :
1264 0 : #[derive(Debug, Serialize, Deserialize)]
1265 : pub struct DownloadRemoteLayersTaskSpawnRequest {
1266 : pub max_concurrent_downloads: NonZeroUsize,
1267 : }
1268 :
1269 0 : #[derive(Debug, Serialize, Deserialize)]
1270 : pub struct IngestAuxFilesRequest {
1271 : pub aux_files: HashMap<String, String>,
1272 : }
1273 :
1274 0 : #[derive(Debug, Serialize, Deserialize)]
1275 : pub struct ListAuxFilesRequest {
1276 : pub lsn: Lsn,
1277 : }
1278 :
1279 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
1280 : pub struct DownloadRemoteLayersTaskInfo {
1281 : pub task_id: String,
1282 : pub state: DownloadRemoteLayersTaskState,
1283 : pub total_layer_count: u64, // stable once `completed`
1284 : pub successful_download_count: u64, // stable once `completed`
1285 : pub failed_download_count: u64, // stable once `completed`
1286 : }
1287 :
1288 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
1289 : pub enum DownloadRemoteLayersTaskState {
1290 : Running,
1291 : Completed,
1292 : ShutDown,
1293 : }
1294 :
1295 0 : #[derive(Debug, Serialize, Deserialize)]
1296 : pub struct TimelineGcRequest {
1297 : pub gc_horizon: Option<u64>,
1298 : }
1299 :
1300 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1301 : pub struct WalRedoManagerProcessStatus {
1302 : pub pid: u32,
1303 : }
1304 :
1305 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1306 : pub struct WalRedoManagerStatus {
1307 : pub last_redo_at: Option<chrono::DateTime<chrono::Utc>>,
1308 : pub process: Option<WalRedoManagerProcessStatus>,
1309 : }
1310 :
1311 : /// The progress of a secondary tenant.
1312 : ///
1313 : /// It is mostly useful when doing a long running download: e.g. initiating
1314 : /// a download job, timing out while waiting for it to run, and then inspecting this status to understand
1315 : /// what's happening.
1316 0 : #[derive(Default, Debug, Serialize, Deserialize, Clone)]
1317 : pub struct SecondaryProgress {
1318 : /// The remote storage LastModified time of the heatmap object we last downloaded.
1319 : pub heatmap_mtime: Option<serde_system_time::SystemTime>,
1320 :
1321 : /// The number of layers currently on-disk
1322 : pub layers_downloaded: usize,
1323 : /// The number of layers in the most recently seen heatmap
1324 : pub layers_total: usize,
1325 :
1326 : /// The number of layer bytes currently on-disk
1327 : pub bytes_downloaded: u64,
1328 : /// The number of layer bytes in the most recently seen heatmap
1329 : pub bytes_total: u64,
1330 : }
1331 :
1332 0 : #[derive(Serialize, Deserialize, Debug)]
1333 : pub struct TenantScanRemoteStorageShard {
1334 : pub tenant_shard_id: TenantShardId,
1335 : pub generation: Option<u32>,
1336 : }
1337 :
1338 0 : #[derive(Serialize, Deserialize, Debug, Default)]
1339 : pub struct TenantScanRemoteStorageResponse {
1340 : pub shards: Vec<TenantScanRemoteStorageShard>,
1341 : }
1342 :
1343 0 : #[derive(Serialize, Deserialize, Debug, Clone)]
1344 : #[serde(rename_all = "snake_case")]
1345 : pub enum TenantSorting {
1346 : ResidentSize,
1347 : MaxLogicalSize,
1348 : }
1349 :
1350 : impl Default for TenantSorting {
1351 0 : fn default() -> Self {
1352 0 : Self::ResidentSize
1353 0 : }
1354 : }
1355 :
1356 0 : #[derive(Serialize, Deserialize, Debug, Clone)]
1357 : pub struct TopTenantShardsRequest {
1358 : // How would you like to sort the tenants?
1359 : pub order_by: TenantSorting,
1360 :
1361 : // How many results?
1362 : pub limit: usize,
1363 :
1364 : // Omit tenants with more than this many shards (e.g. if this is the max number of shards
1365 : // that the caller would ever split to)
1366 : pub where_shards_lt: Option<ShardCount>,
1367 :
1368 : // Omit tenants where the ordering metric is less than this (this is an optimization to
1369 : // let us quickly exclude numerous tiny shards)
1370 : pub where_gt: Option<u64>,
1371 : }
1372 :
1373 0 : #[derive(Serialize, Deserialize, Debug, PartialEq, Eq)]
1374 : pub struct TopTenantShardItem {
1375 : pub id: TenantShardId,
1376 :
1377 : /// Total size of layers on local disk for all timelines in this tenant
1378 : pub resident_size: u64,
1379 :
1380 : /// Total size of layers in remote storage for all timelines in this tenant
1381 : pub physical_size: u64,
1382 :
1383 : /// The largest logical size of a timeline within this tenant
1384 : pub max_logical_size: u64,
1385 : }
1386 :
1387 0 : #[derive(Serialize, Deserialize, Debug, Default)]
1388 : pub struct TopTenantShardsResponse {
1389 : pub shards: Vec<TopTenantShardItem>,
1390 : }
1391 :
1392 : pub mod virtual_file {
1393 : #[derive(
1394 : Copy,
1395 : Clone,
1396 : PartialEq,
1397 : Eq,
1398 : Hash,
1399 0 : strum_macros::EnumString,
1400 : strum_macros::Display,
1401 0 : serde_with::DeserializeFromStr,
1402 : serde_with::SerializeDisplay,
1403 : Debug,
1404 : )]
1405 : #[strum(serialize_all = "kebab-case")]
1406 : pub enum IoEngineKind {
1407 : StdFs,
1408 : #[cfg(target_os = "linux")]
1409 : TokioEpollUring,
1410 : }
1411 :
1412 : /// Direct IO modes for a pageserver.
1413 : #[derive(
1414 : Copy,
1415 : Clone,
1416 : PartialEq,
1417 : Eq,
1418 : Hash,
1419 0 : strum_macros::EnumString,
1420 : strum_macros::Display,
1421 0 : serde_with::DeserializeFromStr,
1422 : serde_with::SerializeDisplay,
1423 : Debug,
1424 : )]
1425 : #[strum(serialize_all = "kebab-case")]
1426 : #[repr(u8)]
1427 : pub enum IoMode {
1428 : /// Uses buffered IO.
1429 : Buffered,
1430 : /// Uses direct IO, error out if the operation fails.
1431 : #[cfg(target_os = "linux")]
1432 : Direct,
1433 : }
1434 :
1435 : impl IoMode {
1436 476 : pub const fn preferred() -> Self {
1437 476 : Self::Buffered
1438 476 : }
1439 : }
1440 :
1441 : impl TryFrom<u8> for IoMode {
1442 : type Error = u8;
1443 :
1444 5064 : fn try_from(value: u8) -> Result<Self, Self::Error> {
1445 5064 : Ok(match value {
1446 5064 : v if v == (IoMode::Buffered as u8) => IoMode::Buffered,
1447 : #[cfg(target_os = "linux")]
1448 0 : v if v == (IoMode::Direct as u8) => IoMode::Direct,
1449 0 : x => return Err(x),
1450 : })
1451 5064 : }
1452 : }
1453 : }
1454 :
1455 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1456 : pub struct ScanDisposableKeysResponse {
1457 : pub disposable_count: usize,
1458 : pub not_disposable_count: usize,
1459 : }
1460 :
1461 : // Wrapped in libpq CopyData
1462 : #[derive(PartialEq, Eq, Debug)]
1463 : pub enum PagestreamFeMessage {
1464 : Exists(PagestreamExistsRequest),
1465 : Nblocks(PagestreamNblocksRequest),
1466 : GetPage(PagestreamGetPageRequest),
1467 : DbSize(PagestreamDbSizeRequest),
1468 : GetSlruSegment(PagestreamGetSlruSegmentRequest),
1469 : #[cfg(feature = "testing")]
1470 : Test(PagestreamTestRequest),
1471 : }
1472 :
1473 : // Wrapped in libpq CopyData
1474 : #[derive(strum_macros::EnumProperty)]
1475 : pub enum PagestreamBeMessage {
1476 : Exists(PagestreamExistsResponse),
1477 : Nblocks(PagestreamNblocksResponse),
1478 : GetPage(PagestreamGetPageResponse),
1479 : Error(PagestreamErrorResponse),
1480 : DbSize(PagestreamDbSizeResponse),
1481 : GetSlruSegment(PagestreamGetSlruSegmentResponse),
1482 : #[cfg(feature = "testing")]
1483 : Test(PagestreamTestResponse),
1484 : }
1485 :
1486 : // Keep in sync with `pagestore_client.h`
1487 : #[repr(u8)]
1488 : enum PagestreamFeMessageTag {
1489 : Exists = 0,
1490 : Nblocks = 1,
1491 : GetPage = 2,
1492 : DbSize = 3,
1493 : GetSlruSegment = 4,
1494 : /* future tags above this line */
1495 : /// For testing purposes, not available in production.
1496 : #[cfg(feature = "testing")]
1497 : Test = 99,
1498 : }
1499 :
1500 : // Keep in sync with `pagestore_client.h`
1501 : #[repr(u8)]
1502 : enum PagestreamBeMessageTag {
1503 : Exists = 100,
1504 : Nblocks = 101,
1505 : GetPage = 102,
1506 : Error = 103,
1507 : DbSize = 104,
1508 : GetSlruSegment = 105,
1509 : /* future tags above this line */
1510 : /// For testing purposes, not available in production.
1511 : #[cfg(feature = "testing")]
1512 : Test = 199,
1513 : }
1514 :
1515 : impl TryFrom<u8> for PagestreamFeMessageTag {
1516 : type Error = u8;
1517 4 : fn try_from(value: u8) -> Result<Self, u8> {
1518 4 : match value {
1519 1 : 0 => Ok(PagestreamFeMessageTag::Exists),
1520 1 : 1 => Ok(PagestreamFeMessageTag::Nblocks),
1521 1 : 2 => Ok(PagestreamFeMessageTag::GetPage),
1522 1 : 3 => Ok(PagestreamFeMessageTag::DbSize),
1523 0 : 4 => Ok(PagestreamFeMessageTag::GetSlruSegment),
1524 : #[cfg(feature = "testing")]
1525 0 : 99 => Ok(PagestreamFeMessageTag::Test),
1526 0 : _ => Err(value),
1527 : }
1528 4 : }
1529 : }
1530 :
1531 : impl TryFrom<u8> for PagestreamBeMessageTag {
1532 : type Error = u8;
1533 0 : fn try_from(value: u8) -> Result<Self, u8> {
1534 0 : match value {
1535 0 : 100 => Ok(PagestreamBeMessageTag::Exists),
1536 0 : 101 => Ok(PagestreamBeMessageTag::Nblocks),
1537 0 : 102 => Ok(PagestreamBeMessageTag::GetPage),
1538 0 : 103 => Ok(PagestreamBeMessageTag::Error),
1539 0 : 104 => Ok(PagestreamBeMessageTag::DbSize),
1540 0 : 105 => Ok(PagestreamBeMessageTag::GetSlruSegment),
1541 : #[cfg(feature = "testing")]
1542 0 : 199 => Ok(PagestreamBeMessageTag::Test),
1543 0 : _ => Err(value),
1544 : }
1545 0 : }
1546 : }
1547 :
1548 : // A GetPage request contains two LSN values:
1549 : //
1550 : // request_lsn: Get the page version at this point in time. Lsn::Max is a special value that means
1551 : // "get the latest version present". It's used by the primary server, which knows that no one else
1552 : // is writing WAL. 'not_modified_since' must be set to a proper value even if request_lsn is
1553 : // Lsn::Max. Standby servers use the current replay LSN as the request LSN.
1554 : //
1555 : // not_modified_since: Hint to the pageserver that the client knows that the page has not been
1556 : // modified between 'not_modified_since' and the request LSN. It's always correct to set
1557 : // 'not_modified_since equal' to 'request_lsn' (unless Lsn::Max is used as the 'request_lsn'), but
1558 : // passing an earlier LSN can speed up the request, by allowing the pageserver to process the
1559 : // request without waiting for 'request_lsn' to arrive.
1560 : //
1561 : // The now-defunct V1 interface contained only one LSN, and a boolean 'latest' flag. The V1 interface was
1562 : // sufficient for the primary; the 'lsn' was equivalent to the 'not_modified_since' value, and
1563 : // 'latest' was set to true. The V2 interface was added because there was no correct way for a
1564 : // standby to request a page at a particular non-latest LSN, and also include the
1565 : // 'not_modified_since' hint. That led to an awkward choice of either using an old LSN in the
1566 : // request, if the standby knows that the page hasn't been modified since, and risk getting an error
1567 : // if that LSN has fallen behind the GC horizon, or requesting the current replay LSN, which could
1568 : // require the pageserver unnecessarily to wait for the WAL to arrive up to that point. The new V2
1569 : // interface allows sending both LSNs, and let the pageserver do the right thing. There was no
1570 : // difference in the responses between V1 and V2.
1571 : //
1572 : // V3 version of protocol adds request ID to all requests. This request ID is also included in response
1573 : // as well as other fields from requests, which allows to verify that we receive response for our request.
1574 : // We copy fields from request to response to make checking more reliable: request ID is formed from process ID
1575 : // and local counter, so in principle there can be duplicated requests IDs if process PID is reused.
1576 : //
1577 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
1578 : pub enum PagestreamProtocolVersion {
1579 : V2,
1580 : V3,
1581 : }
1582 :
1583 : pub type RequestId = u64;
1584 :
1585 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
1586 : pub struct PagestreamRequest {
1587 : pub reqid: RequestId,
1588 : pub request_lsn: Lsn,
1589 : pub not_modified_since: Lsn,
1590 : }
1591 :
1592 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
1593 : pub struct PagestreamExistsRequest {
1594 : pub hdr: PagestreamRequest,
1595 : pub rel: RelTag,
1596 : }
1597 :
1598 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
1599 : pub struct PagestreamNblocksRequest {
1600 : pub hdr: PagestreamRequest,
1601 : pub rel: RelTag,
1602 : }
1603 :
1604 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
1605 : pub struct PagestreamGetPageRequest {
1606 : pub hdr: PagestreamRequest,
1607 : pub rel: RelTag,
1608 : pub blkno: u32,
1609 : }
1610 :
1611 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
1612 : pub struct PagestreamDbSizeRequest {
1613 : pub hdr: PagestreamRequest,
1614 : pub dbnode: u32,
1615 : }
1616 :
1617 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
1618 : pub struct PagestreamGetSlruSegmentRequest {
1619 : pub hdr: PagestreamRequest,
1620 : pub kind: u8,
1621 : pub segno: u32,
1622 : }
1623 :
1624 : #[derive(Debug)]
1625 : pub struct PagestreamExistsResponse {
1626 : pub req: PagestreamExistsRequest,
1627 : pub exists: bool,
1628 : }
1629 :
1630 : #[derive(Debug)]
1631 : pub struct PagestreamNblocksResponse {
1632 : pub req: PagestreamNblocksRequest,
1633 : pub n_blocks: u32,
1634 : }
1635 :
1636 : #[derive(Debug)]
1637 : pub struct PagestreamGetPageResponse {
1638 : pub req: PagestreamGetPageRequest,
1639 : pub page: Bytes,
1640 : }
1641 :
1642 : #[derive(Debug)]
1643 : pub struct PagestreamGetSlruSegmentResponse {
1644 : pub req: PagestreamGetSlruSegmentRequest,
1645 : pub segment: Bytes,
1646 : }
1647 :
1648 : #[derive(Debug)]
1649 : pub struct PagestreamErrorResponse {
1650 : pub req: PagestreamRequest,
1651 : pub message: String,
1652 : }
1653 :
1654 : #[derive(Debug)]
1655 : pub struct PagestreamDbSizeResponse {
1656 : pub req: PagestreamDbSizeRequest,
1657 : pub db_size: i64,
1658 : }
1659 :
1660 : #[cfg(feature = "testing")]
1661 : #[derive(Debug, PartialEq, Eq, Clone)]
1662 : pub struct PagestreamTestRequest {
1663 : pub hdr: PagestreamRequest,
1664 : pub batch_key: u64,
1665 : pub message: String,
1666 : }
1667 :
1668 : #[cfg(feature = "testing")]
1669 : #[derive(Debug)]
1670 : pub struct PagestreamTestResponse {
1671 : pub req: PagestreamTestRequest,
1672 : }
1673 :
1674 : // This is a cut-down version of TenantHistorySize from the pageserver crate, omitting fields
1675 : // that require pageserver-internal types. It is sufficient to get the total size.
1676 0 : #[derive(Serialize, Deserialize, Debug)]
1677 : pub struct TenantHistorySize {
1678 : pub id: TenantId,
1679 : /// Size is a mixture of WAL and logical size, so the unit is bytes.
1680 : ///
1681 : /// Will be none if `?inputs_only=true` was given.
1682 : pub size: Option<u64>,
1683 : }
1684 :
1685 : impl PagestreamFeMessage {
1686 : /// Serialize a compute -> pageserver message. This is currently only used in testing
1687 : /// tools. Always uses protocol version 3.
1688 4 : pub fn serialize(&self) -> Bytes {
1689 4 : let mut bytes = BytesMut::new();
1690 4 :
1691 4 : match self {
1692 1 : Self::Exists(req) => {
1693 1 : bytes.put_u8(PagestreamFeMessageTag::Exists as u8);
1694 1 : bytes.put_u64(req.hdr.reqid);
1695 1 : bytes.put_u64(req.hdr.request_lsn.0);
1696 1 : bytes.put_u64(req.hdr.not_modified_since.0);
1697 1 : bytes.put_u32(req.rel.spcnode);
1698 1 : bytes.put_u32(req.rel.dbnode);
1699 1 : bytes.put_u32(req.rel.relnode);
1700 1 : bytes.put_u8(req.rel.forknum);
1701 1 : }
1702 :
1703 1 : Self::Nblocks(req) => {
1704 1 : bytes.put_u8(PagestreamFeMessageTag::Nblocks as u8);
1705 1 : bytes.put_u64(req.hdr.reqid);
1706 1 : bytes.put_u64(req.hdr.request_lsn.0);
1707 1 : bytes.put_u64(req.hdr.not_modified_since.0);
1708 1 : bytes.put_u32(req.rel.spcnode);
1709 1 : bytes.put_u32(req.rel.dbnode);
1710 1 : bytes.put_u32(req.rel.relnode);
1711 1 : bytes.put_u8(req.rel.forknum);
1712 1 : }
1713 :
1714 1 : Self::GetPage(req) => {
1715 1 : bytes.put_u8(PagestreamFeMessageTag::GetPage as u8);
1716 1 : bytes.put_u64(req.hdr.reqid);
1717 1 : bytes.put_u64(req.hdr.request_lsn.0);
1718 1 : bytes.put_u64(req.hdr.not_modified_since.0);
1719 1 : bytes.put_u32(req.rel.spcnode);
1720 1 : bytes.put_u32(req.rel.dbnode);
1721 1 : bytes.put_u32(req.rel.relnode);
1722 1 : bytes.put_u8(req.rel.forknum);
1723 1 : bytes.put_u32(req.blkno);
1724 1 : }
1725 :
1726 1 : Self::DbSize(req) => {
1727 1 : bytes.put_u8(PagestreamFeMessageTag::DbSize as u8);
1728 1 : bytes.put_u64(req.hdr.reqid);
1729 1 : bytes.put_u64(req.hdr.request_lsn.0);
1730 1 : bytes.put_u64(req.hdr.not_modified_since.0);
1731 1 : bytes.put_u32(req.dbnode);
1732 1 : }
1733 :
1734 0 : Self::GetSlruSegment(req) => {
1735 0 : bytes.put_u8(PagestreamFeMessageTag::GetSlruSegment as u8);
1736 0 : bytes.put_u64(req.hdr.reqid);
1737 0 : bytes.put_u64(req.hdr.request_lsn.0);
1738 0 : bytes.put_u64(req.hdr.not_modified_since.0);
1739 0 : bytes.put_u8(req.kind);
1740 0 : bytes.put_u32(req.segno);
1741 0 : }
1742 : #[cfg(feature = "testing")]
1743 0 : Self::Test(req) => {
1744 0 : bytes.put_u8(PagestreamFeMessageTag::Test as u8);
1745 0 : bytes.put_u64(req.hdr.reqid);
1746 0 : bytes.put_u64(req.hdr.request_lsn.0);
1747 0 : bytes.put_u64(req.hdr.not_modified_since.0);
1748 0 : bytes.put_u64(req.batch_key);
1749 0 : let message = req.message.as_bytes();
1750 0 : bytes.put_u64(message.len() as u64);
1751 0 : bytes.put_slice(message);
1752 0 : }
1753 : }
1754 :
1755 4 : bytes.into()
1756 4 : }
1757 :
1758 4 : pub fn parse<R: std::io::Read>(
1759 4 : body: &mut R,
1760 4 : protocol_version: PagestreamProtocolVersion,
1761 4 : ) -> anyhow::Result<PagestreamFeMessage> {
1762 : // these correspond to the NeonMessageTag enum in pagestore_client.h
1763 : //
1764 : // TODO: consider using protobuf or serde bincode for less error prone
1765 : // serialization.
1766 4 : let msg_tag = body.read_u8()?;
1767 4 : let (reqid, request_lsn, not_modified_since) = match protocol_version {
1768 : PagestreamProtocolVersion::V2 => (
1769 : 0,
1770 0 : Lsn::from(body.read_u64::<BigEndian>()?),
1771 0 : Lsn::from(body.read_u64::<BigEndian>()?),
1772 : ),
1773 : PagestreamProtocolVersion::V3 => (
1774 4 : body.read_u64::<BigEndian>()?,
1775 4 : Lsn::from(body.read_u64::<BigEndian>()?),
1776 4 : Lsn::from(body.read_u64::<BigEndian>()?),
1777 : ),
1778 : };
1779 :
1780 4 : match PagestreamFeMessageTag::try_from(msg_tag)
1781 4 : .map_err(|tag: u8| anyhow::anyhow!("invalid tag {tag}"))?
1782 : {
1783 : PagestreamFeMessageTag::Exists => {
1784 : Ok(PagestreamFeMessage::Exists(PagestreamExistsRequest {
1785 1 : hdr: PagestreamRequest {
1786 1 : reqid,
1787 1 : request_lsn,
1788 1 : not_modified_since,
1789 1 : },
1790 1 : rel: RelTag {
1791 1 : spcnode: body.read_u32::<BigEndian>()?,
1792 1 : dbnode: body.read_u32::<BigEndian>()?,
1793 1 : relnode: body.read_u32::<BigEndian>()?,
1794 1 : forknum: body.read_u8()?,
1795 : },
1796 : }))
1797 : }
1798 : PagestreamFeMessageTag::Nblocks => {
1799 : Ok(PagestreamFeMessage::Nblocks(PagestreamNblocksRequest {
1800 1 : hdr: PagestreamRequest {
1801 1 : reqid,
1802 1 : request_lsn,
1803 1 : not_modified_since,
1804 1 : },
1805 1 : rel: RelTag {
1806 1 : spcnode: body.read_u32::<BigEndian>()?,
1807 1 : dbnode: body.read_u32::<BigEndian>()?,
1808 1 : relnode: body.read_u32::<BigEndian>()?,
1809 1 : forknum: body.read_u8()?,
1810 : },
1811 : }))
1812 : }
1813 : PagestreamFeMessageTag::GetPage => {
1814 : Ok(PagestreamFeMessage::GetPage(PagestreamGetPageRequest {
1815 1 : hdr: PagestreamRequest {
1816 1 : reqid,
1817 1 : request_lsn,
1818 1 : not_modified_since,
1819 1 : },
1820 1 : rel: RelTag {
1821 1 : spcnode: body.read_u32::<BigEndian>()?,
1822 1 : dbnode: body.read_u32::<BigEndian>()?,
1823 1 : relnode: body.read_u32::<BigEndian>()?,
1824 1 : forknum: body.read_u8()?,
1825 : },
1826 1 : blkno: body.read_u32::<BigEndian>()?,
1827 : }))
1828 : }
1829 : PagestreamFeMessageTag::DbSize => {
1830 : Ok(PagestreamFeMessage::DbSize(PagestreamDbSizeRequest {
1831 1 : hdr: PagestreamRequest {
1832 1 : reqid,
1833 1 : request_lsn,
1834 1 : not_modified_since,
1835 1 : },
1836 1 : dbnode: body.read_u32::<BigEndian>()?,
1837 : }))
1838 : }
1839 : PagestreamFeMessageTag::GetSlruSegment => Ok(PagestreamFeMessage::GetSlruSegment(
1840 : PagestreamGetSlruSegmentRequest {
1841 0 : hdr: PagestreamRequest {
1842 0 : reqid,
1843 0 : request_lsn,
1844 0 : not_modified_since,
1845 0 : },
1846 0 : kind: body.read_u8()?,
1847 0 : segno: body.read_u32::<BigEndian>()?,
1848 : },
1849 : )),
1850 : #[cfg(feature = "testing")]
1851 : PagestreamFeMessageTag::Test => Ok(PagestreamFeMessage::Test(PagestreamTestRequest {
1852 0 : hdr: PagestreamRequest {
1853 0 : reqid,
1854 0 : request_lsn,
1855 0 : not_modified_since,
1856 0 : },
1857 0 : batch_key: body.read_u64::<BigEndian>()?,
1858 : message: {
1859 0 : let len = body.read_u64::<BigEndian>()?;
1860 0 : let mut buf = vec![0; len as usize];
1861 0 : body.read_exact(&mut buf)?;
1862 0 : String::from_utf8(buf)?
1863 : },
1864 : })),
1865 : }
1866 4 : }
1867 : }
1868 :
1869 : impl PagestreamBeMessage {
1870 0 : pub fn serialize(&self, protocol_version: PagestreamProtocolVersion) -> Bytes {
1871 0 : let mut bytes = BytesMut::new();
1872 :
1873 : use PagestreamBeMessageTag as Tag;
1874 0 : match protocol_version {
1875 : PagestreamProtocolVersion::V2 => {
1876 0 : match self {
1877 0 : Self::Exists(resp) => {
1878 0 : bytes.put_u8(Tag::Exists as u8);
1879 0 : bytes.put_u8(resp.exists as u8);
1880 0 : }
1881 :
1882 0 : Self::Nblocks(resp) => {
1883 0 : bytes.put_u8(Tag::Nblocks as u8);
1884 0 : bytes.put_u32(resp.n_blocks);
1885 0 : }
1886 :
1887 0 : Self::GetPage(resp) => {
1888 0 : bytes.put_u8(Tag::GetPage as u8);
1889 0 : bytes.put(&resp.page[..])
1890 : }
1891 :
1892 0 : Self::Error(resp) => {
1893 0 : bytes.put_u8(Tag::Error as u8);
1894 0 : bytes.put(resp.message.as_bytes());
1895 0 : bytes.put_u8(0); // null terminator
1896 0 : }
1897 0 : Self::DbSize(resp) => {
1898 0 : bytes.put_u8(Tag::DbSize as u8);
1899 0 : bytes.put_i64(resp.db_size);
1900 0 : }
1901 :
1902 0 : Self::GetSlruSegment(resp) => {
1903 0 : bytes.put_u8(Tag::GetSlruSegment as u8);
1904 0 : bytes.put_u32((resp.segment.len() / BLCKSZ as usize) as u32);
1905 0 : bytes.put(&resp.segment[..]);
1906 0 : }
1907 :
1908 : #[cfg(feature = "testing")]
1909 0 : Self::Test(resp) => {
1910 0 : bytes.put_u8(Tag::Test as u8);
1911 0 : bytes.put_u64(resp.req.batch_key);
1912 0 : let message = resp.req.message.as_bytes();
1913 0 : bytes.put_u64(message.len() as u64);
1914 0 : bytes.put_slice(message);
1915 0 : }
1916 : }
1917 : }
1918 : PagestreamProtocolVersion::V3 => {
1919 0 : match self {
1920 0 : Self::Exists(resp) => {
1921 0 : bytes.put_u8(Tag::Exists as u8);
1922 0 : bytes.put_u64(resp.req.hdr.reqid);
1923 0 : bytes.put_u64(resp.req.hdr.request_lsn.0);
1924 0 : bytes.put_u64(resp.req.hdr.not_modified_since.0);
1925 0 : bytes.put_u32(resp.req.rel.spcnode);
1926 0 : bytes.put_u32(resp.req.rel.dbnode);
1927 0 : bytes.put_u32(resp.req.rel.relnode);
1928 0 : bytes.put_u8(resp.req.rel.forknum);
1929 0 : bytes.put_u8(resp.exists as u8);
1930 0 : }
1931 :
1932 0 : Self::Nblocks(resp) => {
1933 0 : bytes.put_u8(Tag::Nblocks as u8);
1934 0 : bytes.put_u64(resp.req.hdr.reqid);
1935 0 : bytes.put_u64(resp.req.hdr.request_lsn.0);
1936 0 : bytes.put_u64(resp.req.hdr.not_modified_since.0);
1937 0 : bytes.put_u32(resp.req.rel.spcnode);
1938 0 : bytes.put_u32(resp.req.rel.dbnode);
1939 0 : bytes.put_u32(resp.req.rel.relnode);
1940 0 : bytes.put_u8(resp.req.rel.forknum);
1941 0 : bytes.put_u32(resp.n_blocks);
1942 0 : }
1943 :
1944 0 : Self::GetPage(resp) => {
1945 0 : bytes.put_u8(Tag::GetPage as u8);
1946 0 : bytes.put_u64(resp.req.hdr.reqid);
1947 0 : bytes.put_u64(resp.req.hdr.request_lsn.0);
1948 0 : bytes.put_u64(resp.req.hdr.not_modified_since.0);
1949 0 : bytes.put_u32(resp.req.rel.spcnode);
1950 0 : bytes.put_u32(resp.req.rel.dbnode);
1951 0 : bytes.put_u32(resp.req.rel.relnode);
1952 0 : bytes.put_u8(resp.req.rel.forknum);
1953 0 : bytes.put_u32(resp.req.blkno);
1954 0 : bytes.put(&resp.page[..])
1955 : }
1956 :
1957 0 : Self::Error(resp) => {
1958 0 : bytes.put_u8(Tag::Error as u8);
1959 0 : bytes.put_u64(resp.req.reqid);
1960 0 : bytes.put_u64(resp.req.request_lsn.0);
1961 0 : bytes.put_u64(resp.req.not_modified_since.0);
1962 0 : bytes.put(resp.message.as_bytes());
1963 0 : bytes.put_u8(0); // null terminator
1964 0 : }
1965 0 : Self::DbSize(resp) => {
1966 0 : bytes.put_u8(Tag::DbSize as u8);
1967 0 : bytes.put_u64(resp.req.hdr.reqid);
1968 0 : bytes.put_u64(resp.req.hdr.request_lsn.0);
1969 0 : bytes.put_u64(resp.req.hdr.not_modified_since.0);
1970 0 : bytes.put_u32(resp.req.dbnode);
1971 0 : bytes.put_i64(resp.db_size);
1972 0 : }
1973 :
1974 0 : Self::GetSlruSegment(resp) => {
1975 0 : bytes.put_u8(Tag::GetSlruSegment as u8);
1976 0 : bytes.put_u64(resp.req.hdr.reqid);
1977 0 : bytes.put_u64(resp.req.hdr.request_lsn.0);
1978 0 : bytes.put_u64(resp.req.hdr.not_modified_since.0);
1979 0 : bytes.put_u8(resp.req.kind);
1980 0 : bytes.put_u32(resp.req.segno);
1981 0 : bytes.put_u32((resp.segment.len() / BLCKSZ as usize) as u32);
1982 0 : bytes.put(&resp.segment[..]);
1983 0 : }
1984 :
1985 : #[cfg(feature = "testing")]
1986 0 : Self::Test(resp) => {
1987 0 : bytes.put_u8(Tag::Test as u8);
1988 0 : bytes.put_u64(resp.req.hdr.reqid);
1989 0 : bytes.put_u64(resp.req.hdr.request_lsn.0);
1990 0 : bytes.put_u64(resp.req.hdr.not_modified_since.0);
1991 0 : bytes.put_u64(resp.req.batch_key);
1992 0 : let message = resp.req.message.as_bytes();
1993 0 : bytes.put_u64(message.len() as u64);
1994 0 : bytes.put_slice(message);
1995 0 : }
1996 : }
1997 : }
1998 : }
1999 0 : bytes.into()
2000 0 : }
2001 :
2002 0 : pub fn deserialize(buf: Bytes) -> anyhow::Result<Self> {
2003 0 : let mut buf = buf.reader();
2004 0 : let msg_tag = buf.read_u8()?;
2005 :
2006 : use PagestreamBeMessageTag as Tag;
2007 0 : let ok =
2008 0 : match Tag::try_from(msg_tag).map_err(|tag: u8| anyhow::anyhow!("invalid tag {tag}"))? {
2009 : Tag::Exists => {
2010 0 : let reqid = buf.read_u64::<BigEndian>()?;
2011 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2012 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2013 0 : let rel = RelTag {
2014 0 : spcnode: buf.read_u32::<BigEndian>()?,
2015 0 : dbnode: buf.read_u32::<BigEndian>()?,
2016 0 : relnode: buf.read_u32::<BigEndian>()?,
2017 0 : forknum: buf.read_u8()?,
2018 : };
2019 0 : let exists = buf.read_u8()? != 0;
2020 0 : Self::Exists(PagestreamExistsResponse {
2021 0 : req: PagestreamExistsRequest {
2022 0 : hdr: PagestreamRequest {
2023 0 : reqid,
2024 0 : request_lsn,
2025 0 : not_modified_since,
2026 0 : },
2027 0 : rel,
2028 0 : },
2029 0 : exists,
2030 0 : })
2031 : }
2032 : Tag::Nblocks => {
2033 0 : let reqid = buf.read_u64::<BigEndian>()?;
2034 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2035 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2036 0 : let rel = RelTag {
2037 0 : spcnode: buf.read_u32::<BigEndian>()?,
2038 0 : dbnode: buf.read_u32::<BigEndian>()?,
2039 0 : relnode: buf.read_u32::<BigEndian>()?,
2040 0 : forknum: buf.read_u8()?,
2041 : };
2042 0 : let n_blocks = buf.read_u32::<BigEndian>()?;
2043 0 : Self::Nblocks(PagestreamNblocksResponse {
2044 0 : req: PagestreamNblocksRequest {
2045 0 : hdr: PagestreamRequest {
2046 0 : reqid,
2047 0 : request_lsn,
2048 0 : not_modified_since,
2049 0 : },
2050 0 : rel,
2051 0 : },
2052 0 : n_blocks,
2053 0 : })
2054 : }
2055 : Tag::GetPage => {
2056 0 : let reqid = buf.read_u64::<BigEndian>()?;
2057 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2058 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2059 0 : let rel = RelTag {
2060 0 : spcnode: buf.read_u32::<BigEndian>()?,
2061 0 : dbnode: buf.read_u32::<BigEndian>()?,
2062 0 : relnode: buf.read_u32::<BigEndian>()?,
2063 0 : forknum: buf.read_u8()?,
2064 : };
2065 0 : let blkno = buf.read_u32::<BigEndian>()?;
2066 0 : let mut page = vec![0; 8192]; // TODO: use MaybeUninit
2067 0 : buf.read_exact(&mut page)?;
2068 0 : Self::GetPage(PagestreamGetPageResponse {
2069 0 : req: PagestreamGetPageRequest {
2070 0 : hdr: PagestreamRequest {
2071 0 : reqid,
2072 0 : request_lsn,
2073 0 : not_modified_since,
2074 0 : },
2075 0 : rel,
2076 0 : blkno,
2077 0 : },
2078 0 : page: page.into(),
2079 0 : })
2080 : }
2081 : Tag::Error => {
2082 0 : let reqid = buf.read_u64::<BigEndian>()?;
2083 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2084 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2085 0 : let mut msg = Vec::new();
2086 0 : buf.read_until(0, &mut msg)?;
2087 0 : let cstring = std::ffi::CString::from_vec_with_nul(msg)?;
2088 0 : let rust_str = cstring.to_str()?;
2089 0 : Self::Error(PagestreamErrorResponse {
2090 0 : req: PagestreamRequest {
2091 0 : reqid,
2092 0 : request_lsn,
2093 0 : not_modified_since,
2094 0 : },
2095 0 : message: rust_str.to_owned(),
2096 0 : })
2097 : }
2098 : Tag::DbSize => {
2099 0 : let reqid = buf.read_u64::<BigEndian>()?;
2100 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2101 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2102 0 : let dbnode = buf.read_u32::<BigEndian>()?;
2103 0 : let db_size = buf.read_i64::<BigEndian>()?;
2104 0 : Self::DbSize(PagestreamDbSizeResponse {
2105 0 : req: PagestreamDbSizeRequest {
2106 0 : hdr: PagestreamRequest {
2107 0 : reqid,
2108 0 : request_lsn,
2109 0 : not_modified_since,
2110 0 : },
2111 0 : dbnode,
2112 0 : },
2113 0 : db_size,
2114 0 : })
2115 : }
2116 : Tag::GetSlruSegment => {
2117 0 : let reqid = buf.read_u64::<BigEndian>()?;
2118 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2119 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2120 0 : let kind = buf.read_u8()?;
2121 0 : let segno = buf.read_u32::<BigEndian>()?;
2122 0 : let n_blocks = buf.read_u32::<BigEndian>()?;
2123 0 : let mut segment = vec![0; n_blocks as usize * BLCKSZ as usize];
2124 0 : buf.read_exact(&mut segment)?;
2125 0 : Self::GetSlruSegment(PagestreamGetSlruSegmentResponse {
2126 0 : req: PagestreamGetSlruSegmentRequest {
2127 0 : hdr: PagestreamRequest {
2128 0 : reqid,
2129 0 : request_lsn,
2130 0 : not_modified_since,
2131 0 : },
2132 0 : kind,
2133 0 : segno,
2134 0 : },
2135 0 : segment: segment.into(),
2136 0 : })
2137 : }
2138 : #[cfg(feature = "testing")]
2139 : Tag::Test => {
2140 0 : let reqid = buf.read_u64::<BigEndian>()?;
2141 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2142 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2143 0 : let batch_key = buf.read_u64::<BigEndian>()?;
2144 0 : let len = buf.read_u64::<BigEndian>()?;
2145 0 : let mut msg = vec![0; len as usize];
2146 0 : buf.read_exact(&mut msg)?;
2147 0 : let message = String::from_utf8(msg)?;
2148 0 : Self::Test(PagestreamTestResponse {
2149 0 : req: PagestreamTestRequest {
2150 0 : hdr: PagestreamRequest {
2151 0 : reqid,
2152 0 : request_lsn,
2153 0 : not_modified_since,
2154 0 : },
2155 0 : batch_key,
2156 0 : message,
2157 0 : },
2158 0 : })
2159 : }
2160 : };
2161 0 : let remaining = buf.into_inner();
2162 0 : if !remaining.is_empty() {
2163 0 : anyhow::bail!(
2164 0 : "remaining bytes in msg with tag={msg_tag}: {}",
2165 0 : remaining.len()
2166 0 : );
2167 0 : }
2168 0 : Ok(ok)
2169 0 : }
2170 :
2171 0 : pub fn kind(&self) -> &'static str {
2172 0 : match self {
2173 0 : Self::Exists(_) => "Exists",
2174 0 : Self::Nblocks(_) => "Nblocks",
2175 0 : Self::GetPage(_) => "GetPage",
2176 0 : Self::Error(_) => "Error",
2177 0 : Self::DbSize(_) => "DbSize",
2178 0 : Self::GetSlruSegment(_) => "GetSlruSegment",
2179 : #[cfg(feature = "testing")]
2180 0 : Self::Test(_) => "Test",
2181 : }
2182 0 : }
2183 : }
2184 :
2185 0 : #[derive(Debug, Serialize, Deserialize)]
2186 : pub struct PageTraceEvent {
2187 : pub key: CompactKey,
2188 : pub effective_lsn: Lsn,
2189 : pub time: SystemTime,
2190 : }
2191 :
2192 : impl Default for PageTraceEvent {
2193 0 : fn default() -> Self {
2194 0 : Self {
2195 0 : key: Default::default(),
2196 0 : effective_lsn: Default::default(),
2197 0 : time: std::time::UNIX_EPOCH,
2198 0 : }
2199 0 : }
2200 : }
2201 :
2202 : #[cfg(test)]
2203 : mod tests {
2204 : use serde_json::json;
2205 : use std::str::FromStr;
2206 :
2207 : use super::*;
2208 :
2209 : #[test]
2210 1 : fn test_pagestream() {
2211 1 : // Test serialization/deserialization of PagestreamFeMessage
2212 1 : let messages = vec![
2213 1 : PagestreamFeMessage::Exists(PagestreamExistsRequest {
2214 1 : hdr: PagestreamRequest {
2215 1 : reqid: 0,
2216 1 : request_lsn: Lsn(4),
2217 1 : not_modified_since: Lsn(3),
2218 1 : },
2219 1 : rel: RelTag {
2220 1 : forknum: 1,
2221 1 : spcnode: 2,
2222 1 : dbnode: 3,
2223 1 : relnode: 4,
2224 1 : },
2225 1 : }),
2226 1 : PagestreamFeMessage::Nblocks(PagestreamNblocksRequest {
2227 1 : hdr: PagestreamRequest {
2228 1 : reqid: 0,
2229 1 : request_lsn: Lsn(4),
2230 1 : not_modified_since: Lsn(4),
2231 1 : },
2232 1 : rel: RelTag {
2233 1 : forknum: 1,
2234 1 : spcnode: 2,
2235 1 : dbnode: 3,
2236 1 : relnode: 4,
2237 1 : },
2238 1 : }),
2239 1 : PagestreamFeMessage::GetPage(PagestreamGetPageRequest {
2240 1 : hdr: PagestreamRequest {
2241 1 : reqid: 0,
2242 1 : request_lsn: Lsn(4),
2243 1 : not_modified_since: Lsn(3),
2244 1 : },
2245 1 : rel: RelTag {
2246 1 : forknum: 1,
2247 1 : spcnode: 2,
2248 1 : dbnode: 3,
2249 1 : relnode: 4,
2250 1 : },
2251 1 : blkno: 7,
2252 1 : }),
2253 1 : PagestreamFeMessage::DbSize(PagestreamDbSizeRequest {
2254 1 : hdr: PagestreamRequest {
2255 1 : reqid: 0,
2256 1 : request_lsn: Lsn(4),
2257 1 : not_modified_since: Lsn(3),
2258 1 : },
2259 1 : dbnode: 7,
2260 1 : }),
2261 1 : ];
2262 5 : for msg in messages {
2263 4 : let bytes = msg.serialize();
2264 4 : let reconstructed =
2265 4 : PagestreamFeMessage::parse(&mut bytes.reader(), PagestreamProtocolVersion::V3)
2266 4 : .unwrap();
2267 4 : assert!(msg == reconstructed);
2268 : }
2269 1 : }
2270 :
2271 : #[test]
2272 1 : fn test_tenantinfo_serde() {
2273 1 : // Test serialization/deserialization of TenantInfo
2274 1 : let original_active = TenantInfo {
2275 1 : id: TenantShardId::unsharded(TenantId::generate()),
2276 1 : state: TenantState::Active,
2277 1 : current_physical_size: Some(42),
2278 1 : attachment_status: TenantAttachmentStatus::Attached,
2279 1 : generation: 1,
2280 1 : gc_blocking: None,
2281 1 : };
2282 1 : let expected_active = json!({
2283 1 : "id": original_active.id.to_string(),
2284 1 : "state": {
2285 1 : "slug": "Active",
2286 1 : },
2287 1 : "current_physical_size": 42,
2288 1 : "attachment_status": {
2289 1 : "slug":"attached",
2290 1 : },
2291 1 : "generation" : 1
2292 1 : });
2293 1 :
2294 1 : let original_broken = TenantInfo {
2295 1 : id: TenantShardId::unsharded(TenantId::generate()),
2296 1 : state: TenantState::Broken {
2297 1 : reason: "reason".into(),
2298 1 : backtrace: "backtrace info".into(),
2299 1 : },
2300 1 : current_physical_size: Some(42),
2301 1 : attachment_status: TenantAttachmentStatus::Attached,
2302 1 : generation: 1,
2303 1 : gc_blocking: None,
2304 1 : };
2305 1 : let expected_broken = json!({
2306 1 : "id": original_broken.id.to_string(),
2307 1 : "state": {
2308 1 : "slug": "Broken",
2309 1 : "data": {
2310 1 : "backtrace": "backtrace info",
2311 1 : "reason": "reason",
2312 1 : }
2313 1 : },
2314 1 : "current_physical_size": 42,
2315 1 : "attachment_status": {
2316 1 : "slug":"attached",
2317 1 : },
2318 1 : "generation" : 1
2319 1 : });
2320 1 :
2321 1 : assert_eq!(
2322 1 : serde_json::to_value(&original_active).unwrap(),
2323 1 : expected_active
2324 1 : );
2325 :
2326 1 : assert_eq!(
2327 1 : serde_json::to_value(&original_broken).unwrap(),
2328 1 : expected_broken
2329 1 : );
2330 1 : assert!(format!("{:?}", &original_broken.state).contains("reason"));
2331 1 : assert!(format!("{:?}", &original_broken.state).contains("backtrace info"));
2332 1 : }
2333 :
2334 : #[test]
2335 1 : fn test_reject_unknown_field() {
2336 1 : let id = TenantId::generate();
2337 1 : let config_request = json!({
2338 1 : "tenant_id": id.to_string(),
2339 1 : "unknown_field": "unknown_value".to_string(),
2340 1 : });
2341 1 : let err = serde_json::from_value::<TenantConfigRequest>(config_request).unwrap_err();
2342 1 : assert!(
2343 1 : err.to_string().contains("unknown field `unknown_field`"),
2344 0 : "expect unknown field `unknown_field` error, got: {}",
2345 : err
2346 : );
2347 1 : }
2348 :
2349 : #[test]
2350 1 : fn tenantstatus_activating_serde() {
2351 1 : let states = [TenantState::Activating(ActivatingFrom::Attaching)];
2352 1 : let expected = "[{\"slug\":\"Activating\",\"data\":\"Attaching\"}]";
2353 1 :
2354 1 : let actual = serde_json::to_string(&states).unwrap();
2355 1 :
2356 1 : assert_eq!(actual, expected);
2357 :
2358 1 : let parsed = serde_json::from_str::<Vec<TenantState>>(&actual).unwrap();
2359 1 :
2360 1 : assert_eq!(states.as_slice(), &parsed);
2361 1 : }
2362 :
2363 : #[test]
2364 1 : fn tenantstatus_activating_strum() {
2365 1 : // tests added, because we use these for metrics
2366 1 : let examples = [
2367 1 : (line!(), TenantState::Attaching, "Attaching"),
2368 1 : (
2369 1 : line!(),
2370 1 : TenantState::Activating(ActivatingFrom::Attaching),
2371 1 : "Activating",
2372 1 : ),
2373 1 : (line!(), TenantState::Active, "Active"),
2374 1 : (
2375 1 : line!(),
2376 1 : TenantState::Stopping {
2377 1 : progress: utils::completion::Barrier::default(),
2378 1 : },
2379 1 : "Stopping",
2380 1 : ),
2381 1 : (
2382 1 : line!(),
2383 1 : TenantState::Broken {
2384 1 : reason: "Example".into(),
2385 1 : backtrace: "Looooong backtrace".into(),
2386 1 : },
2387 1 : "Broken",
2388 1 : ),
2389 1 : ];
2390 :
2391 6 : for (line, rendered, expected) in examples {
2392 5 : let actual: &'static str = rendered.into();
2393 5 : assert_eq!(actual, expected, "example on {line}");
2394 : }
2395 1 : }
2396 :
2397 : #[test]
2398 1 : fn test_image_compression_algorithm_parsing() {
2399 : use ImageCompressionAlgorithm::*;
2400 1 : let cases = [
2401 1 : ("disabled", Disabled),
2402 1 : ("zstd", Zstd { level: None }),
2403 1 : ("zstd(18)", Zstd { level: Some(18) }),
2404 1 : ("zstd(-3)", Zstd { level: Some(-3) }),
2405 1 : ];
2406 :
2407 5 : for (display, expected) in cases {
2408 4 : assert_eq!(
2409 4 : ImageCompressionAlgorithm::from_str(display).unwrap(),
2410 : expected,
2411 0 : "parsing works"
2412 : );
2413 4 : assert_eq!(format!("{expected}"), display, "Display FromStr roundtrip");
2414 :
2415 4 : let ser = serde_json::to_string(&expected).expect("serialization");
2416 4 : assert_eq!(
2417 4 : serde_json::from_str::<ImageCompressionAlgorithm>(&ser).unwrap(),
2418 : expected,
2419 0 : "serde roundtrip"
2420 : );
2421 :
2422 4 : assert_eq!(
2423 4 : serde_json::Value::String(display.to_string()),
2424 4 : serde_json::to_value(expected).unwrap(),
2425 0 : "Display is the serde serialization"
2426 : );
2427 : }
2428 1 : }
2429 :
2430 : #[test]
2431 1 : fn test_tenant_config_patch_request_serde() {
2432 1 : let patch_request = TenantConfigPatchRequest {
2433 1 : tenant_id: TenantId::from_str("17c6d121946a61e5ab0fe5a2fd4d8215").unwrap(),
2434 1 : config: TenantConfigPatch {
2435 1 : checkpoint_distance: FieldPatch::Upsert(42),
2436 1 : gc_horizon: FieldPatch::Remove,
2437 1 : compaction_threshold: FieldPatch::Noop,
2438 1 : ..TenantConfigPatch::default()
2439 1 : },
2440 1 : };
2441 1 :
2442 1 : let json = serde_json::to_string(&patch_request).unwrap();
2443 1 :
2444 1 : let expected = r#"{"tenant_id":"17c6d121946a61e5ab0fe5a2fd4d8215","checkpoint_distance":42,"gc_horizon":null}"#;
2445 1 : assert_eq!(json, expected);
2446 :
2447 1 : let decoded: TenantConfigPatchRequest = serde_json::from_str(&json).unwrap();
2448 1 : assert_eq!(decoded.tenant_id, patch_request.tenant_id);
2449 1 : assert_eq!(decoded.config, patch_request.config);
2450 :
2451 : // Now apply the patch to a config to demonstrate semantics
2452 :
2453 1 : let base = TenantConfig {
2454 1 : checkpoint_distance: Some(28),
2455 1 : gc_horizon: Some(100),
2456 1 : compaction_target_size: Some(1024),
2457 1 : ..Default::default()
2458 1 : };
2459 1 :
2460 1 : let expected = TenantConfig {
2461 1 : checkpoint_distance: Some(42),
2462 1 : gc_horizon: None,
2463 1 : ..base.clone()
2464 1 : };
2465 1 :
2466 1 : let patched = base.apply_patch(decoded.config);
2467 1 :
2468 1 : assert_eq!(patched, expected);
2469 1 : }
2470 : }
|
