Line data Source code
1 : pub mod detach_ancestor;
2 : pub mod partitioning;
3 : pub mod utilization;
4 :
5 : pub use utilization::PageserverUtilization;
6 :
7 : use std::{
8 : borrow::Cow,
9 : collections::HashMap,
10 : io::{BufRead, Read},
11 : num::{NonZeroU64, NonZeroUsize},
12 : sync::atomic::AtomicUsize,
13 : time::{Duration, SystemTime},
14 : };
15 :
16 : use byteorder::{BigEndian, ReadBytesExt};
17 : use postgres_ffi::BLCKSZ;
18 : use serde::{Deserialize, Serialize};
19 : use serde_with::serde_as;
20 : use utils::{
21 : completion,
22 : history_buffer::HistoryBufferWithDropCounter,
23 : id::{NodeId, TenantId, TimelineId},
24 : lsn::Lsn,
25 : serde_system_time,
26 : };
27 :
28 : use crate::controller_api::PlacementPolicy;
29 : use crate::{
30 : reltag::RelTag,
31 : shard::{ShardCount, ShardStripeSize, TenantShardId},
32 : };
33 : use anyhow::bail;
34 : use bytes::{Buf, BufMut, Bytes, BytesMut};
35 :
36 : /// The state of a tenant in this pageserver.
37 : ///
38 : /// ```mermaid
39 : /// stateDiagram-v2
40 : ///
41 : /// [*] --> Loading: spawn_load()
42 : /// [*] --> Attaching: spawn_attach()
43 : ///
44 : /// Loading --> Activating: activate()
45 : /// Attaching --> Activating: activate()
46 : /// Activating --> Active: infallible
47 : ///
48 : /// Loading --> Broken: load() failure
49 : /// Attaching --> Broken: attach() failure
50 : ///
51 : /// Active --> Stopping: set_stopping(), part of shutdown & detach
52 : /// Stopping --> Broken: late error in remove_tenant_from_memory
53 : ///
54 : /// Broken --> [*]: ignore / detach / shutdown
55 : /// Stopping --> [*]: remove_from_memory complete
56 : ///
57 : /// Active --> Broken: cfg(testing)-only tenant break point
58 : /// ```
59 : #[derive(
60 : Clone,
61 : PartialEq,
62 : Eq,
63 2 : serde::Serialize,
64 12 : serde::Deserialize,
65 0 : strum_macros::Display,
66 : strum_macros::EnumVariantNames,
67 0 : strum_macros::AsRefStr,
68 324 : strum_macros::IntoStaticStr,
69 : )]
70 : #[serde(tag = "slug", content = "data")]
71 : pub enum TenantState {
72 : /// This tenant is being loaded from local disk.
73 : ///
74 : /// `set_stopping()` and `set_broken()` do not work in this state and wait for it to pass.
75 : Loading,
76 : /// This tenant is being attached to the pageserver.
77 : ///
78 : /// `set_stopping()` and `set_broken()` do not work in this state and wait for it to pass.
79 : Attaching,
80 : /// The tenant is transitioning from Loading/Attaching to Active.
81 : ///
82 : /// While in this state, the individual timelines are being activated.
83 : ///
84 : /// `set_stopping()` and `set_broken()` do not work in this state and wait for it to pass.
85 : Activating(ActivatingFrom),
86 : /// The tenant has finished activating and is open for business.
87 : ///
88 : /// Transitions out of this state are possible through `set_stopping()` and `set_broken()`.
89 : Active,
90 : /// The tenant is recognized by pageserver, but it is being detached or the
91 : /// system is being shut down.
92 : ///
93 : /// Transitions out of this state are possible through `set_broken()`.
94 : Stopping {
95 : // Because of https://github.com/serde-rs/serde/issues/2105 this has to be a named field,
96 : // otherwise it will not be skipped during deserialization
97 : #[serde(skip)]
98 : progress: completion::Barrier,
99 : },
100 : /// The tenant is recognized by the pageserver, but can no longer be used for
101 : /// any operations.
102 : ///
103 : /// If the tenant fails to load or attach, it will transition to this state
104 : /// and it is guaranteed that no background tasks are running in its name.
105 : ///
106 : /// The other way to transition into this state is from `Stopping` state
107 : /// through `set_broken()` called from `remove_tenant_from_memory()`. That happens
108 : /// if the cleanup future executed by `remove_tenant_from_memory()` fails.
109 : Broken { reason: String, backtrace: String },
110 : }
111 :
112 : impl TenantState {
113 0 : pub fn attachment_status(&self) -> TenantAttachmentStatus {
114 : use TenantAttachmentStatus::*;
115 :
116 : // Below TenantState::Activating is used as "transient" or "transparent" state for
117 : // attachment_status determining.
118 0 : match self {
119 : // The attach procedure writes the marker file before adding the Attaching tenant to the tenants map.
120 : // So, technically, we can return Attached here.
121 : // However, as soon as Console observes Attached, it will proceed with the Postgres-level health check.
122 : // But, our attach task might still be fetching the remote timelines, etc.
123 : // So, return `Maybe` while Attaching, making Console wait for the attach task to finish.
124 0 : Self::Attaching | Self::Activating(ActivatingFrom::Attaching) => Maybe,
125 : // tenant mgr startup distinguishes attaching from loading via marker file.
126 0 : Self::Loading | Self::Activating(ActivatingFrom::Loading) => Attached,
127 : // We only reach Active after successful load / attach.
128 : // So, call atttachment status Attached.
129 0 : Self::Active => Attached,
130 : // If the (initial or resumed) attach procedure fails, the tenant becomes Broken.
131 : // However, it also becomes Broken if the regular load fails.
132 : // From Console's perspective there's no practical difference
133 : // because attachment_status is polled by console only during attach operation execution.
134 0 : Self::Broken { reason, .. } => Failed {
135 0 : reason: reason.to_owned(),
136 0 : },
137 : // Why is Stopping a Maybe case? Because, during pageserver shutdown,
138 : // we set the Stopping state irrespective of whether the tenant
139 : // has finished attaching or not.
140 0 : Self::Stopping { .. } => Maybe,
141 : }
142 0 : }
143 :
144 0 : pub fn broken_from_reason(reason: String) -> Self {
145 0 : let backtrace_str: String = format!("{}", std::backtrace::Backtrace::force_capture());
146 0 : Self::Broken {
147 0 : reason,
148 0 : backtrace: backtrace_str,
149 0 : }
150 0 : }
151 : }
152 :
153 : impl std::fmt::Debug for TenantState {
154 4 : fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
155 4 : match self {
156 4 : Self::Broken { reason, backtrace } if !reason.is_empty() => {
157 4 : write!(f, "Broken due to: {reason}. Backtrace:\n{backtrace}")
158 : }
159 0 : _ => write!(f, "{self}"),
160 : }
161 4 : }
162 : }
163 :
164 : /// A temporary lease to a specific lsn inside a timeline.
165 : /// Access to the lsn is guaranteed by the pageserver until the expiration indicated by `valid_until`.
166 : #[serde_as]
167 0 : #[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
168 : pub struct LsnLease {
169 : #[serde_as(as = "SystemTimeAsRfc3339Millis")]
170 : pub valid_until: SystemTime,
171 : }
172 :
173 : serde_with::serde_conv!(
174 : SystemTimeAsRfc3339Millis,
175 : SystemTime,
176 0 : |time: &SystemTime| humantime::format_rfc3339_millis(*time).to_string(),
177 0 : |value: String| -> Result<_, humantime::TimestampError> { humantime::parse_rfc3339(&value) }
178 : );
179 :
180 : impl LsnLease {
181 : /// The default length for an explicit LSN lease request (10 minutes).
182 : pub const DEFAULT_LENGTH: Duration = Duration::from_secs(10 * 60);
183 :
184 : /// The default length for an implicit LSN lease granted during
185 : /// `get_lsn_by_timestamp` request (1 minutes).
186 : pub const DEFAULT_LENGTH_FOR_TS: Duration = Duration::from_secs(60);
187 :
188 : /// Checks whether the lease is expired.
189 6 : pub fn is_expired(&self, now: &SystemTime) -> bool {
190 6 : now > &self.valid_until
191 6 : }
192 : }
193 :
194 : /// The only [`TenantState`] variants we could be `TenantState::Activating` from.
195 8 : #[derive(Clone, Copy, Debug, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
196 : pub enum ActivatingFrom {
197 : /// Arrived to [`TenantState::Activating`] from [`TenantState::Loading`]
198 : Loading,
199 : /// Arrived to [`TenantState::Activating`] from [`TenantState::Attaching`]
200 : Attaching,
201 : }
202 :
203 : /// A state of a timeline in pageserver's memory.
204 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
205 : pub enum TimelineState {
206 : /// The timeline is recognized by the pageserver but is not yet operational.
207 : /// In particular, the walreceiver connection loop is not running for this timeline.
208 : /// It will eventually transition to state Active or Broken.
209 : Loading,
210 : /// The timeline is fully operational.
211 : /// It can be queried, and the walreceiver connection loop is running.
212 : Active,
213 : /// The timeline was previously Loading or Active but is shutting down.
214 : /// It cannot transition back into any other state.
215 : Stopping,
216 : /// The timeline is broken and not operational (previous states: Loading or Active).
217 : Broken { reason: String, backtrace: String },
218 : }
219 :
220 0 : #[derive(Serialize, Deserialize, Clone)]
221 : pub struct TimelineCreateRequest {
222 : pub new_timeline_id: TimelineId,
223 : #[serde(default)]
224 : pub ancestor_timeline_id: Option<TimelineId>,
225 : #[serde(default)]
226 : pub existing_initdb_timeline_id: Option<TimelineId>,
227 : #[serde(default)]
228 : pub ancestor_start_lsn: Option<Lsn>,
229 : pub pg_version: Option<u32>,
230 : }
231 :
232 0 : #[derive(Serialize, Deserialize)]
233 : pub struct TenantShardSplitRequest {
234 : pub new_shard_count: u8,
235 :
236 : // A tenant's stripe size is only meaningful the first time their shard count goes
237 : // above 1: therefore during a split from 1->N shards, we may modify the stripe size.
238 : //
239 : // If this is set while the stripe count is being increased from an already >1 value,
240 : // then the request will fail with 400.
241 : pub new_stripe_size: Option<ShardStripeSize>,
242 : }
243 :
244 0 : #[derive(Serialize, Deserialize)]
245 : pub struct TenantShardSplitResponse {
246 : pub new_shards: Vec<TenantShardId>,
247 : }
248 :
249 : /// Parameters that apply to all shards in a tenant. Used during tenant creation.
250 0 : #[derive(Serialize, Deserialize, Debug)]
251 : #[serde(deny_unknown_fields)]
252 : pub struct ShardParameters {
253 : pub count: ShardCount,
254 : pub stripe_size: ShardStripeSize,
255 : }
256 :
257 : impl ShardParameters {
258 : pub const DEFAULT_STRIPE_SIZE: ShardStripeSize = ShardStripeSize(256 * 1024 / 8);
259 :
260 0 : pub fn is_unsharded(&self) -> bool {
261 0 : self.count.is_unsharded()
262 0 : }
263 : }
264 :
265 : impl Default for ShardParameters {
266 166 : fn default() -> Self {
267 166 : Self {
268 166 : count: ShardCount::new(0),
269 166 : stripe_size: Self::DEFAULT_STRIPE_SIZE,
270 166 : }
271 166 : }
272 : }
273 :
274 6 : #[derive(Serialize, Deserialize, Debug)]
275 : #[serde(deny_unknown_fields)]
276 : pub struct TenantCreateRequest {
277 : pub new_tenant_id: TenantShardId,
278 : #[serde(default)]
279 : #[serde(skip_serializing_if = "Option::is_none")]
280 : pub generation: Option<u32>,
281 :
282 : // If omitted, create a single shard with TenantShardId::unsharded()
283 : #[serde(default)]
284 : #[serde(skip_serializing_if = "ShardParameters::is_unsharded")]
285 : pub shard_parameters: ShardParameters,
286 :
287 : // This parameter is only meaningful in requests sent to the storage controller
288 : #[serde(default)]
289 : #[serde(skip_serializing_if = "Option::is_none")]
290 : pub placement_policy: Option<PlacementPolicy>,
291 :
292 : #[serde(flatten)]
293 : pub config: TenantConfig, // as we have a flattened field, we should reject all unknown fields in it
294 : }
295 :
296 0 : #[derive(Deserialize, Debug)]
297 : #[serde(deny_unknown_fields)]
298 : pub struct TenantLoadRequest {
299 : #[serde(default)]
300 : #[serde(skip_serializing_if = "Option::is_none")]
301 : pub generation: Option<u32>,
302 : }
303 :
304 : impl std::ops::Deref for TenantCreateRequest {
305 : type Target = TenantConfig;
306 :
307 0 : fn deref(&self) -> &Self::Target {
308 0 : &self.config
309 0 : }
310 : }
311 :
312 : /// An alternative representation of `pageserver::tenant::TenantConf` with
313 : /// simpler types.
314 6 : #[derive(Serialize, Deserialize, Debug, Default, Clone, Eq, PartialEq)]
315 : pub struct TenantConfig {
316 : pub checkpoint_distance: Option<u64>,
317 : pub checkpoint_timeout: Option<String>,
318 : pub compaction_target_size: Option<u64>,
319 : pub compaction_period: Option<String>,
320 : pub compaction_threshold: Option<usize>,
321 : // defer parsing compaction_algorithm, like eviction_policy
322 : pub compaction_algorithm: Option<CompactionAlgorithmSettings>,
323 : pub gc_horizon: Option<u64>,
324 : pub gc_period: Option<String>,
325 : pub image_creation_threshold: Option<usize>,
326 : pub pitr_interval: Option<String>,
327 : pub walreceiver_connect_timeout: Option<String>,
328 : pub lagging_wal_timeout: Option<String>,
329 : pub max_lsn_wal_lag: Option<NonZeroU64>,
330 : pub trace_read_requests: Option<bool>,
331 : pub eviction_policy: Option<EvictionPolicy>,
332 : pub min_resident_size_override: Option<u64>,
333 : pub evictions_low_residence_duration_metric_threshold: Option<String>,
334 : pub heatmap_period: Option<String>,
335 : pub lazy_slru_download: Option<bool>,
336 : pub timeline_get_throttle: Option<ThrottleConfig>,
337 : pub image_layer_creation_check_threshold: Option<u8>,
338 : pub switch_aux_file_policy: Option<AuxFilePolicy>,
339 : pub lsn_lease_length: Option<String>,
340 : pub lsn_lease_length_for_ts: Option<String>,
341 : }
342 :
343 : /// The policy for the aux file storage. It can be switched through `switch_aux_file_policy`
344 : /// tenant config. When the first aux file written, the policy will be persisted in the
345 : /// `index_part.json` file and has a limited migration path.
346 : ///
347 : /// Currently, we only allow the following migration path:
348 : ///
349 : /// Unset -> V1
350 : /// -> V2
351 : /// -> CrossValidation -> V2
352 : #[derive(
353 : Eq,
354 : PartialEq,
355 : Debug,
356 : Copy,
357 : Clone,
358 8 : strum_macros::EnumString,
359 21 : strum_macros::Display,
360 0 : serde_with::DeserializeFromStr,
361 : serde_with::SerializeDisplay,
362 : )]
363 : #[strum(serialize_all = "kebab-case")]
364 : pub enum AuxFilePolicy {
365 : /// V1 aux file policy: store everything in AUX_FILE_KEY
366 : #[strum(ascii_case_insensitive)]
367 : V1,
368 : /// V2 aux file policy: store in the AUX_FILE keyspace
369 : #[strum(ascii_case_insensitive)]
370 : V2,
371 : /// Cross validation runs both formats on the write path and does validation
372 : /// on the read path.
373 : #[strum(ascii_case_insensitive)]
374 : CrossValidation,
375 : }
376 :
377 : impl AuxFilePolicy {
378 54 : pub fn is_valid_migration_path(from: Option<Self>, to: Self) -> bool {
379 34 : matches!(
380 54 : (from, to),
381 : (None, _) | (Some(AuxFilePolicy::CrossValidation), AuxFilePolicy::V2)
382 : )
383 54 : }
384 :
385 : /// If a tenant writes aux files without setting `switch_aux_policy`, this value will be used.
386 370 : pub fn default_tenant_config() -> Self {
387 370 : Self::V1
388 370 : }
389 : }
390 :
391 : /// The aux file policy memory flag. Users can store `Option<AuxFilePolicy>` into this atomic flag. 0 == unspecified.
392 : pub struct AtomicAuxFilePolicy(AtomicUsize);
393 :
394 : impl AtomicAuxFilePolicy {
395 378 : pub fn new(policy: Option<AuxFilePolicy>) -> Self {
396 378 : Self(AtomicUsize::new(
397 378 : policy.map(AuxFilePolicy::to_usize).unwrap_or_default(),
398 378 : ))
399 378 : }
400 :
401 306 : pub fn load(&self) -> Option<AuxFilePolicy> {
402 306 : match self.0.load(std::sync::atomic::Ordering::Acquire) {
403 240 : 0 => None,
404 66 : other => Some(AuxFilePolicy::from_usize(other)),
405 : }
406 306 : }
407 :
408 22 : pub fn store(&self, policy: Option<AuxFilePolicy>) {
409 22 : self.0.store(
410 22 : policy.map(AuxFilePolicy::to_usize).unwrap_or_default(),
411 22 : std::sync::atomic::Ordering::Release,
412 22 : );
413 22 : }
414 : }
415 :
416 : impl AuxFilePolicy {
417 20 : pub fn to_usize(self) -> usize {
418 20 : match self {
419 14 : Self::V1 => 1,
420 2 : Self::CrossValidation => 2,
421 4 : Self::V2 => 3,
422 : }
423 20 : }
424 :
425 66 : pub fn try_from_usize(this: usize) -> Option<Self> {
426 66 : match this {
427 36 : 1 => Some(Self::V1),
428 6 : 2 => Some(Self::CrossValidation),
429 24 : 3 => Some(Self::V2),
430 0 : _ => None,
431 : }
432 66 : }
433 :
434 66 : pub fn from_usize(this: usize) -> Self {
435 66 : Self::try_from_usize(this).unwrap()
436 66 : }
437 : }
438 :
439 4 : #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
440 : #[serde(tag = "kind")]
441 : pub enum EvictionPolicy {
442 : NoEviction,
443 : LayerAccessThreshold(EvictionPolicyLayerAccessThreshold),
444 : OnlyImitiate(EvictionPolicyLayerAccessThreshold),
445 : }
446 :
447 : impl EvictionPolicy {
448 0 : pub fn discriminant_str(&self) -> &'static str {
449 0 : match self {
450 0 : EvictionPolicy::NoEviction => "NoEviction",
451 0 : EvictionPolicy::LayerAccessThreshold(_) => "LayerAccessThreshold",
452 0 : EvictionPolicy::OnlyImitiate(_) => "OnlyImitiate",
453 : }
454 0 : }
455 : }
456 :
457 : #[derive(
458 : Eq,
459 : PartialEq,
460 : Debug,
461 : Copy,
462 : Clone,
463 0 : strum_macros::EnumString,
464 0 : strum_macros::Display,
465 0 : serde_with::DeserializeFromStr,
466 : serde_with::SerializeDisplay,
467 : )]
468 : #[strum(serialize_all = "kebab-case")]
469 : pub enum CompactionAlgorithm {
470 : Legacy,
471 : Tiered,
472 : }
473 :
474 0 : #[derive(Eq, PartialEq, Debug, Clone, Serialize, Deserialize)]
475 : pub struct CompactionAlgorithmSettings {
476 : pub kind: CompactionAlgorithm,
477 : }
478 :
479 20 : #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
480 : pub struct EvictionPolicyLayerAccessThreshold {
481 : #[serde(with = "humantime_serde")]
482 : pub period: Duration,
483 : #[serde(with = "humantime_serde")]
484 : pub threshold: Duration,
485 : }
486 :
487 0 : #[derive(Debug, Serialize, Deserialize, Clone, PartialEq, Eq)]
488 : pub struct ThrottleConfig {
489 : pub task_kinds: Vec<String>, // TaskKind
490 : pub initial: usize,
491 : #[serde(with = "humantime_serde")]
492 : pub refill_interval: Duration,
493 : pub refill_amount: NonZeroUsize,
494 : pub max: usize,
495 : pub fair: bool,
496 : }
497 :
498 : impl ThrottleConfig {
499 352 : pub fn disabled() -> Self {
500 352 : Self {
501 352 : task_kinds: vec![], // effectively disables the throttle
502 352 : // other values don't matter with emtpy `task_kinds`.
503 352 : initial: 0,
504 352 : refill_interval: Duration::from_millis(1),
505 352 : refill_amount: NonZeroUsize::new(1).unwrap(),
506 352 : max: 1,
507 352 : fair: true,
508 352 : }
509 352 : }
510 : /// The requests per second allowed by the given config.
511 0 : pub fn steady_rps(&self) -> f64 {
512 0 : (self.refill_amount.get() as f64) / (self.refill_interval.as_secs_f64())
513 0 : }
514 : }
515 :
516 : /// A flattened analog of a `pagesever::tenant::LocationMode`, which
517 : /// lists out all possible states (and the virtual "Detached" state)
518 : /// in a flat form rather than using rust-style enums.
519 0 : #[derive(Serialize, Deserialize, Debug, Clone, Copy, Eq, PartialEq)]
520 : pub enum LocationConfigMode {
521 : AttachedSingle,
522 : AttachedMulti,
523 : AttachedStale,
524 : Secondary,
525 : Detached,
526 : }
527 :
528 0 : #[derive(Serialize, Deserialize, Debug, Clone, Eq, PartialEq)]
529 : pub struct LocationConfigSecondary {
530 : pub warm: bool,
531 : }
532 :
533 : /// An alternative representation of `pageserver::tenant::LocationConf`,
534 : /// for use in external-facing APIs.
535 0 : #[derive(Serialize, Deserialize, Debug, Clone, Eq, PartialEq)]
536 : pub struct LocationConfig {
537 : pub mode: LocationConfigMode,
538 : /// If attaching, in what generation?
539 : #[serde(default)]
540 : pub generation: Option<u32>,
541 :
542 : // If requesting mode `Secondary`, configuration for that.
543 : #[serde(default)]
544 : pub secondary_conf: Option<LocationConfigSecondary>,
545 :
546 : // Shard parameters: if shard_count is nonzero, then other shard_* fields
547 : // must be set accurately.
548 : #[serde(default)]
549 : pub shard_number: u8,
550 : #[serde(default)]
551 : pub shard_count: u8,
552 : #[serde(default)]
553 : pub shard_stripe_size: u32,
554 :
555 : // This configuration only affects attached mode, but should be provided irrespective
556 : // of the mode, as a secondary location might transition on startup if the response
557 : // to the `/re-attach` control plane API requests it.
558 : pub tenant_conf: TenantConfig,
559 : }
560 :
561 0 : #[derive(Serialize, Deserialize)]
562 : pub struct LocationConfigListResponse {
563 : pub tenant_shards: Vec<(TenantShardId, Option<LocationConfig>)>,
564 : }
565 :
566 0 : #[derive(Serialize, Deserialize)]
567 : #[serde(transparent)]
568 : pub struct TenantCreateResponse(pub TenantId);
569 :
570 : #[derive(Serialize)]
571 : pub struct StatusResponse {
572 : pub id: NodeId,
573 : }
574 :
575 0 : #[derive(Serialize, Deserialize, Debug)]
576 : #[serde(deny_unknown_fields)]
577 : pub struct TenantLocationConfigRequest {
578 : #[serde(flatten)]
579 : pub config: LocationConfig, // as we have a flattened field, we should reject all unknown fields in it
580 : }
581 :
582 0 : #[derive(Serialize, Deserialize, Debug)]
583 : #[serde(deny_unknown_fields)]
584 : pub struct TenantTimeTravelRequest {
585 : pub shard_counts: Vec<ShardCount>,
586 : }
587 :
588 0 : #[derive(Serialize, Deserialize, Debug)]
589 : #[serde(deny_unknown_fields)]
590 : pub struct TenantShardLocation {
591 : pub shard_id: TenantShardId,
592 : pub node_id: NodeId,
593 : }
594 :
595 0 : #[derive(Serialize, Deserialize, Debug)]
596 : #[serde(deny_unknown_fields)]
597 : pub struct TenantLocationConfigResponse {
598 : pub shards: Vec<TenantShardLocation>,
599 : // If the shards' ShardCount count is >1, stripe_size will be set.
600 : pub stripe_size: Option<ShardStripeSize>,
601 : }
602 :
603 6 : #[derive(Serialize, Deserialize, Debug)]
604 : #[serde(deny_unknown_fields)]
605 : pub struct TenantConfigRequest {
606 : pub tenant_id: TenantId,
607 : #[serde(flatten)]
608 : pub config: TenantConfig, // as we have a flattened field, we should reject all unknown fields in it
609 : }
610 :
611 : impl std::ops::Deref for TenantConfigRequest {
612 : type Target = TenantConfig;
613 :
614 0 : fn deref(&self) -> &Self::Target {
615 0 : &self.config
616 0 : }
617 : }
618 :
619 : impl TenantConfigRequest {
620 0 : pub fn new(tenant_id: TenantId) -> TenantConfigRequest {
621 0 : let config = TenantConfig::default();
622 0 : TenantConfigRequest { tenant_id, config }
623 0 : }
624 : }
625 :
626 6 : #[derive(Debug, Deserialize)]
627 : pub struct TenantAttachRequest {
628 : #[serde(default)]
629 : pub config: TenantAttachConfig,
630 : #[serde(default)]
631 : pub generation: Option<u32>,
632 : }
633 :
634 : /// Newtype to enforce deny_unknown_fields on TenantConfig for
635 : /// its usage inside `TenantAttachRequest`.
636 2 : #[derive(Debug, Serialize, Deserialize, Default)]
637 : #[serde(deny_unknown_fields)]
638 : pub struct TenantAttachConfig {
639 : #[serde(flatten)]
640 : allowing_unknown_fields: TenantConfig,
641 : }
642 :
643 : impl std::ops::Deref for TenantAttachConfig {
644 : type Target = TenantConfig;
645 :
646 0 : fn deref(&self) -> &Self::Target {
647 0 : &self.allowing_unknown_fields
648 0 : }
649 : }
650 :
651 : /// See [`TenantState::attachment_status`] and the OpenAPI docs for context.
652 0 : #[derive(Serialize, Deserialize, Clone)]
653 : #[serde(tag = "slug", content = "data", rename_all = "snake_case")]
654 : pub enum TenantAttachmentStatus {
655 : Maybe,
656 : Attached,
657 : Failed { reason: String },
658 : }
659 :
660 0 : #[derive(Serialize, Deserialize, Clone)]
661 : pub struct TenantInfo {
662 : pub id: TenantShardId,
663 : // NB: intentionally not part of OpenAPI, we don't want to commit to a specific set of TenantState's
664 : pub state: TenantState,
665 : /// Sum of the size of all layer files.
666 : /// If a layer is present in both local FS and S3, it counts only once.
667 : pub current_physical_size: Option<u64>, // physical size is only included in `tenant_status` endpoint
668 : pub attachment_status: TenantAttachmentStatus,
669 : #[serde(skip_serializing_if = "Option::is_none")]
670 : pub generation: Option<u32>,
671 : }
672 :
673 0 : #[derive(Serialize, Deserialize, Clone)]
674 : pub struct TenantDetails {
675 : #[serde(flatten)]
676 : pub tenant_info: TenantInfo,
677 :
678 : pub walredo: Option<WalRedoManagerStatus>,
679 :
680 : pub timelines: Vec<TimelineId>,
681 : }
682 :
683 : /// This represents the output of the "timeline_detail" and "timeline_list" API calls.
684 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
685 : pub struct TimelineInfo {
686 : pub tenant_id: TenantShardId,
687 : pub timeline_id: TimelineId,
688 :
689 : pub ancestor_timeline_id: Option<TimelineId>,
690 : pub ancestor_lsn: Option<Lsn>,
691 : pub last_record_lsn: Lsn,
692 : pub prev_record_lsn: Option<Lsn>,
693 : pub latest_gc_cutoff_lsn: Lsn,
694 : pub disk_consistent_lsn: Lsn,
695 :
696 : /// The LSN that we have succesfully uploaded to remote storage
697 : pub remote_consistent_lsn: Lsn,
698 :
699 : /// The LSN that we are advertizing to safekeepers
700 : pub remote_consistent_lsn_visible: Lsn,
701 :
702 : /// The LSN from the start of the root timeline (never changes)
703 : pub initdb_lsn: Lsn,
704 :
705 : pub current_logical_size: u64,
706 : pub current_logical_size_is_accurate: bool,
707 :
708 : pub directory_entries_counts: Vec<u64>,
709 :
710 : /// Sum of the size of all layer files.
711 : /// If a layer is present in both local FS and S3, it counts only once.
712 : pub current_physical_size: Option<u64>, // is None when timeline is Unloaded
713 : pub current_logical_size_non_incremental: Option<u64>,
714 :
715 : pub timeline_dir_layer_file_size_sum: Option<u64>,
716 :
717 : pub wal_source_connstr: Option<String>,
718 : pub last_received_msg_lsn: Option<Lsn>,
719 : /// the timestamp (in microseconds) of the last received message
720 : pub last_received_msg_ts: Option<u128>,
721 : pub pg_version: u32,
722 :
723 : pub state: TimelineState,
724 :
725 : pub walreceiver_status: String,
726 :
727 : /// The last aux file policy being used on this timeline
728 : pub last_aux_file_policy: Option<AuxFilePolicy>,
729 : }
730 :
731 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
732 : pub struct LayerMapInfo {
733 : pub in_memory_layers: Vec<InMemoryLayerInfo>,
734 : pub historic_layers: Vec<HistoricLayerInfo>,
735 : }
736 :
737 0 : #[derive(Debug, Hash, PartialEq, Eq, Clone, Copy, Serialize, Deserialize, enum_map::Enum)]
738 : #[repr(usize)]
739 : pub enum LayerAccessKind {
740 : GetValueReconstructData,
741 : Iter,
742 : KeyIter,
743 : Dump,
744 : }
745 :
746 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
747 : pub struct LayerAccessStatFullDetails {
748 : pub when_millis_since_epoch: u64,
749 : pub task_kind: Cow<'static, str>,
750 : pub access_kind: LayerAccessKind,
751 : }
752 :
753 : /// An event that impacts the layer's residence status.
754 : #[serde_as]
755 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
756 : pub struct LayerResidenceEvent {
757 : /// The time when the event occurred.
758 : /// NB: this timestamp is captured while the residence status changes.
759 : /// So, it might be behind/ahead of the actual residence change by a short amount of time.
760 : ///
761 : #[serde(rename = "timestamp_millis_since_epoch")]
762 : #[serde_as(as = "serde_with::TimestampMilliSeconds")]
763 : pub timestamp: SystemTime,
764 : /// The new residence status of the layer.
765 : pub status: LayerResidenceStatus,
766 : /// The reason why we had to record this event.
767 : pub reason: LayerResidenceEventReason,
768 : }
769 :
770 : /// The reason for recording a given [`LayerResidenceEvent`].
771 0 : #[derive(Debug, Clone, Copy, Serialize, Deserialize)]
772 : pub enum LayerResidenceEventReason {
773 : /// The layer map is being populated, e.g. during timeline load or attach.
774 : /// This includes [`RemoteLayer`] objects created in [`reconcile_with_remote`].
775 : /// We need to record such events because there is no persistent storage for the events.
776 : ///
777 : // https://github.com/rust-lang/rust/issues/74481
778 : /// [`RemoteLayer`]: ../../tenant/storage_layer/struct.RemoteLayer.html
779 : /// [`reconcile_with_remote`]: ../../tenant/struct.Timeline.html#method.reconcile_with_remote
780 : LayerLoad,
781 : /// We just created the layer (e.g., freeze_and_flush or compaction).
782 : /// Such layers are always [`LayerResidenceStatus::Resident`].
783 : LayerCreate,
784 : /// We on-demand downloaded or evicted the given layer.
785 : ResidenceChange,
786 : }
787 :
788 : /// The residence status of the layer, after the given [`LayerResidenceEvent`].
789 0 : #[derive(Debug, Clone, Copy, Serialize, Deserialize)]
790 : pub enum LayerResidenceStatus {
791 : /// Residence status for a layer file that exists locally.
792 : /// It may also exist on the remote, we don't care here.
793 : Resident,
794 : /// Residence status for a layer file that only exists on the remote.
795 : Evicted,
796 : }
797 :
798 : impl LayerResidenceEvent {
799 3184 : pub fn new(status: LayerResidenceStatus, reason: LayerResidenceEventReason) -> Self {
800 3184 : Self {
801 3184 : status,
802 3184 : reason,
803 3184 : timestamp: SystemTime::now(),
804 3184 : }
805 3184 : }
806 : }
807 :
808 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
809 : pub struct LayerAccessStats {
810 : pub access_count_by_access_kind: HashMap<LayerAccessKind, u64>,
811 : pub task_kind_access_flag: Vec<Cow<'static, str>>,
812 : pub first: Option<LayerAccessStatFullDetails>,
813 : pub accesses_history: HistoryBufferWithDropCounter<LayerAccessStatFullDetails, 16>,
814 : pub residence_events_history: HistoryBufferWithDropCounter<LayerResidenceEvent, 16>,
815 : }
816 :
817 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
818 : #[serde(tag = "kind")]
819 : pub enum InMemoryLayerInfo {
820 : Open { lsn_start: Lsn },
821 : Frozen { lsn_start: Lsn, lsn_end: Lsn },
822 : }
823 :
824 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
825 : #[serde(tag = "kind")]
826 : pub enum HistoricLayerInfo {
827 : Delta {
828 : layer_file_name: String,
829 : layer_file_size: u64,
830 :
831 : lsn_start: Lsn,
832 : lsn_end: Lsn,
833 : remote: bool,
834 : access_stats: LayerAccessStats,
835 :
836 : l0: bool,
837 : },
838 : Image {
839 : layer_file_name: String,
840 : layer_file_size: u64,
841 :
842 : lsn_start: Lsn,
843 : remote: bool,
844 : access_stats: LayerAccessStats,
845 : },
846 : }
847 :
848 : impl HistoricLayerInfo {
849 0 : pub fn layer_file_name(&self) -> &str {
850 0 : match self {
851 : HistoricLayerInfo::Delta {
852 0 : layer_file_name, ..
853 0 : } => layer_file_name,
854 : HistoricLayerInfo::Image {
855 0 : layer_file_name, ..
856 0 : } => layer_file_name,
857 : }
858 0 : }
859 0 : pub fn is_remote(&self) -> bool {
860 0 : match self {
861 0 : HistoricLayerInfo::Delta { remote, .. } => *remote,
862 0 : HistoricLayerInfo::Image { remote, .. } => *remote,
863 : }
864 0 : }
865 0 : pub fn set_remote(&mut self, value: bool) {
866 0 : let field = match self {
867 0 : HistoricLayerInfo::Delta { remote, .. } => remote,
868 0 : HistoricLayerInfo::Image { remote, .. } => remote,
869 : };
870 0 : *field = value;
871 0 : }
872 0 : pub fn layer_file_size(&self) -> u64 {
873 0 : match self {
874 : HistoricLayerInfo::Delta {
875 0 : layer_file_size, ..
876 0 : } => *layer_file_size,
877 : HistoricLayerInfo::Image {
878 0 : layer_file_size, ..
879 0 : } => *layer_file_size,
880 : }
881 0 : }
882 : }
883 :
884 0 : #[derive(Debug, Serialize, Deserialize)]
885 : pub struct DownloadRemoteLayersTaskSpawnRequest {
886 : pub max_concurrent_downloads: NonZeroUsize,
887 : }
888 :
889 0 : #[derive(Debug, Serialize, Deserialize)]
890 : pub struct IngestAuxFilesRequest {
891 : pub aux_files: HashMap<String, String>,
892 : }
893 :
894 0 : #[derive(Debug, Serialize, Deserialize)]
895 : pub struct ListAuxFilesRequest {
896 : pub lsn: Lsn,
897 : }
898 :
899 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
900 : pub struct DownloadRemoteLayersTaskInfo {
901 : pub task_id: String,
902 : pub state: DownloadRemoteLayersTaskState,
903 : pub total_layer_count: u64, // stable once `completed`
904 : pub successful_download_count: u64, // stable once `completed`
905 : pub failed_download_count: u64, // stable once `completed`
906 : }
907 :
908 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
909 : pub enum DownloadRemoteLayersTaskState {
910 : Running,
911 : Completed,
912 : ShutDown,
913 : }
914 :
915 0 : #[derive(Debug, Serialize, Deserialize)]
916 : pub struct TimelineGcRequest {
917 : pub gc_horizon: Option<u64>,
918 : }
919 :
920 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
921 : pub struct WalRedoManagerProcessStatus {
922 : pub pid: u32,
923 : }
924 :
925 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
926 : pub struct WalRedoManagerStatus {
927 : pub last_redo_at: Option<chrono::DateTime<chrono::Utc>>,
928 : pub process: Option<WalRedoManagerProcessStatus>,
929 : }
930 :
931 : /// The progress of a secondary tenant is mostly useful when doing a long running download: e.g. initiating
932 : /// a download job, timing out while waiting for it to run, and then inspecting this status to understand
933 : /// what's happening.
934 0 : #[derive(Default, Debug, Serialize, Deserialize, Clone)]
935 : pub struct SecondaryProgress {
936 : /// The remote storage LastModified time of the heatmap object we last downloaded.
937 : pub heatmap_mtime: Option<serde_system_time::SystemTime>,
938 :
939 : /// The number of layers currently on-disk
940 : pub layers_downloaded: usize,
941 : /// The number of layers in the most recently seen heatmap
942 : pub layers_total: usize,
943 :
944 : /// The number of layer bytes currently on-disk
945 : pub bytes_downloaded: u64,
946 : /// The number of layer bytes in the most recently seen heatmap
947 : pub bytes_total: u64,
948 : }
949 :
950 0 : #[derive(Serialize, Deserialize, Debug)]
951 : pub struct TenantScanRemoteStorageShard {
952 : pub tenant_shard_id: TenantShardId,
953 : pub generation: Option<u32>,
954 : }
955 :
956 0 : #[derive(Serialize, Deserialize, Debug, Default)]
957 : pub struct TenantScanRemoteStorageResponse {
958 : pub shards: Vec<TenantScanRemoteStorageShard>,
959 : }
960 :
961 0 : #[derive(Serialize, Deserialize, Debug, Clone)]
962 : #[serde(rename_all = "snake_case")]
963 : pub enum TenantSorting {
964 : ResidentSize,
965 : MaxLogicalSize,
966 : }
967 :
968 : impl Default for TenantSorting {
969 0 : fn default() -> Self {
970 0 : Self::ResidentSize
971 0 : }
972 : }
973 :
974 0 : #[derive(Serialize, Deserialize, Debug, Clone)]
975 : pub struct TopTenantShardsRequest {
976 : // How would you like to sort the tenants?
977 : pub order_by: TenantSorting,
978 :
979 : // How many results?
980 : pub limit: usize,
981 :
982 : // Omit tenants with more than this many shards (e.g. if this is the max number of shards
983 : // that the caller would ever split to)
984 : pub where_shards_lt: Option<ShardCount>,
985 :
986 : // Omit tenants where the ordering metric is less than this (this is an optimization to
987 : // let us quickly exclude numerous tiny shards)
988 : pub where_gt: Option<u64>,
989 : }
990 :
991 0 : #[derive(Serialize, Deserialize, Debug, PartialEq, Eq)]
992 : pub struct TopTenantShardItem {
993 : pub id: TenantShardId,
994 :
995 : /// Total size of layers on local disk for all timelines in this tenant
996 : pub resident_size: u64,
997 :
998 : /// Total size of layers in remote storage for all timelines in this tenant
999 : pub physical_size: u64,
1000 :
1001 : /// The largest logical size of a timeline within this tenant
1002 : pub max_logical_size: u64,
1003 : }
1004 :
1005 0 : #[derive(Serialize, Deserialize, Debug, Default)]
1006 : pub struct TopTenantShardsResponse {
1007 : pub shards: Vec<TopTenantShardItem>,
1008 : }
1009 :
1010 : pub mod virtual_file {
1011 : #[derive(
1012 : Copy,
1013 : Clone,
1014 : PartialEq,
1015 : Eq,
1016 : Hash,
1017 344 : strum_macros::EnumString,
1018 0 : strum_macros::Display,
1019 0 : serde_with::DeserializeFromStr,
1020 : serde_with::SerializeDisplay,
1021 : Debug,
1022 : )]
1023 : #[strum(serialize_all = "kebab-case")]
1024 : pub enum IoEngineKind {
1025 : StdFs,
1026 : #[cfg(target_os = "linux")]
1027 : TokioEpollUring,
1028 : }
1029 : }
1030 :
1031 : // Wrapped in libpq CopyData
1032 : #[derive(PartialEq, Eq, Debug)]
1033 : pub enum PagestreamFeMessage {
1034 : Exists(PagestreamExistsRequest),
1035 : Nblocks(PagestreamNblocksRequest),
1036 : GetPage(PagestreamGetPageRequest),
1037 : DbSize(PagestreamDbSizeRequest),
1038 : GetSlruSegment(PagestreamGetSlruSegmentRequest),
1039 : }
1040 :
1041 : // Wrapped in libpq CopyData
1042 0 : #[derive(strum_macros::EnumProperty)]
1043 : pub enum PagestreamBeMessage {
1044 : Exists(PagestreamExistsResponse),
1045 : Nblocks(PagestreamNblocksResponse),
1046 : GetPage(PagestreamGetPageResponse),
1047 : Error(PagestreamErrorResponse),
1048 : DbSize(PagestreamDbSizeResponse),
1049 : GetSlruSegment(PagestreamGetSlruSegmentResponse),
1050 : }
1051 :
1052 : // Keep in sync with `pagestore_client.h`
1053 : #[repr(u8)]
1054 : enum PagestreamBeMessageTag {
1055 : Exists = 100,
1056 : Nblocks = 101,
1057 : GetPage = 102,
1058 : Error = 103,
1059 : DbSize = 104,
1060 : GetSlruSegment = 105,
1061 : }
1062 : impl TryFrom<u8> for PagestreamBeMessageTag {
1063 : type Error = u8;
1064 0 : fn try_from(value: u8) -> Result<Self, u8> {
1065 0 : match value {
1066 0 : 100 => Ok(PagestreamBeMessageTag::Exists),
1067 0 : 101 => Ok(PagestreamBeMessageTag::Nblocks),
1068 0 : 102 => Ok(PagestreamBeMessageTag::GetPage),
1069 0 : 103 => Ok(PagestreamBeMessageTag::Error),
1070 0 : 104 => Ok(PagestreamBeMessageTag::DbSize),
1071 0 : 105 => Ok(PagestreamBeMessageTag::GetSlruSegment),
1072 0 : _ => Err(value),
1073 : }
1074 0 : }
1075 : }
1076 :
1077 : // In the V2 protocol version, a GetPage request contains two LSN values:
1078 : //
1079 : // request_lsn: Get the page version at this point in time. Lsn::Max is a special value that means
1080 : // "get the latest version present". It's used by the primary server, which knows that no one else
1081 : // is writing WAL. 'not_modified_since' must be set to a proper value even if request_lsn is
1082 : // Lsn::Max. Standby servers use the current replay LSN as the request LSN.
1083 : //
1084 : // not_modified_since: Hint to the pageserver that the client knows that the page has not been
1085 : // modified between 'not_modified_since' and the request LSN. It's always correct to set
1086 : // 'not_modified_since equal' to 'request_lsn' (unless Lsn::Max is used as the 'request_lsn'), but
1087 : // passing an earlier LSN can speed up the request, by allowing the pageserver to process the
1088 : // request without waiting for 'request_lsn' to arrive.
1089 : //
1090 : // The legacy V1 interface contained only one LSN, and a boolean 'latest' flag. The V1 interface was
1091 : // sufficient for the primary; the 'lsn' was equivalent to the 'not_modified_since' value, and
1092 : // 'latest' was set to true. The V2 interface was added because there was no correct way for a
1093 : // standby to request a page at a particular non-latest LSN, and also include the
1094 : // 'not_modified_since' hint. That led to an awkward choice of either using an old LSN in the
1095 : // request, if the standby knows that the page hasn't been modified since, and risk getting an error
1096 : // if that LSN has fallen behind the GC horizon, or requesting the current replay LSN, which could
1097 : // require the pageserver unnecessarily to wait for the WAL to arrive up to that point. The new V2
1098 : // interface allows sending both LSNs, and let the pageserver do the right thing. There is no
1099 : // difference in the responses between V1 and V2.
1100 : //
1101 : // The Request structs below reflect the V2 interface. If V1 is used, the parse function
1102 : // maps the old format requests to the new format.
1103 : //
1104 : #[derive(Clone, Copy)]
1105 : pub enum PagestreamProtocolVersion {
1106 : V1,
1107 : V2,
1108 : }
1109 :
1110 : #[derive(Debug, PartialEq, Eq)]
1111 : pub struct PagestreamExistsRequest {
1112 : pub request_lsn: Lsn,
1113 : pub not_modified_since: Lsn,
1114 : pub rel: RelTag,
1115 : }
1116 :
1117 : #[derive(Debug, PartialEq, Eq)]
1118 : pub struct PagestreamNblocksRequest {
1119 : pub request_lsn: Lsn,
1120 : pub not_modified_since: Lsn,
1121 : pub rel: RelTag,
1122 : }
1123 :
1124 : #[derive(Debug, PartialEq, Eq)]
1125 : pub struct PagestreamGetPageRequest {
1126 : pub request_lsn: Lsn,
1127 : pub not_modified_since: Lsn,
1128 : pub rel: RelTag,
1129 : pub blkno: u32,
1130 : }
1131 :
1132 : #[derive(Debug, PartialEq, Eq)]
1133 : pub struct PagestreamDbSizeRequest {
1134 : pub request_lsn: Lsn,
1135 : pub not_modified_since: Lsn,
1136 : pub dbnode: u32,
1137 : }
1138 :
1139 : #[derive(Debug, PartialEq, Eq)]
1140 : pub struct PagestreamGetSlruSegmentRequest {
1141 : pub request_lsn: Lsn,
1142 : pub not_modified_since: Lsn,
1143 : pub kind: u8,
1144 : pub segno: u32,
1145 : }
1146 :
1147 : #[derive(Debug)]
1148 : pub struct PagestreamExistsResponse {
1149 : pub exists: bool,
1150 : }
1151 :
1152 : #[derive(Debug)]
1153 : pub struct PagestreamNblocksResponse {
1154 : pub n_blocks: u32,
1155 : }
1156 :
1157 : #[derive(Debug)]
1158 : pub struct PagestreamGetPageResponse {
1159 : pub page: Bytes,
1160 : }
1161 :
1162 : #[derive(Debug)]
1163 : pub struct PagestreamGetSlruSegmentResponse {
1164 : pub segment: Bytes,
1165 : }
1166 :
1167 : #[derive(Debug)]
1168 : pub struct PagestreamErrorResponse {
1169 : pub message: String,
1170 : }
1171 :
1172 : #[derive(Debug)]
1173 : pub struct PagestreamDbSizeResponse {
1174 : pub db_size: i64,
1175 : }
1176 :
1177 : // This is a cut-down version of TenantHistorySize from the pageserver crate, omitting fields
1178 : // that require pageserver-internal types. It is sufficient to get the total size.
1179 0 : #[derive(Serialize, Deserialize, Debug)]
1180 : pub struct TenantHistorySize {
1181 : pub id: TenantId,
1182 : /// Size is a mixture of WAL and logical size, so the unit is bytes.
1183 : ///
1184 : /// Will be none if `?inputs_only=true` was given.
1185 : pub size: Option<u64>,
1186 : }
1187 :
1188 : impl PagestreamFeMessage {
1189 : /// Serialize a compute -> pageserver message. This is currently only used in testing
1190 : /// tools. Always uses protocol version 2.
1191 8 : pub fn serialize(&self) -> Bytes {
1192 8 : let mut bytes = BytesMut::new();
1193 8 :
1194 8 : match self {
1195 2 : Self::Exists(req) => {
1196 2 : bytes.put_u8(0);
1197 2 : bytes.put_u64(req.request_lsn.0);
1198 2 : bytes.put_u64(req.not_modified_since.0);
1199 2 : bytes.put_u32(req.rel.spcnode);
1200 2 : bytes.put_u32(req.rel.dbnode);
1201 2 : bytes.put_u32(req.rel.relnode);
1202 2 : bytes.put_u8(req.rel.forknum);
1203 2 : }
1204 :
1205 2 : Self::Nblocks(req) => {
1206 2 : bytes.put_u8(1);
1207 2 : bytes.put_u64(req.request_lsn.0);
1208 2 : bytes.put_u64(req.not_modified_since.0);
1209 2 : bytes.put_u32(req.rel.spcnode);
1210 2 : bytes.put_u32(req.rel.dbnode);
1211 2 : bytes.put_u32(req.rel.relnode);
1212 2 : bytes.put_u8(req.rel.forknum);
1213 2 : }
1214 :
1215 2 : Self::GetPage(req) => {
1216 2 : bytes.put_u8(2);
1217 2 : bytes.put_u64(req.request_lsn.0);
1218 2 : bytes.put_u64(req.not_modified_since.0);
1219 2 : bytes.put_u32(req.rel.spcnode);
1220 2 : bytes.put_u32(req.rel.dbnode);
1221 2 : bytes.put_u32(req.rel.relnode);
1222 2 : bytes.put_u8(req.rel.forknum);
1223 2 : bytes.put_u32(req.blkno);
1224 2 : }
1225 :
1226 2 : Self::DbSize(req) => {
1227 2 : bytes.put_u8(3);
1228 2 : bytes.put_u64(req.request_lsn.0);
1229 2 : bytes.put_u64(req.not_modified_since.0);
1230 2 : bytes.put_u32(req.dbnode);
1231 2 : }
1232 :
1233 0 : Self::GetSlruSegment(req) => {
1234 0 : bytes.put_u8(4);
1235 0 : bytes.put_u64(req.request_lsn.0);
1236 0 : bytes.put_u64(req.not_modified_since.0);
1237 0 : bytes.put_u8(req.kind);
1238 0 : bytes.put_u32(req.segno);
1239 0 : }
1240 : }
1241 :
1242 8 : bytes.into()
1243 8 : }
1244 :
1245 8 : pub fn parse<R: std::io::Read>(
1246 8 : body: &mut R,
1247 8 : protocol_version: PagestreamProtocolVersion,
1248 8 : ) -> anyhow::Result<PagestreamFeMessage> {
1249 : // these correspond to the NeonMessageTag enum in pagestore_client.h
1250 : //
1251 : // TODO: consider using protobuf or serde bincode for less error prone
1252 : // serialization.
1253 8 : let msg_tag = body.read_u8()?;
1254 :
1255 8 : let (request_lsn, not_modified_since) = match protocol_version {
1256 : PagestreamProtocolVersion::V2 => (
1257 8 : Lsn::from(body.read_u64::<BigEndian>()?),
1258 8 : Lsn::from(body.read_u64::<BigEndian>()?),
1259 : ),
1260 : PagestreamProtocolVersion::V1 => {
1261 : // In the old protocol, each message starts with a boolean 'latest' flag,
1262 : // followed by 'lsn'. Convert that to the two LSNs, 'request_lsn' and
1263 : // 'not_modified_since', used in the new protocol version.
1264 0 : let latest = body.read_u8()? != 0;
1265 0 : let request_lsn = Lsn::from(body.read_u64::<BigEndian>()?);
1266 0 : if latest {
1267 0 : (Lsn::MAX, request_lsn) // get latest version
1268 : } else {
1269 0 : (request_lsn, request_lsn) // get version at specified LSN
1270 : }
1271 : }
1272 : };
1273 :
1274 : // The rest of the messages are the same between V1 and V2
1275 8 : match msg_tag {
1276 : 0 => Ok(PagestreamFeMessage::Exists(PagestreamExistsRequest {
1277 2 : request_lsn,
1278 2 : not_modified_since,
1279 2 : rel: RelTag {
1280 2 : spcnode: body.read_u32::<BigEndian>()?,
1281 2 : dbnode: body.read_u32::<BigEndian>()?,
1282 2 : relnode: body.read_u32::<BigEndian>()?,
1283 2 : forknum: body.read_u8()?,
1284 : },
1285 : })),
1286 : 1 => Ok(PagestreamFeMessage::Nblocks(PagestreamNblocksRequest {
1287 2 : request_lsn,
1288 2 : not_modified_since,
1289 2 : rel: RelTag {
1290 2 : spcnode: body.read_u32::<BigEndian>()?,
1291 2 : dbnode: body.read_u32::<BigEndian>()?,
1292 2 : relnode: body.read_u32::<BigEndian>()?,
1293 2 : forknum: body.read_u8()?,
1294 : },
1295 : })),
1296 : 2 => Ok(PagestreamFeMessage::GetPage(PagestreamGetPageRequest {
1297 2 : request_lsn,
1298 2 : not_modified_since,
1299 2 : rel: RelTag {
1300 2 : spcnode: body.read_u32::<BigEndian>()?,
1301 2 : dbnode: body.read_u32::<BigEndian>()?,
1302 2 : relnode: body.read_u32::<BigEndian>()?,
1303 2 : forknum: body.read_u8()?,
1304 : },
1305 2 : blkno: body.read_u32::<BigEndian>()?,
1306 : })),
1307 : 3 => Ok(PagestreamFeMessage::DbSize(PagestreamDbSizeRequest {
1308 2 : request_lsn,
1309 2 : not_modified_since,
1310 2 : dbnode: body.read_u32::<BigEndian>()?,
1311 : })),
1312 : 4 => Ok(PagestreamFeMessage::GetSlruSegment(
1313 : PagestreamGetSlruSegmentRequest {
1314 0 : request_lsn,
1315 0 : not_modified_since,
1316 0 : kind: body.read_u8()?,
1317 0 : segno: body.read_u32::<BigEndian>()?,
1318 : },
1319 : )),
1320 0 : _ => bail!("unknown smgr message tag: {:?}", msg_tag),
1321 : }
1322 8 : }
1323 : }
1324 :
1325 : impl PagestreamBeMessage {
1326 0 : pub fn serialize(&self) -> Bytes {
1327 0 : let mut bytes = BytesMut::new();
1328 0 :
1329 0 : use PagestreamBeMessageTag as Tag;
1330 0 : match self {
1331 0 : Self::Exists(resp) => {
1332 0 : bytes.put_u8(Tag::Exists as u8);
1333 0 : bytes.put_u8(resp.exists as u8);
1334 0 : }
1335 :
1336 0 : Self::Nblocks(resp) => {
1337 0 : bytes.put_u8(Tag::Nblocks as u8);
1338 0 : bytes.put_u32(resp.n_blocks);
1339 0 : }
1340 :
1341 0 : Self::GetPage(resp) => {
1342 0 : bytes.put_u8(Tag::GetPage as u8);
1343 0 : bytes.put(&resp.page[..]);
1344 0 : }
1345 :
1346 0 : Self::Error(resp) => {
1347 0 : bytes.put_u8(Tag::Error as u8);
1348 0 : bytes.put(resp.message.as_bytes());
1349 0 : bytes.put_u8(0); // null terminator
1350 0 : }
1351 0 : Self::DbSize(resp) => {
1352 0 : bytes.put_u8(Tag::DbSize as u8);
1353 0 : bytes.put_i64(resp.db_size);
1354 0 : }
1355 :
1356 0 : Self::GetSlruSegment(resp) => {
1357 0 : bytes.put_u8(Tag::GetSlruSegment as u8);
1358 0 : bytes.put_u32((resp.segment.len() / BLCKSZ as usize) as u32);
1359 0 : bytes.put(&resp.segment[..]);
1360 0 : }
1361 : }
1362 :
1363 0 : bytes.into()
1364 0 : }
1365 :
1366 0 : pub fn deserialize(buf: Bytes) -> anyhow::Result<Self> {
1367 0 : let mut buf = buf.reader();
1368 0 : let msg_tag = buf.read_u8()?;
1369 :
1370 : use PagestreamBeMessageTag as Tag;
1371 0 : let ok =
1372 0 : match Tag::try_from(msg_tag).map_err(|tag: u8| anyhow::anyhow!("invalid tag {tag}"))? {
1373 : Tag::Exists => {
1374 0 : let exists = buf.read_u8()?;
1375 0 : Self::Exists(PagestreamExistsResponse {
1376 0 : exists: exists != 0,
1377 0 : })
1378 : }
1379 : Tag::Nblocks => {
1380 0 : let n_blocks = buf.read_u32::<BigEndian>()?;
1381 0 : Self::Nblocks(PagestreamNblocksResponse { n_blocks })
1382 : }
1383 : Tag::GetPage => {
1384 0 : let mut page = vec![0; 8192]; // TODO: use MaybeUninit
1385 0 : buf.read_exact(&mut page)?;
1386 0 : PagestreamBeMessage::GetPage(PagestreamGetPageResponse { page: page.into() })
1387 : }
1388 : Tag::Error => {
1389 0 : let mut msg = Vec::new();
1390 0 : buf.read_until(0, &mut msg)?;
1391 0 : let cstring = std::ffi::CString::from_vec_with_nul(msg)?;
1392 0 : let rust_str = cstring.to_str()?;
1393 0 : PagestreamBeMessage::Error(PagestreamErrorResponse {
1394 0 : message: rust_str.to_owned(),
1395 0 : })
1396 : }
1397 : Tag::DbSize => {
1398 0 : let db_size = buf.read_i64::<BigEndian>()?;
1399 0 : Self::DbSize(PagestreamDbSizeResponse { db_size })
1400 : }
1401 : Tag::GetSlruSegment => {
1402 0 : let n_blocks = buf.read_u32::<BigEndian>()?;
1403 0 : let mut segment = vec![0; n_blocks as usize * BLCKSZ as usize];
1404 0 : buf.read_exact(&mut segment)?;
1405 0 : Self::GetSlruSegment(PagestreamGetSlruSegmentResponse {
1406 0 : segment: segment.into(),
1407 0 : })
1408 : }
1409 : };
1410 0 : let remaining = buf.into_inner();
1411 0 : if !remaining.is_empty() {
1412 0 : anyhow::bail!(
1413 0 : "remaining bytes in msg with tag={msg_tag}: {}",
1414 0 : remaining.len()
1415 0 : );
1416 0 : }
1417 0 : Ok(ok)
1418 0 : }
1419 :
1420 0 : pub fn kind(&self) -> &'static str {
1421 0 : match self {
1422 0 : Self::Exists(_) => "Exists",
1423 0 : Self::Nblocks(_) => "Nblocks",
1424 0 : Self::GetPage(_) => "GetPage",
1425 0 : Self::Error(_) => "Error",
1426 0 : Self::DbSize(_) => "DbSize",
1427 0 : Self::GetSlruSegment(_) => "GetSlruSegment",
1428 : }
1429 0 : }
1430 : }
1431 :
1432 : #[cfg(test)]
1433 : mod tests {
1434 : use serde_json::json;
1435 : use std::str::FromStr;
1436 :
1437 : use super::*;
1438 :
1439 : #[test]
1440 2 : fn test_pagestream() {
1441 2 : // Test serialization/deserialization of PagestreamFeMessage
1442 2 : let messages = vec![
1443 2 : PagestreamFeMessage::Exists(PagestreamExistsRequest {
1444 2 : request_lsn: Lsn(4),
1445 2 : not_modified_since: Lsn(3),
1446 2 : rel: RelTag {
1447 2 : forknum: 1,
1448 2 : spcnode: 2,
1449 2 : dbnode: 3,
1450 2 : relnode: 4,
1451 2 : },
1452 2 : }),
1453 2 : PagestreamFeMessage::Nblocks(PagestreamNblocksRequest {
1454 2 : request_lsn: Lsn(4),
1455 2 : not_modified_since: Lsn(4),
1456 2 : rel: RelTag {
1457 2 : forknum: 1,
1458 2 : spcnode: 2,
1459 2 : dbnode: 3,
1460 2 : relnode: 4,
1461 2 : },
1462 2 : }),
1463 2 : PagestreamFeMessage::GetPage(PagestreamGetPageRequest {
1464 2 : request_lsn: Lsn(4),
1465 2 : not_modified_since: Lsn(3),
1466 2 : rel: RelTag {
1467 2 : forknum: 1,
1468 2 : spcnode: 2,
1469 2 : dbnode: 3,
1470 2 : relnode: 4,
1471 2 : },
1472 2 : blkno: 7,
1473 2 : }),
1474 2 : PagestreamFeMessage::DbSize(PagestreamDbSizeRequest {
1475 2 : request_lsn: Lsn(4),
1476 2 : not_modified_since: Lsn(3),
1477 2 : dbnode: 7,
1478 2 : }),
1479 2 : ];
1480 10 : for msg in messages {
1481 8 : let bytes = msg.serialize();
1482 8 : let reconstructed =
1483 8 : PagestreamFeMessage::parse(&mut bytes.reader(), PagestreamProtocolVersion::V2)
1484 8 : .unwrap();
1485 8 : assert!(msg == reconstructed);
1486 : }
1487 2 : }
1488 :
1489 : #[test]
1490 2 : fn test_tenantinfo_serde() {
1491 2 : // Test serialization/deserialization of TenantInfo
1492 2 : let original_active = TenantInfo {
1493 2 : id: TenantShardId::unsharded(TenantId::generate()),
1494 2 : state: TenantState::Active,
1495 2 : current_physical_size: Some(42),
1496 2 : attachment_status: TenantAttachmentStatus::Attached,
1497 2 : generation: None,
1498 2 : };
1499 2 : let expected_active = json!({
1500 2 : "id": original_active.id.to_string(),
1501 2 : "state": {
1502 2 : "slug": "Active",
1503 2 : },
1504 2 : "current_physical_size": 42,
1505 2 : "attachment_status": {
1506 2 : "slug":"attached",
1507 2 : }
1508 2 : });
1509 2 :
1510 2 : let original_broken = TenantInfo {
1511 2 : id: TenantShardId::unsharded(TenantId::generate()),
1512 2 : state: TenantState::Broken {
1513 2 : reason: "reason".into(),
1514 2 : backtrace: "backtrace info".into(),
1515 2 : },
1516 2 : current_physical_size: Some(42),
1517 2 : attachment_status: TenantAttachmentStatus::Attached,
1518 2 : generation: None,
1519 2 : };
1520 2 : let expected_broken = json!({
1521 2 : "id": original_broken.id.to_string(),
1522 2 : "state": {
1523 2 : "slug": "Broken",
1524 2 : "data": {
1525 2 : "backtrace": "backtrace info",
1526 2 : "reason": "reason",
1527 2 : }
1528 2 : },
1529 2 : "current_physical_size": 42,
1530 2 : "attachment_status": {
1531 2 : "slug":"attached",
1532 2 : }
1533 2 : });
1534 2 :
1535 2 : assert_eq!(
1536 2 : serde_json::to_value(&original_active).unwrap(),
1537 2 : expected_active
1538 2 : );
1539 :
1540 2 : assert_eq!(
1541 2 : serde_json::to_value(&original_broken).unwrap(),
1542 2 : expected_broken
1543 2 : );
1544 2 : assert!(format!("{:?}", &original_broken.state).contains("reason"));
1545 2 : assert!(format!("{:?}", &original_broken.state).contains("backtrace info"));
1546 2 : }
1547 :
1548 : #[test]
1549 2 : fn test_reject_unknown_field() {
1550 2 : let id = TenantId::generate();
1551 2 : let create_request = json!({
1552 2 : "new_tenant_id": id.to_string(),
1553 2 : "unknown_field": "unknown_value".to_string(),
1554 2 : });
1555 2 : let err = serde_json::from_value::<TenantCreateRequest>(create_request).unwrap_err();
1556 2 : assert!(
1557 2 : err.to_string().contains("unknown field `unknown_field`"),
1558 0 : "expect unknown field `unknown_field` error, got: {}",
1559 : err
1560 : );
1561 :
1562 2 : let id = TenantId::generate();
1563 2 : let config_request = json!({
1564 2 : "tenant_id": id.to_string(),
1565 2 : "unknown_field": "unknown_value".to_string(),
1566 2 : });
1567 2 : let err = serde_json::from_value::<TenantConfigRequest>(config_request).unwrap_err();
1568 2 : assert!(
1569 2 : err.to_string().contains("unknown field `unknown_field`"),
1570 0 : "expect unknown field `unknown_field` error, got: {}",
1571 : err
1572 : );
1573 :
1574 2 : let attach_request = json!({
1575 2 : "config": {
1576 2 : "unknown_field": "unknown_value".to_string(),
1577 2 : },
1578 2 : });
1579 2 : let err = serde_json::from_value::<TenantAttachRequest>(attach_request).unwrap_err();
1580 2 : assert!(
1581 2 : err.to_string().contains("unknown field `unknown_field`"),
1582 0 : "expect unknown field `unknown_field` error, got: {}",
1583 : err
1584 : );
1585 2 : }
1586 :
1587 : #[test]
1588 2 : fn tenantstatus_activating_serde() {
1589 2 : let states = [
1590 2 : TenantState::Activating(ActivatingFrom::Loading),
1591 2 : TenantState::Activating(ActivatingFrom::Attaching),
1592 2 : ];
1593 2 : let expected = "[{\"slug\":\"Activating\",\"data\":\"Loading\"},{\"slug\":\"Activating\",\"data\":\"Attaching\"}]";
1594 2 :
1595 2 : let actual = serde_json::to_string(&states).unwrap();
1596 2 :
1597 2 : assert_eq!(actual, expected);
1598 :
1599 2 : let parsed = serde_json::from_str::<Vec<TenantState>>(&actual).unwrap();
1600 2 :
1601 2 : assert_eq!(states.as_slice(), &parsed);
1602 2 : }
1603 :
1604 : #[test]
1605 2 : fn tenantstatus_activating_strum() {
1606 2 : // tests added, because we use these for metrics
1607 2 : let examples = [
1608 2 : (line!(), TenantState::Loading, "Loading"),
1609 2 : (line!(), TenantState::Attaching, "Attaching"),
1610 2 : (
1611 2 : line!(),
1612 2 : TenantState::Activating(ActivatingFrom::Loading),
1613 2 : "Activating",
1614 2 : ),
1615 2 : (
1616 2 : line!(),
1617 2 : TenantState::Activating(ActivatingFrom::Attaching),
1618 2 : "Activating",
1619 2 : ),
1620 2 : (line!(), TenantState::Active, "Active"),
1621 2 : (
1622 2 : line!(),
1623 2 : TenantState::Stopping {
1624 2 : progress: utils::completion::Barrier::default(),
1625 2 : },
1626 2 : "Stopping",
1627 2 : ),
1628 2 : (
1629 2 : line!(),
1630 2 : TenantState::Broken {
1631 2 : reason: "Example".into(),
1632 2 : backtrace: "Looooong backtrace".into(),
1633 2 : },
1634 2 : "Broken",
1635 2 : ),
1636 2 : ];
1637 :
1638 16 : for (line, rendered, expected) in examples {
1639 14 : let actual: &'static str = rendered.into();
1640 14 : assert_eq!(actual, expected, "example on {line}");
1641 : }
1642 2 : }
1643 :
1644 : #[test]
1645 2 : fn test_aux_file_migration_path() {
1646 2 : assert!(AuxFilePolicy::is_valid_migration_path(
1647 2 : None,
1648 2 : AuxFilePolicy::V1
1649 2 : ));
1650 2 : assert!(AuxFilePolicy::is_valid_migration_path(
1651 2 : None,
1652 2 : AuxFilePolicy::V2
1653 2 : ));
1654 2 : assert!(AuxFilePolicy::is_valid_migration_path(
1655 2 : None,
1656 2 : AuxFilePolicy::CrossValidation
1657 2 : ));
1658 : // Self-migration is not a valid migration path, and the caller should handle it by itself.
1659 2 : assert!(!AuxFilePolicy::is_valid_migration_path(
1660 2 : Some(AuxFilePolicy::V1),
1661 2 : AuxFilePolicy::V1
1662 2 : ));
1663 2 : assert!(!AuxFilePolicy::is_valid_migration_path(
1664 2 : Some(AuxFilePolicy::V2),
1665 2 : AuxFilePolicy::V2
1666 2 : ));
1667 2 : assert!(!AuxFilePolicy::is_valid_migration_path(
1668 2 : Some(AuxFilePolicy::CrossValidation),
1669 2 : AuxFilePolicy::CrossValidation
1670 2 : ));
1671 : // Migrations not allowed
1672 2 : assert!(!AuxFilePolicy::is_valid_migration_path(
1673 2 : Some(AuxFilePolicy::CrossValidation),
1674 2 : AuxFilePolicy::V1
1675 2 : ));
1676 2 : assert!(!AuxFilePolicy::is_valid_migration_path(
1677 2 : Some(AuxFilePolicy::V1),
1678 2 : AuxFilePolicy::V2
1679 2 : ));
1680 2 : assert!(!AuxFilePolicy::is_valid_migration_path(
1681 2 : Some(AuxFilePolicy::V2),
1682 2 : AuxFilePolicy::V1
1683 2 : ));
1684 2 : assert!(!AuxFilePolicy::is_valid_migration_path(
1685 2 : Some(AuxFilePolicy::V2),
1686 2 : AuxFilePolicy::CrossValidation
1687 2 : ));
1688 2 : assert!(!AuxFilePolicy::is_valid_migration_path(
1689 2 : Some(AuxFilePolicy::V1),
1690 2 : AuxFilePolicy::CrossValidation
1691 2 : ));
1692 : // Migrations allowed
1693 2 : assert!(AuxFilePolicy::is_valid_migration_path(
1694 2 : Some(AuxFilePolicy::CrossValidation),
1695 2 : AuxFilePolicy::V2
1696 2 : ));
1697 2 : }
1698 :
1699 : #[test]
1700 2 : fn test_aux_parse() {
1701 2 : assert_eq!(AuxFilePolicy::from_str("V2").unwrap(), AuxFilePolicy::V2);
1702 2 : assert_eq!(AuxFilePolicy::from_str("v2").unwrap(), AuxFilePolicy::V2);
1703 2 : assert_eq!(
1704 2 : AuxFilePolicy::from_str("cross-validation").unwrap(),
1705 2 : AuxFilePolicy::CrossValidation
1706 2 : );
1707 2 : }
1708 : }
|
