Line data Source code
1 : use camino::Utf8PathBuf;
2 :
3 : #[cfg(test)]
4 : mod tests;
5 :
6 : use const_format::formatcp;
7 : pub const DEFAULT_PG_LISTEN_PORT: u16 = 64000;
8 : pub const DEFAULT_PG_LISTEN_ADDR: &str = formatcp!("127.0.0.1:{DEFAULT_PG_LISTEN_PORT}");
9 : pub const DEFAULT_HTTP_LISTEN_PORT: u16 = 9898;
10 : pub const DEFAULT_HTTP_LISTEN_ADDR: &str = formatcp!("127.0.0.1:{DEFAULT_HTTP_LISTEN_PORT}");
11 :
12 : use std::collections::HashMap;
13 : use std::num::{NonZeroU64, NonZeroUsize};
14 : use std::str::FromStr;
15 : use std::time::Duration;
16 :
17 : use postgres_backend::AuthType;
18 : use remote_storage::RemoteStorageConfig;
19 : use serde_with::serde_as;
20 : use utils::logging::LogFormat;
21 : use utils::postgres_client::PostgresClientProtocol;
22 :
23 : use crate::models::{ImageCompressionAlgorithm, LsnLease};
24 :
25 : // Certain metadata (e.g. externally-addressable name, AZ) is delivered
26 : // as a separate structure. This information is not neeed by the pageserver
27 : // itself, it is only used for registering the pageserver with the control
28 : // plane and/or storage controller.
29 : //
30 9 : #[derive(PartialEq, Eq, Debug, serde::Serialize, serde::Deserialize)]
31 : pub struct NodeMetadata {
32 : #[serde(rename = "host")]
33 : pub postgres_host: String,
34 : #[serde(rename = "port")]
35 : pub postgres_port: u16,
36 : pub http_host: String,
37 : pub http_port: u16,
38 : pub https_port: Option<u16>,
39 :
40 : // Deployment tools may write fields to the metadata file beyond what we
41 : // use in this type: this type intentionally only names fields that require.
42 : #[serde(flatten)]
43 : pub other: HashMap<String, serde_json::Value>,
44 : }
45 :
46 : /// `pageserver.toml`
47 : ///
48 : /// We use serde derive with `#[serde(default)]` to generate a deserializer
49 : /// that fills in the default values for each config field.
50 : ///
51 : /// If there cannot be a static default value because we need to make runtime
52 : /// checks to determine the default, make it an `Option` (which defaults to None).
53 : /// The runtime check should be done in the consuming crate, i.e., `pageserver`.
54 : ///
55 : /// Unknown fields are silently ignored during deserialization.
56 : /// The alternative, which we used in the past, was to set `deny_unknown_fields`,
57 : /// which fails deserialization, and hence pageserver startup, if there is an unknown field.
58 : /// The reason we don't do that anymore is that it complicates
59 : /// usage of config fields for feature flagging, which we commonly do for
60 : /// region-by-region rollouts.
61 : /// The complications mainly arise because the `pageserver.toml` contents on a
62 : /// prod server have a separate lifecycle from the pageserver binary.
63 : /// For instance, `pageserver.toml` contents today are defined in the internal
64 : /// infra repo, and thus introducing a new config field to pageserver and
65 : /// rolling it out to prod servers are separate commits in separate repos
66 : /// that can't be made or rolled back atomically.
67 : /// Rollbacks in particular pose a risk with deny_unknown_fields because
68 : /// the old pageserver binary may reject a new config field, resulting in
69 : /// an outage unless the person doing the pageserver rollback remembers
70 : /// to also revert the commit that added the config field in to the
71 : /// `pageserver.toml` templates in the internal infra repo.
72 : /// (A pre-deploy config check would eliminate this risk during rollbacks,
73 : /// cf [here](https://github.com/neondatabase/cloud/issues/24349).)
74 : /// In addition to this compatibility problem during emergency rollbacks,
75 : /// deny_unknown_fields adds further complications when decomissioning a feature
76 : /// flag: with deny_unknown_fields, we can't remove a flag from the [`ConfigToml`]
77 : /// until all prod servers' `pageserver.toml` files have been updated to a version
78 : /// that doesn't specify the flag. Otherwise new software would fail to start up.
79 : /// This adds the requirement for an intermediate step where the new config field
80 : /// is accepted but ignored, prolonging the decomissioning process by an entire
81 : /// release cycle.
82 : /// By contrast with unknown fields silently ignored, decomissioning a feature
83 : /// flag is a one-step process: we can skip the intermediate step and straight
84 : /// remove the field from the [`ConfigToml`]. We leave the field in the
85 : /// `pageserver.toml` files on prod servers until we reach certainty that we
86 : /// will not roll back to old software whose behavior was dependent on config.
87 : /// Then we can remove the field from the templates in the internal infra repo.
88 : /// This process is [documented internally](
89 : /// https://docs.neon.build/storage/pageserver_configuration.html).
90 : ///
91 : /// Note that above relaxed compatbility for the config format does NOT APPLY
92 : /// TO THE STORAGE FORMAT. As general guidance, when introducing storage format
93 : /// changes, ensure that the potential rollback target version will be compatible
94 : /// with the new format. This must hold regardless of what flags are set in in the `pageserver.toml`:
95 : /// any format version that exists in an environment must be compatible with the software that runs there.
96 : /// Use a pageserver.toml flag only to gate whether software _writes_ the new format.
97 : /// For more compatibility considerations, refer to [internal docs](
98 : /// https://docs.neon.build/storage/compat.html?highlight=compat#format-versions--compatibility)
99 : #[serde_as]
100 0 : #[derive(Clone, Debug, serde::Deserialize, serde::Serialize)]
101 : #[serde(default)]
102 : pub struct ConfigToml {
103 : // types mapped 1:1 into the runtime PageServerConfig type
104 : pub listen_pg_addr: String,
105 : pub listen_http_addr: String,
106 : pub listen_https_addr: Option<String>,
107 : pub ssl_key_file: Utf8PathBuf,
108 : pub ssl_cert_file: Utf8PathBuf,
109 : #[serde(with = "humantime_serde")]
110 : pub ssl_cert_reload_period: Duration,
111 : pub ssl_ca_file: Option<Utf8PathBuf>,
112 : pub availability_zone: Option<String>,
113 : #[serde(with = "humantime_serde")]
114 : pub wait_lsn_timeout: Duration,
115 : #[serde(with = "humantime_serde")]
116 : pub wal_redo_timeout: Duration,
117 : pub superuser: String,
118 : pub locale: String,
119 : pub page_cache_size: usize,
120 : pub max_file_descriptors: usize,
121 : pub pg_distrib_dir: Option<Utf8PathBuf>,
122 : #[serde_as(as = "serde_with::DisplayFromStr")]
123 : pub http_auth_type: AuthType,
124 : #[serde_as(as = "serde_with::DisplayFromStr")]
125 : pub pg_auth_type: AuthType,
126 : pub auth_validation_public_key_path: Option<Utf8PathBuf>,
127 : pub remote_storage: Option<RemoteStorageConfig>,
128 : pub tenant_config: TenantConfigToml,
129 : #[serde_as(as = "serde_with::DisplayFromStr")]
130 : pub broker_endpoint: storage_broker::Uri,
131 : #[serde(with = "humantime_serde")]
132 : pub broker_keepalive_interval: Duration,
133 : #[serde_as(as = "serde_with::DisplayFromStr")]
134 : pub log_format: LogFormat,
135 : pub concurrent_tenant_warmup: NonZeroUsize,
136 : pub concurrent_tenant_size_logical_size_queries: NonZeroUsize,
137 : #[serde(with = "humantime_serde")]
138 : pub metric_collection_interval: Duration,
139 : pub metric_collection_endpoint: Option<reqwest::Url>,
140 : pub metric_collection_bucket: Option<RemoteStorageConfig>,
141 : #[serde(with = "humantime_serde")]
142 : pub synthetic_size_calculation_interval: Duration,
143 : pub disk_usage_based_eviction: Option<DiskUsageEvictionTaskConfig>,
144 : pub test_remote_failures: u64,
145 : pub ondemand_download_behavior_treat_error_as_warn: bool,
146 : #[serde(with = "humantime_serde")]
147 : pub background_task_maximum_delay: Duration,
148 : pub control_plane_api: Option<reqwest::Url>,
149 : pub control_plane_api_token: Option<String>,
150 : pub control_plane_emergency_mode: bool,
151 : /// Unstable feature: subject to change or removal without notice.
152 : /// See <https://github.com/neondatabase/neon/pull/9218>.
153 : pub import_pgdata_upcall_api: Option<reqwest::Url>,
154 : /// Unstable feature: subject to change or removal without notice.
155 : /// See <https://github.com/neondatabase/neon/pull/9218>.
156 : pub import_pgdata_upcall_api_token: Option<String>,
157 : /// Unstable feature: subject to change or removal without notice.
158 : /// See <https://github.com/neondatabase/neon/pull/9218>.
159 : pub import_pgdata_aws_endpoint_url: Option<reqwest::Url>,
160 : pub heatmap_upload_concurrency: usize,
161 : pub secondary_download_concurrency: usize,
162 : pub virtual_file_io_engine: Option<crate::models::virtual_file::IoEngineKind>,
163 : pub ingest_batch_size: u64,
164 : pub max_vectored_read_bytes: MaxVectoredReadBytes,
165 : pub image_compression: ImageCompressionAlgorithm,
166 : pub timeline_offloading: bool,
167 : pub ephemeral_bytes_per_memory_kb: usize,
168 : pub l0_flush: Option<crate::models::L0FlushConfig>,
169 : pub virtual_file_io_mode: Option<crate::models::virtual_file::IoMode>,
170 : #[serde(skip_serializing_if = "Option::is_none")]
171 : pub no_sync: Option<bool>,
172 : pub wal_receiver_protocol: PostgresClientProtocol,
173 : pub page_service_pipelining: PageServicePipeliningConfig,
174 : pub get_vectored_concurrent_io: GetVectoredConcurrentIo,
175 : pub enable_read_path_debugging: Option<bool>,
176 : #[serde(skip_serializing_if = "Option::is_none")]
177 : pub validate_wal_contiguity: Option<bool>,
178 : #[serde(skip_serializing_if = "Option::is_none")]
179 : pub load_previous_heatmap: Option<bool>,
180 : #[serde(skip_serializing_if = "Option::is_none")]
181 : pub generate_unarchival_heatmap: Option<bool>,
182 : pub tracing: Option<Tracing>,
183 : pub enable_tls_page_service_api: bool,
184 : }
185 :
186 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
187 : pub struct DiskUsageEvictionTaskConfig {
188 : pub max_usage_pct: utils::serde_percent::Percent,
189 : pub min_avail_bytes: u64,
190 : #[serde(with = "humantime_serde")]
191 : pub period: Duration,
192 : #[cfg(feature = "testing")]
193 : pub mock_statvfs: Option<statvfs::mock::Behavior>,
194 : /// Select sorting for evicted layers
195 : #[serde(default)]
196 : pub eviction_order: EvictionOrder,
197 : }
198 :
199 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
200 : #[serde(tag = "mode", rename_all = "kebab-case")]
201 : pub enum PageServicePipeliningConfig {
202 : Serial,
203 : Pipelined(PageServicePipeliningConfigPipelined),
204 : }
205 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
206 : pub struct PageServicePipeliningConfigPipelined {
207 : /// Causes runtime errors if larger than max get_vectored batch size.
208 : pub max_batch_size: NonZeroUsize,
209 : pub execution: PageServiceProtocolPipelinedExecutionStrategy,
210 : // The default below is such that new versions of the software can start
211 : // with the old configuration.
212 : #[serde(default)]
213 : pub batching: PageServiceProtocolPipelinedBatchingStrategy,
214 : }
215 :
216 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
217 : #[serde(rename_all = "kebab-case")]
218 : pub enum PageServiceProtocolPipelinedExecutionStrategy {
219 : ConcurrentFutures,
220 : Tasks,
221 : }
222 :
223 0 : #[derive(Default, Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
224 : #[serde(rename_all = "kebab-case")]
225 : pub enum PageServiceProtocolPipelinedBatchingStrategy {
226 : /// All get page requests in a batch will be at the same LSN
227 : #[default]
228 : UniformLsn,
229 : /// Get page requests in a batch may be at different LSN
230 : ///
231 : /// One key cannot be present more than once at different LSNs in
232 : /// the same batch.
233 : ScatteredLsn,
234 : }
235 :
236 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
237 : #[serde(tag = "mode", rename_all = "kebab-case")]
238 : pub enum GetVectoredConcurrentIo {
239 : /// The read path is fully sequential: layers are visited
240 : /// one after the other and IOs are issued and waited upon
241 : /// from the same task that traverses the layers.
242 : Sequential,
243 : /// The read path still traverses layers sequentially, and
244 : /// index blocks will be read into the PS PageCache from
245 : /// that task, with waiting.
246 : /// But data IOs are dispatched and waited upon from a sidecar
247 : /// task so that the traversing task can continue to traverse
248 : /// layers while the IOs are in flight.
249 : /// If the PS PageCache miss rate is low, this improves
250 : /// throughput dramatically.
251 : SidecarTask,
252 : }
253 :
254 0 : #[derive(Debug, Copy, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
255 : pub struct Ratio {
256 : pub numerator: usize,
257 : pub denominator: usize,
258 : }
259 :
260 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
261 : pub struct OtelExporterConfig {
262 : pub endpoint: String,
263 : pub protocol: OtelExporterProtocol,
264 : #[serde(with = "humantime_serde")]
265 : pub timeout: Duration,
266 : }
267 :
268 0 : #[derive(Debug, Copy, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
269 : #[serde(rename_all = "kebab-case")]
270 : pub enum OtelExporterProtocol {
271 : Grpc,
272 : HttpBinary,
273 : HttpJson,
274 : }
275 :
276 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
277 : pub struct Tracing {
278 : pub sampling_ratio: Ratio,
279 : pub export_config: OtelExporterConfig,
280 : }
281 :
282 : impl From<&OtelExporterConfig> for tracing_utils::ExportConfig {
283 0 : fn from(val: &OtelExporterConfig) -> Self {
284 0 : tracing_utils::ExportConfig {
285 0 : endpoint: Some(val.endpoint.clone()),
286 0 : protocol: val.protocol.into(),
287 0 : timeout: val.timeout,
288 0 : }
289 0 : }
290 : }
291 :
292 : impl From<OtelExporterProtocol> for tracing_utils::Protocol {
293 0 : fn from(val: OtelExporterProtocol) -> Self {
294 0 : match val {
295 0 : OtelExporterProtocol::Grpc => tracing_utils::Protocol::Grpc,
296 0 : OtelExporterProtocol::HttpJson => tracing_utils::Protocol::HttpJson,
297 0 : OtelExporterProtocol::HttpBinary => tracing_utils::Protocol::HttpBinary,
298 : }
299 0 : }
300 : }
301 :
302 : pub mod statvfs {
303 : pub mod mock {
304 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
305 : #[serde(tag = "type")]
306 : pub enum Behavior {
307 : Success {
308 : blocksize: u64,
309 : total_blocks: u64,
310 : name_filter: Option<utils::serde_regex::Regex>,
311 : },
312 : #[cfg(feature = "testing")]
313 : Failure { mocked_error: MockedError },
314 : }
315 :
316 : #[cfg(feature = "testing")]
317 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
318 : #[allow(clippy::upper_case_acronyms)]
319 : pub enum MockedError {
320 : EIO,
321 : }
322 :
323 : #[cfg(feature = "testing")]
324 : impl From<MockedError> for nix::Error {
325 0 : fn from(e: MockedError) -> Self {
326 0 : match e {
327 0 : MockedError::EIO => nix::Error::EIO,
328 0 : }
329 0 : }
330 : }
331 : }
332 : }
333 :
334 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
335 : #[serde(tag = "type", content = "args")]
336 : pub enum EvictionOrder {
337 : RelativeAccessed {
338 : highest_layer_count_loses_first: bool,
339 : },
340 : }
341 :
342 : impl Default for EvictionOrder {
343 12 : fn default() -> Self {
344 12 : Self::RelativeAccessed {
345 12 : highest_layer_count_loses_first: true,
346 12 : }
347 12 : }
348 : }
349 :
350 0 : #[derive(Copy, Clone, Debug, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
351 : #[serde(transparent)]
352 : pub struct MaxVectoredReadBytes(pub NonZeroUsize);
353 :
354 : /// Tenant-level configuration values, used for various purposes.
355 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
356 : #[serde(default)]
357 : pub struct TenantConfigToml {
358 : // Flush out an inmemory layer, if it's holding WAL older than this
359 : // This puts a backstop on how much WAL needs to be re-digested if the
360 : // page server crashes.
361 : // This parameter actually determines L0 layer file size.
362 : pub checkpoint_distance: u64,
363 : // Inmemory layer is also flushed at least once in checkpoint_timeout to
364 : // eventually upload WAL after activity is stopped.
365 : #[serde(with = "humantime_serde")]
366 : pub checkpoint_timeout: Duration,
367 : // Target file size, when creating image and delta layers.
368 : // This parameter determines L1 layer file size.
369 : pub compaction_target_size: u64,
370 : // How often to check if there's compaction work to be done.
371 : // Duration::ZERO means automatic compaction is disabled.
372 : #[serde(with = "humantime_serde")]
373 : pub compaction_period: Duration,
374 : /// Level0 delta layer threshold for compaction.
375 : pub compaction_threshold: usize,
376 : /// Controls the amount of L0 included in a single compaction iteration.
377 : /// The unit is `checkpoint_distance`, i.e., a size.
378 : /// We add L0s to the set of layers to compact until their cumulative
379 : /// size exceeds `compaction_upper_limit * checkpoint_distance`.
380 : pub compaction_upper_limit: usize,
381 : pub compaction_algorithm: crate::models::CompactionAlgorithmSettings,
382 : /// If true, enable shard ancestor compaction (enabled by default).
383 : pub compaction_shard_ancestor: bool,
384 : /// If true, compact down L0 across all tenant timelines before doing regular compaction. L0
385 : /// compaction must be responsive to avoid read amp during heavy ingestion. Defaults to true.
386 : pub compaction_l0_first: bool,
387 : /// If true, use a separate semaphore (i.e. concurrency limit) for the L0 compaction pass. Only
388 : /// has an effect if `compaction_l0_first` is true. Defaults to true.
389 : pub compaction_l0_semaphore: bool,
390 : /// Level0 delta layer threshold at which to delay layer flushes such that they take 2x as long,
391 : /// and block on layer flushes during ephemeral layer rolls, for compaction backpressure. This
392 : /// helps compaction keep up with WAL ingestion, and avoids read amplification blowing up.
393 : /// Should be >compaction_threshold. 0 to disable. Defaults to 3x compaction_threshold.
394 : pub l0_flush_delay_threshold: Option<usize>,
395 : /// Level0 delta layer threshold at which to stall layer flushes. Must be >compaction_threshold
396 : /// to avoid deadlock. 0 to disable. Disabled by default.
397 : pub l0_flush_stall_threshold: Option<usize>,
398 : // Determines how much history is retained, to allow
399 : // branching and read replicas at an older point in time.
400 : // The unit is #of bytes of WAL.
401 : // Page versions older than this are garbage collected away.
402 : pub gc_horizon: u64,
403 : // Interval at which garbage collection is triggered.
404 : // Duration::ZERO means automatic GC is disabled
405 : #[serde(with = "humantime_serde")]
406 : pub gc_period: Duration,
407 : // Delta layer churn threshold to create L1 image layers.
408 : pub image_creation_threshold: usize,
409 : // Determines how much history is retained, to allow
410 : // branching and read replicas at an older point in time.
411 : // The unit is time.
412 : // Page versions older than this are garbage collected away.
413 : #[serde(with = "humantime_serde")]
414 : pub pitr_interval: Duration,
415 : /// Maximum amount of time to wait while opening a connection to receive wal, before erroring.
416 : #[serde(with = "humantime_serde")]
417 : pub walreceiver_connect_timeout: Duration,
418 : /// Considers safekeepers stalled after no WAL updates were received longer than this threshold.
419 : /// A stalled safekeeper will be changed to a newer one when it appears.
420 : #[serde(with = "humantime_serde")]
421 : pub lagging_wal_timeout: Duration,
422 : /// Considers safekeepers lagging when their WAL is behind another safekeeper for more than this threshold.
423 : /// A lagging safekeeper will be changed after `lagging_wal_timeout` time elapses since the last WAL update,
424 : /// to avoid eager reconnects.
425 : pub max_lsn_wal_lag: NonZeroU64,
426 : pub eviction_policy: crate::models::EvictionPolicy,
427 : pub min_resident_size_override: Option<u64>,
428 : // See the corresponding metric's help string.
429 : #[serde(with = "humantime_serde")]
430 : pub evictions_low_residence_duration_metric_threshold: Duration,
431 :
432 : /// If non-zero, the period between uploads of a heatmap from attached tenants. This
433 : /// may be disabled if a Tenant will not have secondary locations: only secondary
434 : /// locations will use the heatmap uploaded by attached locations.
435 : #[serde(with = "humantime_serde")]
436 : pub heatmap_period: Duration,
437 :
438 : /// If true then SLRU segments are dowloaded on demand, if false SLRU segments are included in basebackup
439 : pub lazy_slru_download: bool,
440 :
441 : pub timeline_get_throttle: crate::models::ThrottleConfig,
442 :
443 : // How much WAL must be ingested before checking again whether a new image layer is required.
444 : // Expresed in multiples of checkpoint distance.
445 : pub image_layer_creation_check_threshold: u8,
446 :
447 : // How many multiples of L0 `compaction_threshold` will preempt image layer creation and do L0 compaction.
448 : // Set to 0 to disable preemption.
449 : pub image_creation_preempt_threshold: usize,
450 :
451 : /// The length for an explicit LSN lease request.
452 : /// Layers needed to reconstruct pages at LSN will not be GC-ed during this interval.
453 : #[serde(with = "humantime_serde")]
454 : pub lsn_lease_length: Duration,
455 :
456 : /// The length for an implicit LSN lease granted as part of `get_lsn_by_timestamp` request.
457 : /// Layers needed to reconstruct pages at LSN will not be GC-ed during this interval.
458 : #[serde(with = "humantime_serde")]
459 : pub lsn_lease_length_for_ts: Duration,
460 :
461 : /// Enable auto-offloading of timelines.
462 : /// (either this flag or the pageserver-global one need to be set)
463 : pub timeline_offloading: bool,
464 :
465 : pub wal_receiver_protocol_override: Option<PostgresClientProtocol>,
466 :
467 : /// Enable rel_size_v2 for this tenant. Once enabled, the tenant will persist this information into
468 : /// `index_part.json`, and it cannot be reversed.
469 : pub rel_size_v2_enabled: bool,
470 :
471 : // gc-compaction related configs
472 : /// Enable automatic gc-compaction trigger on this tenant.
473 : pub gc_compaction_enabled: bool,
474 : /// Enable verification of gc-compaction results.
475 : pub gc_compaction_verification: bool,
476 : /// The initial threshold for gc-compaction in KB. Once the total size of layers below the gc-horizon is above this threshold,
477 : /// gc-compaction will be triggered.
478 : pub gc_compaction_initial_threshold_kb: u64,
479 : /// The ratio that triggers the auto gc-compaction. If (the total size of layers between L2 LSN and gc-horizon) / (size below the L2 LSN)
480 : /// is above this ratio, gc-compaction will be triggered.
481 : pub gc_compaction_ratio_percent: u64,
482 : /// Tenant level performance sampling ratio override. Controls the ratio of get page requests
483 : /// that will get perf sampling for the tenant.
484 : pub sampling_ratio: Option<Ratio>,
485 : }
486 :
487 : pub mod defaults {
488 : pub use storage_broker::DEFAULT_ENDPOINT as BROKER_DEFAULT_ENDPOINT;
489 :
490 : use crate::models::ImageCompressionAlgorithm;
491 :
492 : pub const DEFAULT_WAIT_LSN_TIMEOUT: &str = "300 s";
493 : pub const DEFAULT_WAL_REDO_TIMEOUT: &str = "60 s";
494 :
495 : pub const DEFAULT_SUPERUSER: &str = "cloud_admin";
496 : pub const DEFAULT_LOCALE: &str = if cfg!(target_os = "macos") {
497 : "C"
498 : } else {
499 : "C.UTF-8"
500 : };
501 :
502 : pub const DEFAULT_PAGE_CACHE_SIZE: usize = 8192;
503 : pub const DEFAULT_MAX_FILE_DESCRIPTORS: usize = 100;
504 :
505 : pub const DEFAULT_LOG_FORMAT: &str = "plain";
506 :
507 : pub const DEFAULT_CONCURRENT_TENANT_WARMUP: usize = 8;
508 :
509 : pub const DEFAULT_CONCURRENT_TENANT_SIZE_LOGICAL_SIZE_QUERIES: usize = 1;
510 :
511 : pub const DEFAULT_METRIC_COLLECTION_INTERVAL: &str = "10 min";
512 : pub const DEFAULT_METRIC_COLLECTION_ENDPOINT: Option<reqwest::Url> = None;
513 : pub const DEFAULT_SYNTHETIC_SIZE_CALCULATION_INTERVAL: &str = "10 min";
514 : pub const DEFAULT_BACKGROUND_TASK_MAXIMUM_DELAY: &str = "10s";
515 :
516 : pub const DEFAULT_HEATMAP_UPLOAD_CONCURRENCY: usize = 8;
517 : pub const DEFAULT_SECONDARY_DOWNLOAD_CONCURRENCY: usize = 1;
518 :
519 : pub const DEFAULT_INGEST_BATCH_SIZE: u64 = 100;
520 :
521 : /// Soft limit for the maximum size of a vectored read.
522 : ///
523 : /// This is determined by the largest NeonWalRecord that can exist (minus dbdir and reldir keys
524 : /// which are bounded by the blob io limits only). As of this writing, that is a `NeonWalRecord::ClogSetCommitted` record,
525 : /// with 32k xids. That's the max number of XIDS on a single CLOG page. The size of such a record
526 : /// is `sizeof(Transactionid) * 32768 + (some fixed overhead from 'timestamp`, the Vec length and whatever extra serde serialization adds)`.
527 : /// That is, slightly above 128 kB.
528 : pub const DEFAULT_MAX_VECTORED_READ_BYTES: usize = 130 * 1024; // 130 KiB
529 :
530 : pub const DEFAULT_IMAGE_COMPRESSION: ImageCompressionAlgorithm =
531 : ImageCompressionAlgorithm::Zstd { level: Some(1) };
532 :
533 : pub const DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB: usize = 0;
534 :
535 : pub const DEFAULT_IO_BUFFER_ALIGNMENT: usize = 512;
536 :
537 : pub const DEFAULT_WAL_RECEIVER_PROTOCOL: utils::postgres_client::PostgresClientProtocol =
538 : utils::postgres_client::PostgresClientProtocol::Vanilla;
539 :
540 : pub const DEFAULT_SSL_KEY_FILE: &str = "server.key";
541 : pub const DEFAULT_SSL_CERT_FILE: &str = "server.crt";
542 : }
543 :
544 : impl Default for ConfigToml {
545 1500 : fn default() -> Self {
546 : use defaults::*;
547 :
548 : Self {
549 1500 : listen_pg_addr: (DEFAULT_PG_LISTEN_ADDR.to_string()),
550 1500 : listen_http_addr: (DEFAULT_HTTP_LISTEN_ADDR.to_string()),
551 1500 : listen_https_addr: (None),
552 1500 : ssl_key_file: Utf8PathBuf::from(DEFAULT_SSL_KEY_FILE),
553 1500 : ssl_cert_file: Utf8PathBuf::from(DEFAULT_SSL_CERT_FILE),
554 1500 : ssl_cert_reload_period: Duration::from_secs(60),
555 1500 : ssl_ca_file: None,
556 1500 : availability_zone: (None),
557 1500 : wait_lsn_timeout: (humantime::parse_duration(DEFAULT_WAIT_LSN_TIMEOUT)
558 1500 : .expect("cannot parse default wait lsn timeout")),
559 1500 : wal_redo_timeout: (humantime::parse_duration(DEFAULT_WAL_REDO_TIMEOUT)
560 1500 : .expect("cannot parse default wal redo timeout")),
561 1500 : superuser: (DEFAULT_SUPERUSER.to_string()),
562 1500 : locale: DEFAULT_LOCALE.to_string(),
563 1500 : page_cache_size: (DEFAULT_PAGE_CACHE_SIZE),
564 1500 : max_file_descriptors: (DEFAULT_MAX_FILE_DESCRIPTORS),
565 1500 : pg_distrib_dir: None, // Utf8PathBuf::from("./pg_install"), // TODO: formely, this was std::env::current_dir()
566 1500 : http_auth_type: (AuthType::Trust),
567 1500 : pg_auth_type: (AuthType::Trust),
568 1500 : auth_validation_public_key_path: (None),
569 1500 : remote_storage: None,
570 1500 : broker_endpoint: (storage_broker::DEFAULT_ENDPOINT
571 1500 : .parse()
572 1500 : .expect("failed to parse default broker endpoint")),
573 1500 : broker_keepalive_interval: (humantime::parse_duration(
574 1500 : storage_broker::DEFAULT_KEEPALIVE_INTERVAL,
575 1500 : )
576 1500 : .expect("cannot parse default keepalive interval")),
577 1500 : log_format: (LogFormat::from_str(DEFAULT_LOG_FORMAT).unwrap()),
578 1500 :
579 1500 : concurrent_tenant_warmup: (NonZeroUsize::new(DEFAULT_CONCURRENT_TENANT_WARMUP)
580 1500 : .expect("Invalid default constant")),
581 1500 : concurrent_tenant_size_logical_size_queries: NonZeroUsize::new(
582 1500 : DEFAULT_CONCURRENT_TENANT_SIZE_LOGICAL_SIZE_QUERIES,
583 1500 : )
584 1500 : .unwrap(),
585 1500 : metric_collection_interval: (humantime::parse_duration(
586 1500 : DEFAULT_METRIC_COLLECTION_INTERVAL,
587 1500 : )
588 1500 : .expect("cannot parse default metric collection interval")),
589 1500 : synthetic_size_calculation_interval: (humantime::parse_duration(
590 1500 : DEFAULT_SYNTHETIC_SIZE_CALCULATION_INTERVAL,
591 1500 : )
592 1500 : .expect("cannot parse default synthetic size calculation interval")),
593 1500 : metric_collection_endpoint: (DEFAULT_METRIC_COLLECTION_ENDPOINT),
594 1500 :
595 1500 : metric_collection_bucket: (None),
596 1500 :
597 1500 : disk_usage_based_eviction: (None),
598 1500 :
599 1500 : test_remote_failures: (0),
600 1500 :
601 1500 : ondemand_download_behavior_treat_error_as_warn: (false),
602 1500 :
603 1500 : background_task_maximum_delay: (humantime::parse_duration(
604 1500 : DEFAULT_BACKGROUND_TASK_MAXIMUM_DELAY,
605 1500 : )
606 1500 : .unwrap()),
607 1500 :
608 1500 : control_plane_api: (None),
609 1500 : control_plane_api_token: (None),
610 1500 : control_plane_emergency_mode: (false),
611 1500 :
612 1500 : import_pgdata_upcall_api: (None),
613 1500 : import_pgdata_upcall_api_token: (None),
614 1500 : import_pgdata_aws_endpoint_url: (None),
615 1500 :
616 1500 : heatmap_upload_concurrency: (DEFAULT_HEATMAP_UPLOAD_CONCURRENCY),
617 1500 : secondary_download_concurrency: (DEFAULT_SECONDARY_DOWNLOAD_CONCURRENCY),
618 1500 :
619 1500 : ingest_batch_size: (DEFAULT_INGEST_BATCH_SIZE),
620 1500 :
621 1500 : virtual_file_io_engine: None,
622 1500 :
623 1500 : max_vectored_read_bytes: (MaxVectoredReadBytes(
624 1500 : NonZeroUsize::new(DEFAULT_MAX_VECTORED_READ_BYTES).unwrap(),
625 1500 : )),
626 1500 : image_compression: (DEFAULT_IMAGE_COMPRESSION),
627 1500 : timeline_offloading: true,
628 1500 : ephemeral_bytes_per_memory_kb: (DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB),
629 1500 : l0_flush: None,
630 1500 : virtual_file_io_mode: None,
631 1500 : tenant_config: TenantConfigToml::default(),
632 1500 : no_sync: None,
633 1500 : wal_receiver_protocol: DEFAULT_WAL_RECEIVER_PROTOCOL,
634 1500 : page_service_pipelining: if !cfg!(test) {
635 1500 : PageServicePipeliningConfig::Serial
636 : } else {
637 : // Do not turn this into the default until scattered reads have been
638 : // validated and rolled-out fully.
639 0 : PageServicePipeliningConfig::Pipelined(PageServicePipeliningConfigPipelined {
640 0 : max_batch_size: NonZeroUsize::new(32).unwrap(),
641 0 : execution: PageServiceProtocolPipelinedExecutionStrategy::ConcurrentFutures,
642 0 : batching: PageServiceProtocolPipelinedBatchingStrategy::ScatteredLsn,
643 0 : })
644 : },
645 1500 : get_vectored_concurrent_io: if !cfg!(test) {
646 1500 : GetVectoredConcurrentIo::Sequential
647 : } else {
648 0 : GetVectoredConcurrentIo::SidecarTask
649 : },
650 1500 : enable_read_path_debugging: if cfg!(test) || cfg!(feature = "testing") {
651 1500 : Some(true)
652 : } else {
653 0 : None
654 : },
655 1500 : validate_wal_contiguity: None,
656 1500 : load_previous_heatmap: None,
657 1500 : generate_unarchival_heatmap: None,
658 1500 : tracing: None,
659 1500 : enable_tls_page_service_api: false,
660 1500 : }
661 1500 : }
662 : }
663 :
664 : pub mod tenant_conf_defaults {
665 :
666 : // FIXME: This current value is very low. I would imagine something like 1 GB or 10 GB
667 : // would be more appropriate. But a low value forces the code to be exercised more,
668 : // which is good for now to trigger bugs.
669 : // This parameter actually determines L0 layer file size.
670 : pub const DEFAULT_CHECKPOINT_DISTANCE: u64 = 256 * 1024 * 1024;
671 : pub const DEFAULT_CHECKPOINT_TIMEOUT: &str = "10 m";
672 :
673 : // FIXME the below configs are only used by legacy algorithm. The new algorithm
674 : // has different parameters.
675 :
676 : // Target file size, when creating image and delta layers.
677 : // This parameter determines L1 layer file size.
678 : pub const DEFAULT_COMPACTION_TARGET_SIZE: u64 = 128 * 1024 * 1024;
679 :
680 : pub const DEFAULT_COMPACTION_PERIOD: &str = "20 s";
681 : pub const DEFAULT_COMPACTION_THRESHOLD: usize = 10;
682 : pub const DEFAULT_COMPACTION_SHARD_ANCESTOR: bool = true;
683 :
684 : // This value needs to be tuned to avoid OOM. We have 3/4*CPUs threads for L0 compaction, that's
685 : // 3/4*8=6 on most of our pageservers. Compacting 10 layers requires a maximum of
686 : // DEFAULT_CHECKPOINT_DISTANCE*10 memory, that's 2560MB. So with this config, we can get a maximum peak
687 : // compaction usage of 15360MB.
688 : pub const DEFAULT_COMPACTION_UPPER_LIMIT: usize = 10;
689 : // Enable L0 compaction pass and semaphore by default. L0 compaction must be responsive to avoid
690 : // read amp.
691 : pub const DEFAULT_COMPACTION_L0_FIRST: bool = true;
692 : pub const DEFAULT_COMPACTION_L0_SEMAPHORE: bool = true;
693 :
694 : pub const DEFAULT_COMPACTION_ALGORITHM: crate::models::CompactionAlgorithm =
695 : crate::models::CompactionAlgorithm::Legacy;
696 :
697 : pub const DEFAULT_GC_HORIZON: u64 = 64 * 1024 * 1024;
698 :
699 : // Large DEFAULT_GC_PERIOD is fine as long as PITR_INTERVAL is larger.
700 : // If there's a need to decrease this value, first make sure that GC
701 : // doesn't hold a layer map write lock for non-trivial operations.
702 : // Relevant: https://github.com/neondatabase/neon/issues/3394
703 : pub const DEFAULT_GC_PERIOD: &str = "1 hr";
704 : pub const DEFAULT_IMAGE_CREATION_THRESHOLD: usize = 3;
705 : // Currently, any value other than 0 will trigger image layer creation preemption immediately with L0 backpressure
706 : // without looking at the exact number of L0 layers.
707 : // It was expected to have the following behavior:
708 : // > If there are more than threshold * compaction_threshold (that is 3 * 10 in the default config) L0 layers, image
709 : // > layer creation will end immediately. Set to 0 to disable.
710 : pub const DEFAULT_IMAGE_CREATION_PREEMPT_THRESHOLD: usize = 3;
711 : pub const DEFAULT_PITR_INTERVAL: &str = "7 days";
712 : pub const DEFAULT_WALRECEIVER_CONNECT_TIMEOUT: &str = "10 seconds";
713 : pub const DEFAULT_WALRECEIVER_LAGGING_WAL_TIMEOUT: &str = "10 seconds";
714 : // The default limit on WAL lag should be set to avoid causing disconnects under high throughput
715 : // scenarios: since the broker stats are updated ~1/s, a value of 1GiB should be sufficient for
716 : // throughputs up to 1GiB/s per timeline.
717 : pub const DEFAULT_MAX_WALRECEIVER_LSN_WAL_LAG: u64 = 1024 * 1024 * 1024;
718 : pub const DEFAULT_EVICTIONS_LOW_RESIDENCE_DURATION_METRIC_THRESHOLD: &str = "24 hour";
719 : // By default ingest enough WAL for two new L0 layers before checking if new image
720 : // image layers should be created.
721 : pub const DEFAULT_IMAGE_LAYER_CREATION_CHECK_THRESHOLD: u8 = 2;
722 : pub const DEFAULT_GC_COMPACTION_ENABLED: bool = false;
723 : pub const DEFAULT_GC_COMPACTION_VERIFICATION: bool = true;
724 : pub const DEFAULT_GC_COMPACTION_INITIAL_THRESHOLD_KB: u64 = 5 * 1024 * 1024; // 5GB
725 : pub const DEFAULT_GC_COMPACTION_RATIO_PERCENT: u64 = 100;
726 : }
727 :
728 : impl Default for TenantConfigToml {
729 1500 : fn default() -> Self {
730 : use tenant_conf_defaults::*;
731 1500 : Self {
732 1500 : checkpoint_distance: DEFAULT_CHECKPOINT_DISTANCE,
733 1500 : checkpoint_timeout: humantime::parse_duration(DEFAULT_CHECKPOINT_TIMEOUT)
734 1500 : .expect("cannot parse default checkpoint timeout"),
735 1500 : compaction_target_size: DEFAULT_COMPACTION_TARGET_SIZE,
736 1500 : compaction_period: humantime::parse_duration(DEFAULT_COMPACTION_PERIOD)
737 1500 : .expect("cannot parse default compaction period"),
738 1500 : compaction_threshold: DEFAULT_COMPACTION_THRESHOLD,
739 1500 : compaction_upper_limit: DEFAULT_COMPACTION_UPPER_LIMIT,
740 1500 : compaction_algorithm: crate::models::CompactionAlgorithmSettings {
741 1500 : kind: DEFAULT_COMPACTION_ALGORITHM,
742 1500 : },
743 1500 : compaction_shard_ancestor: DEFAULT_COMPACTION_SHARD_ANCESTOR,
744 1500 : compaction_l0_first: DEFAULT_COMPACTION_L0_FIRST,
745 1500 : compaction_l0_semaphore: DEFAULT_COMPACTION_L0_SEMAPHORE,
746 1500 : l0_flush_delay_threshold: None,
747 1500 : l0_flush_stall_threshold: None,
748 1500 : gc_horizon: DEFAULT_GC_HORIZON,
749 1500 : gc_period: humantime::parse_duration(DEFAULT_GC_PERIOD)
750 1500 : .expect("cannot parse default gc period"),
751 1500 : image_creation_threshold: DEFAULT_IMAGE_CREATION_THRESHOLD,
752 1500 : pitr_interval: humantime::parse_duration(DEFAULT_PITR_INTERVAL)
753 1500 : .expect("cannot parse default PITR interval"),
754 1500 : walreceiver_connect_timeout: humantime::parse_duration(
755 1500 : DEFAULT_WALRECEIVER_CONNECT_TIMEOUT,
756 1500 : )
757 1500 : .expect("cannot parse default walreceiver connect timeout"),
758 1500 : lagging_wal_timeout: humantime::parse_duration(DEFAULT_WALRECEIVER_LAGGING_WAL_TIMEOUT)
759 1500 : .expect("cannot parse default walreceiver lagging wal timeout"),
760 1500 : max_lsn_wal_lag: NonZeroU64::new(DEFAULT_MAX_WALRECEIVER_LSN_WAL_LAG)
761 1500 : .expect("cannot parse default max walreceiver Lsn wal lag"),
762 1500 : eviction_policy: crate::models::EvictionPolicy::NoEviction,
763 1500 : min_resident_size_override: None,
764 1500 : evictions_low_residence_duration_metric_threshold: humantime::parse_duration(
765 1500 : DEFAULT_EVICTIONS_LOW_RESIDENCE_DURATION_METRIC_THRESHOLD,
766 1500 : )
767 1500 : .expect("cannot parse default evictions_low_residence_duration_metric_threshold"),
768 1500 : heatmap_period: Duration::ZERO,
769 1500 : lazy_slru_download: false,
770 1500 : timeline_get_throttle: crate::models::ThrottleConfig::disabled(),
771 1500 : image_layer_creation_check_threshold: DEFAULT_IMAGE_LAYER_CREATION_CHECK_THRESHOLD,
772 1500 : image_creation_preempt_threshold: DEFAULT_IMAGE_CREATION_PREEMPT_THRESHOLD,
773 1500 : lsn_lease_length: LsnLease::DEFAULT_LENGTH,
774 1500 : lsn_lease_length_for_ts: LsnLease::DEFAULT_LENGTH_FOR_TS,
775 1500 : timeline_offloading: true,
776 1500 : wal_receiver_protocol_override: None,
777 1500 : rel_size_v2_enabled: false,
778 1500 : gc_compaction_enabled: DEFAULT_GC_COMPACTION_ENABLED,
779 1500 : gc_compaction_verification: DEFAULT_GC_COMPACTION_VERIFICATION,
780 1500 : gc_compaction_initial_threshold_kb: DEFAULT_GC_COMPACTION_INITIAL_THRESHOLD_KB,
781 1500 : gc_compaction_ratio_percent: DEFAULT_GC_COMPACTION_RATIO_PERCENT,
782 1500 : sampling_ratio: None,
783 1500 : }
784 1500 : }
785 : }
|
