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::key::{CompactKey, Key};
27 : use crate::reltag::RelTag;
28 : use crate::shard::{ShardCount, ShardStripeSize, TenantShardId};
29 :
30 : /// The state of a tenant in this pageserver.
31 : ///
32 : /// ```mermaid
33 : /// stateDiagram-v2
34 : ///
35 : /// [*] --> Attaching: spawn_attach()
36 : ///
37 : /// Attaching --> Activating: activate()
38 : /// Activating --> Active: infallible
39 : ///
40 : /// Attaching --> Broken: attach() failure
41 : ///
42 : /// Active --> Stopping: set_stopping(), part of shutdown & detach
43 : /// Stopping --> Broken: late error in remove_tenant_from_memory
44 : ///
45 : /// Broken --> [*]: ignore / detach / shutdown
46 : /// Stopping --> [*]: remove_from_memory complete
47 : ///
48 : /// Active --> Broken: cfg(testing)-only tenant break point
49 : /// ```
50 : #[derive(
51 : Clone,
52 : PartialEq,
53 : Eq,
54 0 : serde::Serialize,
55 1 : serde::Deserialize,
56 : strum_macros::Display,
57 : strum_macros::VariantNames,
58 : strum_macros::AsRefStr,
59 : strum_macros::IntoStaticStr,
60 : )]
61 : #[serde(tag = "slug", content = "data")]
62 : pub enum TenantState {
63 : /// This tenant is being attached to the pageserver.
64 : ///
65 : /// `set_stopping()` and `set_broken()` do not work in this state and wait for it to pass.
66 : Attaching,
67 : /// The tenant is transitioning from Loading/Attaching to Active.
68 : ///
69 : /// While in this state, the individual timelines are being activated.
70 : ///
71 : /// `set_stopping()` and `set_broken()` do not work in this state and wait for it to pass.
72 : Activating(ActivatingFrom),
73 : /// The tenant has finished activating and is open for business.
74 : ///
75 : /// Transitions out of this state are possible through `set_stopping()` and `set_broken()`.
76 : Active,
77 : /// The tenant is recognized by pageserver, but it is being detached or the
78 : /// system is being shut down.
79 : ///
80 : /// Transitions out of this state are possible through `set_broken()`.
81 : Stopping {
82 : // Because of https://github.com/serde-rs/serde/issues/2105 this has to be a named field,
83 : // otherwise it will not be skipped during deserialization
84 : #[serde(skip)]
85 : progress: completion::Barrier,
86 : },
87 : /// The tenant is recognized by the pageserver, but can no longer be used for
88 : /// any operations.
89 : ///
90 : /// If the tenant fails to load or attach, it will transition to this state
91 : /// and it is guaranteed that no background tasks are running in its name.
92 : ///
93 : /// The other way to transition into this state is from `Stopping` state
94 : /// through `set_broken()` called from `remove_tenant_from_memory()`. That happens
95 : /// if the cleanup future executed by `remove_tenant_from_memory()` fails.
96 : Broken { reason: String, backtrace: String },
97 : }
98 :
99 : impl TenantState {
100 0 : pub fn attachment_status(&self) -> TenantAttachmentStatus {
101 : use TenantAttachmentStatus::*;
102 :
103 : // Below TenantState::Activating is used as "transient" or "transparent" state for
104 : // attachment_status determining.
105 0 : match self {
106 : // The attach procedure writes the marker file before adding the Attaching tenant to the tenants map.
107 : // So, technically, we can return Attached here.
108 : // However, as soon as Console observes Attached, it will proceed with the Postgres-level health check.
109 : // But, our attach task might still be fetching the remote timelines, etc.
110 : // So, return `Maybe` while Attaching, making Console wait for the attach task to finish.
111 0 : Self::Attaching | Self::Activating(ActivatingFrom::Attaching) => Maybe,
112 : // We only reach Active after successful load / attach.
113 : // So, call atttachment status Attached.
114 0 : Self::Active => Attached,
115 : // If the (initial or resumed) attach procedure fails, the tenant becomes Broken.
116 : // However, it also becomes Broken if the regular load fails.
117 : // From Console's perspective there's no practical difference
118 : // because attachment_status is polled by console only during attach operation execution.
119 0 : Self::Broken { reason, .. } => Failed {
120 0 : reason: reason.to_owned(),
121 0 : },
122 : // Why is Stopping a Maybe case? Because, during pageserver shutdown,
123 : // we set the Stopping state irrespective of whether the tenant
124 : // has finished attaching or not.
125 0 : Self::Stopping { .. } => Maybe,
126 : }
127 0 : }
128 :
129 0 : pub fn broken_from_reason(reason: String) -> Self {
130 0 : let backtrace_str: String = format!("{}", std::backtrace::Backtrace::force_capture());
131 0 : Self::Broken {
132 0 : reason,
133 0 : backtrace: backtrace_str,
134 0 : }
135 0 : }
136 : }
137 :
138 : impl std::fmt::Debug for TenantState {
139 2 : fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
140 2 : match self {
141 2 : Self::Broken { reason, backtrace } if !reason.is_empty() => {
142 2 : write!(f, "Broken due to: {reason}. Backtrace:\n{backtrace}")
143 : }
144 0 : _ => write!(f, "{self}"),
145 : }
146 2 : }
147 : }
148 :
149 : /// A temporary lease to a specific lsn inside a timeline.
150 : /// Access to the lsn is guaranteed by the pageserver until the expiration indicated by `valid_until`.
151 : #[serde_as]
152 0 : #[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
153 : pub struct LsnLease {
154 : #[serde_as(as = "SystemTimeAsRfc3339Millis")]
155 : pub valid_until: SystemTime,
156 : }
157 :
158 : serde_with::serde_conv!(
159 : SystemTimeAsRfc3339Millis,
160 : SystemTime,
161 0 : |time: &SystemTime| humantime::format_rfc3339_millis(*time).to_string(),
162 0 : |value: String| -> Result<_, humantime::TimestampError> { humantime::parse_rfc3339(&value) }
163 : );
164 :
165 : impl LsnLease {
166 : /// The default length for an explicit LSN lease request (10 minutes).
167 : pub const DEFAULT_LENGTH: Duration = Duration::from_secs(10 * 60);
168 :
169 : /// The default length for an implicit LSN lease granted during
170 : /// `get_lsn_by_timestamp` request (1 minutes).
171 : pub const DEFAULT_LENGTH_FOR_TS: Duration = Duration::from_secs(60);
172 :
173 : /// Checks whether the lease is expired.
174 12 : pub fn is_expired(&self, now: &SystemTime) -> bool {
175 12 : now > &self.valid_until
176 12 : }
177 : }
178 :
179 : /// The only [`TenantState`] variants we could be `TenantState::Activating` from.
180 : ///
181 : /// XXX: We used to have more variants here, but now it's just one, which makes this rather
182 : /// useless. Remove, once we've checked that there's no client code left that looks at this.
183 1 : #[derive(Clone, Copy, Debug, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
184 : pub enum ActivatingFrom {
185 : /// Arrived to [`TenantState::Activating`] from [`TenantState::Attaching`]
186 : Attaching,
187 : }
188 :
189 : /// A state of a timeline in pageserver's memory.
190 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
191 : pub enum TimelineState {
192 : /// The timeline is recognized by the pageserver but is not yet operational.
193 : /// In particular, the walreceiver connection loop is not running for this timeline.
194 : /// It will eventually transition to state Active or Broken.
195 : Loading,
196 : /// The timeline is fully operational.
197 : /// It can be queried, and the walreceiver connection loop is running.
198 : Active,
199 : /// The timeline was previously Loading or Active but is shutting down.
200 : /// It cannot transition back into any other state.
201 : Stopping,
202 : /// The timeline is broken and not operational (previous states: Loading or Active).
203 : Broken { reason: String, backtrace: String },
204 : }
205 :
206 : #[serde_with::serde_as]
207 0 : #[derive(Debug, Clone, serde::Deserialize, serde::Serialize)]
208 : pub struct CompactLsnRange {
209 : pub start: Lsn,
210 : pub end: Lsn,
211 : }
212 :
213 : #[serde_with::serde_as]
214 0 : #[derive(Debug, Clone, serde::Deserialize, serde::Serialize)]
215 : pub struct CompactKeyRange {
216 : #[serde_as(as = "serde_with::DisplayFromStr")]
217 : pub start: Key,
218 : #[serde_as(as = "serde_with::DisplayFromStr")]
219 : pub end: Key,
220 : }
221 :
222 : impl From<Range<Lsn>> for CompactLsnRange {
223 12 : fn from(range: Range<Lsn>) -> Self {
224 12 : Self {
225 12 : start: range.start,
226 12 : end: range.end,
227 12 : }
228 12 : }
229 : }
230 :
231 : impl From<Range<Key>> for CompactKeyRange {
232 32 : fn from(range: Range<Key>) -> Self {
233 32 : Self {
234 32 : start: range.start,
235 32 : end: range.end,
236 32 : }
237 32 : }
238 : }
239 :
240 : impl From<CompactLsnRange> for Range<Lsn> {
241 20 : fn from(range: CompactLsnRange) -> Self {
242 20 : range.start..range.end
243 20 : }
244 : }
245 :
246 : impl From<CompactKeyRange> for Range<Key> {
247 32 : fn from(range: CompactKeyRange) -> Self {
248 32 : range.start..range.end
249 32 : }
250 : }
251 :
252 : impl CompactLsnRange {
253 8 : pub fn above(lsn: Lsn) -> Self {
254 8 : Self {
255 8 : start: lsn,
256 8 : end: Lsn::MAX,
257 8 : }
258 8 : }
259 : }
260 :
261 : #[derive(Debug, Clone, Serialize)]
262 : pub struct CompactInfoResponse {
263 : pub compact_key_range: Option<CompactKeyRange>,
264 : pub compact_lsn_range: Option<CompactLsnRange>,
265 : pub sub_compaction: bool,
266 : pub running: bool,
267 : pub job_id: usize,
268 : }
269 :
270 0 : #[derive(Serialize, Deserialize, Clone)]
271 : pub struct TimelineCreateRequest {
272 : pub new_timeline_id: TimelineId,
273 : #[serde(flatten)]
274 : pub mode: TimelineCreateRequestMode,
275 : }
276 :
277 : /// Storage controller specific extensions to [`TimelineInfo`].
278 0 : #[derive(Serialize, Deserialize, Clone)]
279 : pub struct TimelineCreateResponseStorcon {
280 : #[serde(flatten)]
281 : pub timeline_info: TimelineInfo,
282 :
283 : pub safekeepers: Option<SafekeepersInfo>,
284 : }
285 :
286 : /// Safekeepers as returned in timeline creation request to storcon or pushed to
287 : /// cplane in the post migration hook.
288 0 : #[derive(Serialize, Deserialize, Clone)]
289 : pub struct SafekeepersInfo {
290 : pub tenant_id: TenantId,
291 : pub timeline_id: TimelineId,
292 : pub generation: u32,
293 : pub safekeepers: Vec<SafekeeperInfo>,
294 : }
295 :
296 0 : #[derive(Serialize, Deserialize, Clone)]
297 : pub struct SafekeeperInfo {
298 : pub id: NodeId,
299 : pub hostname: String,
300 : }
301 :
302 0 : #[derive(Serialize, Deserialize, Clone)]
303 : #[serde(untagged)]
304 : pub enum TimelineCreateRequestMode {
305 : Branch {
306 : ancestor_timeline_id: TimelineId,
307 : #[serde(default)]
308 : ancestor_start_lsn: Option<Lsn>,
309 : // TODO: cplane sets this, but, the branching code always
310 : // inherits the ancestor's pg_version. Earlier code wasn't
311 : // using a flattened enum, so, it was an accepted field, and
312 : // we continue to accept it by having it here.
313 : pg_version: Option<u32>,
314 : },
315 : ImportPgdata {
316 : import_pgdata: TimelineCreateRequestModeImportPgdata,
317 : },
318 : // NB: Bootstrap is all-optional, and thus the serde(untagged) will cause serde to stop at Bootstrap.
319 : // (serde picks the first matching enum variant, in declaration order).
320 : Bootstrap {
321 : #[serde(default)]
322 : existing_initdb_timeline_id: Option<TimelineId>,
323 : pg_version: Option<u32>,
324 : },
325 : }
326 :
327 0 : #[derive(Serialize, Deserialize, Clone)]
328 : pub struct TimelineCreateRequestModeImportPgdata {
329 : pub location: ImportPgdataLocation,
330 : pub idempotency_key: ImportPgdataIdempotencyKey,
331 : }
332 :
333 0 : #[derive(Serialize, Deserialize, Clone, Debug)]
334 : pub enum ImportPgdataLocation {
335 : #[cfg(feature = "testing")]
336 : LocalFs { path: Utf8PathBuf },
337 : AwsS3 {
338 : region: String,
339 : bucket: String,
340 : /// A better name for this would be `prefix`; changing requires coordination with cplane.
341 : /// See <https://github.com/neondatabase/cloud/issues/20646>.
342 : key: String,
343 : },
344 : }
345 :
346 0 : #[derive(Serialize, Deserialize, Clone)]
347 : #[serde(transparent)]
348 : pub struct ImportPgdataIdempotencyKey(pub String);
349 :
350 : impl ImportPgdataIdempotencyKey {
351 0 : pub fn random() -> Self {
352 : use rand::Rng;
353 : use rand::distributions::Alphanumeric;
354 0 : Self(
355 0 : rand::thread_rng()
356 0 : .sample_iter(&Alphanumeric)
357 0 : .take(20)
358 0 : .map(char::from)
359 0 : .collect(),
360 0 : )
361 0 : }
362 : }
363 :
364 0 : #[derive(Serialize, Deserialize, Clone)]
365 : pub struct LsnLeaseRequest {
366 : pub lsn: Lsn,
367 : }
368 :
369 0 : #[derive(Serialize, Deserialize)]
370 : pub struct TenantShardSplitRequest {
371 : pub new_shard_count: u8,
372 :
373 : // A tenant's stripe size is only meaningful the first time their shard count goes
374 : // above 1: therefore during a split from 1->N shards, we may modify the stripe size.
375 : //
376 : // If this is set while the stripe count is being increased from an already >1 value,
377 : // then the request will fail with 400.
378 : pub new_stripe_size: Option<ShardStripeSize>,
379 : }
380 :
381 0 : #[derive(Serialize, Deserialize)]
382 : pub struct TenantShardSplitResponse {
383 : pub new_shards: Vec<TenantShardId>,
384 : }
385 :
386 : /// Parameters that apply to all shards in a tenant. Used during tenant creation.
387 0 : #[derive(Serialize, Deserialize, Debug)]
388 : #[serde(deny_unknown_fields)]
389 : pub struct ShardParameters {
390 : pub count: ShardCount,
391 : pub stripe_size: ShardStripeSize,
392 : }
393 :
394 : impl ShardParameters {
395 : pub const DEFAULT_STRIPE_SIZE: ShardStripeSize = ShardStripeSize(256 * 1024 / 8);
396 :
397 0 : pub fn is_unsharded(&self) -> bool {
398 0 : self.count.is_unsharded()
399 0 : }
400 : }
401 :
402 : impl Default for ShardParameters {
403 453 : fn default() -> Self {
404 453 : Self {
405 453 : count: ShardCount::new(0),
406 453 : stripe_size: Self::DEFAULT_STRIPE_SIZE,
407 453 : }
408 453 : }
409 : }
410 :
411 : #[derive(Debug, Default, Clone, Eq, PartialEq)]
412 : pub enum FieldPatch<T> {
413 : Upsert(T),
414 : Remove,
415 : #[default]
416 : Noop,
417 : }
418 :
419 : impl<T> FieldPatch<T> {
420 70 : fn is_noop(&self) -> bool {
421 70 : matches!(self, FieldPatch::Noop)
422 70 : }
423 :
424 35 : pub fn apply(self, target: &mut Option<T>) {
425 35 : match self {
426 1 : Self::Upsert(v) => *target = Some(v),
427 1 : Self::Remove => *target = None,
428 33 : Self::Noop => {}
429 : }
430 35 : }
431 :
432 10 : pub fn map<U, E, F: FnOnce(T) -> Result<U, E>>(self, map: F) -> Result<FieldPatch<U>, E> {
433 10 : match self {
434 0 : Self::Upsert(v) => Ok(FieldPatch::<U>::Upsert(map(v)?)),
435 0 : Self::Remove => Ok(FieldPatch::<U>::Remove),
436 10 : Self::Noop => Ok(FieldPatch::<U>::Noop),
437 : }
438 10 : }
439 : }
440 :
441 : impl<'de, T: Deserialize<'de>> Deserialize<'de> for FieldPatch<T> {
442 2 : fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
443 2 : where
444 2 : D: Deserializer<'de>,
445 2 : {
446 2 : Option::deserialize(deserializer).map(|opt| match opt {
447 1 : None => FieldPatch::Remove,
448 1 : Some(val) => FieldPatch::Upsert(val),
449 2 : })
450 2 : }
451 : }
452 :
453 : impl<T: Serialize> Serialize for FieldPatch<T> {
454 2 : fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
455 2 : where
456 2 : S: Serializer,
457 2 : {
458 2 : match self {
459 1 : FieldPatch::Upsert(val) => serializer.serialize_some(val),
460 1 : FieldPatch::Remove => serializer.serialize_none(),
461 0 : FieldPatch::Noop => unreachable!(),
462 : }
463 2 : }
464 : }
465 :
466 2 : #[derive(Serialize, Deserialize, Debug, Default, Clone, Eq, PartialEq)]
467 : #[serde(default)]
468 : pub struct TenantConfigPatch {
469 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
470 : pub checkpoint_distance: FieldPatch<u64>,
471 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
472 : pub checkpoint_timeout: FieldPatch<String>,
473 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
474 : pub compaction_target_size: FieldPatch<u64>,
475 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
476 : pub compaction_period: FieldPatch<String>,
477 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
478 : pub compaction_threshold: FieldPatch<usize>,
479 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
480 : pub compaction_upper_limit: FieldPatch<usize>,
481 : // defer parsing compaction_algorithm, like eviction_policy
482 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
483 : pub compaction_algorithm: FieldPatch<CompactionAlgorithmSettings>,
484 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
485 : pub compaction_l0_first: FieldPatch<bool>,
486 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
487 : pub compaction_l0_semaphore: FieldPatch<bool>,
488 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
489 : pub l0_flush_delay_threshold: FieldPatch<usize>,
490 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
491 : pub l0_flush_stall_threshold: FieldPatch<usize>,
492 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
493 : pub l0_flush_wait_upload: FieldPatch<bool>,
494 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
495 : pub gc_horizon: FieldPatch<u64>,
496 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
497 : pub gc_period: FieldPatch<String>,
498 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
499 : pub image_creation_threshold: FieldPatch<usize>,
500 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
501 : pub pitr_interval: FieldPatch<String>,
502 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
503 : pub walreceiver_connect_timeout: FieldPatch<String>,
504 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
505 : pub lagging_wal_timeout: FieldPatch<String>,
506 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
507 : pub max_lsn_wal_lag: FieldPatch<NonZeroU64>,
508 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
509 : pub eviction_policy: FieldPatch<EvictionPolicy>,
510 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
511 : pub min_resident_size_override: FieldPatch<u64>,
512 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
513 : pub evictions_low_residence_duration_metric_threshold: FieldPatch<String>,
514 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
515 : pub heatmap_period: FieldPatch<String>,
516 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
517 : pub lazy_slru_download: FieldPatch<bool>,
518 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
519 : pub timeline_get_throttle: FieldPatch<ThrottleConfig>,
520 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
521 : pub image_layer_creation_check_threshold: FieldPatch<u8>,
522 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
523 : pub image_creation_preempt_threshold: FieldPatch<usize>,
524 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
525 : pub lsn_lease_length: FieldPatch<String>,
526 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
527 : pub lsn_lease_length_for_ts: FieldPatch<String>,
528 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
529 : pub timeline_offloading: FieldPatch<bool>,
530 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
531 : pub wal_receiver_protocol_override: FieldPatch<PostgresClientProtocol>,
532 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
533 : pub rel_size_v2_enabled: FieldPatch<bool>,
534 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
535 : pub gc_compaction_enabled: FieldPatch<bool>,
536 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
537 : pub gc_compaction_initial_threshold_kb: FieldPatch<u64>,
538 : #[serde(skip_serializing_if = "FieldPatch::is_noop")]
539 : pub gc_compaction_ratio_percent: FieldPatch<u64>,
540 : }
541 :
542 : /// An alternative representation of `pageserver::tenant::TenantConf` with
543 : /// simpler types.
544 40 : #[derive(Serialize, Deserialize, Debug, Default, Clone, Eq, PartialEq)]
545 : pub struct TenantConfig {
546 : pub checkpoint_distance: Option<u64>,
547 : #[serde(default)]
548 : #[serde(with = "humantime_serde")]
549 : pub checkpoint_timeout: Option<Duration>,
550 : pub compaction_target_size: Option<u64>,
551 : #[serde(default)]
552 : #[serde(with = "humantime_serde")]
553 : pub compaction_period: Option<Duration>,
554 : pub compaction_threshold: Option<usize>,
555 : pub compaction_upper_limit: Option<usize>,
556 : // defer parsing compaction_algorithm, like eviction_policy
557 : pub compaction_algorithm: Option<CompactionAlgorithmSettings>,
558 : pub compaction_l0_first: Option<bool>,
559 : pub compaction_l0_semaphore: Option<bool>,
560 : pub l0_flush_delay_threshold: Option<usize>,
561 : pub l0_flush_stall_threshold: Option<usize>,
562 : pub l0_flush_wait_upload: Option<bool>,
563 : pub gc_horizon: Option<u64>,
564 : #[serde(default)]
565 : #[serde(with = "humantime_serde")]
566 : pub gc_period: Option<Duration>,
567 : pub image_creation_threshold: Option<usize>,
568 : #[serde(default)]
569 : #[serde(with = "humantime_serde")]
570 : pub pitr_interval: Option<Duration>,
571 : #[serde(default)]
572 : #[serde(with = "humantime_serde")]
573 : pub walreceiver_connect_timeout: Option<Duration>,
574 : #[serde(default)]
575 : #[serde(with = "humantime_serde")]
576 : pub lagging_wal_timeout: Option<Duration>,
577 : pub max_lsn_wal_lag: Option<NonZeroU64>,
578 : pub eviction_policy: Option<EvictionPolicy>,
579 : pub min_resident_size_override: Option<u64>,
580 : #[serde(default)]
581 : #[serde(with = "humantime_serde")]
582 : pub evictions_low_residence_duration_metric_threshold: Option<Duration>,
583 : #[serde(default)]
584 : #[serde(with = "humantime_serde")]
585 : pub heatmap_period: Option<Duration>,
586 : pub lazy_slru_download: Option<bool>,
587 : pub timeline_get_throttle: Option<ThrottleConfig>,
588 : pub image_layer_creation_check_threshold: Option<u8>,
589 : pub image_creation_preempt_threshold: Option<usize>,
590 : #[serde(default)]
591 : #[serde(with = "humantime_serde")]
592 : pub lsn_lease_length: Option<Duration>,
593 : #[serde(default)]
594 : #[serde(with = "humantime_serde")]
595 : pub lsn_lease_length_for_ts: Option<Duration>,
596 : pub timeline_offloading: Option<bool>,
597 : pub wal_receiver_protocol_override: Option<PostgresClientProtocol>,
598 : pub rel_size_v2_enabled: Option<bool>,
599 : pub gc_compaction_enabled: Option<bool>,
600 : pub gc_compaction_initial_threshold_kb: Option<u64>,
601 : pub gc_compaction_ratio_percent: Option<u64>,
602 : }
603 :
604 : impl TenantConfig {
605 1 : pub fn apply_patch(
606 1 : self,
607 1 : patch: TenantConfigPatch,
608 1 : ) -> Result<TenantConfig, humantime::DurationError> {
609 1 : let Self {
610 1 : mut checkpoint_distance,
611 1 : mut checkpoint_timeout,
612 1 : mut compaction_target_size,
613 1 : mut compaction_period,
614 1 : mut compaction_threshold,
615 1 : mut compaction_upper_limit,
616 1 : mut compaction_algorithm,
617 1 : mut compaction_l0_first,
618 1 : mut compaction_l0_semaphore,
619 1 : mut l0_flush_delay_threshold,
620 1 : mut l0_flush_stall_threshold,
621 1 : mut l0_flush_wait_upload,
622 1 : mut gc_horizon,
623 1 : mut gc_period,
624 1 : mut image_creation_threshold,
625 1 : mut pitr_interval,
626 1 : mut walreceiver_connect_timeout,
627 1 : mut lagging_wal_timeout,
628 1 : mut max_lsn_wal_lag,
629 1 : mut eviction_policy,
630 1 : mut min_resident_size_override,
631 1 : mut evictions_low_residence_duration_metric_threshold,
632 1 : mut heatmap_period,
633 1 : mut lazy_slru_download,
634 1 : mut timeline_get_throttle,
635 1 : mut image_layer_creation_check_threshold,
636 1 : mut image_creation_preempt_threshold,
637 1 : mut lsn_lease_length,
638 1 : mut lsn_lease_length_for_ts,
639 1 : mut timeline_offloading,
640 1 : mut wal_receiver_protocol_override,
641 1 : mut rel_size_v2_enabled,
642 1 : mut gc_compaction_enabled,
643 1 : mut gc_compaction_initial_threshold_kb,
644 1 : mut gc_compaction_ratio_percent,
645 1 : } = self;
646 1 :
647 1 : patch.checkpoint_distance.apply(&mut checkpoint_distance);
648 1 : patch
649 1 : .checkpoint_timeout
650 1 : .map(|v| humantime::parse_duration(&v))?
651 1 : .apply(&mut checkpoint_timeout);
652 1 : patch
653 1 : .compaction_target_size
654 1 : .apply(&mut compaction_target_size);
655 1 : patch
656 1 : .compaction_period
657 1 : .map(|v| humantime::parse_duration(&v))?
658 1 : .apply(&mut compaction_period);
659 1 : patch.compaction_threshold.apply(&mut compaction_threshold);
660 1 : patch
661 1 : .compaction_upper_limit
662 1 : .apply(&mut compaction_upper_limit);
663 1 : patch.compaction_algorithm.apply(&mut compaction_algorithm);
664 1 : patch.compaction_l0_first.apply(&mut compaction_l0_first);
665 1 : patch
666 1 : .compaction_l0_semaphore
667 1 : .apply(&mut compaction_l0_semaphore);
668 1 : patch
669 1 : .l0_flush_delay_threshold
670 1 : .apply(&mut l0_flush_delay_threshold);
671 1 : patch
672 1 : .l0_flush_stall_threshold
673 1 : .apply(&mut l0_flush_stall_threshold);
674 1 : patch.l0_flush_wait_upload.apply(&mut l0_flush_wait_upload);
675 1 : patch.gc_horizon.apply(&mut gc_horizon);
676 1 : patch
677 1 : .gc_period
678 1 : .map(|v| humantime::parse_duration(&v))?
679 1 : .apply(&mut gc_period);
680 1 : patch
681 1 : .image_creation_threshold
682 1 : .apply(&mut image_creation_threshold);
683 1 : patch
684 1 : .pitr_interval
685 1 : .map(|v| humantime::parse_duration(&v))?
686 1 : .apply(&mut pitr_interval);
687 1 : patch
688 1 : .walreceiver_connect_timeout
689 1 : .map(|v| humantime::parse_duration(&v))?
690 1 : .apply(&mut walreceiver_connect_timeout);
691 1 : patch
692 1 : .lagging_wal_timeout
693 1 : .map(|v| humantime::parse_duration(&v))?
694 1 : .apply(&mut lagging_wal_timeout);
695 1 : patch.max_lsn_wal_lag.apply(&mut max_lsn_wal_lag);
696 1 : patch.eviction_policy.apply(&mut eviction_policy);
697 1 : patch
698 1 : .min_resident_size_override
699 1 : .apply(&mut min_resident_size_override);
700 1 : patch
701 1 : .evictions_low_residence_duration_metric_threshold
702 1 : .map(|v| humantime::parse_duration(&v))?
703 1 : .apply(&mut evictions_low_residence_duration_metric_threshold);
704 1 : patch
705 1 : .heatmap_period
706 1 : .map(|v| humantime::parse_duration(&v))?
707 1 : .apply(&mut heatmap_period);
708 1 : patch.lazy_slru_download.apply(&mut lazy_slru_download);
709 1 : patch
710 1 : .timeline_get_throttle
711 1 : .apply(&mut timeline_get_throttle);
712 1 : patch
713 1 : .image_layer_creation_check_threshold
714 1 : .apply(&mut image_layer_creation_check_threshold);
715 1 : patch
716 1 : .image_creation_preempt_threshold
717 1 : .apply(&mut image_creation_preempt_threshold);
718 1 : patch
719 1 : .lsn_lease_length
720 1 : .map(|v| humantime::parse_duration(&v))?
721 1 : .apply(&mut lsn_lease_length);
722 1 : patch
723 1 : .lsn_lease_length_for_ts
724 1 : .map(|v| humantime::parse_duration(&v))?
725 1 : .apply(&mut lsn_lease_length_for_ts);
726 1 : patch.timeline_offloading.apply(&mut timeline_offloading);
727 1 : patch
728 1 : .wal_receiver_protocol_override
729 1 : .apply(&mut wal_receiver_protocol_override);
730 1 : patch.rel_size_v2_enabled.apply(&mut rel_size_v2_enabled);
731 1 : patch
732 1 : .gc_compaction_enabled
733 1 : .apply(&mut gc_compaction_enabled);
734 1 : patch
735 1 : .gc_compaction_initial_threshold_kb
736 1 : .apply(&mut gc_compaction_initial_threshold_kb);
737 1 : patch
738 1 : .gc_compaction_ratio_percent
739 1 : .apply(&mut gc_compaction_ratio_percent);
740 1 :
741 1 : Ok(Self {
742 1 : checkpoint_distance,
743 1 : checkpoint_timeout,
744 1 : compaction_target_size,
745 1 : compaction_period,
746 1 : compaction_threshold,
747 1 : compaction_upper_limit,
748 1 : compaction_algorithm,
749 1 : compaction_l0_first,
750 1 : compaction_l0_semaphore,
751 1 : l0_flush_delay_threshold,
752 1 : l0_flush_stall_threshold,
753 1 : l0_flush_wait_upload,
754 1 : gc_horizon,
755 1 : gc_period,
756 1 : image_creation_threshold,
757 1 : pitr_interval,
758 1 : walreceiver_connect_timeout,
759 1 : lagging_wal_timeout,
760 1 : max_lsn_wal_lag,
761 1 : eviction_policy,
762 1 : min_resident_size_override,
763 1 : evictions_low_residence_duration_metric_threshold,
764 1 : heatmap_period,
765 1 : lazy_slru_download,
766 1 : timeline_get_throttle,
767 1 : image_layer_creation_check_threshold,
768 1 : image_creation_preempt_threshold,
769 1 : lsn_lease_length,
770 1 : lsn_lease_length_for_ts,
771 1 : timeline_offloading,
772 1 : wal_receiver_protocol_override,
773 1 : rel_size_v2_enabled,
774 1 : gc_compaction_enabled,
775 1 : gc_compaction_initial_threshold_kb,
776 1 : gc_compaction_ratio_percent,
777 1 : })
778 1 : }
779 : }
780 :
781 : /// The policy for the aux file storage.
782 : ///
783 : /// It can be switched through `switch_aux_file_policy` tenant config.
784 : /// When the first aux file written, the policy will be persisted in the
785 : /// `index_part.json` file and has a limited migration path.
786 : ///
787 : /// Currently, we only allow the following migration path:
788 : ///
789 : /// Unset -> V1
790 : /// -> V2
791 : /// -> CrossValidation -> V2
792 : #[derive(
793 : Eq,
794 : PartialEq,
795 : Debug,
796 : Copy,
797 : Clone,
798 0 : strum_macros::EnumString,
799 : strum_macros::Display,
800 4 : serde_with::DeserializeFromStr,
801 : serde_with::SerializeDisplay,
802 : )]
803 : #[strum(serialize_all = "kebab-case")]
804 : pub enum AuxFilePolicy {
805 : /// V1 aux file policy: store everything in AUX_FILE_KEY
806 : #[strum(ascii_case_insensitive)]
807 : V1,
808 : /// V2 aux file policy: store in the AUX_FILE keyspace
809 : #[strum(ascii_case_insensitive)]
810 : V2,
811 : /// Cross validation runs both formats on the write path and does validation
812 : /// on the read path.
813 : #[strum(ascii_case_insensitive)]
814 : CrossValidation,
815 : }
816 :
817 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
818 : #[serde(tag = "kind")]
819 : pub enum EvictionPolicy {
820 : NoEviction,
821 : LayerAccessThreshold(EvictionPolicyLayerAccessThreshold),
822 : OnlyImitiate(EvictionPolicyLayerAccessThreshold),
823 : }
824 :
825 : impl EvictionPolicy {
826 0 : pub fn discriminant_str(&self) -> &'static str {
827 0 : match self {
828 0 : EvictionPolicy::NoEviction => "NoEviction",
829 0 : EvictionPolicy::LayerAccessThreshold(_) => "LayerAccessThreshold",
830 0 : EvictionPolicy::OnlyImitiate(_) => "OnlyImitiate",
831 : }
832 0 : }
833 : }
834 :
835 : #[derive(
836 : Eq,
837 : PartialEq,
838 : Debug,
839 : Copy,
840 : Clone,
841 0 : strum_macros::EnumString,
842 : strum_macros::Display,
843 0 : serde_with::DeserializeFromStr,
844 : serde_with::SerializeDisplay,
845 : )]
846 : #[strum(serialize_all = "kebab-case")]
847 : pub enum CompactionAlgorithm {
848 : Legacy,
849 : Tiered,
850 : }
851 :
852 : #[derive(
853 4 : Debug, Clone, Copy, PartialEq, Eq, serde_with::DeserializeFromStr, serde_with::SerializeDisplay,
854 : )]
855 : pub enum ImageCompressionAlgorithm {
856 : // Disabled for writes, support decompressing during read path
857 : Disabled,
858 : /// Zstandard compression. Level 0 means and None mean the same (default level). Levels can be negative as well.
859 : /// For details, see the [manual](http://facebook.github.io/zstd/zstd_manual.html).
860 : Zstd {
861 : level: Option<i8>,
862 : },
863 : }
864 :
865 : impl FromStr for ImageCompressionAlgorithm {
866 : type Err = anyhow::Error;
867 8 : fn from_str(s: &str) -> Result<Self, Self::Err> {
868 8 : let mut components = s.split(['(', ')']);
869 8 : let first = components
870 8 : .next()
871 8 : .ok_or_else(|| anyhow::anyhow!("empty string"))?;
872 8 : match first {
873 8 : "disabled" => Ok(ImageCompressionAlgorithm::Disabled),
874 6 : "zstd" => {
875 6 : let level = if let Some(v) = components.next() {
876 4 : let v: i8 = v.parse()?;
877 4 : Some(v)
878 : } else {
879 2 : None
880 : };
881 :
882 6 : Ok(ImageCompressionAlgorithm::Zstd { level })
883 : }
884 0 : _ => anyhow::bail!("invalid specifier '{first}'"),
885 : }
886 8 : }
887 : }
888 :
889 : impl Display for ImageCompressionAlgorithm {
890 12 : fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
891 12 : match self {
892 3 : ImageCompressionAlgorithm::Disabled => write!(f, "disabled"),
893 9 : ImageCompressionAlgorithm::Zstd { level } => {
894 9 : if let Some(level) = level {
895 6 : write!(f, "zstd({})", level)
896 : } else {
897 3 : write!(f, "zstd")
898 : }
899 : }
900 : }
901 12 : }
902 : }
903 :
904 0 : #[derive(Eq, PartialEq, Debug, Clone, Serialize, Deserialize)]
905 : pub struct CompactionAlgorithmSettings {
906 : pub kind: CompactionAlgorithm,
907 : }
908 :
909 8 : #[derive(Debug, PartialEq, Eq, Clone, Deserialize, Serialize)]
910 : #[serde(tag = "mode", rename_all = "kebab-case", deny_unknown_fields)]
911 : pub enum L0FlushConfig {
912 : #[serde(rename_all = "snake_case")]
913 : Direct { max_concurrency: NonZeroUsize },
914 : }
915 :
916 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
917 : pub struct EvictionPolicyLayerAccessThreshold {
918 : #[serde(with = "humantime_serde")]
919 : pub period: Duration,
920 : #[serde(with = "humantime_serde")]
921 : pub threshold: Duration,
922 : }
923 :
924 6 : #[derive(Debug, Serialize, Deserialize, Clone, PartialEq, Eq)]
925 : pub struct ThrottleConfig {
926 : /// See [`ThrottleConfigTaskKinds`] for why we do the serde `rename`.
927 : #[serde(rename = "task_kinds")]
928 : pub enabled: ThrottleConfigTaskKinds,
929 : pub initial: u32,
930 : #[serde(with = "humantime_serde")]
931 : pub refill_interval: Duration,
932 : pub refill_amount: NonZeroU32,
933 : pub max: u32,
934 : }
935 :
936 : /// Before <https://github.com/neondatabase/neon/pull/9962>
937 : /// the throttle was a per `Timeline::get`/`Timeline::get_vectored` call.
938 : /// The `task_kinds` field controlled which Pageserver "Task Kind"s
939 : /// were subject to the throttle.
940 : ///
941 : /// After that PR, the throttle is applied at pagestream request level
942 : /// and the `task_kinds` field does not apply since the only task kind
943 : /// that us subject to the throttle is that of the page service.
944 : ///
945 : /// However, we don't want to make a breaking config change right now
946 : /// because it means we have to migrate all the tenant configs.
947 : /// This will be done in a future PR.
948 : ///
949 : /// In the meantime, we use emptiness / non-emptsiness of the `task_kinds`
950 : /// field to determine if the throttle is enabled or not.
951 1 : #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq)]
952 : #[serde(transparent)]
953 : pub struct ThrottleConfigTaskKinds(Vec<String>);
954 :
955 : impl ThrottleConfigTaskKinds {
956 925 : pub fn disabled() -> Self {
957 925 : Self(vec![])
958 925 : }
959 454 : pub fn is_enabled(&self) -> bool {
960 454 : !self.0.is_empty()
961 454 : }
962 : }
963 :
964 : impl ThrottleConfig {
965 925 : pub fn disabled() -> Self {
966 925 : Self {
967 925 : enabled: ThrottleConfigTaskKinds::disabled(),
968 925 : // other values don't matter with emtpy `task_kinds`.
969 925 : initial: 0,
970 925 : refill_interval: Duration::from_millis(1),
971 925 : refill_amount: NonZeroU32::new(1).unwrap(),
972 925 : max: 1,
973 925 : }
974 925 : }
975 : /// The requests per second allowed by the given config.
976 0 : pub fn steady_rps(&self) -> f64 {
977 0 : (self.refill_amount.get() as f64) / (self.refill_interval.as_secs_f64())
978 0 : }
979 : }
980 :
981 : #[cfg(test)]
982 : mod throttle_config_tests {
983 : use super::*;
984 :
985 : #[test]
986 1 : fn test_disabled_is_disabled() {
987 1 : let config = ThrottleConfig::disabled();
988 1 : assert!(!config.enabled.is_enabled());
989 1 : }
990 : #[test]
991 1 : fn test_enabled_backwards_compat() {
992 1 : let input = serde_json::json!({
993 1 : "task_kinds": ["PageRequestHandler"],
994 1 : "initial": 40000,
995 1 : "refill_interval": "50ms",
996 1 : "refill_amount": 1000,
997 1 : "max": 40000,
998 1 : "fair": true
999 1 : });
1000 1 : let config: ThrottleConfig = serde_json::from_value(input).unwrap();
1001 1 : assert!(config.enabled.is_enabled());
1002 1 : }
1003 : }
1004 :
1005 : /// A flattened analog of a `pagesever::tenant::LocationMode`, which
1006 : /// lists out all possible states (and the virtual "Detached" state)
1007 : /// in a flat form rather than using rust-style enums.
1008 0 : #[derive(Serialize, Deserialize, Debug, Clone, Copy, Eq, PartialEq)]
1009 : pub enum LocationConfigMode {
1010 : AttachedSingle,
1011 : AttachedMulti,
1012 : AttachedStale,
1013 : Secondary,
1014 : Detached,
1015 : }
1016 :
1017 0 : #[derive(Serialize, Deserialize, Debug, Clone, Eq, PartialEq)]
1018 : pub struct LocationConfigSecondary {
1019 : pub warm: bool,
1020 : }
1021 :
1022 : /// An alternative representation of `pageserver::tenant::LocationConf`,
1023 : /// for use in external-facing APIs.
1024 0 : #[derive(Serialize, Deserialize, Debug, Clone, Eq, PartialEq)]
1025 : pub struct LocationConfig {
1026 : pub mode: LocationConfigMode,
1027 : /// If attaching, in what generation?
1028 : #[serde(default)]
1029 : pub generation: Option<u32>,
1030 :
1031 : // If requesting mode `Secondary`, configuration for that.
1032 : #[serde(default)]
1033 : pub secondary_conf: Option<LocationConfigSecondary>,
1034 :
1035 : // Shard parameters: if shard_count is nonzero, then other shard_* fields
1036 : // must be set accurately.
1037 : #[serde(default)]
1038 : pub shard_number: u8,
1039 : #[serde(default)]
1040 : pub shard_count: u8,
1041 : #[serde(default)]
1042 : pub shard_stripe_size: u32,
1043 :
1044 : // This configuration only affects attached mode, but should be provided irrespective
1045 : // of the mode, as a secondary location might transition on startup if the response
1046 : // to the `/re-attach` control plane API requests it.
1047 : pub tenant_conf: TenantConfig,
1048 : }
1049 :
1050 0 : #[derive(Serialize, Deserialize)]
1051 : pub struct LocationConfigListResponse {
1052 : pub tenant_shards: Vec<(TenantShardId, Option<LocationConfig>)>,
1053 : }
1054 :
1055 : #[derive(Serialize)]
1056 : pub struct StatusResponse {
1057 : pub id: NodeId,
1058 : }
1059 :
1060 0 : #[derive(Serialize, Deserialize, Debug)]
1061 : #[serde(deny_unknown_fields)]
1062 : pub struct TenantLocationConfigRequest {
1063 : #[serde(flatten)]
1064 : pub config: LocationConfig, // as we have a flattened field, we should reject all unknown fields in it
1065 : }
1066 :
1067 0 : #[derive(Serialize, Deserialize, Debug)]
1068 : #[serde(deny_unknown_fields)]
1069 : pub struct TenantTimeTravelRequest {
1070 : pub shard_counts: Vec<ShardCount>,
1071 : }
1072 :
1073 0 : #[derive(Serialize, Deserialize, Debug)]
1074 : #[serde(deny_unknown_fields)]
1075 : pub struct TenantShardLocation {
1076 : pub shard_id: TenantShardId,
1077 : pub node_id: NodeId,
1078 : }
1079 :
1080 0 : #[derive(Serialize, Deserialize, Debug)]
1081 : #[serde(deny_unknown_fields)]
1082 : pub struct TenantLocationConfigResponse {
1083 : pub shards: Vec<TenantShardLocation>,
1084 : // If the shards' ShardCount count is >1, stripe_size will be set.
1085 : pub stripe_size: Option<ShardStripeSize>,
1086 : }
1087 :
1088 2 : #[derive(Serialize, Deserialize, Debug)]
1089 : #[serde(deny_unknown_fields)]
1090 : pub struct TenantConfigRequest {
1091 : pub tenant_id: TenantId,
1092 : #[serde(flatten)]
1093 : pub config: TenantConfig, // as we have a flattened field, we should reject all unknown fields in it
1094 : }
1095 :
1096 : impl std::ops::Deref for TenantConfigRequest {
1097 : type Target = TenantConfig;
1098 :
1099 0 : fn deref(&self) -> &Self::Target {
1100 0 : &self.config
1101 0 : }
1102 : }
1103 :
1104 : impl TenantConfigRequest {
1105 0 : pub fn new(tenant_id: TenantId) -> TenantConfigRequest {
1106 0 : let config = TenantConfig::default();
1107 0 : TenantConfigRequest { tenant_id, config }
1108 0 : }
1109 : }
1110 :
1111 3 : #[derive(Serialize, Deserialize, Debug)]
1112 : #[serde(deny_unknown_fields)]
1113 : pub struct TenantConfigPatchRequest {
1114 : pub tenant_id: TenantId,
1115 : #[serde(flatten)]
1116 : pub config: TenantConfigPatch, // as we have a flattened field, we should reject all unknown fields in it
1117 : }
1118 :
1119 0 : #[derive(Serialize, Deserialize, Debug)]
1120 : pub struct TenantWaitLsnRequest {
1121 : #[serde(flatten)]
1122 : pub timelines: HashMap<TimelineId, Lsn>,
1123 : pub timeout: Duration,
1124 : }
1125 :
1126 : /// See [`TenantState::attachment_status`] and the OpenAPI docs for context.
1127 0 : #[derive(Serialize, Deserialize, Clone)]
1128 : #[serde(tag = "slug", content = "data", rename_all = "snake_case")]
1129 : pub enum TenantAttachmentStatus {
1130 : Maybe,
1131 : Attached,
1132 : Failed { reason: String },
1133 : }
1134 :
1135 0 : #[derive(Serialize, Deserialize, Clone)]
1136 : pub struct TenantInfo {
1137 : pub id: TenantShardId,
1138 : // NB: intentionally not part of OpenAPI, we don't want to commit to a specific set of TenantState's
1139 : pub state: TenantState,
1140 : /// Sum of the size of all layer files.
1141 : /// If a layer is present in both local FS and S3, it counts only once.
1142 : pub current_physical_size: Option<u64>, // physical size is only included in `tenant_status` endpoint
1143 : pub attachment_status: TenantAttachmentStatus,
1144 : pub generation: u32,
1145 :
1146 : /// Opaque explanation if gc is being blocked.
1147 : ///
1148 : /// Only looked up for the individual tenant detail, not the listing.
1149 : #[serde(skip_serializing_if = "Option::is_none")]
1150 : pub gc_blocking: Option<String>,
1151 : }
1152 :
1153 0 : #[derive(Serialize, Deserialize, Clone)]
1154 : pub struct TenantDetails {
1155 : #[serde(flatten)]
1156 : pub tenant_info: TenantInfo,
1157 :
1158 : pub walredo: Option<WalRedoManagerStatus>,
1159 :
1160 : pub timelines: Vec<TimelineId>,
1161 : }
1162 :
1163 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone, Copy, Debug)]
1164 : pub enum TimelineArchivalState {
1165 : Archived,
1166 : Unarchived,
1167 : }
1168 :
1169 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone)]
1170 : pub struct TimelineArchivalConfigRequest {
1171 : pub state: TimelineArchivalState,
1172 : }
1173 :
1174 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone)]
1175 : pub struct TimelinePatchIndexPartRequest {
1176 : pub rel_size_migration: Option<RelSizeMigration>,
1177 : pub gc_compaction_last_completed_lsn: Option<Lsn>,
1178 : pub applied_gc_cutoff_lsn: Option<Lsn>,
1179 : #[serde(default)]
1180 : pub force_index_update: bool,
1181 : }
1182 :
1183 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
1184 : pub struct TimelinesInfoAndOffloaded {
1185 : pub timelines: Vec<TimelineInfo>,
1186 : pub offloaded: Vec<OffloadedTimelineInfo>,
1187 : }
1188 :
1189 : /// Analog of [`TimelineInfo`] for offloaded timelines.
1190 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
1191 : pub struct OffloadedTimelineInfo {
1192 : pub tenant_id: TenantShardId,
1193 : pub timeline_id: TimelineId,
1194 : /// Whether the timeline has a parent it has been branched off from or not
1195 : pub ancestor_timeline_id: Option<TimelineId>,
1196 : /// Whether to retain the branch lsn at the ancestor or not
1197 : pub ancestor_retain_lsn: Option<Lsn>,
1198 : /// The time point when the timeline was archived
1199 : pub archived_at: chrono::DateTime<chrono::Utc>,
1200 : }
1201 :
1202 12 : #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
1203 : #[serde(rename_all = "camelCase")]
1204 : pub enum RelSizeMigration {
1205 : /// The tenant is using the old rel_size format.
1206 : /// Note that this enum is persisted as `Option<RelSizeMigration>` in the index part, so
1207 : /// `None` is the same as `Some(RelSizeMigration::Legacy)`.
1208 : Legacy,
1209 : /// The tenant is migrating to the new rel_size format. Both old and new rel_size format are
1210 : /// persisted in the index part. The read path will read both formats and merge them.
1211 : Migrating,
1212 : /// The tenant has migrated to the new rel_size format. Only the new rel_size format is persisted
1213 : /// in the index part, and the read path will not read the old format.
1214 : Migrated,
1215 : }
1216 :
1217 : /// This represents the output of the "timeline_detail" and "timeline_list" API calls.
1218 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
1219 : pub struct TimelineInfo {
1220 : pub tenant_id: TenantShardId,
1221 : pub timeline_id: TimelineId,
1222 :
1223 : pub ancestor_timeline_id: Option<TimelineId>,
1224 : pub ancestor_lsn: Option<Lsn>,
1225 : pub last_record_lsn: Lsn,
1226 : pub prev_record_lsn: Option<Lsn>,
1227 :
1228 : /// Legacy field, retained for one version to enable old storage controller to
1229 : /// decode (it was a mandatory field).
1230 : #[serde(default, rename = "latest_gc_cutoff_lsn")]
1231 : pub _unused: Lsn,
1232 :
1233 : /// The LSN up to which GC has advanced: older data may still exist but it is not available for clients.
1234 : /// This LSN is not suitable for deciding where to create branches etc: use [`TimelineInfo::min_readable_lsn`] instead,
1235 : /// as it is easier to reason about.
1236 : #[serde(default)]
1237 : pub applied_gc_cutoff_lsn: Lsn,
1238 :
1239 : /// The upper bound of data which is either already GC'ed, or elegible to be GC'ed at any time based on PITR interval.
1240 : /// This LSN represents the "end of history" for this timeline, and callers should use it to figure out the oldest
1241 : /// LSN at which it is legal to create a branch or ephemeral endpoint.
1242 : ///
1243 : /// Note that holders of valid LSN leases may be able to create branches and read pages earlier
1244 : /// than this LSN, but new leases may not be taken out earlier than this LSN.
1245 : #[serde(default)]
1246 : pub min_readable_lsn: Lsn,
1247 :
1248 : pub disk_consistent_lsn: Lsn,
1249 :
1250 : /// The LSN that we have succesfully uploaded to remote storage
1251 : pub remote_consistent_lsn: Lsn,
1252 :
1253 : /// The LSN that we are advertizing to safekeepers
1254 : pub remote_consistent_lsn_visible: Lsn,
1255 :
1256 : /// The LSN from the start of the root timeline (never changes)
1257 : pub initdb_lsn: Lsn,
1258 :
1259 : pub current_logical_size: u64,
1260 : pub current_logical_size_is_accurate: bool,
1261 :
1262 : pub directory_entries_counts: Vec<u64>,
1263 :
1264 : /// Sum of the size of all layer files.
1265 : /// If a layer is present in both local FS and S3, it counts only once.
1266 : pub current_physical_size: Option<u64>, // is None when timeline is Unloaded
1267 : pub current_logical_size_non_incremental: Option<u64>,
1268 :
1269 : /// How many bytes of WAL are within this branch's pitr_interval. If the pitr_interval goes
1270 : /// beyond the branch's branch point, we only count up to the branch point.
1271 : pub pitr_history_size: u64,
1272 :
1273 : /// Whether this branch's branch point is within its ancestor's PITR interval (i.e. any
1274 : /// ancestor data used by this branch would have been retained anyway). If this is false, then
1275 : /// this branch may be imposing a cost on the ancestor by causing it to retain layers that it would
1276 : /// otherwise be able to GC.
1277 : pub within_ancestor_pitr: bool,
1278 :
1279 : pub timeline_dir_layer_file_size_sum: Option<u64>,
1280 :
1281 : pub wal_source_connstr: Option<String>,
1282 : pub last_received_msg_lsn: Option<Lsn>,
1283 : /// the timestamp (in microseconds) of the last received message
1284 : pub last_received_msg_ts: Option<u128>,
1285 : pub pg_version: u32,
1286 :
1287 : pub state: TimelineState,
1288 :
1289 : pub walreceiver_status: String,
1290 :
1291 : // ALWAYS add new fields at the end of the struct with `Option` to ensure forward/backward compatibility.
1292 : // Backward compatibility: you will get a JSON not containing the newly-added field.
1293 : // Forward compatibility: a previous version of the pageserver will receive a JSON. serde::Deserialize does
1294 : // not deny unknown fields by default so it's safe to set the field to some value, though it won't be
1295 : // read.
1296 : /// Whether the timeline is archived.
1297 : pub is_archived: Option<bool>,
1298 :
1299 : /// The status of the rel_size migration.
1300 : pub rel_size_migration: Option<RelSizeMigration>,
1301 : }
1302 :
1303 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1304 : pub struct LayerMapInfo {
1305 : pub in_memory_layers: Vec<InMemoryLayerInfo>,
1306 : pub historic_layers: Vec<HistoricLayerInfo>,
1307 : }
1308 :
1309 : /// The residence status of a layer
1310 0 : #[derive(Debug, Clone, Copy, Serialize, Deserialize)]
1311 : pub enum LayerResidenceStatus {
1312 : /// Residence status for a layer file that exists locally.
1313 : /// It may also exist on the remote, we don't care here.
1314 : Resident,
1315 : /// Residence status for a layer file that only exists on the remote.
1316 : Evicted,
1317 : }
1318 :
1319 : #[serde_as]
1320 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1321 : pub struct LayerAccessStats {
1322 : #[serde_as(as = "serde_with::TimestampMilliSeconds")]
1323 : pub access_time: SystemTime,
1324 :
1325 : #[serde_as(as = "serde_with::TimestampMilliSeconds")]
1326 : pub residence_time: SystemTime,
1327 :
1328 : pub visible: bool,
1329 : }
1330 :
1331 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1332 : #[serde(tag = "kind")]
1333 : pub enum InMemoryLayerInfo {
1334 : Open { lsn_start: Lsn },
1335 : Frozen { lsn_start: Lsn, lsn_end: Lsn },
1336 : }
1337 :
1338 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1339 : #[serde(tag = "kind")]
1340 : pub enum HistoricLayerInfo {
1341 : Delta {
1342 : layer_file_name: String,
1343 : layer_file_size: u64,
1344 :
1345 : lsn_start: Lsn,
1346 : lsn_end: Lsn,
1347 : remote: bool,
1348 : access_stats: LayerAccessStats,
1349 :
1350 : l0: bool,
1351 : },
1352 : Image {
1353 : layer_file_name: String,
1354 : layer_file_size: u64,
1355 :
1356 : lsn_start: Lsn,
1357 : remote: bool,
1358 : access_stats: LayerAccessStats,
1359 : },
1360 : }
1361 :
1362 : impl HistoricLayerInfo {
1363 0 : pub fn layer_file_name(&self) -> &str {
1364 0 : match self {
1365 : HistoricLayerInfo::Delta {
1366 0 : layer_file_name, ..
1367 0 : } => layer_file_name,
1368 : HistoricLayerInfo::Image {
1369 0 : layer_file_name, ..
1370 0 : } => layer_file_name,
1371 : }
1372 0 : }
1373 0 : pub fn is_remote(&self) -> bool {
1374 0 : match self {
1375 0 : HistoricLayerInfo::Delta { remote, .. } => *remote,
1376 0 : HistoricLayerInfo::Image { remote, .. } => *remote,
1377 : }
1378 0 : }
1379 0 : pub fn set_remote(&mut self, value: bool) {
1380 0 : let field = match self {
1381 0 : HistoricLayerInfo::Delta { remote, .. } => remote,
1382 0 : HistoricLayerInfo::Image { remote, .. } => remote,
1383 : };
1384 0 : *field = value;
1385 0 : }
1386 0 : pub fn layer_file_size(&self) -> u64 {
1387 0 : match self {
1388 : HistoricLayerInfo::Delta {
1389 0 : layer_file_size, ..
1390 0 : } => *layer_file_size,
1391 : HistoricLayerInfo::Image {
1392 0 : layer_file_size, ..
1393 0 : } => *layer_file_size,
1394 : }
1395 0 : }
1396 : }
1397 :
1398 0 : #[derive(Debug, Serialize, Deserialize)]
1399 : pub struct DownloadRemoteLayersTaskSpawnRequest {
1400 : pub max_concurrent_downloads: NonZeroUsize,
1401 : }
1402 :
1403 0 : #[derive(Debug, Serialize, Deserialize)]
1404 : pub struct IngestAuxFilesRequest {
1405 : pub aux_files: HashMap<String, String>,
1406 : }
1407 :
1408 0 : #[derive(Debug, Serialize, Deserialize)]
1409 : pub struct ListAuxFilesRequest {
1410 : pub lsn: Lsn,
1411 : }
1412 :
1413 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
1414 : pub struct DownloadRemoteLayersTaskInfo {
1415 : pub task_id: String,
1416 : pub state: DownloadRemoteLayersTaskState,
1417 : pub total_layer_count: u64, // stable once `completed`
1418 : pub successful_download_count: u64, // stable once `completed`
1419 : pub failed_download_count: u64, // stable once `completed`
1420 : }
1421 :
1422 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
1423 : pub enum DownloadRemoteLayersTaskState {
1424 : Running,
1425 : Completed,
1426 : ShutDown,
1427 : }
1428 :
1429 0 : #[derive(Debug, Serialize, Deserialize)]
1430 : pub struct TimelineGcRequest {
1431 : pub gc_horizon: Option<u64>,
1432 : }
1433 :
1434 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1435 : pub struct WalRedoManagerProcessStatus {
1436 : pub pid: u32,
1437 : }
1438 :
1439 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1440 : pub struct WalRedoManagerStatus {
1441 : pub last_redo_at: Option<chrono::DateTime<chrono::Utc>>,
1442 : pub process: Option<WalRedoManagerProcessStatus>,
1443 : }
1444 :
1445 : /// The progress of a secondary tenant.
1446 : ///
1447 : /// It is mostly useful when doing a long running download: e.g. initiating
1448 : /// a download job, timing out while waiting for it to run, and then inspecting this status to understand
1449 : /// what's happening.
1450 0 : #[derive(Default, Debug, Serialize, Deserialize, Clone)]
1451 : pub struct SecondaryProgress {
1452 : /// The remote storage LastModified time of the heatmap object we last downloaded.
1453 : pub heatmap_mtime: Option<serde_system_time::SystemTime>,
1454 :
1455 : /// The number of layers currently on-disk
1456 : pub layers_downloaded: usize,
1457 : /// The number of layers in the most recently seen heatmap
1458 : pub layers_total: usize,
1459 :
1460 : /// The number of layer bytes currently on-disk
1461 : pub bytes_downloaded: u64,
1462 : /// The number of layer bytes in the most recently seen heatmap
1463 : pub bytes_total: u64,
1464 : }
1465 :
1466 0 : #[derive(Serialize, Deserialize, Debug)]
1467 : pub struct TenantScanRemoteStorageShard {
1468 : pub tenant_shard_id: TenantShardId,
1469 : pub generation: Option<u32>,
1470 : }
1471 :
1472 0 : #[derive(Serialize, Deserialize, Debug, Default)]
1473 : pub struct TenantScanRemoteStorageResponse {
1474 : pub shards: Vec<TenantScanRemoteStorageShard>,
1475 : }
1476 :
1477 0 : #[derive(Serialize, Deserialize, Debug, Clone)]
1478 : #[serde(rename_all = "snake_case")]
1479 : pub enum TenantSorting {
1480 : /// Total size of layers on local disk for all timelines in a shard.
1481 : ResidentSize,
1482 : /// The logical size of the largest timeline within a _tenant_ (not shard). Only tracked on
1483 : /// shard 0, contains the sum across all shards.
1484 : MaxLogicalSize,
1485 : /// The logical size of the largest timeline within a _tenant_ (not shard), divided by number of
1486 : /// shards. Only tracked on shard 0, and estimates the per-shard logical size.
1487 : MaxLogicalSizePerShard,
1488 : }
1489 :
1490 : impl Default for TenantSorting {
1491 0 : fn default() -> Self {
1492 0 : Self::ResidentSize
1493 0 : }
1494 : }
1495 :
1496 0 : #[derive(Serialize, Deserialize, Debug, Clone)]
1497 : pub struct TopTenantShardsRequest {
1498 : // How would you like to sort the tenants?
1499 : pub order_by: TenantSorting,
1500 :
1501 : // How many results?
1502 : pub limit: usize,
1503 :
1504 : // Omit tenants with more than this many shards (e.g. if this is the max number of shards
1505 : // that the caller would ever split to)
1506 : pub where_shards_lt: Option<ShardCount>,
1507 :
1508 : // Omit tenants where the ordering metric is less than this (this is an optimization to
1509 : // let us quickly exclude numerous tiny shards)
1510 : pub where_gt: Option<u64>,
1511 : }
1512 :
1513 0 : #[derive(Serialize, Deserialize, Debug, PartialEq, Eq)]
1514 : pub struct TopTenantShardItem {
1515 : pub id: TenantShardId,
1516 :
1517 : /// Total size of layers on local disk for all timelines in this shard.
1518 : pub resident_size: u64,
1519 :
1520 : /// Total size of layers in remote storage for all timelines in this shard.
1521 : pub physical_size: u64,
1522 :
1523 : /// The largest logical size of a timeline within this _tenant_ (not shard). This is only
1524 : /// tracked on shard 0, and contains the sum of the logical size across all shards.
1525 : pub max_logical_size: u64,
1526 :
1527 : /// The largest logical size of a timeline within this _tenant_ (not shard) divided by number of
1528 : /// shards. This is only tracked on shard 0, and is only an estimate as we divide it evenly by
1529 : /// shard count, rounded up.
1530 : pub max_logical_size_per_shard: u64,
1531 : }
1532 :
1533 0 : #[derive(Serialize, Deserialize, Debug, Default)]
1534 : pub struct TopTenantShardsResponse {
1535 : pub shards: Vec<TopTenantShardItem>,
1536 : }
1537 :
1538 : pub mod virtual_file {
1539 : #[derive(
1540 : Copy,
1541 : Clone,
1542 : PartialEq,
1543 : Eq,
1544 : Hash,
1545 0 : strum_macros::EnumString,
1546 : strum_macros::Display,
1547 0 : serde_with::DeserializeFromStr,
1548 : serde_with::SerializeDisplay,
1549 : Debug,
1550 : )]
1551 : #[strum(serialize_all = "kebab-case")]
1552 : pub enum IoEngineKind {
1553 : StdFs,
1554 : #[cfg(target_os = "linux")]
1555 : TokioEpollUring,
1556 : }
1557 :
1558 : /// Direct IO modes for a pageserver.
1559 : #[derive(
1560 : Copy,
1561 : Clone,
1562 : PartialEq,
1563 : Eq,
1564 : Hash,
1565 0 : strum_macros::EnumString,
1566 : strum_macros::Display,
1567 0 : serde_with::DeserializeFromStr,
1568 : serde_with::SerializeDisplay,
1569 : Debug,
1570 : )]
1571 : #[strum(serialize_all = "kebab-case")]
1572 : #[repr(u8)]
1573 : pub enum IoMode {
1574 : /// Uses buffered IO.
1575 : Buffered,
1576 : /// Uses direct IO, error out if the operation fails.
1577 : #[cfg(target_os = "linux")]
1578 : Direct,
1579 : }
1580 :
1581 : impl IoMode {
1582 488 : pub const fn preferred() -> Self {
1583 488 : Self::Buffered
1584 488 : }
1585 : }
1586 :
1587 : impl TryFrom<u8> for IoMode {
1588 : type Error = u8;
1589 :
1590 5096 : fn try_from(value: u8) -> Result<Self, Self::Error> {
1591 5096 : Ok(match value {
1592 5096 : v if v == (IoMode::Buffered as u8) => IoMode::Buffered,
1593 : #[cfg(target_os = "linux")]
1594 0 : v if v == (IoMode::Direct as u8) => IoMode::Direct,
1595 0 : x => return Err(x),
1596 : })
1597 5096 : }
1598 : }
1599 : }
1600 :
1601 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
1602 : pub struct ScanDisposableKeysResponse {
1603 : pub disposable_count: usize,
1604 : pub not_disposable_count: usize,
1605 : }
1606 :
1607 : // Wrapped in libpq CopyData
1608 : #[derive(PartialEq, Eq, Debug)]
1609 : pub enum PagestreamFeMessage {
1610 : Exists(PagestreamExistsRequest),
1611 : Nblocks(PagestreamNblocksRequest),
1612 : GetPage(PagestreamGetPageRequest),
1613 : DbSize(PagestreamDbSizeRequest),
1614 : GetSlruSegment(PagestreamGetSlruSegmentRequest),
1615 : #[cfg(feature = "testing")]
1616 : Test(PagestreamTestRequest),
1617 : }
1618 :
1619 : // Wrapped in libpq CopyData
1620 : #[derive(strum_macros::EnumProperty)]
1621 : pub enum PagestreamBeMessage {
1622 : Exists(PagestreamExistsResponse),
1623 : Nblocks(PagestreamNblocksResponse),
1624 : GetPage(PagestreamGetPageResponse),
1625 : Error(PagestreamErrorResponse),
1626 : DbSize(PagestreamDbSizeResponse),
1627 : GetSlruSegment(PagestreamGetSlruSegmentResponse),
1628 : #[cfg(feature = "testing")]
1629 : Test(PagestreamTestResponse),
1630 : }
1631 :
1632 : // Keep in sync with `pagestore_client.h`
1633 : #[repr(u8)]
1634 : enum PagestreamFeMessageTag {
1635 : Exists = 0,
1636 : Nblocks = 1,
1637 : GetPage = 2,
1638 : DbSize = 3,
1639 : GetSlruSegment = 4,
1640 : /* future tags above this line */
1641 : /// For testing purposes, not available in production.
1642 : #[cfg(feature = "testing")]
1643 : Test = 99,
1644 : }
1645 :
1646 : // Keep in sync with `pagestore_client.h`
1647 : #[repr(u8)]
1648 : enum PagestreamBeMessageTag {
1649 : Exists = 100,
1650 : Nblocks = 101,
1651 : GetPage = 102,
1652 : Error = 103,
1653 : DbSize = 104,
1654 : GetSlruSegment = 105,
1655 : /* future tags above this line */
1656 : /// For testing purposes, not available in production.
1657 : #[cfg(feature = "testing")]
1658 : Test = 199,
1659 : }
1660 :
1661 : impl TryFrom<u8> for PagestreamFeMessageTag {
1662 : type Error = u8;
1663 4 : fn try_from(value: u8) -> Result<Self, u8> {
1664 4 : match value {
1665 1 : 0 => Ok(PagestreamFeMessageTag::Exists),
1666 1 : 1 => Ok(PagestreamFeMessageTag::Nblocks),
1667 1 : 2 => Ok(PagestreamFeMessageTag::GetPage),
1668 1 : 3 => Ok(PagestreamFeMessageTag::DbSize),
1669 0 : 4 => Ok(PagestreamFeMessageTag::GetSlruSegment),
1670 : #[cfg(feature = "testing")]
1671 0 : 99 => Ok(PagestreamFeMessageTag::Test),
1672 0 : _ => Err(value),
1673 : }
1674 4 : }
1675 : }
1676 :
1677 : impl TryFrom<u8> for PagestreamBeMessageTag {
1678 : type Error = u8;
1679 0 : fn try_from(value: u8) -> Result<Self, u8> {
1680 0 : match value {
1681 0 : 100 => Ok(PagestreamBeMessageTag::Exists),
1682 0 : 101 => Ok(PagestreamBeMessageTag::Nblocks),
1683 0 : 102 => Ok(PagestreamBeMessageTag::GetPage),
1684 0 : 103 => Ok(PagestreamBeMessageTag::Error),
1685 0 : 104 => Ok(PagestreamBeMessageTag::DbSize),
1686 0 : 105 => Ok(PagestreamBeMessageTag::GetSlruSegment),
1687 : #[cfg(feature = "testing")]
1688 0 : 199 => Ok(PagestreamBeMessageTag::Test),
1689 0 : _ => Err(value),
1690 : }
1691 0 : }
1692 : }
1693 :
1694 : // A GetPage request contains two LSN values:
1695 : //
1696 : // request_lsn: Get the page version at this point in time. Lsn::Max is a special value that means
1697 : // "get the latest version present". It's used by the primary server, which knows that no one else
1698 : // is writing WAL. 'not_modified_since' must be set to a proper value even if request_lsn is
1699 : // Lsn::Max. Standby servers use the current replay LSN as the request LSN.
1700 : //
1701 : // not_modified_since: Hint to the pageserver that the client knows that the page has not been
1702 : // modified between 'not_modified_since' and the request LSN. It's always correct to set
1703 : // 'not_modified_since equal' to 'request_lsn' (unless Lsn::Max is used as the 'request_lsn'), but
1704 : // passing an earlier LSN can speed up the request, by allowing the pageserver to process the
1705 : // request without waiting for 'request_lsn' to arrive.
1706 : //
1707 : // The now-defunct V1 interface contained only one LSN, and a boolean 'latest' flag. The V1 interface was
1708 : // sufficient for the primary; the 'lsn' was equivalent to the 'not_modified_since' value, and
1709 : // 'latest' was set to true. The V2 interface was added because there was no correct way for a
1710 : // standby to request a page at a particular non-latest LSN, and also include the
1711 : // 'not_modified_since' hint. That led to an awkward choice of either using an old LSN in the
1712 : // request, if the standby knows that the page hasn't been modified since, and risk getting an error
1713 : // if that LSN has fallen behind the GC horizon, or requesting the current replay LSN, which could
1714 : // require the pageserver unnecessarily to wait for the WAL to arrive up to that point. The new V2
1715 : // interface allows sending both LSNs, and let the pageserver do the right thing. There was no
1716 : // difference in the responses between V1 and V2.
1717 : //
1718 : // V3 version of protocol adds request ID to all requests. This request ID is also included in response
1719 : // as well as other fields from requests, which allows to verify that we receive response for our request.
1720 : // We copy fields from request to response to make checking more reliable: request ID is formed from process ID
1721 : // and local counter, so in principle there can be duplicated requests IDs if process PID is reused.
1722 : //
1723 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
1724 : pub enum PagestreamProtocolVersion {
1725 : V2,
1726 : V3,
1727 : }
1728 :
1729 : pub type RequestId = u64;
1730 :
1731 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
1732 : pub struct PagestreamRequest {
1733 : pub reqid: RequestId,
1734 : pub request_lsn: Lsn,
1735 : pub not_modified_since: Lsn,
1736 : }
1737 :
1738 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
1739 : pub struct PagestreamExistsRequest {
1740 : pub hdr: PagestreamRequest,
1741 : pub rel: RelTag,
1742 : }
1743 :
1744 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
1745 : pub struct PagestreamNblocksRequest {
1746 : pub hdr: PagestreamRequest,
1747 : pub rel: RelTag,
1748 : }
1749 :
1750 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
1751 : pub struct PagestreamGetPageRequest {
1752 : pub hdr: PagestreamRequest,
1753 : pub rel: RelTag,
1754 : pub blkno: u32,
1755 : }
1756 :
1757 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
1758 : pub struct PagestreamDbSizeRequest {
1759 : pub hdr: PagestreamRequest,
1760 : pub dbnode: u32,
1761 : }
1762 :
1763 : #[derive(Debug, PartialEq, Eq, Clone, Copy)]
1764 : pub struct PagestreamGetSlruSegmentRequest {
1765 : pub hdr: PagestreamRequest,
1766 : pub kind: u8,
1767 : pub segno: u32,
1768 : }
1769 :
1770 : #[derive(Debug)]
1771 : pub struct PagestreamExistsResponse {
1772 : pub req: PagestreamExistsRequest,
1773 : pub exists: bool,
1774 : }
1775 :
1776 : #[derive(Debug)]
1777 : pub struct PagestreamNblocksResponse {
1778 : pub req: PagestreamNblocksRequest,
1779 : pub n_blocks: u32,
1780 : }
1781 :
1782 : #[derive(Debug)]
1783 : pub struct PagestreamGetPageResponse {
1784 : pub req: PagestreamGetPageRequest,
1785 : pub page: Bytes,
1786 : }
1787 :
1788 : #[derive(Debug)]
1789 : pub struct PagestreamGetSlruSegmentResponse {
1790 : pub req: PagestreamGetSlruSegmentRequest,
1791 : pub segment: Bytes,
1792 : }
1793 :
1794 : #[derive(Debug)]
1795 : pub struct PagestreamErrorResponse {
1796 : pub req: PagestreamRequest,
1797 : pub message: String,
1798 : }
1799 :
1800 : #[derive(Debug)]
1801 : pub struct PagestreamDbSizeResponse {
1802 : pub req: PagestreamDbSizeRequest,
1803 : pub db_size: i64,
1804 : }
1805 :
1806 : #[cfg(feature = "testing")]
1807 : #[derive(Debug, PartialEq, Eq, Clone)]
1808 : pub struct PagestreamTestRequest {
1809 : pub hdr: PagestreamRequest,
1810 : pub batch_key: u64,
1811 : pub message: String,
1812 : }
1813 :
1814 : #[cfg(feature = "testing")]
1815 : #[derive(Debug)]
1816 : pub struct PagestreamTestResponse {
1817 : pub req: PagestreamTestRequest,
1818 : }
1819 :
1820 : // This is a cut-down version of TenantHistorySize from the pageserver crate, omitting fields
1821 : // that require pageserver-internal types. It is sufficient to get the total size.
1822 0 : #[derive(Serialize, Deserialize, Debug)]
1823 : pub struct TenantHistorySize {
1824 : pub id: TenantId,
1825 : /// Size is a mixture of WAL and logical size, so the unit is bytes.
1826 : ///
1827 : /// Will be none if `?inputs_only=true` was given.
1828 : pub size: Option<u64>,
1829 : }
1830 :
1831 : impl PagestreamFeMessage {
1832 : /// Serialize a compute -> pageserver message. This is currently only used in testing
1833 : /// tools. Always uses protocol version 3.
1834 4 : pub fn serialize(&self) -> Bytes {
1835 4 : let mut bytes = BytesMut::new();
1836 4 :
1837 4 : match self {
1838 1 : Self::Exists(req) => {
1839 1 : bytes.put_u8(PagestreamFeMessageTag::Exists as u8);
1840 1 : bytes.put_u64(req.hdr.reqid);
1841 1 : bytes.put_u64(req.hdr.request_lsn.0);
1842 1 : bytes.put_u64(req.hdr.not_modified_since.0);
1843 1 : bytes.put_u32(req.rel.spcnode);
1844 1 : bytes.put_u32(req.rel.dbnode);
1845 1 : bytes.put_u32(req.rel.relnode);
1846 1 : bytes.put_u8(req.rel.forknum);
1847 1 : }
1848 :
1849 1 : Self::Nblocks(req) => {
1850 1 : bytes.put_u8(PagestreamFeMessageTag::Nblocks as u8);
1851 1 : bytes.put_u64(req.hdr.reqid);
1852 1 : bytes.put_u64(req.hdr.request_lsn.0);
1853 1 : bytes.put_u64(req.hdr.not_modified_since.0);
1854 1 : bytes.put_u32(req.rel.spcnode);
1855 1 : bytes.put_u32(req.rel.dbnode);
1856 1 : bytes.put_u32(req.rel.relnode);
1857 1 : bytes.put_u8(req.rel.forknum);
1858 1 : }
1859 :
1860 1 : Self::GetPage(req) => {
1861 1 : bytes.put_u8(PagestreamFeMessageTag::GetPage as u8);
1862 1 : bytes.put_u64(req.hdr.reqid);
1863 1 : bytes.put_u64(req.hdr.request_lsn.0);
1864 1 : bytes.put_u64(req.hdr.not_modified_since.0);
1865 1 : bytes.put_u32(req.rel.spcnode);
1866 1 : bytes.put_u32(req.rel.dbnode);
1867 1 : bytes.put_u32(req.rel.relnode);
1868 1 : bytes.put_u8(req.rel.forknum);
1869 1 : bytes.put_u32(req.blkno);
1870 1 : }
1871 :
1872 1 : Self::DbSize(req) => {
1873 1 : bytes.put_u8(PagestreamFeMessageTag::DbSize as u8);
1874 1 : bytes.put_u64(req.hdr.reqid);
1875 1 : bytes.put_u64(req.hdr.request_lsn.0);
1876 1 : bytes.put_u64(req.hdr.not_modified_since.0);
1877 1 : bytes.put_u32(req.dbnode);
1878 1 : }
1879 :
1880 0 : Self::GetSlruSegment(req) => {
1881 0 : bytes.put_u8(PagestreamFeMessageTag::GetSlruSegment as u8);
1882 0 : bytes.put_u64(req.hdr.reqid);
1883 0 : bytes.put_u64(req.hdr.request_lsn.0);
1884 0 : bytes.put_u64(req.hdr.not_modified_since.0);
1885 0 : bytes.put_u8(req.kind);
1886 0 : bytes.put_u32(req.segno);
1887 0 : }
1888 : #[cfg(feature = "testing")]
1889 0 : Self::Test(req) => {
1890 0 : bytes.put_u8(PagestreamFeMessageTag::Test as u8);
1891 0 : bytes.put_u64(req.hdr.reqid);
1892 0 : bytes.put_u64(req.hdr.request_lsn.0);
1893 0 : bytes.put_u64(req.hdr.not_modified_since.0);
1894 0 : bytes.put_u64(req.batch_key);
1895 0 : let message = req.message.as_bytes();
1896 0 : bytes.put_u64(message.len() as u64);
1897 0 : bytes.put_slice(message);
1898 0 : }
1899 : }
1900 :
1901 4 : bytes.into()
1902 4 : }
1903 :
1904 4 : pub fn parse<R: std::io::Read>(
1905 4 : body: &mut R,
1906 4 : protocol_version: PagestreamProtocolVersion,
1907 4 : ) -> anyhow::Result<PagestreamFeMessage> {
1908 : // these correspond to the NeonMessageTag enum in pagestore_client.h
1909 : //
1910 : // TODO: consider using protobuf or serde bincode for less error prone
1911 : // serialization.
1912 4 : let msg_tag = body.read_u8()?;
1913 4 : let (reqid, request_lsn, not_modified_since) = match protocol_version {
1914 : PagestreamProtocolVersion::V2 => (
1915 : 0,
1916 0 : Lsn::from(body.read_u64::<BigEndian>()?),
1917 0 : Lsn::from(body.read_u64::<BigEndian>()?),
1918 : ),
1919 : PagestreamProtocolVersion::V3 => (
1920 4 : body.read_u64::<BigEndian>()?,
1921 4 : Lsn::from(body.read_u64::<BigEndian>()?),
1922 4 : Lsn::from(body.read_u64::<BigEndian>()?),
1923 : ),
1924 : };
1925 :
1926 4 : match PagestreamFeMessageTag::try_from(msg_tag)
1927 4 : .map_err(|tag: u8| anyhow::anyhow!("invalid tag {tag}"))?
1928 : {
1929 : PagestreamFeMessageTag::Exists => {
1930 : Ok(PagestreamFeMessage::Exists(PagestreamExistsRequest {
1931 1 : hdr: PagestreamRequest {
1932 1 : reqid,
1933 1 : request_lsn,
1934 1 : not_modified_since,
1935 1 : },
1936 1 : rel: RelTag {
1937 1 : spcnode: body.read_u32::<BigEndian>()?,
1938 1 : dbnode: body.read_u32::<BigEndian>()?,
1939 1 : relnode: body.read_u32::<BigEndian>()?,
1940 1 : forknum: body.read_u8()?,
1941 : },
1942 : }))
1943 : }
1944 : PagestreamFeMessageTag::Nblocks => {
1945 : Ok(PagestreamFeMessage::Nblocks(PagestreamNblocksRequest {
1946 1 : hdr: PagestreamRequest {
1947 1 : reqid,
1948 1 : request_lsn,
1949 1 : not_modified_since,
1950 1 : },
1951 1 : rel: RelTag {
1952 1 : spcnode: body.read_u32::<BigEndian>()?,
1953 1 : dbnode: body.read_u32::<BigEndian>()?,
1954 1 : relnode: body.read_u32::<BigEndian>()?,
1955 1 : forknum: body.read_u8()?,
1956 : },
1957 : }))
1958 : }
1959 : PagestreamFeMessageTag::GetPage => {
1960 : Ok(PagestreamFeMessage::GetPage(PagestreamGetPageRequest {
1961 1 : hdr: PagestreamRequest {
1962 1 : reqid,
1963 1 : request_lsn,
1964 1 : not_modified_since,
1965 1 : },
1966 1 : rel: RelTag {
1967 1 : spcnode: body.read_u32::<BigEndian>()?,
1968 1 : dbnode: body.read_u32::<BigEndian>()?,
1969 1 : relnode: body.read_u32::<BigEndian>()?,
1970 1 : forknum: body.read_u8()?,
1971 : },
1972 1 : blkno: body.read_u32::<BigEndian>()?,
1973 : }))
1974 : }
1975 : PagestreamFeMessageTag::DbSize => {
1976 : Ok(PagestreamFeMessage::DbSize(PagestreamDbSizeRequest {
1977 1 : hdr: PagestreamRequest {
1978 1 : reqid,
1979 1 : request_lsn,
1980 1 : not_modified_since,
1981 1 : },
1982 1 : dbnode: body.read_u32::<BigEndian>()?,
1983 : }))
1984 : }
1985 : PagestreamFeMessageTag::GetSlruSegment => Ok(PagestreamFeMessage::GetSlruSegment(
1986 : PagestreamGetSlruSegmentRequest {
1987 0 : hdr: PagestreamRequest {
1988 0 : reqid,
1989 0 : request_lsn,
1990 0 : not_modified_since,
1991 0 : },
1992 0 : kind: body.read_u8()?,
1993 0 : segno: body.read_u32::<BigEndian>()?,
1994 : },
1995 : )),
1996 : #[cfg(feature = "testing")]
1997 : PagestreamFeMessageTag::Test => Ok(PagestreamFeMessage::Test(PagestreamTestRequest {
1998 0 : hdr: PagestreamRequest {
1999 0 : reqid,
2000 0 : request_lsn,
2001 0 : not_modified_since,
2002 0 : },
2003 0 : batch_key: body.read_u64::<BigEndian>()?,
2004 : message: {
2005 0 : let len = body.read_u64::<BigEndian>()?;
2006 0 : let mut buf = vec![0; len as usize];
2007 0 : body.read_exact(&mut buf)?;
2008 0 : String::from_utf8(buf)?
2009 : },
2010 : })),
2011 : }
2012 4 : }
2013 : }
2014 :
2015 : impl PagestreamBeMessage {
2016 0 : pub fn serialize(&self, protocol_version: PagestreamProtocolVersion) -> Bytes {
2017 0 : let mut bytes = BytesMut::new();
2018 :
2019 : use PagestreamBeMessageTag as Tag;
2020 0 : match protocol_version {
2021 : PagestreamProtocolVersion::V2 => {
2022 0 : match self {
2023 0 : Self::Exists(resp) => {
2024 0 : bytes.put_u8(Tag::Exists as u8);
2025 0 : bytes.put_u8(resp.exists as u8);
2026 0 : }
2027 :
2028 0 : Self::Nblocks(resp) => {
2029 0 : bytes.put_u8(Tag::Nblocks as u8);
2030 0 : bytes.put_u32(resp.n_blocks);
2031 0 : }
2032 :
2033 0 : Self::GetPage(resp) => {
2034 0 : bytes.put_u8(Tag::GetPage as u8);
2035 0 : bytes.put(&resp.page[..])
2036 : }
2037 :
2038 0 : Self::Error(resp) => {
2039 0 : bytes.put_u8(Tag::Error as u8);
2040 0 : bytes.put(resp.message.as_bytes());
2041 0 : bytes.put_u8(0); // null terminator
2042 0 : }
2043 0 : Self::DbSize(resp) => {
2044 0 : bytes.put_u8(Tag::DbSize as u8);
2045 0 : bytes.put_i64(resp.db_size);
2046 0 : }
2047 :
2048 0 : Self::GetSlruSegment(resp) => {
2049 0 : bytes.put_u8(Tag::GetSlruSegment as u8);
2050 0 : bytes.put_u32((resp.segment.len() / BLCKSZ as usize) as u32);
2051 0 : bytes.put(&resp.segment[..]);
2052 0 : }
2053 :
2054 : #[cfg(feature = "testing")]
2055 0 : Self::Test(resp) => {
2056 0 : bytes.put_u8(Tag::Test as u8);
2057 0 : bytes.put_u64(resp.req.batch_key);
2058 0 : let message = resp.req.message.as_bytes();
2059 0 : bytes.put_u64(message.len() as u64);
2060 0 : bytes.put_slice(message);
2061 0 : }
2062 : }
2063 : }
2064 : PagestreamProtocolVersion::V3 => {
2065 0 : match self {
2066 0 : Self::Exists(resp) => {
2067 0 : bytes.put_u8(Tag::Exists as u8);
2068 0 : bytes.put_u64(resp.req.hdr.reqid);
2069 0 : bytes.put_u64(resp.req.hdr.request_lsn.0);
2070 0 : bytes.put_u64(resp.req.hdr.not_modified_since.0);
2071 0 : bytes.put_u32(resp.req.rel.spcnode);
2072 0 : bytes.put_u32(resp.req.rel.dbnode);
2073 0 : bytes.put_u32(resp.req.rel.relnode);
2074 0 : bytes.put_u8(resp.req.rel.forknum);
2075 0 : bytes.put_u8(resp.exists as u8);
2076 0 : }
2077 :
2078 0 : Self::Nblocks(resp) => {
2079 0 : bytes.put_u8(Tag::Nblocks as u8);
2080 0 : bytes.put_u64(resp.req.hdr.reqid);
2081 0 : bytes.put_u64(resp.req.hdr.request_lsn.0);
2082 0 : bytes.put_u64(resp.req.hdr.not_modified_since.0);
2083 0 : bytes.put_u32(resp.req.rel.spcnode);
2084 0 : bytes.put_u32(resp.req.rel.dbnode);
2085 0 : bytes.put_u32(resp.req.rel.relnode);
2086 0 : bytes.put_u8(resp.req.rel.forknum);
2087 0 : bytes.put_u32(resp.n_blocks);
2088 0 : }
2089 :
2090 0 : Self::GetPage(resp) => {
2091 0 : bytes.put_u8(Tag::GetPage as u8);
2092 0 : bytes.put_u64(resp.req.hdr.reqid);
2093 0 : bytes.put_u64(resp.req.hdr.request_lsn.0);
2094 0 : bytes.put_u64(resp.req.hdr.not_modified_since.0);
2095 0 : bytes.put_u32(resp.req.rel.spcnode);
2096 0 : bytes.put_u32(resp.req.rel.dbnode);
2097 0 : bytes.put_u32(resp.req.rel.relnode);
2098 0 : bytes.put_u8(resp.req.rel.forknum);
2099 0 : bytes.put_u32(resp.req.blkno);
2100 0 : bytes.put(&resp.page[..])
2101 : }
2102 :
2103 0 : Self::Error(resp) => {
2104 0 : bytes.put_u8(Tag::Error as u8);
2105 0 : bytes.put_u64(resp.req.reqid);
2106 0 : bytes.put_u64(resp.req.request_lsn.0);
2107 0 : bytes.put_u64(resp.req.not_modified_since.0);
2108 0 : bytes.put(resp.message.as_bytes());
2109 0 : bytes.put_u8(0); // null terminator
2110 0 : }
2111 0 : Self::DbSize(resp) => {
2112 0 : bytes.put_u8(Tag::DbSize as u8);
2113 0 : bytes.put_u64(resp.req.hdr.reqid);
2114 0 : bytes.put_u64(resp.req.hdr.request_lsn.0);
2115 0 : bytes.put_u64(resp.req.hdr.not_modified_since.0);
2116 0 : bytes.put_u32(resp.req.dbnode);
2117 0 : bytes.put_i64(resp.db_size);
2118 0 : }
2119 :
2120 0 : Self::GetSlruSegment(resp) => {
2121 0 : bytes.put_u8(Tag::GetSlruSegment as u8);
2122 0 : bytes.put_u64(resp.req.hdr.reqid);
2123 0 : bytes.put_u64(resp.req.hdr.request_lsn.0);
2124 0 : bytes.put_u64(resp.req.hdr.not_modified_since.0);
2125 0 : bytes.put_u8(resp.req.kind);
2126 0 : bytes.put_u32(resp.req.segno);
2127 0 : bytes.put_u32((resp.segment.len() / BLCKSZ as usize) as u32);
2128 0 : bytes.put(&resp.segment[..]);
2129 0 : }
2130 :
2131 : #[cfg(feature = "testing")]
2132 0 : Self::Test(resp) => {
2133 0 : bytes.put_u8(Tag::Test as u8);
2134 0 : bytes.put_u64(resp.req.hdr.reqid);
2135 0 : bytes.put_u64(resp.req.hdr.request_lsn.0);
2136 0 : bytes.put_u64(resp.req.hdr.not_modified_since.0);
2137 0 : bytes.put_u64(resp.req.batch_key);
2138 0 : let message = resp.req.message.as_bytes();
2139 0 : bytes.put_u64(message.len() as u64);
2140 0 : bytes.put_slice(message);
2141 0 : }
2142 : }
2143 : }
2144 : }
2145 0 : bytes.into()
2146 0 : }
2147 :
2148 0 : pub fn deserialize(buf: Bytes) -> anyhow::Result<Self> {
2149 0 : let mut buf = buf.reader();
2150 0 : let msg_tag = buf.read_u8()?;
2151 :
2152 : use PagestreamBeMessageTag as Tag;
2153 0 : let ok =
2154 0 : match Tag::try_from(msg_tag).map_err(|tag: u8| anyhow::anyhow!("invalid tag {tag}"))? {
2155 : Tag::Exists => {
2156 0 : let reqid = buf.read_u64::<BigEndian>()?;
2157 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2158 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2159 0 : let rel = RelTag {
2160 0 : spcnode: buf.read_u32::<BigEndian>()?,
2161 0 : dbnode: buf.read_u32::<BigEndian>()?,
2162 0 : relnode: buf.read_u32::<BigEndian>()?,
2163 0 : forknum: buf.read_u8()?,
2164 : };
2165 0 : let exists = buf.read_u8()? != 0;
2166 0 : Self::Exists(PagestreamExistsResponse {
2167 0 : req: PagestreamExistsRequest {
2168 0 : hdr: PagestreamRequest {
2169 0 : reqid,
2170 0 : request_lsn,
2171 0 : not_modified_since,
2172 0 : },
2173 0 : rel,
2174 0 : },
2175 0 : exists,
2176 0 : })
2177 : }
2178 : Tag::Nblocks => {
2179 0 : let reqid = buf.read_u64::<BigEndian>()?;
2180 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2181 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2182 0 : let rel = RelTag {
2183 0 : spcnode: buf.read_u32::<BigEndian>()?,
2184 0 : dbnode: buf.read_u32::<BigEndian>()?,
2185 0 : relnode: buf.read_u32::<BigEndian>()?,
2186 0 : forknum: buf.read_u8()?,
2187 : };
2188 0 : let n_blocks = buf.read_u32::<BigEndian>()?;
2189 0 : Self::Nblocks(PagestreamNblocksResponse {
2190 0 : req: PagestreamNblocksRequest {
2191 0 : hdr: PagestreamRequest {
2192 0 : reqid,
2193 0 : request_lsn,
2194 0 : not_modified_since,
2195 0 : },
2196 0 : rel,
2197 0 : },
2198 0 : n_blocks,
2199 0 : })
2200 : }
2201 : Tag::GetPage => {
2202 0 : let reqid = buf.read_u64::<BigEndian>()?;
2203 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2204 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2205 0 : let rel = RelTag {
2206 0 : spcnode: buf.read_u32::<BigEndian>()?,
2207 0 : dbnode: buf.read_u32::<BigEndian>()?,
2208 0 : relnode: buf.read_u32::<BigEndian>()?,
2209 0 : forknum: buf.read_u8()?,
2210 : };
2211 0 : let blkno = buf.read_u32::<BigEndian>()?;
2212 0 : let mut page = vec![0; 8192]; // TODO: use MaybeUninit
2213 0 : buf.read_exact(&mut page)?;
2214 0 : Self::GetPage(PagestreamGetPageResponse {
2215 0 : req: PagestreamGetPageRequest {
2216 0 : hdr: PagestreamRequest {
2217 0 : reqid,
2218 0 : request_lsn,
2219 0 : not_modified_since,
2220 0 : },
2221 0 : rel,
2222 0 : blkno,
2223 0 : },
2224 0 : page: page.into(),
2225 0 : })
2226 : }
2227 : Tag::Error => {
2228 0 : let reqid = buf.read_u64::<BigEndian>()?;
2229 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2230 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2231 0 : let mut msg = Vec::new();
2232 0 : buf.read_until(0, &mut msg)?;
2233 0 : let cstring = std::ffi::CString::from_vec_with_nul(msg)?;
2234 0 : let rust_str = cstring.to_str()?;
2235 0 : Self::Error(PagestreamErrorResponse {
2236 0 : req: PagestreamRequest {
2237 0 : reqid,
2238 0 : request_lsn,
2239 0 : not_modified_since,
2240 0 : },
2241 0 : message: rust_str.to_owned(),
2242 0 : })
2243 : }
2244 : Tag::DbSize => {
2245 0 : let reqid = buf.read_u64::<BigEndian>()?;
2246 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2247 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2248 0 : let dbnode = buf.read_u32::<BigEndian>()?;
2249 0 : let db_size = buf.read_i64::<BigEndian>()?;
2250 0 : Self::DbSize(PagestreamDbSizeResponse {
2251 0 : req: PagestreamDbSizeRequest {
2252 0 : hdr: PagestreamRequest {
2253 0 : reqid,
2254 0 : request_lsn,
2255 0 : not_modified_since,
2256 0 : },
2257 0 : dbnode,
2258 0 : },
2259 0 : db_size,
2260 0 : })
2261 : }
2262 : Tag::GetSlruSegment => {
2263 0 : let reqid = buf.read_u64::<BigEndian>()?;
2264 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2265 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2266 0 : let kind = buf.read_u8()?;
2267 0 : let segno = buf.read_u32::<BigEndian>()?;
2268 0 : let n_blocks = buf.read_u32::<BigEndian>()?;
2269 0 : let mut segment = vec![0; n_blocks as usize * BLCKSZ as usize];
2270 0 : buf.read_exact(&mut segment)?;
2271 0 : Self::GetSlruSegment(PagestreamGetSlruSegmentResponse {
2272 0 : req: PagestreamGetSlruSegmentRequest {
2273 0 : hdr: PagestreamRequest {
2274 0 : reqid,
2275 0 : request_lsn,
2276 0 : not_modified_since,
2277 0 : },
2278 0 : kind,
2279 0 : segno,
2280 0 : },
2281 0 : segment: segment.into(),
2282 0 : })
2283 : }
2284 : #[cfg(feature = "testing")]
2285 : Tag::Test => {
2286 0 : let reqid = buf.read_u64::<BigEndian>()?;
2287 0 : let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
2288 0 : let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
2289 0 : let batch_key = buf.read_u64::<BigEndian>()?;
2290 0 : let len = buf.read_u64::<BigEndian>()?;
2291 0 : let mut msg = vec![0; len as usize];
2292 0 : buf.read_exact(&mut msg)?;
2293 0 : let message = String::from_utf8(msg)?;
2294 0 : Self::Test(PagestreamTestResponse {
2295 0 : req: PagestreamTestRequest {
2296 0 : hdr: PagestreamRequest {
2297 0 : reqid,
2298 0 : request_lsn,
2299 0 : not_modified_since,
2300 0 : },
2301 0 : batch_key,
2302 0 : message,
2303 0 : },
2304 0 : })
2305 : }
2306 : };
2307 0 : let remaining = buf.into_inner();
2308 0 : if !remaining.is_empty() {
2309 0 : anyhow::bail!(
2310 0 : "remaining bytes in msg with tag={msg_tag}: {}",
2311 0 : remaining.len()
2312 0 : );
2313 0 : }
2314 0 : Ok(ok)
2315 0 : }
2316 :
2317 0 : pub fn kind(&self) -> &'static str {
2318 0 : match self {
2319 0 : Self::Exists(_) => "Exists",
2320 0 : Self::Nblocks(_) => "Nblocks",
2321 0 : Self::GetPage(_) => "GetPage",
2322 0 : Self::Error(_) => "Error",
2323 0 : Self::DbSize(_) => "DbSize",
2324 0 : Self::GetSlruSegment(_) => "GetSlruSegment",
2325 : #[cfg(feature = "testing")]
2326 0 : Self::Test(_) => "Test",
2327 : }
2328 0 : }
2329 : }
2330 :
2331 0 : #[derive(Debug, Serialize, Deserialize)]
2332 : pub struct PageTraceEvent {
2333 : pub key: CompactKey,
2334 : pub effective_lsn: Lsn,
2335 : pub time: SystemTime,
2336 : }
2337 :
2338 : impl Default for PageTraceEvent {
2339 0 : fn default() -> Self {
2340 0 : Self {
2341 0 : key: Default::default(),
2342 0 : effective_lsn: Default::default(),
2343 0 : time: std::time::UNIX_EPOCH,
2344 0 : }
2345 0 : }
2346 : }
2347 :
2348 : #[cfg(test)]
2349 : mod tests {
2350 : use std::str::FromStr;
2351 :
2352 : use serde_json::json;
2353 :
2354 : use super::*;
2355 :
2356 : #[test]
2357 1 : fn test_pagestream() {
2358 1 : // Test serialization/deserialization of PagestreamFeMessage
2359 1 : let messages = vec![
2360 1 : PagestreamFeMessage::Exists(PagestreamExistsRequest {
2361 1 : hdr: PagestreamRequest {
2362 1 : reqid: 0,
2363 1 : request_lsn: Lsn(4),
2364 1 : not_modified_since: Lsn(3),
2365 1 : },
2366 1 : rel: RelTag {
2367 1 : forknum: 1,
2368 1 : spcnode: 2,
2369 1 : dbnode: 3,
2370 1 : relnode: 4,
2371 1 : },
2372 1 : }),
2373 1 : PagestreamFeMessage::Nblocks(PagestreamNblocksRequest {
2374 1 : hdr: PagestreamRequest {
2375 1 : reqid: 0,
2376 1 : request_lsn: Lsn(4),
2377 1 : not_modified_since: Lsn(4),
2378 1 : },
2379 1 : rel: RelTag {
2380 1 : forknum: 1,
2381 1 : spcnode: 2,
2382 1 : dbnode: 3,
2383 1 : relnode: 4,
2384 1 : },
2385 1 : }),
2386 1 : PagestreamFeMessage::GetPage(PagestreamGetPageRequest {
2387 1 : hdr: PagestreamRequest {
2388 1 : reqid: 0,
2389 1 : request_lsn: Lsn(4),
2390 1 : not_modified_since: Lsn(3),
2391 1 : },
2392 1 : rel: RelTag {
2393 1 : forknum: 1,
2394 1 : spcnode: 2,
2395 1 : dbnode: 3,
2396 1 : relnode: 4,
2397 1 : },
2398 1 : blkno: 7,
2399 1 : }),
2400 1 : PagestreamFeMessage::DbSize(PagestreamDbSizeRequest {
2401 1 : hdr: PagestreamRequest {
2402 1 : reqid: 0,
2403 1 : request_lsn: Lsn(4),
2404 1 : not_modified_since: Lsn(3),
2405 1 : },
2406 1 : dbnode: 7,
2407 1 : }),
2408 1 : ];
2409 5 : for msg in messages {
2410 4 : let bytes = msg.serialize();
2411 4 : let reconstructed =
2412 4 : PagestreamFeMessage::parse(&mut bytes.reader(), PagestreamProtocolVersion::V3)
2413 4 : .unwrap();
2414 4 : assert!(msg == reconstructed);
2415 : }
2416 1 : }
2417 :
2418 : #[test]
2419 1 : fn test_tenantinfo_serde() {
2420 1 : // Test serialization/deserialization of TenantInfo
2421 1 : let original_active = TenantInfo {
2422 1 : id: TenantShardId::unsharded(TenantId::generate()),
2423 1 : state: TenantState::Active,
2424 1 : current_physical_size: Some(42),
2425 1 : attachment_status: TenantAttachmentStatus::Attached,
2426 1 : generation: 1,
2427 1 : gc_blocking: None,
2428 1 : };
2429 1 : let expected_active = json!({
2430 1 : "id": original_active.id.to_string(),
2431 1 : "state": {
2432 1 : "slug": "Active",
2433 1 : },
2434 1 : "current_physical_size": 42,
2435 1 : "attachment_status": {
2436 1 : "slug":"attached",
2437 1 : },
2438 1 : "generation" : 1
2439 1 : });
2440 1 :
2441 1 : let original_broken = TenantInfo {
2442 1 : id: TenantShardId::unsharded(TenantId::generate()),
2443 1 : state: TenantState::Broken {
2444 1 : reason: "reason".into(),
2445 1 : backtrace: "backtrace info".into(),
2446 1 : },
2447 1 : current_physical_size: Some(42),
2448 1 : attachment_status: TenantAttachmentStatus::Attached,
2449 1 : generation: 1,
2450 1 : gc_blocking: None,
2451 1 : };
2452 1 : let expected_broken = json!({
2453 1 : "id": original_broken.id.to_string(),
2454 1 : "state": {
2455 1 : "slug": "Broken",
2456 1 : "data": {
2457 1 : "backtrace": "backtrace info",
2458 1 : "reason": "reason",
2459 1 : }
2460 1 : },
2461 1 : "current_physical_size": 42,
2462 1 : "attachment_status": {
2463 1 : "slug":"attached",
2464 1 : },
2465 1 : "generation" : 1
2466 1 : });
2467 1 :
2468 1 : assert_eq!(
2469 1 : serde_json::to_value(&original_active).unwrap(),
2470 1 : expected_active
2471 1 : );
2472 :
2473 1 : assert_eq!(
2474 1 : serde_json::to_value(&original_broken).unwrap(),
2475 1 : expected_broken
2476 1 : );
2477 1 : assert!(format!("{:?}", &original_broken.state).contains("reason"));
2478 1 : assert!(format!("{:?}", &original_broken.state).contains("backtrace info"));
2479 1 : }
2480 :
2481 : #[test]
2482 1 : fn test_reject_unknown_field() {
2483 1 : let id = TenantId::generate();
2484 1 : let config_request = json!({
2485 1 : "tenant_id": id.to_string(),
2486 1 : "unknown_field": "unknown_value".to_string(),
2487 1 : });
2488 1 : let err = serde_json::from_value::<TenantConfigRequest>(config_request).unwrap_err();
2489 1 : assert!(
2490 1 : err.to_string().contains("unknown field `unknown_field`"),
2491 0 : "expect unknown field `unknown_field` error, got: {}",
2492 : err
2493 : );
2494 1 : }
2495 :
2496 : #[test]
2497 1 : fn tenantstatus_activating_serde() {
2498 1 : let states = [TenantState::Activating(ActivatingFrom::Attaching)];
2499 1 : let expected = "[{\"slug\":\"Activating\",\"data\":\"Attaching\"}]";
2500 1 :
2501 1 : let actual = serde_json::to_string(&states).unwrap();
2502 1 :
2503 1 : assert_eq!(actual, expected);
2504 :
2505 1 : let parsed = serde_json::from_str::<Vec<TenantState>>(&actual).unwrap();
2506 1 :
2507 1 : assert_eq!(states.as_slice(), &parsed);
2508 1 : }
2509 :
2510 : #[test]
2511 1 : fn tenantstatus_activating_strum() {
2512 1 : // tests added, because we use these for metrics
2513 1 : let examples = [
2514 1 : (line!(), TenantState::Attaching, "Attaching"),
2515 1 : (
2516 1 : line!(),
2517 1 : TenantState::Activating(ActivatingFrom::Attaching),
2518 1 : "Activating",
2519 1 : ),
2520 1 : (line!(), TenantState::Active, "Active"),
2521 1 : (
2522 1 : line!(),
2523 1 : TenantState::Stopping {
2524 1 : progress: utils::completion::Barrier::default(),
2525 1 : },
2526 1 : "Stopping",
2527 1 : ),
2528 1 : (
2529 1 : line!(),
2530 1 : TenantState::Broken {
2531 1 : reason: "Example".into(),
2532 1 : backtrace: "Looooong backtrace".into(),
2533 1 : },
2534 1 : "Broken",
2535 1 : ),
2536 1 : ];
2537 :
2538 6 : for (line, rendered, expected) in examples {
2539 5 : let actual: &'static str = rendered.into();
2540 5 : assert_eq!(actual, expected, "example on {line}");
2541 : }
2542 1 : }
2543 :
2544 : #[test]
2545 1 : fn test_image_compression_algorithm_parsing() {
2546 : use ImageCompressionAlgorithm::*;
2547 1 : let cases = [
2548 1 : ("disabled", Disabled),
2549 1 : ("zstd", Zstd { level: None }),
2550 1 : ("zstd(18)", Zstd { level: Some(18) }),
2551 1 : ("zstd(-3)", Zstd { level: Some(-3) }),
2552 1 : ];
2553 :
2554 5 : for (display, expected) in cases {
2555 4 : assert_eq!(
2556 4 : ImageCompressionAlgorithm::from_str(display).unwrap(),
2557 : expected,
2558 0 : "parsing works"
2559 : );
2560 4 : assert_eq!(format!("{expected}"), display, "Display FromStr roundtrip");
2561 :
2562 4 : let ser = serde_json::to_string(&expected).expect("serialization");
2563 4 : assert_eq!(
2564 4 : serde_json::from_str::<ImageCompressionAlgorithm>(&ser).unwrap(),
2565 : expected,
2566 0 : "serde roundtrip"
2567 : );
2568 :
2569 4 : assert_eq!(
2570 4 : serde_json::Value::String(display.to_string()),
2571 4 : serde_json::to_value(expected).unwrap(),
2572 0 : "Display is the serde serialization"
2573 : );
2574 : }
2575 1 : }
2576 :
2577 : #[test]
2578 1 : fn test_tenant_config_patch_request_serde() {
2579 1 : let patch_request = TenantConfigPatchRequest {
2580 1 : tenant_id: TenantId::from_str("17c6d121946a61e5ab0fe5a2fd4d8215").unwrap(),
2581 1 : config: TenantConfigPatch {
2582 1 : checkpoint_distance: FieldPatch::Upsert(42),
2583 1 : gc_horizon: FieldPatch::Remove,
2584 1 : compaction_threshold: FieldPatch::Noop,
2585 1 : ..TenantConfigPatch::default()
2586 1 : },
2587 1 : };
2588 1 :
2589 1 : let json = serde_json::to_string(&patch_request).unwrap();
2590 1 :
2591 1 : let expected = r#"{"tenant_id":"17c6d121946a61e5ab0fe5a2fd4d8215","checkpoint_distance":42,"gc_horizon":null}"#;
2592 1 : assert_eq!(json, expected);
2593 :
2594 1 : let decoded: TenantConfigPatchRequest = serde_json::from_str(&json).unwrap();
2595 1 : assert_eq!(decoded.tenant_id, patch_request.tenant_id);
2596 1 : assert_eq!(decoded.config, patch_request.config);
2597 :
2598 : // Now apply the patch to a config to demonstrate semantics
2599 :
2600 1 : let base = TenantConfig {
2601 1 : checkpoint_distance: Some(28),
2602 1 : gc_horizon: Some(100),
2603 1 : compaction_target_size: Some(1024),
2604 1 : ..Default::default()
2605 1 : };
2606 1 :
2607 1 : let expected = TenantConfig {
2608 1 : checkpoint_distance: Some(42),
2609 1 : gc_horizon: None,
2610 1 : ..base.clone()
2611 1 : };
2612 1 :
2613 1 : let patched = base.apply_patch(decoded.config).unwrap();
2614 1 :
2615 1 : assert_eq!(patched, expected);
2616 1 : }
2617 : }
|
