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 : collections::HashMap,
9 : fmt::Display,
10 : io::{BufRead, Read},
11 : num::{NonZeroU32, NonZeroU64, NonZeroUsize},
12 : str::FromStr,
13 : sync::atomic::AtomicUsize,
14 : time::{Duration, SystemTime},
15 : };
16 :
17 : use byteorder::{BigEndian, ReadBytesExt};
18 : use postgres_ffi::BLCKSZ;
19 : use serde::{Deserialize, Serialize};
20 : use serde_with::serde_as;
21 : use utils::{
22 : completion,
23 : id::{NodeId, TenantId, TimelineId},
24 : lsn::Lsn,
25 : serde_system_time,
26 : };
27 :
28 : use crate::{
29 : reltag::RelTag,
30 : shard::{ShardCount, ShardStripeSize, TenantShardId},
31 : };
32 : use anyhow::bail;
33 : use bytes::{Buf, BufMut, Bytes, BytesMut};
34 :
35 : /// The state of a tenant in this pageserver.
36 : ///
37 : /// ```mermaid
38 : /// stateDiagram-v2
39 : ///
40 : /// [*] --> Attaching: spawn_attach()
41 : ///
42 : /// Attaching --> Activating: activate()
43 : /// Activating --> Active: infallible
44 : ///
45 : /// Attaching --> Broken: attach() failure
46 : ///
47 : /// Active --> Stopping: set_stopping(), part of shutdown & detach
48 : /// Stopping --> Broken: late error in remove_tenant_from_memory
49 : ///
50 : /// Broken --> [*]: ignore / detach / shutdown
51 : /// Stopping --> [*]: remove_from_memory complete
52 : ///
53 : /// Active --> Broken: cfg(testing)-only tenant break point
54 : /// ```
55 : #[derive(
56 : Clone,
57 : PartialEq,
58 : Eq,
59 1 : serde::Serialize,
60 3 : serde::Deserialize,
61 0 : strum_macros::Display,
62 : strum_macros::VariantNames,
63 0 : strum_macros::AsRefStr,
64 375 : strum_macros::IntoStaticStr,
65 : )]
66 : #[serde(tag = "slug", content = "data")]
67 : pub enum TenantState {
68 : /// This tenant is being attached to the pageserver.
69 : ///
70 : /// `set_stopping()` and `set_broken()` do not work in this state and wait for it to pass.
71 : Attaching,
72 : /// The tenant is transitioning from Loading/Attaching to Active.
73 : ///
74 : /// While in this state, the individual timelines are being activated.
75 : ///
76 : /// `set_stopping()` and `set_broken()` do not work in this state and wait for it to pass.
77 : Activating(ActivatingFrom),
78 : /// The tenant has finished activating and is open for business.
79 : ///
80 : /// Transitions out of this state are possible through `set_stopping()` and `set_broken()`.
81 : Active,
82 : /// The tenant is recognized by pageserver, but it is being detached or the
83 : /// system is being shut down.
84 : ///
85 : /// Transitions out of this state are possible through `set_broken()`.
86 : Stopping {
87 : // Because of https://github.com/serde-rs/serde/issues/2105 this has to be a named field,
88 : // otherwise it will not be skipped during deserialization
89 : #[serde(skip)]
90 : progress: completion::Barrier,
91 : },
92 : /// The tenant is recognized by the pageserver, but can no longer be used for
93 : /// any operations.
94 : ///
95 : /// If the tenant fails to load or attach, it will transition to this state
96 : /// and it is guaranteed that no background tasks are running in its name.
97 : ///
98 : /// The other way to transition into this state is from `Stopping` state
99 : /// through `set_broken()` called from `remove_tenant_from_memory()`. That happens
100 : /// if the cleanup future executed by `remove_tenant_from_memory()` fails.
101 : Broken { reason: String, backtrace: String },
102 : }
103 :
104 : impl TenantState {
105 0 : pub fn attachment_status(&self) -> TenantAttachmentStatus {
106 : use TenantAttachmentStatus::*;
107 :
108 : // Below TenantState::Activating is used as "transient" or "transparent" state for
109 : // attachment_status determining.
110 0 : match self {
111 : // The attach procedure writes the marker file before adding the Attaching tenant to the tenants map.
112 : // So, technically, we can return Attached here.
113 : // However, as soon as Console observes Attached, it will proceed with the Postgres-level health check.
114 : // But, our attach task might still be fetching the remote timelines, etc.
115 : // So, return `Maybe` while Attaching, making Console wait for the attach task to finish.
116 0 : Self::Attaching | Self::Activating(ActivatingFrom::Attaching) => Maybe,
117 : // We only reach Active after successful load / attach.
118 : // So, call atttachment status Attached.
119 0 : Self::Active => Attached,
120 : // If the (initial or resumed) attach procedure fails, the tenant becomes Broken.
121 : // However, it also becomes Broken if the regular load fails.
122 : // From Console's perspective there's no practical difference
123 : // because attachment_status is polled by console only during attach operation execution.
124 0 : Self::Broken { reason, .. } => Failed {
125 0 : reason: reason.to_owned(),
126 0 : },
127 : // Why is Stopping a Maybe case? Because, during pageserver shutdown,
128 : // we set the Stopping state irrespective of whether the tenant
129 : // has finished attaching or not.
130 0 : Self::Stopping { .. } => Maybe,
131 : }
132 0 : }
133 :
134 0 : pub fn broken_from_reason(reason: String) -> Self {
135 0 : let backtrace_str: String = format!("{}", std::backtrace::Backtrace::force_capture());
136 0 : Self::Broken {
137 0 : reason,
138 0 : backtrace: backtrace_str,
139 0 : }
140 0 : }
141 : }
142 :
143 : impl std::fmt::Debug for TenantState {
144 2 : fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
145 2 : match self {
146 2 : Self::Broken { reason, backtrace } if !reason.is_empty() => {
147 2 : write!(f, "Broken due to: {reason}. Backtrace:\n{backtrace}")
148 : }
149 0 : _ => write!(f, "{self}"),
150 : }
151 2 : }
152 : }
153 :
154 : /// A temporary lease to a specific lsn inside a timeline.
155 : /// Access to the lsn is guaranteed by the pageserver until the expiration indicated by `valid_until`.
156 : #[serde_as]
157 0 : #[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
158 : pub struct LsnLease {
159 : #[serde_as(as = "SystemTimeAsRfc3339Millis")]
160 : pub valid_until: SystemTime,
161 : }
162 :
163 : serde_with::serde_conv!(
164 : SystemTimeAsRfc3339Millis,
165 : SystemTime,
166 0 : |time: &SystemTime| humantime::format_rfc3339_millis(*time).to_string(),
167 0 : |value: String| -> Result<_, humantime::TimestampError> { humantime::parse_rfc3339(&value) }
168 : );
169 :
170 : impl LsnLease {
171 : /// The default length for an explicit LSN lease request (10 minutes).
172 : pub const DEFAULT_LENGTH: Duration = Duration::from_secs(10 * 60);
173 :
174 : /// The default length for an implicit LSN lease granted during
175 : /// `get_lsn_by_timestamp` request (1 minutes).
176 : pub const DEFAULT_LENGTH_FOR_TS: Duration = Duration::from_secs(60);
177 :
178 : /// Checks whether the lease is expired.
179 6 : pub fn is_expired(&self, now: &SystemTime) -> bool {
180 6 : now > &self.valid_until
181 6 : }
182 : }
183 :
184 : /// The only [`TenantState`] variants we could be `TenantState::Activating` from.
185 : ///
186 : /// XXX: We used to have more variants here, but now it's just one, which makes this rather
187 : /// useless. Remove, once we've checked that there's no client code left that looks at this.
188 2 : #[derive(Clone, Copy, Debug, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
189 : pub enum ActivatingFrom {
190 : /// Arrived to [`TenantState::Activating`] from [`TenantState::Attaching`]
191 : Attaching,
192 : }
193 :
194 : /// A state of a timeline in pageserver's memory.
195 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
196 : pub enum TimelineState {
197 : /// The timeline is recognized by the pageserver but is not yet operational.
198 : /// In particular, the walreceiver connection loop is not running for this timeline.
199 : /// It will eventually transition to state Active or Broken.
200 : Loading,
201 : /// The timeline is fully operational.
202 : /// It can be queried, and the walreceiver connection loop is running.
203 : Active,
204 : /// The timeline was previously Loading or Active but is shutting down.
205 : /// It cannot transition back into any other state.
206 : Stopping,
207 : /// The timeline is broken and not operational (previous states: Loading or Active).
208 : Broken { reason: String, backtrace: String },
209 : }
210 :
211 0 : #[derive(Serialize, Deserialize, Clone)]
212 : pub struct TimelineCreateRequest {
213 : pub new_timeline_id: TimelineId,
214 : #[serde(default)]
215 : pub ancestor_timeline_id: Option<TimelineId>,
216 : #[serde(default)]
217 : pub existing_initdb_timeline_id: Option<TimelineId>,
218 : #[serde(default)]
219 : pub ancestor_start_lsn: Option<Lsn>,
220 : pub pg_version: Option<u32>,
221 : }
222 :
223 0 : #[derive(Serialize, Deserialize, Clone)]
224 : pub struct LsnLeaseRequest {
225 : pub lsn: Lsn,
226 : }
227 :
228 0 : #[derive(Serialize, Deserialize)]
229 : pub struct TenantShardSplitRequest {
230 : pub new_shard_count: u8,
231 :
232 : // A tenant's stripe size is only meaningful the first time their shard count goes
233 : // above 1: therefore during a split from 1->N shards, we may modify the stripe size.
234 : //
235 : // If this is set while the stripe count is being increased from an already >1 value,
236 : // then the request will fail with 400.
237 : pub new_stripe_size: Option<ShardStripeSize>,
238 : }
239 :
240 0 : #[derive(Serialize, Deserialize)]
241 : pub struct TenantShardSplitResponse {
242 : pub new_shards: Vec<TenantShardId>,
243 : }
244 :
245 : /// Parameters that apply to all shards in a tenant. Used during tenant creation.
246 0 : #[derive(Serialize, Deserialize, Debug)]
247 : #[serde(deny_unknown_fields)]
248 : pub struct ShardParameters {
249 : pub count: ShardCount,
250 : pub stripe_size: ShardStripeSize,
251 : }
252 :
253 : impl ShardParameters {
254 : pub const DEFAULT_STRIPE_SIZE: ShardStripeSize = ShardStripeSize(256 * 1024 / 8);
255 :
256 0 : pub fn is_unsharded(&self) -> bool {
257 0 : self.count.is_unsharded()
258 0 : }
259 : }
260 :
261 : impl Default for ShardParameters {
262 187 : fn default() -> Self {
263 187 : Self {
264 187 : count: ShardCount::new(0),
265 187 : stripe_size: Self::DEFAULT_STRIPE_SIZE,
266 187 : }
267 187 : }
268 : }
269 :
270 : /// An alternative representation of `pageserver::tenant::TenantConf` with
271 : /// simpler types.
272 2 : #[derive(Serialize, Deserialize, Debug, Default, Clone, Eq, PartialEq)]
273 : pub struct TenantConfig {
274 : pub checkpoint_distance: Option<u64>,
275 : pub checkpoint_timeout: Option<String>,
276 : pub compaction_target_size: Option<u64>,
277 : pub compaction_period: Option<String>,
278 : pub compaction_threshold: Option<usize>,
279 : // defer parsing compaction_algorithm, like eviction_policy
280 : pub compaction_algorithm: Option<CompactionAlgorithmSettings>,
281 : pub gc_horizon: Option<u64>,
282 : pub gc_period: Option<String>,
283 : pub image_creation_threshold: Option<usize>,
284 : pub pitr_interval: Option<String>,
285 : pub walreceiver_connect_timeout: Option<String>,
286 : pub lagging_wal_timeout: Option<String>,
287 : pub max_lsn_wal_lag: Option<NonZeroU64>,
288 : pub eviction_policy: Option<EvictionPolicy>,
289 : pub min_resident_size_override: Option<u64>,
290 : pub evictions_low_residence_duration_metric_threshold: Option<String>,
291 : pub heatmap_period: Option<String>,
292 : pub lazy_slru_download: Option<bool>,
293 : pub timeline_get_throttle: Option<ThrottleConfig>,
294 : pub image_layer_creation_check_threshold: Option<u8>,
295 : pub switch_aux_file_policy: Option<AuxFilePolicy>,
296 : pub lsn_lease_length: Option<String>,
297 : pub lsn_lease_length_for_ts: Option<String>,
298 : }
299 :
300 : /// The policy for the aux file storage.
301 : ///
302 : /// It can be switched through `switch_aux_file_policy` tenant config.
303 : /// When the first aux file written, the policy will be persisted in the
304 : /// `index_part.json` file and has a limited migration path.
305 : ///
306 : /// Currently, we only allow the following migration path:
307 : ///
308 : /// Unset -> V1
309 : /// -> V2
310 : /// -> CrossValidation -> V2
311 : #[derive(
312 : Eq,
313 : PartialEq,
314 : Debug,
315 : Copy,
316 : Clone,
317 5 : strum_macros::EnumString,
318 0 : strum_macros::Display,
319 0 : serde_with::DeserializeFromStr,
320 : serde_with::SerializeDisplay,
321 : )]
322 : #[strum(serialize_all = "kebab-case")]
323 : pub enum AuxFilePolicy {
324 : /// V1 aux file policy: store everything in AUX_FILE_KEY
325 : #[strum(ascii_case_insensitive)]
326 : V1,
327 : /// V2 aux file policy: store in the AUX_FILE keyspace
328 : #[strum(ascii_case_insensitive)]
329 : V2,
330 : /// Cross validation runs both formats on the write path and does validation
331 : /// on the read path.
332 : #[strum(ascii_case_insensitive)]
333 : CrossValidation,
334 : }
335 :
336 : impl AuxFilePolicy {
337 12 : pub fn is_valid_migration_path(from: Option<Self>, to: Self) -> bool {
338 8 : matches!(
339 12 : (from, to),
340 : (None, _) | (Some(AuxFilePolicy::CrossValidation), AuxFilePolicy::V2)
341 : )
342 12 : }
343 :
344 : /// If a tenant writes aux files without setting `switch_aux_policy`, this value will be used.
345 382 : pub fn default_tenant_config() -> Self {
346 382 : Self::V2
347 382 : }
348 : }
349 :
350 : /// The aux file policy memory flag. Users can store `Option<AuxFilePolicy>` into this atomic flag. 0 == unspecified.
351 : pub struct AtomicAuxFilePolicy(AtomicUsize);
352 :
353 : impl AtomicAuxFilePolicy {
354 0 : pub fn new(policy: Option<AuxFilePolicy>) -> Self {
355 0 : Self(AtomicUsize::new(
356 0 : policy.map(AuxFilePolicy::to_usize).unwrap_or_default(),
357 0 : ))
358 0 : }
359 :
360 0 : pub fn load(&self) -> Option<AuxFilePolicy> {
361 0 : match self.0.load(std::sync::atomic::Ordering::Acquire) {
362 0 : 0 => None,
363 0 : other => Some(AuxFilePolicy::from_usize(other)),
364 : }
365 0 : }
366 :
367 0 : pub fn store(&self, policy: Option<AuxFilePolicy>) {
368 0 : self.0.store(
369 0 : policy.map(AuxFilePolicy::to_usize).unwrap_or_default(),
370 0 : std::sync::atomic::Ordering::Release,
371 0 : );
372 0 : }
373 : }
374 :
375 : impl AuxFilePolicy {
376 0 : pub fn to_usize(self) -> usize {
377 0 : match self {
378 0 : Self::V1 => 1,
379 0 : Self::CrossValidation => 2,
380 0 : Self::V2 => 3,
381 : }
382 0 : }
383 :
384 0 : pub fn try_from_usize(this: usize) -> Option<Self> {
385 0 : match this {
386 0 : 1 => Some(Self::V1),
387 0 : 2 => Some(Self::CrossValidation),
388 0 : 3 => Some(Self::V2),
389 0 : _ => None,
390 : }
391 0 : }
392 :
393 0 : pub fn from_usize(this: usize) -> Self {
394 0 : Self::try_from_usize(this).unwrap()
395 0 : }
396 : }
397 :
398 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
399 : #[serde(tag = "kind")]
400 : pub enum EvictionPolicy {
401 : NoEviction,
402 : LayerAccessThreshold(EvictionPolicyLayerAccessThreshold),
403 : OnlyImitiate(EvictionPolicyLayerAccessThreshold),
404 : }
405 :
406 : impl EvictionPolicy {
407 0 : pub fn discriminant_str(&self) -> &'static str {
408 0 : match self {
409 0 : EvictionPolicy::NoEviction => "NoEviction",
410 0 : EvictionPolicy::LayerAccessThreshold(_) => "LayerAccessThreshold",
411 0 : EvictionPolicy::OnlyImitiate(_) => "OnlyImitiate",
412 : }
413 0 : }
414 : }
415 :
416 : #[derive(
417 : Eq,
418 : PartialEq,
419 : Debug,
420 : Copy,
421 : Clone,
422 0 : strum_macros::EnumString,
423 0 : strum_macros::Display,
424 0 : serde_with::DeserializeFromStr,
425 : serde_with::SerializeDisplay,
426 : )]
427 : #[strum(serialize_all = "kebab-case")]
428 : pub enum CompactionAlgorithm {
429 : Legacy,
430 : Tiered,
431 : }
432 :
433 : #[derive(
434 0 : Debug, Clone, Copy, PartialEq, Eq, serde_with::DeserializeFromStr, serde_with::SerializeDisplay,
435 : )]
436 : pub enum ImageCompressionAlgorithm {
437 : // Disabled for writes, support decompressing during read path
438 : Disabled,
439 : /// Zstandard compression. Level 0 means and None mean the same (default level). Levels can be negative as well.
440 : /// For details, see the [manual](http://facebook.github.io/zstd/zstd_manual.html).
441 : Zstd {
442 : level: Option<i8>,
443 : },
444 : }
445 :
446 : impl FromStr for ImageCompressionAlgorithm {
447 : type Err = anyhow::Error;
448 8 : fn from_str(s: &str) -> Result<Self, Self::Err> {
449 8 : let mut components = s.split(['(', ')']);
450 8 : let first = components
451 8 : .next()
452 8 : .ok_or_else(|| anyhow::anyhow!("empty string"))?;
453 8 : match first {
454 8 : "disabled" => Ok(ImageCompressionAlgorithm::Disabled),
455 6 : "zstd" => {
456 6 : let level = if let Some(v) = components.next() {
457 4 : let v: i8 = v.parse()?;
458 4 : Some(v)
459 : } else {
460 2 : None
461 : };
462 :
463 6 : Ok(ImageCompressionAlgorithm::Zstd { level })
464 : }
465 0 : _ => anyhow::bail!("invalid specifier '{first}'"),
466 : }
467 8 : }
468 : }
469 :
470 : impl Display for ImageCompressionAlgorithm {
471 12 : fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
472 12 : match self {
473 3 : ImageCompressionAlgorithm::Disabled => write!(f, "disabled"),
474 9 : ImageCompressionAlgorithm::Zstd { level } => {
475 9 : if let Some(level) = level {
476 6 : write!(f, "zstd({})", level)
477 : } else {
478 3 : write!(f, "zstd")
479 : }
480 : }
481 : }
482 12 : }
483 : }
484 :
485 0 : #[derive(Eq, PartialEq, Debug, Clone, Serialize, Deserialize)]
486 : pub struct CompactionAlgorithmSettings {
487 : pub kind: CompactionAlgorithm,
488 : }
489 :
490 6 : #[derive(Debug, PartialEq, Eq, Clone, Deserialize, Serialize)]
491 : #[serde(tag = "mode", rename_all = "kebab-case", deny_unknown_fields)]
492 : pub enum L0FlushConfig {
493 : #[serde(rename_all = "snake_case")]
494 : Direct { max_concurrency: NonZeroUsize },
495 : }
496 :
497 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
498 : pub struct EvictionPolicyLayerAccessThreshold {
499 : #[serde(with = "humantime_serde")]
500 : pub period: Duration,
501 : #[serde(with = "humantime_serde")]
502 : pub threshold: Duration,
503 : }
504 :
505 0 : #[derive(Debug, Serialize, Deserialize, Clone, PartialEq, Eq)]
506 : pub struct ThrottleConfig {
507 : pub task_kinds: Vec<String>, // TaskKind
508 : pub initial: u32,
509 : #[serde(with = "humantime_serde")]
510 : pub refill_interval: Duration,
511 : pub refill_amount: NonZeroU32,
512 : pub max: u32,
513 : }
514 :
515 : impl ThrottleConfig {
516 382 : pub fn disabled() -> Self {
517 382 : Self {
518 382 : task_kinds: vec![], // effectively disables the throttle
519 382 : // other values don't matter with emtpy `task_kinds`.
520 382 : initial: 0,
521 382 : refill_interval: Duration::from_millis(1),
522 382 : refill_amount: NonZeroU32::new(1).unwrap(),
523 382 : max: 1,
524 382 : }
525 382 : }
526 : /// The requests per second allowed by the given config.
527 0 : pub fn steady_rps(&self) -> f64 {
528 0 : (self.refill_amount.get() as f64) / (self.refill_interval.as_secs_f64())
529 0 : }
530 : }
531 :
532 : /// A flattened analog of a `pagesever::tenant::LocationMode`, which
533 : /// lists out all possible states (and the virtual "Detached" state)
534 : /// in a flat form rather than using rust-style enums.
535 0 : #[derive(Serialize, Deserialize, Debug, Clone, Copy, Eq, PartialEq)]
536 : pub enum LocationConfigMode {
537 : AttachedSingle,
538 : AttachedMulti,
539 : AttachedStale,
540 : Secondary,
541 : Detached,
542 : }
543 :
544 0 : #[derive(Serialize, Deserialize, Debug, Clone, Eq, PartialEq)]
545 : pub struct LocationConfigSecondary {
546 : pub warm: bool,
547 : }
548 :
549 : /// An alternative representation of `pageserver::tenant::LocationConf`,
550 : /// for use in external-facing APIs.
551 0 : #[derive(Serialize, Deserialize, Debug, Clone, Eq, PartialEq)]
552 : pub struct LocationConfig {
553 : pub mode: LocationConfigMode,
554 : /// If attaching, in what generation?
555 : #[serde(default)]
556 : pub generation: Option<u32>,
557 :
558 : // If requesting mode `Secondary`, configuration for that.
559 : #[serde(default)]
560 : pub secondary_conf: Option<LocationConfigSecondary>,
561 :
562 : // Shard parameters: if shard_count is nonzero, then other shard_* fields
563 : // must be set accurately.
564 : #[serde(default)]
565 : pub shard_number: u8,
566 : #[serde(default)]
567 : pub shard_count: u8,
568 : #[serde(default)]
569 : pub shard_stripe_size: u32,
570 :
571 : // This configuration only affects attached mode, but should be provided irrespective
572 : // of the mode, as a secondary location might transition on startup if the response
573 : // to the `/re-attach` control plane API requests it.
574 : pub tenant_conf: TenantConfig,
575 : }
576 :
577 0 : #[derive(Serialize, Deserialize)]
578 : pub struct LocationConfigListResponse {
579 : pub tenant_shards: Vec<(TenantShardId, Option<LocationConfig>)>,
580 : }
581 :
582 : #[derive(Serialize)]
583 : pub struct StatusResponse {
584 : pub id: NodeId,
585 : }
586 :
587 0 : #[derive(Serialize, Deserialize, Debug)]
588 : #[serde(deny_unknown_fields)]
589 : pub struct TenantLocationConfigRequest {
590 : #[serde(flatten)]
591 : pub config: LocationConfig, // as we have a flattened field, we should reject all unknown fields in it
592 : }
593 :
594 0 : #[derive(Serialize, Deserialize, Debug)]
595 : #[serde(deny_unknown_fields)]
596 : pub struct TenantTimeTravelRequest {
597 : pub shard_counts: Vec<ShardCount>,
598 : }
599 :
600 0 : #[derive(Serialize, Deserialize, Debug)]
601 : #[serde(deny_unknown_fields)]
602 : pub struct TenantShardLocation {
603 : pub shard_id: TenantShardId,
604 : pub node_id: NodeId,
605 : }
606 :
607 0 : #[derive(Serialize, Deserialize, Debug)]
608 : #[serde(deny_unknown_fields)]
609 : pub struct TenantLocationConfigResponse {
610 : pub shards: Vec<TenantShardLocation>,
611 : // If the shards' ShardCount count is >1, stripe_size will be set.
612 : pub stripe_size: Option<ShardStripeSize>,
613 : }
614 :
615 3 : #[derive(Serialize, Deserialize, Debug)]
616 : #[serde(deny_unknown_fields)]
617 : pub struct TenantConfigRequest {
618 : pub tenant_id: TenantId,
619 : #[serde(flatten)]
620 : pub config: TenantConfig, // as we have a flattened field, we should reject all unknown fields in it
621 : }
622 :
623 : impl std::ops::Deref for TenantConfigRequest {
624 : type Target = TenantConfig;
625 :
626 0 : fn deref(&self) -> &Self::Target {
627 0 : &self.config
628 0 : }
629 : }
630 :
631 : impl TenantConfigRequest {
632 0 : pub fn new(tenant_id: TenantId) -> TenantConfigRequest {
633 0 : let config = TenantConfig::default();
634 0 : TenantConfigRequest { tenant_id, config }
635 0 : }
636 : }
637 :
638 : /// See [`TenantState::attachment_status`] and the OpenAPI docs for context.
639 0 : #[derive(Serialize, Deserialize, Clone)]
640 : #[serde(tag = "slug", content = "data", rename_all = "snake_case")]
641 : pub enum TenantAttachmentStatus {
642 : Maybe,
643 : Attached,
644 : Failed { reason: String },
645 : }
646 :
647 0 : #[derive(Serialize, Deserialize, Clone)]
648 : pub struct TenantInfo {
649 : pub id: TenantShardId,
650 : // NB: intentionally not part of OpenAPI, we don't want to commit to a specific set of TenantState's
651 : pub state: TenantState,
652 : /// Sum of the size of all layer files.
653 : /// If a layer is present in both local FS and S3, it counts only once.
654 : pub current_physical_size: Option<u64>, // physical size is only included in `tenant_status` endpoint
655 : pub attachment_status: TenantAttachmentStatus,
656 : pub generation: u32,
657 :
658 : /// Opaque explanation if gc is being blocked.
659 : ///
660 : /// Only looked up for the individual tenant detail, not the listing. This is purely for
661 : /// debugging, not included in openapi.
662 : #[serde(skip_serializing_if = "Option::is_none")]
663 : pub gc_blocking: Option<String>,
664 : }
665 :
666 0 : #[derive(Serialize, Deserialize, Clone)]
667 : pub struct TenantDetails {
668 : #[serde(flatten)]
669 : pub tenant_info: TenantInfo,
670 :
671 : pub walredo: Option<WalRedoManagerStatus>,
672 :
673 : pub timelines: Vec<TimelineId>,
674 : }
675 :
676 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone, Copy, Debug)]
677 : pub enum TimelineArchivalState {
678 : Archived,
679 : Unarchived,
680 : }
681 :
682 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone)]
683 : pub struct TimelineArchivalConfigRequest {
684 : pub state: TimelineArchivalState,
685 : }
686 :
687 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
688 : pub struct TimelinesInfoAndOffloaded {
689 : pub timelines: Vec<TimelineInfo>,
690 : pub offloaded: Vec<OffloadedTimelineInfo>,
691 : }
692 :
693 : /// Analog of [`TimelineInfo`] for offloaded timelines.
694 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
695 : pub struct OffloadedTimelineInfo {
696 : pub tenant_id: TenantShardId,
697 : pub timeline_id: TimelineId,
698 : /// Whether the timeline has a parent it has been branched off from or not
699 : pub ancestor_timeline_id: Option<TimelineId>,
700 : /// Whether to retain the branch lsn at the ancestor or not
701 : pub ancestor_retain_lsn: Option<Lsn>,
702 : }
703 :
704 : /// This represents the output of the "timeline_detail" and "timeline_list" API calls.
705 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
706 : pub struct TimelineInfo {
707 : pub tenant_id: TenantShardId,
708 : pub timeline_id: TimelineId,
709 :
710 : pub ancestor_timeline_id: Option<TimelineId>,
711 : pub ancestor_lsn: Option<Lsn>,
712 : pub last_record_lsn: Lsn,
713 : pub prev_record_lsn: Option<Lsn>,
714 : pub latest_gc_cutoff_lsn: Lsn,
715 : pub disk_consistent_lsn: Lsn,
716 :
717 : /// The LSN that we have succesfully uploaded to remote storage
718 : pub remote_consistent_lsn: Lsn,
719 :
720 : /// The LSN that we are advertizing to safekeepers
721 : pub remote_consistent_lsn_visible: Lsn,
722 :
723 : /// The LSN from the start of the root timeline (never changes)
724 : pub initdb_lsn: Lsn,
725 :
726 : pub current_logical_size: u64,
727 : pub current_logical_size_is_accurate: bool,
728 :
729 : pub directory_entries_counts: Vec<u64>,
730 :
731 : /// Sum of the size of all layer files.
732 : /// If a layer is present in both local FS and S3, it counts only once.
733 : pub current_physical_size: Option<u64>, // is None when timeline is Unloaded
734 : pub current_logical_size_non_incremental: Option<u64>,
735 :
736 : /// How many bytes of WAL are within this branch's pitr_interval. If the pitr_interval goes
737 : /// beyond the branch's branch point, we only count up to the branch point.
738 : pub pitr_history_size: u64,
739 :
740 : /// Whether this branch's branch point is within its ancestor's PITR interval (i.e. any
741 : /// ancestor data used by this branch would have been retained anyway). If this is false, then
742 : /// this branch may be imposing a cost on the ancestor by causing it to retain layers that it would
743 : /// otherwise be able to GC.
744 : pub within_ancestor_pitr: bool,
745 :
746 : pub timeline_dir_layer_file_size_sum: Option<u64>,
747 :
748 : pub wal_source_connstr: Option<String>,
749 : pub last_received_msg_lsn: Option<Lsn>,
750 : /// the timestamp (in microseconds) of the last received message
751 : pub last_received_msg_ts: Option<u128>,
752 : pub pg_version: u32,
753 :
754 : pub state: TimelineState,
755 :
756 : pub walreceiver_status: String,
757 :
758 : // ALWAYS add new fields at the end of the struct with `Option` to ensure forward/backward compatibility.
759 : // Backward compatibility: you will get a JSON not containing the newly-added field.
760 : // Forward compatibility: a previous version of the pageserver will receive a JSON. serde::Deserialize does
761 : // not deny unknown fields by default so it's safe to set the field to some value, though it won't be
762 : // read.
763 : pub is_archived: Option<bool>,
764 : }
765 :
766 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
767 : pub struct LayerMapInfo {
768 : pub in_memory_layers: Vec<InMemoryLayerInfo>,
769 : pub historic_layers: Vec<HistoricLayerInfo>,
770 : }
771 :
772 : /// The residence status of a layer
773 0 : #[derive(Debug, Clone, Copy, Serialize, Deserialize)]
774 : pub enum LayerResidenceStatus {
775 : /// Residence status for a layer file that exists locally.
776 : /// It may also exist on the remote, we don't care here.
777 : Resident,
778 : /// Residence status for a layer file that only exists on the remote.
779 : Evicted,
780 : }
781 :
782 : #[serde_as]
783 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
784 : pub struct LayerAccessStats {
785 : #[serde_as(as = "serde_with::TimestampMilliSeconds")]
786 : pub access_time: SystemTime,
787 :
788 : #[serde_as(as = "serde_with::TimestampMilliSeconds")]
789 : pub residence_time: SystemTime,
790 :
791 : pub visible: bool,
792 : }
793 :
794 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
795 : #[serde(tag = "kind")]
796 : pub enum InMemoryLayerInfo {
797 : Open { lsn_start: Lsn },
798 : Frozen { lsn_start: Lsn, lsn_end: Lsn },
799 : }
800 :
801 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
802 : #[serde(tag = "kind")]
803 : pub enum HistoricLayerInfo {
804 : Delta {
805 : layer_file_name: String,
806 : layer_file_size: u64,
807 :
808 : lsn_start: Lsn,
809 : lsn_end: Lsn,
810 : remote: bool,
811 : access_stats: LayerAccessStats,
812 :
813 : l0: bool,
814 : },
815 : Image {
816 : layer_file_name: String,
817 : layer_file_size: u64,
818 :
819 : lsn_start: Lsn,
820 : remote: bool,
821 : access_stats: LayerAccessStats,
822 : },
823 : }
824 :
825 : impl HistoricLayerInfo {
826 0 : pub fn layer_file_name(&self) -> &str {
827 0 : match self {
828 : HistoricLayerInfo::Delta {
829 0 : layer_file_name, ..
830 0 : } => layer_file_name,
831 : HistoricLayerInfo::Image {
832 0 : layer_file_name, ..
833 0 : } => layer_file_name,
834 : }
835 0 : }
836 0 : pub fn is_remote(&self) -> bool {
837 0 : match self {
838 0 : HistoricLayerInfo::Delta { remote, .. } => *remote,
839 0 : HistoricLayerInfo::Image { remote, .. } => *remote,
840 : }
841 0 : }
842 0 : pub fn set_remote(&mut self, value: bool) {
843 0 : let field = match self {
844 0 : HistoricLayerInfo::Delta { remote, .. } => remote,
845 0 : HistoricLayerInfo::Image { remote, .. } => remote,
846 : };
847 0 : *field = value;
848 0 : }
849 0 : pub fn layer_file_size(&self) -> u64 {
850 0 : match self {
851 : HistoricLayerInfo::Delta {
852 0 : layer_file_size, ..
853 0 : } => *layer_file_size,
854 : HistoricLayerInfo::Image {
855 0 : layer_file_size, ..
856 0 : } => *layer_file_size,
857 : }
858 0 : }
859 : }
860 :
861 0 : #[derive(Debug, Serialize, Deserialize)]
862 : pub struct DownloadRemoteLayersTaskSpawnRequest {
863 : pub max_concurrent_downloads: NonZeroUsize,
864 : }
865 :
866 0 : #[derive(Debug, Serialize, Deserialize)]
867 : pub struct IngestAuxFilesRequest {
868 : pub aux_files: HashMap<String, String>,
869 : }
870 :
871 0 : #[derive(Debug, Serialize, Deserialize)]
872 : pub struct ListAuxFilesRequest {
873 : pub lsn: Lsn,
874 : }
875 :
876 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
877 : pub struct DownloadRemoteLayersTaskInfo {
878 : pub task_id: String,
879 : pub state: DownloadRemoteLayersTaskState,
880 : pub total_layer_count: u64, // stable once `completed`
881 : pub successful_download_count: u64, // stable once `completed`
882 : pub failed_download_count: u64, // stable once `completed`
883 : }
884 :
885 0 : #[derive(Debug, Serialize, Deserialize, Clone)]
886 : pub enum DownloadRemoteLayersTaskState {
887 : Running,
888 : Completed,
889 : ShutDown,
890 : }
891 :
892 0 : #[derive(Debug, Serialize, Deserialize)]
893 : pub struct TimelineGcRequest {
894 : pub gc_horizon: Option<u64>,
895 : }
896 :
897 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
898 : pub struct WalRedoManagerProcessStatus {
899 : pub pid: u32,
900 : }
901 :
902 0 : #[derive(Debug, Clone, Serialize, Deserialize)]
903 : pub struct WalRedoManagerStatus {
904 : pub last_redo_at: Option<chrono::DateTime<chrono::Utc>>,
905 : pub process: Option<WalRedoManagerProcessStatus>,
906 : }
907 :
908 : /// The progress of a secondary tenant.
909 : ///
910 : /// It is mostly useful when doing a long running download: e.g. initiating
911 : /// a download job, timing out while waiting for it to run, and then inspecting this status to understand
912 : /// what's happening.
913 0 : #[derive(Default, Debug, Serialize, Deserialize, Clone)]
914 : pub struct SecondaryProgress {
915 : /// The remote storage LastModified time of the heatmap object we last downloaded.
916 : pub heatmap_mtime: Option<serde_system_time::SystemTime>,
917 :
918 : /// The number of layers currently on-disk
919 : pub layers_downloaded: usize,
920 : /// The number of layers in the most recently seen heatmap
921 : pub layers_total: usize,
922 :
923 : /// The number of layer bytes currently on-disk
924 : pub bytes_downloaded: u64,
925 : /// The number of layer bytes in the most recently seen heatmap
926 : pub bytes_total: u64,
927 : }
928 :
929 0 : #[derive(Serialize, Deserialize, Debug)]
930 : pub struct TenantScanRemoteStorageShard {
931 : pub tenant_shard_id: TenantShardId,
932 : pub generation: Option<u32>,
933 : }
934 :
935 0 : #[derive(Serialize, Deserialize, Debug, Default)]
936 : pub struct TenantScanRemoteStorageResponse {
937 : pub shards: Vec<TenantScanRemoteStorageShard>,
938 : }
939 :
940 0 : #[derive(Serialize, Deserialize, Debug, Clone)]
941 : #[serde(rename_all = "snake_case")]
942 : pub enum TenantSorting {
943 : ResidentSize,
944 : MaxLogicalSize,
945 : }
946 :
947 : impl Default for TenantSorting {
948 0 : fn default() -> Self {
949 0 : Self::ResidentSize
950 0 : }
951 : }
952 :
953 0 : #[derive(Serialize, Deserialize, Debug, Clone)]
954 : pub struct TopTenantShardsRequest {
955 : // How would you like to sort the tenants?
956 : pub order_by: TenantSorting,
957 :
958 : // How many results?
959 : pub limit: usize,
960 :
961 : // Omit tenants with more than this many shards (e.g. if this is the max number of shards
962 : // that the caller would ever split to)
963 : pub where_shards_lt: Option<ShardCount>,
964 :
965 : // Omit tenants where the ordering metric is less than this (this is an optimization to
966 : // let us quickly exclude numerous tiny shards)
967 : pub where_gt: Option<u64>,
968 : }
969 :
970 0 : #[derive(Serialize, Deserialize, Debug, PartialEq, Eq)]
971 : pub struct TopTenantShardItem {
972 : pub id: TenantShardId,
973 :
974 : /// Total size of layers on local disk for all timelines in this tenant
975 : pub resident_size: u64,
976 :
977 : /// Total size of layers in remote storage for all timelines in this tenant
978 : pub physical_size: u64,
979 :
980 : /// The largest logical size of a timeline within this tenant
981 : pub max_logical_size: u64,
982 : }
983 :
984 0 : #[derive(Serialize, Deserialize, Debug, Default)]
985 : pub struct TopTenantShardsResponse {
986 : pub shards: Vec<TopTenantShardItem>,
987 : }
988 :
989 : pub mod virtual_file {
990 : #[derive(
991 : Copy,
992 : Clone,
993 : PartialEq,
994 : Eq,
995 : Hash,
996 198 : strum_macros::EnumString,
997 0 : strum_macros::Display,
998 0 : serde_with::DeserializeFromStr,
999 : serde_with::SerializeDisplay,
1000 : Debug,
1001 : )]
1002 : #[strum(serialize_all = "kebab-case")]
1003 : pub enum IoEngineKind {
1004 : StdFs,
1005 : #[cfg(target_os = "linux")]
1006 : TokioEpollUring,
1007 : }
1008 :
1009 : /// Direct IO modes for a pageserver.
1010 : #[derive(
1011 : Copy,
1012 : Clone,
1013 : PartialEq,
1014 : Eq,
1015 : Hash,
1016 0 : strum_macros::EnumString,
1017 0 : strum_macros::Display,
1018 0 : serde_with::DeserializeFromStr,
1019 : serde_with::SerializeDisplay,
1020 : Debug,
1021 : )]
1022 : #[strum(serialize_all = "kebab-case")]
1023 : #[repr(u8)]
1024 : pub enum IoMode {
1025 : /// Uses buffered IO.
1026 : Buffered,
1027 : /// Uses direct IO, error out if the operation fails.
1028 : #[cfg(target_os = "linux")]
1029 : Direct,
1030 : }
1031 :
1032 : impl IoMode {
1033 204 : pub const fn preferred() -> Self {
1034 204 : Self::Buffered
1035 204 : }
1036 : }
1037 :
1038 : impl TryFrom<u8> for IoMode {
1039 : type Error = u8;
1040 :
1041 1148 : fn try_from(value: u8) -> Result<Self, Self::Error> {
1042 1148 : Ok(match value {
1043 1148 : v if v == (IoMode::Buffered as u8) => IoMode::Buffered,
1044 : #[cfg(target_os = "linux")]
1045 0 : v if v == (IoMode::Direct as u8) => IoMode::Direct,
1046 0 : x => return Err(x),
1047 : })
1048 1148 : }
1049 : }
1050 : }
1051 :
1052 : // Wrapped in libpq CopyData
1053 : #[derive(PartialEq, Eq, Debug)]
1054 : pub enum PagestreamFeMessage {
1055 : Exists(PagestreamExistsRequest),
1056 : Nblocks(PagestreamNblocksRequest),
1057 : GetPage(PagestreamGetPageRequest),
1058 : DbSize(PagestreamDbSizeRequest),
1059 : GetSlruSegment(PagestreamGetSlruSegmentRequest),
1060 : }
1061 :
1062 : // Wrapped in libpq CopyData
1063 0 : #[derive(strum_macros::EnumProperty)]
1064 : pub enum PagestreamBeMessage {
1065 : Exists(PagestreamExistsResponse),
1066 : Nblocks(PagestreamNblocksResponse),
1067 : GetPage(PagestreamGetPageResponse),
1068 : Error(PagestreamErrorResponse),
1069 : DbSize(PagestreamDbSizeResponse),
1070 : GetSlruSegment(PagestreamGetSlruSegmentResponse),
1071 : }
1072 :
1073 : // Keep in sync with `pagestore_client.h`
1074 : #[repr(u8)]
1075 : enum PagestreamBeMessageTag {
1076 : Exists = 100,
1077 : Nblocks = 101,
1078 : GetPage = 102,
1079 : Error = 103,
1080 : DbSize = 104,
1081 : GetSlruSegment = 105,
1082 : }
1083 : impl TryFrom<u8> for PagestreamBeMessageTag {
1084 : type Error = u8;
1085 0 : fn try_from(value: u8) -> Result<Self, u8> {
1086 0 : match value {
1087 0 : 100 => Ok(PagestreamBeMessageTag::Exists),
1088 0 : 101 => Ok(PagestreamBeMessageTag::Nblocks),
1089 0 : 102 => Ok(PagestreamBeMessageTag::GetPage),
1090 0 : 103 => Ok(PagestreamBeMessageTag::Error),
1091 0 : 104 => Ok(PagestreamBeMessageTag::DbSize),
1092 0 : 105 => Ok(PagestreamBeMessageTag::GetSlruSegment),
1093 0 : _ => Err(value),
1094 : }
1095 0 : }
1096 : }
1097 :
1098 : // A GetPage request contains two LSN values:
1099 : //
1100 : // request_lsn: Get the page version at this point in time. Lsn::Max is a special value that means
1101 : // "get the latest version present". It's used by the primary server, which knows that no one else
1102 : // is writing WAL. 'not_modified_since' must be set to a proper value even if request_lsn is
1103 : // Lsn::Max. Standby servers use the current replay LSN as the request LSN.
1104 : //
1105 : // not_modified_since: Hint to the pageserver that the client knows that the page has not been
1106 : // modified between 'not_modified_since' and the request LSN. It's always correct to set
1107 : // 'not_modified_since equal' to 'request_lsn' (unless Lsn::Max is used as the 'request_lsn'), but
1108 : // passing an earlier LSN can speed up the request, by allowing the pageserver to process the
1109 : // request without waiting for 'request_lsn' to arrive.
1110 : //
1111 : // The now-defunct V1 interface contained only one LSN, and a boolean 'latest' flag. The V1 interface was
1112 : // sufficient for the primary; the 'lsn' was equivalent to the 'not_modified_since' value, and
1113 : // 'latest' was set to true. The V2 interface was added because there was no correct way for a
1114 : // standby to request a page at a particular non-latest LSN, and also include the
1115 : // 'not_modified_since' hint. That led to an awkward choice of either using an old LSN in the
1116 : // request, if the standby knows that the page hasn't been modified since, and risk getting an error
1117 : // if that LSN has fallen behind the GC horizon, or requesting the current replay LSN, which could
1118 : // require the pageserver unnecessarily to wait for the WAL to arrive up to that point. The new V2
1119 : // interface allows sending both LSNs, and let the pageserver do the right thing. There was no
1120 : // difference in the responses between V1 and V2.
1121 : //
1122 : #[derive(Clone, Copy)]
1123 : pub enum PagestreamProtocolVersion {
1124 : V2,
1125 : }
1126 :
1127 : #[derive(Debug, PartialEq, Eq)]
1128 : pub struct PagestreamExistsRequest {
1129 : pub request_lsn: Lsn,
1130 : pub not_modified_since: Lsn,
1131 : pub rel: RelTag,
1132 : }
1133 :
1134 : #[derive(Debug, PartialEq, Eq)]
1135 : pub struct PagestreamNblocksRequest {
1136 : pub request_lsn: Lsn,
1137 : pub not_modified_since: Lsn,
1138 : pub rel: RelTag,
1139 : }
1140 :
1141 : #[derive(Debug, PartialEq, Eq)]
1142 : pub struct PagestreamGetPageRequest {
1143 : pub request_lsn: Lsn,
1144 : pub not_modified_since: Lsn,
1145 : pub rel: RelTag,
1146 : pub blkno: u32,
1147 : }
1148 :
1149 : #[derive(Debug, PartialEq, Eq)]
1150 : pub struct PagestreamDbSizeRequest {
1151 : pub request_lsn: Lsn,
1152 : pub not_modified_since: Lsn,
1153 : pub dbnode: u32,
1154 : }
1155 :
1156 : #[derive(Debug, PartialEq, Eq)]
1157 : pub struct PagestreamGetSlruSegmentRequest {
1158 : pub request_lsn: Lsn,
1159 : pub not_modified_since: Lsn,
1160 : pub kind: u8,
1161 : pub segno: u32,
1162 : }
1163 :
1164 : #[derive(Debug)]
1165 : pub struct PagestreamExistsResponse {
1166 : pub exists: bool,
1167 : }
1168 :
1169 : #[derive(Debug)]
1170 : pub struct PagestreamNblocksResponse {
1171 : pub n_blocks: u32,
1172 : }
1173 :
1174 : #[derive(Debug)]
1175 : pub struct PagestreamGetPageResponse {
1176 : pub page: Bytes,
1177 : }
1178 :
1179 : #[derive(Debug)]
1180 : pub struct PagestreamGetSlruSegmentResponse {
1181 : pub segment: Bytes,
1182 : }
1183 :
1184 : #[derive(Debug)]
1185 : pub struct PagestreamErrorResponse {
1186 : pub message: String,
1187 : }
1188 :
1189 : #[derive(Debug)]
1190 : pub struct PagestreamDbSizeResponse {
1191 : pub db_size: i64,
1192 : }
1193 :
1194 : // This is a cut-down version of TenantHistorySize from the pageserver crate, omitting fields
1195 : // that require pageserver-internal types. It is sufficient to get the total size.
1196 0 : #[derive(Serialize, Deserialize, Debug)]
1197 : pub struct TenantHistorySize {
1198 : pub id: TenantId,
1199 : /// Size is a mixture of WAL and logical size, so the unit is bytes.
1200 : ///
1201 : /// Will be none if `?inputs_only=true` was given.
1202 : pub size: Option<u64>,
1203 : }
1204 :
1205 : impl PagestreamFeMessage {
1206 : /// Serialize a compute -> pageserver message. This is currently only used in testing
1207 : /// tools. Always uses protocol version 2.
1208 4 : pub fn serialize(&self) -> Bytes {
1209 4 : let mut bytes = BytesMut::new();
1210 4 :
1211 4 : match self {
1212 1 : Self::Exists(req) => {
1213 1 : bytes.put_u8(0);
1214 1 : bytes.put_u64(req.request_lsn.0);
1215 1 : bytes.put_u64(req.not_modified_since.0);
1216 1 : bytes.put_u32(req.rel.spcnode);
1217 1 : bytes.put_u32(req.rel.dbnode);
1218 1 : bytes.put_u32(req.rel.relnode);
1219 1 : bytes.put_u8(req.rel.forknum);
1220 1 : }
1221 :
1222 1 : Self::Nblocks(req) => {
1223 1 : bytes.put_u8(1);
1224 1 : bytes.put_u64(req.request_lsn.0);
1225 1 : bytes.put_u64(req.not_modified_since.0);
1226 1 : bytes.put_u32(req.rel.spcnode);
1227 1 : bytes.put_u32(req.rel.dbnode);
1228 1 : bytes.put_u32(req.rel.relnode);
1229 1 : bytes.put_u8(req.rel.forknum);
1230 1 : }
1231 :
1232 1 : Self::GetPage(req) => {
1233 1 : bytes.put_u8(2);
1234 1 : bytes.put_u64(req.request_lsn.0);
1235 1 : bytes.put_u64(req.not_modified_since.0);
1236 1 : bytes.put_u32(req.rel.spcnode);
1237 1 : bytes.put_u32(req.rel.dbnode);
1238 1 : bytes.put_u32(req.rel.relnode);
1239 1 : bytes.put_u8(req.rel.forknum);
1240 1 : bytes.put_u32(req.blkno);
1241 1 : }
1242 :
1243 1 : Self::DbSize(req) => {
1244 1 : bytes.put_u8(3);
1245 1 : bytes.put_u64(req.request_lsn.0);
1246 1 : bytes.put_u64(req.not_modified_since.0);
1247 1 : bytes.put_u32(req.dbnode);
1248 1 : }
1249 :
1250 0 : Self::GetSlruSegment(req) => {
1251 0 : bytes.put_u8(4);
1252 0 : bytes.put_u64(req.request_lsn.0);
1253 0 : bytes.put_u64(req.not_modified_since.0);
1254 0 : bytes.put_u8(req.kind);
1255 0 : bytes.put_u32(req.segno);
1256 0 : }
1257 : }
1258 :
1259 4 : bytes.into()
1260 4 : }
1261 :
1262 4 : pub fn parse<R: std::io::Read>(body: &mut R) -> anyhow::Result<PagestreamFeMessage> {
1263 : // these correspond to the NeonMessageTag enum in pagestore_client.h
1264 : //
1265 : // TODO: consider using protobuf or serde bincode for less error prone
1266 : // serialization.
1267 4 : let msg_tag = body.read_u8()?;
1268 :
1269 : // these two fields are the same for every request type
1270 4 : let request_lsn = Lsn::from(body.read_u64::<BigEndian>()?);
1271 4 : let not_modified_since = Lsn::from(body.read_u64::<BigEndian>()?);
1272 :
1273 4 : match msg_tag {
1274 : 0 => Ok(PagestreamFeMessage::Exists(PagestreamExistsRequest {
1275 1 : request_lsn,
1276 1 : not_modified_since,
1277 1 : rel: RelTag {
1278 1 : spcnode: body.read_u32::<BigEndian>()?,
1279 1 : dbnode: body.read_u32::<BigEndian>()?,
1280 1 : relnode: body.read_u32::<BigEndian>()?,
1281 1 : forknum: body.read_u8()?,
1282 : },
1283 : })),
1284 : 1 => Ok(PagestreamFeMessage::Nblocks(PagestreamNblocksRequest {
1285 1 : request_lsn,
1286 1 : not_modified_since,
1287 1 : rel: RelTag {
1288 1 : spcnode: body.read_u32::<BigEndian>()?,
1289 1 : dbnode: body.read_u32::<BigEndian>()?,
1290 1 : relnode: body.read_u32::<BigEndian>()?,
1291 1 : forknum: body.read_u8()?,
1292 : },
1293 : })),
1294 : 2 => Ok(PagestreamFeMessage::GetPage(PagestreamGetPageRequest {
1295 1 : request_lsn,
1296 1 : not_modified_since,
1297 1 : rel: RelTag {
1298 1 : spcnode: body.read_u32::<BigEndian>()?,
1299 1 : dbnode: body.read_u32::<BigEndian>()?,
1300 1 : relnode: body.read_u32::<BigEndian>()?,
1301 1 : forknum: body.read_u8()?,
1302 : },
1303 1 : blkno: body.read_u32::<BigEndian>()?,
1304 : })),
1305 : 3 => Ok(PagestreamFeMessage::DbSize(PagestreamDbSizeRequest {
1306 1 : request_lsn,
1307 1 : not_modified_since,
1308 1 : dbnode: body.read_u32::<BigEndian>()?,
1309 : })),
1310 : 4 => Ok(PagestreamFeMessage::GetSlruSegment(
1311 : PagestreamGetSlruSegmentRequest {
1312 0 : request_lsn,
1313 0 : not_modified_since,
1314 0 : kind: body.read_u8()?,
1315 0 : segno: body.read_u32::<BigEndian>()?,
1316 : },
1317 : )),
1318 0 : _ => bail!("unknown smgr message tag: {:?}", msg_tag),
1319 : }
1320 4 : }
1321 : }
1322 :
1323 : impl PagestreamBeMessage {
1324 0 : pub fn serialize(&self) -> Bytes {
1325 0 : let mut bytes = BytesMut::new();
1326 :
1327 : use PagestreamBeMessageTag as Tag;
1328 0 : match self {
1329 0 : Self::Exists(resp) => {
1330 0 : bytes.put_u8(Tag::Exists as u8);
1331 0 : bytes.put_u8(resp.exists as u8);
1332 0 : }
1333 :
1334 0 : Self::Nblocks(resp) => {
1335 0 : bytes.put_u8(Tag::Nblocks as u8);
1336 0 : bytes.put_u32(resp.n_blocks);
1337 0 : }
1338 :
1339 0 : Self::GetPage(resp) => {
1340 0 : bytes.put_u8(Tag::GetPage as u8);
1341 0 : bytes.put(&resp.page[..]);
1342 0 : }
1343 :
1344 0 : Self::Error(resp) => {
1345 0 : bytes.put_u8(Tag::Error as u8);
1346 0 : bytes.put(resp.message.as_bytes());
1347 0 : bytes.put_u8(0); // null terminator
1348 0 : }
1349 0 : Self::DbSize(resp) => {
1350 0 : bytes.put_u8(Tag::DbSize as u8);
1351 0 : bytes.put_i64(resp.db_size);
1352 0 : }
1353 :
1354 0 : Self::GetSlruSegment(resp) => {
1355 0 : bytes.put_u8(Tag::GetSlruSegment as u8);
1356 0 : bytes.put_u32((resp.segment.len() / BLCKSZ as usize) as u32);
1357 0 : bytes.put(&resp.segment[..]);
1358 0 : }
1359 : }
1360 :
1361 0 : bytes.into()
1362 0 : }
1363 :
1364 0 : pub fn deserialize(buf: Bytes) -> anyhow::Result<Self> {
1365 0 : let mut buf = buf.reader();
1366 0 : let msg_tag = buf.read_u8()?;
1367 :
1368 : use PagestreamBeMessageTag as Tag;
1369 0 : let ok =
1370 0 : match Tag::try_from(msg_tag).map_err(|tag: u8| anyhow::anyhow!("invalid tag {tag}"))? {
1371 : Tag::Exists => {
1372 0 : let exists = buf.read_u8()?;
1373 0 : Self::Exists(PagestreamExistsResponse {
1374 0 : exists: exists != 0,
1375 0 : })
1376 : }
1377 : Tag::Nblocks => {
1378 0 : let n_blocks = buf.read_u32::<BigEndian>()?;
1379 0 : Self::Nblocks(PagestreamNblocksResponse { n_blocks })
1380 : }
1381 : Tag::GetPage => {
1382 0 : let mut page = vec![0; 8192]; // TODO: use MaybeUninit
1383 0 : buf.read_exact(&mut page)?;
1384 0 : PagestreamBeMessage::GetPage(PagestreamGetPageResponse { page: page.into() })
1385 : }
1386 : Tag::Error => {
1387 0 : let mut msg = Vec::new();
1388 0 : buf.read_until(0, &mut msg)?;
1389 0 : let cstring = std::ffi::CString::from_vec_with_nul(msg)?;
1390 0 : let rust_str = cstring.to_str()?;
1391 0 : PagestreamBeMessage::Error(PagestreamErrorResponse {
1392 0 : message: rust_str.to_owned(),
1393 0 : })
1394 : }
1395 : Tag::DbSize => {
1396 0 : let db_size = buf.read_i64::<BigEndian>()?;
1397 0 : Self::DbSize(PagestreamDbSizeResponse { db_size })
1398 : }
1399 : Tag::GetSlruSegment => {
1400 0 : let n_blocks = buf.read_u32::<BigEndian>()?;
1401 0 : let mut segment = vec![0; n_blocks as usize * BLCKSZ as usize];
1402 0 : buf.read_exact(&mut segment)?;
1403 0 : Self::GetSlruSegment(PagestreamGetSlruSegmentResponse {
1404 0 : segment: segment.into(),
1405 0 : })
1406 : }
1407 : };
1408 0 : let remaining = buf.into_inner();
1409 0 : if !remaining.is_empty() {
1410 0 : anyhow::bail!(
1411 0 : "remaining bytes in msg with tag={msg_tag}: {}",
1412 0 : remaining.len()
1413 0 : );
1414 0 : }
1415 0 : Ok(ok)
1416 0 : }
1417 :
1418 0 : pub fn kind(&self) -> &'static str {
1419 0 : match self {
1420 0 : Self::Exists(_) => "Exists",
1421 0 : Self::Nblocks(_) => "Nblocks",
1422 0 : Self::GetPage(_) => "GetPage",
1423 0 : Self::Error(_) => "Error",
1424 0 : Self::DbSize(_) => "DbSize",
1425 0 : Self::GetSlruSegment(_) => "GetSlruSegment",
1426 : }
1427 0 : }
1428 : }
1429 :
1430 : #[cfg(test)]
1431 : mod tests {
1432 : use serde_json::json;
1433 : use std::str::FromStr;
1434 :
1435 : use super::*;
1436 :
1437 : #[test]
1438 1 : fn test_pagestream() {
1439 1 : // Test serialization/deserialization of PagestreamFeMessage
1440 1 : let messages = vec![
1441 1 : PagestreamFeMessage::Exists(PagestreamExistsRequest {
1442 1 : request_lsn: Lsn(4),
1443 1 : not_modified_since: Lsn(3),
1444 1 : rel: RelTag {
1445 1 : forknum: 1,
1446 1 : spcnode: 2,
1447 1 : dbnode: 3,
1448 1 : relnode: 4,
1449 1 : },
1450 1 : }),
1451 1 : PagestreamFeMessage::Nblocks(PagestreamNblocksRequest {
1452 1 : request_lsn: Lsn(4),
1453 1 : not_modified_since: Lsn(4),
1454 1 : rel: RelTag {
1455 1 : forknum: 1,
1456 1 : spcnode: 2,
1457 1 : dbnode: 3,
1458 1 : relnode: 4,
1459 1 : },
1460 1 : }),
1461 1 : PagestreamFeMessage::GetPage(PagestreamGetPageRequest {
1462 1 : request_lsn: Lsn(4),
1463 1 : not_modified_since: Lsn(3),
1464 1 : rel: RelTag {
1465 1 : forknum: 1,
1466 1 : spcnode: 2,
1467 1 : dbnode: 3,
1468 1 : relnode: 4,
1469 1 : },
1470 1 : blkno: 7,
1471 1 : }),
1472 1 : PagestreamFeMessage::DbSize(PagestreamDbSizeRequest {
1473 1 : request_lsn: Lsn(4),
1474 1 : not_modified_since: Lsn(3),
1475 1 : dbnode: 7,
1476 1 : }),
1477 1 : ];
1478 5 : for msg in messages {
1479 4 : let bytes = msg.serialize();
1480 4 : let reconstructed = PagestreamFeMessage::parse(&mut bytes.reader()).unwrap();
1481 4 : assert!(msg == reconstructed);
1482 : }
1483 1 : }
1484 :
1485 : #[test]
1486 1 : fn test_tenantinfo_serde() {
1487 1 : // Test serialization/deserialization of TenantInfo
1488 1 : let original_active = TenantInfo {
1489 1 : id: TenantShardId::unsharded(TenantId::generate()),
1490 1 : state: TenantState::Active,
1491 1 : current_physical_size: Some(42),
1492 1 : attachment_status: TenantAttachmentStatus::Attached,
1493 1 : generation: 1,
1494 1 : gc_blocking: None,
1495 1 : };
1496 1 : let expected_active = json!({
1497 1 : "id": original_active.id.to_string(),
1498 1 : "state": {
1499 1 : "slug": "Active",
1500 1 : },
1501 1 : "current_physical_size": 42,
1502 1 : "attachment_status": {
1503 1 : "slug":"attached",
1504 1 : },
1505 1 : "generation" : 1
1506 1 : });
1507 1 :
1508 1 : let original_broken = TenantInfo {
1509 1 : id: TenantShardId::unsharded(TenantId::generate()),
1510 1 : state: TenantState::Broken {
1511 1 : reason: "reason".into(),
1512 1 : backtrace: "backtrace info".into(),
1513 1 : },
1514 1 : current_physical_size: Some(42),
1515 1 : attachment_status: TenantAttachmentStatus::Attached,
1516 1 : generation: 1,
1517 1 : gc_blocking: None,
1518 1 : };
1519 1 : let expected_broken = json!({
1520 1 : "id": original_broken.id.to_string(),
1521 1 : "state": {
1522 1 : "slug": "Broken",
1523 1 : "data": {
1524 1 : "backtrace": "backtrace info",
1525 1 : "reason": "reason",
1526 1 : }
1527 1 : },
1528 1 : "current_physical_size": 42,
1529 1 : "attachment_status": {
1530 1 : "slug":"attached",
1531 1 : },
1532 1 : "generation" : 1
1533 1 : });
1534 1 :
1535 1 : assert_eq!(
1536 1 : serde_json::to_value(&original_active).unwrap(),
1537 1 : expected_active
1538 1 : );
1539 :
1540 1 : assert_eq!(
1541 1 : serde_json::to_value(&original_broken).unwrap(),
1542 1 : expected_broken
1543 1 : );
1544 1 : assert!(format!("{:?}", &original_broken.state).contains("reason"));
1545 1 : assert!(format!("{:?}", &original_broken.state).contains("backtrace info"));
1546 1 : }
1547 :
1548 : #[test]
1549 1 : fn test_reject_unknown_field() {
1550 1 : let id = TenantId::generate();
1551 1 : let config_request = json!({
1552 1 : "tenant_id": id.to_string(),
1553 1 : "unknown_field": "unknown_value".to_string(),
1554 1 : });
1555 1 : let err = serde_json::from_value::<TenantConfigRequest>(config_request).unwrap_err();
1556 1 : assert!(
1557 1 : err.to_string().contains("unknown field `unknown_field`"),
1558 0 : "expect unknown field `unknown_field` error, got: {}",
1559 : err
1560 : );
1561 1 : }
1562 :
1563 : #[test]
1564 1 : fn tenantstatus_activating_serde() {
1565 1 : let states = [TenantState::Activating(ActivatingFrom::Attaching)];
1566 1 : let expected = "[{\"slug\":\"Activating\",\"data\":\"Attaching\"}]";
1567 1 :
1568 1 : let actual = serde_json::to_string(&states).unwrap();
1569 1 :
1570 1 : assert_eq!(actual, expected);
1571 :
1572 1 : let parsed = serde_json::from_str::<Vec<TenantState>>(&actual).unwrap();
1573 1 :
1574 1 : assert_eq!(states.as_slice(), &parsed);
1575 1 : }
1576 :
1577 : #[test]
1578 1 : fn tenantstatus_activating_strum() {
1579 1 : // tests added, because we use these for metrics
1580 1 : let examples = [
1581 1 : (line!(), TenantState::Attaching, "Attaching"),
1582 1 : (
1583 1 : line!(),
1584 1 : TenantState::Activating(ActivatingFrom::Attaching),
1585 1 : "Activating",
1586 1 : ),
1587 1 : (line!(), TenantState::Active, "Active"),
1588 1 : (
1589 1 : line!(),
1590 1 : TenantState::Stopping {
1591 1 : progress: utils::completion::Barrier::default(),
1592 1 : },
1593 1 : "Stopping",
1594 1 : ),
1595 1 : (
1596 1 : line!(),
1597 1 : TenantState::Broken {
1598 1 : reason: "Example".into(),
1599 1 : backtrace: "Looooong backtrace".into(),
1600 1 : },
1601 1 : "Broken",
1602 1 : ),
1603 1 : ];
1604 :
1605 6 : for (line, rendered, expected) in examples {
1606 5 : let actual: &'static str = rendered.into();
1607 5 : assert_eq!(actual, expected, "example on {line}");
1608 : }
1609 1 : }
1610 :
1611 : #[test]
1612 1 : fn test_aux_file_migration_path() {
1613 1 : assert!(AuxFilePolicy::is_valid_migration_path(
1614 1 : None,
1615 1 : AuxFilePolicy::V1
1616 1 : ));
1617 1 : assert!(AuxFilePolicy::is_valid_migration_path(
1618 1 : None,
1619 1 : AuxFilePolicy::V2
1620 1 : ));
1621 1 : assert!(AuxFilePolicy::is_valid_migration_path(
1622 1 : None,
1623 1 : AuxFilePolicy::CrossValidation
1624 1 : ));
1625 : // Self-migration is not a valid migration path, and the caller should handle it by itself.
1626 1 : assert!(!AuxFilePolicy::is_valid_migration_path(
1627 1 : Some(AuxFilePolicy::V1),
1628 1 : AuxFilePolicy::V1
1629 1 : ));
1630 1 : assert!(!AuxFilePolicy::is_valid_migration_path(
1631 1 : Some(AuxFilePolicy::V2),
1632 1 : AuxFilePolicy::V2
1633 1 : ));
1634 1 : assert!(!AuxFilePolicy::is_valid_migration_path(
1635 1 : Some(AuxFilePolicy::CrossValidation),
1636 1 : AuxFilePolicy::CrossValidation
1637 1 : ));
1638 : // Migrations not allowed
1639 1 : assert!(!AuxFilePolicy::is_valid_migration_path(
1640 1 : Some(AuxFilePolicy::CrossValidation),
1641 1 : AuxFilePolicy::V1
1642 1 : ));
1643 1 : assert!(!AuxFilePolicy::is_valid_migration_path(
1644 1 : Some(AuxFilePolicy::V1),
1645 1 : AuxFilePolicy::V2
1646 1 : ));
1647 1 : assert!(!AuxFilePolicy::is_valid_migration_path(
1648 1 : Some(AuxFilePolicy::V2),
1649 1 : AuxFilePolicy::V1
1650 1 : ));
1651 1 : assert!(!AuxFilePolicy::is_valid_migration_path(
1652 1 : Some(AuxFilePolicy::V2),
1653 1 : AuxFilePolicy::CrossValidation
1654 1 : ));
1655 1 : assert!(!AuxFilePolicy::is_valid_migration_path(
1656 1 : Some(AuxFilePolicy::V1),
1657 1 : AuxFilePolicy::CrossValidation
1658 1 : ));
1659 : // Migrations allowed
1660 1 : assert!(AuxFilePolicy::is_valid_migration_path(
1661 1 : Some(AuxFilePolicy::CrossValidation),
1662 1 : AuxFilePolicy::V2
1663 1 : ));
1664 1 : }
1665 :
1666 : #[test]
1667 1 : fn test_aux_parse() {
1668 1 : assert_eq!(AuxFilePolicy::from_str("V2").unwrap(), AuxFilePolicy::V2);
1669 1 : assert_eq!(AuxFilePolicy::from_str("v2").unwrap(), AuxFilePolicy::V2);
1670 1 : assert_eq!(
1671 1 : AuxFilePolicy::from_str("cross-validation").unwrap(),
1672 1 : AuxFilePolicy::CrossValidation
1673 1 : );
1674 1 : }
1675 :
1676 : #[test]
1677 1 : fn test_image_compression_algorithm_parsing() {
1678 : use ImageCompressionAlgorithm::*;
1679 1 : let cases = [
1680 1 : ("disabled", Disabled),
1681 1 : ("zstd", Zstd { level: None }),
1682 1 : ("zstd(18)", Zstd { level: Some(18) }),
1683 1 : ("zstd(-3)", Zstd { level: Some(-3) }),
1684 1 : ];
1685 :
1686 5 : for (display, expected) in cases {
1687 4 : assert_eq!(
1688 4 : ImageCompressionAlgorithm::from_str(display).unwrap(),
1689 : expected,
1690 0 : "parsing works"
1691 : );
1692 4 : assert_eq!(format!("{expected}"), display, "Display FromStr roundtrip");
1693 :
1694 4 : let ser = serde_json::to_string(&expected).expect("serialization");
1695 4 : assert_eq!(
1696 4 : serde_json::from_str::<ImageCompressionAlgorithm>(&ser).unwrap(),
1697 : expected,
1698 0 : "serde roundtrip"
1699 : );
1700 :
1701 4 : assert_eq!(
1702 4 : serde_json::Value::String(display.to_string()),
1703 4 : serde_json::to_value(expected).unwrap(),
1704 0 : "Display is the serde serialization"
1705 : );
1706 : }
1707 1 : }
1708 : }
|
