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 : pub dev_mode: bool,
185 : }
186 :
187 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
188 : pub struct DiskUsageEvictionTaskConfig {
189 : pub max_usage_pct: utils::serde_percent::Percent,
190 : pub min_avail_bytes: u64,
191 : #[serde(with = "humantime_serde")]
192 : pub period: Duration,
193 : #[cfg(feature = "testing")]
194 : pub mock_statvfs: Option<statvfs::mock::Behavior>,
195 : /// Select sorting for evicted layers
196 : #[serde(default)]
197 : pub eviction_order: EvictionOrder,
198 : }
199 :
200 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
201 : #[serde(tag = "mode", rename_all = "kebab-case")]
202 : pub enum PageServicePipeliningConfig {
203 : Serial,
204 : Pipelined(PageServicePipeliningConfigPipelined),
205 : }
206 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
207 : pub struct PageServicePipeliningConfigPipelined {
208 : /// Causes runtime errors if larger than max get_vectored batch size.
209 : pub max_batch_size: NonZeroUsize,
210 : pub execution: PageServiceProtocolPipelinedExecutionStrategy,
211 : // The default below is such that new versions of the software can start
212 : // with the old configuration.
213 : #[serde(default)]
214 : pub batching: PageServiceProtocolPipelinedBatchingStrategy,
215 : }
216 :
217 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
218 : #[serde(rename_all = "kebab-case")]
219 : pub enum PageServiceProtocolPipelinedExecutionStrategy {
220 : ConcurrentFutures,
221 : Tasks,
222 : }
223 :
224 0 : #[derive(Default, Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
225 : #[serde(rename_all = "kebab-case")]
226 : pub enum PageServiceProtocolPipelinedBatchingStrategy {
227 : /// All get page requests in a batch will be at the same LSN
228 : #[default]
229 : UniformLsn,
230 : /// Get page requests in a batch may be at different LSN
231 : ///
232 : /// One key cannot be present more than once at different LSNs in
233 : /// the same batch.
234 : ScatteredLsn,
235 : }
236 :
237 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
238 : #[serde(tag = "mode", rename_all = "kebab-case")]
239 : pub enum GetVectoredConcurrentIo {
240 : /// The read path is fully sequential: layers are visited
241 : /// one after the other and IOs are issued and waited upon
242 : /// from the same task that traverses the layers.
243 : Sequential,
244 : /// The read path still traverses layers sequentially, and
245 : /// index blocks will be read into the PS PageCache from
246 : /// that task, with waiting.
247 : /// But data IOs are dispatched and waited upon from a sidecar
248 : /// task so that the traversing task can continue to traverse
249 : /// layers while the IOs are in flight.
250 : /// If the PS PageCache miss rate is low, this improves
251 : /// throughput dramatically.
252 : SidecarTask,
253 : }
254 :
255 0 : #[derive(Debug, Copy, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
256 : pub struct Ratio {
257 : pub numerator: usize,
258 : pub denominator: usize,
259 : }
260 :
261 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
262 : pub struct OtelExporterConfig {
263 : pub endpoint: String,
264 : pub protocol: OtelExporterProtocol,
265 : #[serde(with = "humantime_serde")]
266 : pub timeout: Duration,
267 : }
268 :
269 0 : #[derive(Debug, Copy, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
270 : #[serde(rename_all = "kebab-case")]
271 : pub enum OtelExporterProtocol {
272 : Grpc,
273 : HttpBinary,
274 : HttpJson,
275 : }
276 :
277 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
278 : pub struct Tracing {
279 : pub sampling_ratio: Ratio,
280 : pub export_config: OtelExporterConfig,
281 : }
282 :
283 : impl From<&OtelExporterConfig> for tracing_utils::ExportConfig {
284 0 : fn from(val: &OtelExporterConfig) -> Self {
285 0 : tracing_utils::ExportConfig {
286 0 : endpoint: Some(val.endpoint.clone()),
287 0 : protocol: val.protocol.into(),
288 0 : timeout: val.timeout,
289 0 : }
290 0 : }
291 : }
292 :
293 : impl From<OtelExporterProtocol> for tracing_utils::Protocol {
294 0 : fn from(val: OtelExporterProtocol) -> Self {
295 0 : match val {
296 0 : OtelExporterProtocol::Grpc => tracing_utils::Protocol::Grpc,
297 0 : OtelExporterProtocol::HttpJson => tracing_utils::Protocol::HttpJson,
298 0 : OtelExporterProtocol::HttpBinary => tracing_utils::Protocol::HttpBinary,
299 : }
300 0 : }
301 : }
302 :
303 : pub mod statvfs {
304 : pub mod mock {
305 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
306 : #[serde(tag = "type")]
307 : pub enum Behavior {
308 : Success {
309 : blocksize: u64,
310 : total_blocks: u64,
311 : name_filter: Option<utils::serde_regex::Regex>,
312 : },
313 : #[cfg(feature = "testing")]
314 : Failure { mocked_error: MockedError },
315 : }
316 :
317 : #[cfg(feature = "testing")]
318 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
319 : #[allow(clippy::upper_case_acronyms)]
320 : pub enum MockedError {
321 : EIO,
322 : }
323 :
324 : #[cfg(feature = "testing")]
325 : impl From<MockedError> for nix::Error {
326 0 : fn from(e: MockedError) -> Self {
327 0 : match e {
328 0 : MockedError::EIO => nix::Error::EIO,
329 0 : }
330 0 : }
331 : }
332 : }
333 : }
334 :
335 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
336 : #[serde(tag = "type", content = "args")]
337 : pub enum EvictionOrder {
338 : RelativeAccessed {
339 : highest_layer_count_loses_first: bool,
340 : },
341 : }
342 :
343 : impl Default for EvictionOrder {
344 12 : fn default() -> Self {
345 12 : Self::RelativeAccessed {
346 12 : highest_layer_count_loses_first: true,
347 12 : }
348 12 : }
349 : }
350 :
351 0 : #[derive(Copy, Clone, Debug, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
352 : #[serde(transparent)]
353 : pub struct MaxVectoredReadBytes(pub NonZeroUsize);
354 :
355 : /// Tenant-level configuration values, used for various purposes.
356 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
357 : #[serde(default)]
358 : pub struct TenantConfigToml {
359 : // Flush out an inmemory layer, if it's holding WAL older than this
360 : // This puts a backstop on how much WAL needs to be re-digested if the
361 : // page server crashes.
362 : // This parameter actually determines L0 layer file size.
363 : pub checkpoint_distance: u64,
364 : // Inmemory layer is also flushed at least once in checkpoint_timeout to
365 : // eventually upload WAL after activity is stopped.
366 : #[serde(with = "humantime_serde")]
367 : pub checkpoint_timeout: Duration,
368 : // Target file size, when creating image and delta layers.
369 : // This parameter determines L1 layer file size.
370 : pub compaction_target_size: u64,
371 : // How often to check if there's compaction work to be done.
372 : // Duration::ZERO means automatic compaction is disabled.
373 : #[serde(with = "humantime_serde")]
374 : pub compaction_period: Duration,
375 : /// Level0 delta layer threshold for compaction.
376 : pub compaction_threshold: usize,
377 : /// Controls the amount of L0 included in a single compaction iteration.
378 : /// The unit is `checkpoint_distance`, i.e., a size.
379 : /// We add L0s to the set of layers to compact until their cumulative
380 : /// size exceeds `compaction_upper_limit * checkpoint_distance`.
381 : pub compaction_upper_limit: usize,
382 : pub compaction_algorithm: crate::models::CompactionAlgorithmSettings,
383 : /// If true, enable shard ancestor compaction (enabled by default).
384 : pub compaction_shard_ancestor: bool,
385 : /// If true, compact down L0 across all tenant timelines before doing regular compaction. L0
386 : /// compaction must be responsive to avoid read amp during heavy ingestion. Defaults to true.
387 : pub compaction_l0_first: bool,
388 : /// If true, use a separate semaphore (i.e. concurrency limit) for the L0 compaction pass. Only
389 : /// has an effect if `compaction_l0_first` is true. Defaults to true.
390 : pub compaction_l0_semaphore: bool,
391 : /// Level0 delta layer threshold at which to delay layer flushes such that they take 2x as long,
392 : /// and block on layer flushes during ephemeral layer rolls, for compaction backpressure. This
393 : /// helps compaction keep up with WAL ingestion, and avoids read amplification blowing up.
394 : /// Should be >compaction_threshold. 0 to disable. Defaults to 3x compaction_threshold.
395 : pub l0_flush_delay_threshold: Option<usize>,
396 : /// Level0 delta layer threshold at which to stall layer flushes. Must be >compaction_threshold
397 : /// to avoid deadlock. 0 to disable. Disabled by default.
398 : pub l0_flush_stall_threshold: Option<usize>,
399 : // Determines how much history is retained, to allow
400 : // branching and read replicas at an older point in time.
401 : // The unit is #of bytes of WAL.
402 : // Page versions older than this are garbage collected away.
403 : pub gc_horizon: u64,
404 : // Interval at which garbage collection is triggered.
405 : // Duration::ZERO means automatic GC is disabled
406 : #[serde(with = "humantime_serde")]
407 : pub gc_period: Duration,
408 : // Delta layer churn threshold to create L1 image layers.
409 : pub image_creation_threshold: usize,
410 : // Determines how much history is retained, to allow
411 : // branching and read replicas at an older point in time.
412 : // The unit is time.
413 : // Page versions older than this are garbage collected away.
414 : #[serde(with = "humantime_serde")]
415 : pub pitr_interval: Duration,
416 : /// Maximum amount of time to wait while opening a connection to receive wal, before erroring.
417 : #[serde(with = "humantime_serde")]
418 : pub walreceiver_connect_timeout: Duration,
419 : /// Considers safekeepers stalled after no WAL updates were received longer than this threshold.
420 : /// A stalled safekeeper will be changed to a newer one when it appears.
421 : #[serde(with = "humantime_serde")]
422 : pub lagging_wal_timeout: Duration,
423 : /// Considers safekeepers lagging when their WAL is behind another safekeeper for more than this threshold.
424 : /// A lagging safekeeper will be changed after `lagging_wal_timeout` time elapses since the last WAL update,
425 : /// to avoid eager reconnects.
426 : pub max_lsn_wal_lag: NonZeroU64,
427 : pub eviction_policy: crate::models::EvictionPolicy,
428 : pub min_resident_size_override: Option<u64>,
429 : // See the corresponding metric's help string.
430 : #[serde(with = "humantime_serde")]
431 : pub evictions_low_residence_duration_metric_threshold: Duration,
432 :
433 : /// If non-zero, the period between uploads of a heatmap from attached tenants. This
434 : /// may be disabled if a Tenant will not have secondary locations: only secondary
435 : /// locations will use the heatmap uploaded by attached locations.
436 : #[serde(with = "humantime_serde")]
437 : pub heatmap_period: Duration,
438 :
439 : /// If true then SLRU segments are dowloaded on demand, if false SLRU segments are included in basebackup
440 : pub lazy_slru_download: bool,
441 :
442 : pub timeline_get_throttle: crate::models::ThrottleConfig,
443 :
444 : // How much WAL must be ingested before checking again whether a new image layer is required.
445 : // Expresed in multiples of checkpoint distance.
446 : pub image_layer_creation_check_threshold: u8,
447 :
448 : // How many multiples of L0 `compaction_threshold` will preempt image layer creation and do L0 compaction.
449 : // Set to 0 to disable preemption.
450 : pub image_creation_preempt_threshold: usize,
451 :
452 : /// The length for an explicit LSN lease request.
453 : /// Layers needed to reconstruct pages at LSN will not be GC-ed during this interval.
454 : #[serde(with = "humantime_serde")]
455 : pub lsn_lease_length: Duration,
456 :
457 : /// The length for an implicit LSN lease granted as part of `get_lsn_by_timestamp` request.
458 : /// Layers needed to reconstruct pages at LSN will not be GC-ed during this interval.
459 : #[serde(with = "humantime_serde")]
460 : pub lsn_lease_length_for_ts: Duration,
461 :
462 : /// Enable auto-offloading of timelines.
463 : /// (either this flag or the pageserver-global one need to be set)
464 : pub timeline_offloading: bool,
465 :
466 : pub wal_receiver_protocol_override: Option<PostgresClientProtocol>,
467 :
468 : /// Enable rel_size_v2 for this tenant. Once enabled, the tenant will persist this information into
469 : /// `index_part.json`, and it cannot be reversed.
470 : pub rel_size_v2_enabled: bool,
471 :
472 : // gc-compaction related configs
473 : /// Enable automatic gc-compaction trigger on this tenant.
474 : pub gc_compaction_enabled: bool,
475 : /// Enable verification of gc-compaction results.
476 : pub gc_compaction_verification: bool,
477 : /// The initial threshold for gc-compaction in KB. Once the total size of layers below the gc-horizon is above this threshold,
478 : /// gc-compaction will be triggered.
479 : pub gc_compaction_initial_threshold_kb: u64,
480 : /// 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)
481 : /// is above this ratio, gc-compaction will be triggered.
482 : pub gc_compaction_ratio_percent: u64,
483 : /// Tenant level performance sampling ratio override. Controls the ratio of get page requests
484 : /// that will get perf sampling for the tenant.
485 : pub sampling_ratio: Option<Ratio>,
486 : }
487 :
488 : pub mod defaults {
489 : pub use storage_broker::DEFAULT_ENDPOINT as BROKER_DEFAULT_ENDPOINT;
490 :
491 : use crate::models::ImageCompressionAlgorithm;
492 :
493 : pub const DEFAULT_WAIT_LSN_TIMEOUT: &str = "300 s";
494 : pub const DEFAULT_WAL_REDO_TIMEOUT: &str = "60 s";
495 :
496 : pub const DEFAULT_SUPERUSER: &str = "cloud_admin";
497 : pub const DEFAULT_LOCALE: &str = if cfg!(target_os = "macos") {
498 : "C"
499 : } else {
500 : "C.UTF-8"
501 : };
502 :
503 : pub const DEFAULT_PAGE_CACHE_SIZE: usize = 8192;
504 : pub const DEFAULT_MAX_FILE_DESCRIPTORS: usize = 100;
505 :
506 : pub const DEFAULT_LOG_FORMAT: &str = "plain";
507 :
508 : pub const DEFAULT_CONCURRENT_TENANT_WARMUP: usize = 8;
509 :
510 : pub const DEFAULT_CONCURRENT_TENANT_SIZE_LOGICAL_SIZE_QUERIES: usize = 1;
511 :
512 : pub const DEFAULT_METRIC_COLLECTION_INTERVAL: &str = "10 min";
513 : pub const DEFAULT_METRIC_COLLECTION_ENDPOINT: Option<reqwest::Url> = None;
514 : pub const DEFAULT_SYNTHETIC_SIZE_CALCULATION_INTERVAL: &str = "10 min";
515 : pub const DEFAULT_BACKGROUND_TASK_MAXIMUM_DELAY: &str = "10s";
516 :
517 : pub const DEFAULT_HEATMAP_UPLOAD_CONCURRENCY: usize = 8;
518 : pub const DEFAULT_SECONDARY_DOWNLOAD_CONCURRENCY: usize = 1;
519 :
520 : pub const DEFAULT_INGEST_BATCH_SIZE: u64 = 100;
521 :
522 : /// Soft limit for the maximum size of a vectored read.
523 : ///
524 : /// This is determined by the largest NeonWalRecord that can exist (minus dbdir and reldir keys
525 : /// which are bounded by the blob io limits only). As of this writing, that is a `NeonWalRecord::ClogSetCommitted` record,
526 : /// with 32k xids. That's the max number of XIDS on a single CLOG page. The size of such a record
527 : /// is `sizeof(Transactionid) * 32768 + (some fixed overhead from 'timestamp`, the Vec length and whatever extra serde serialization adds)`.
528 : /// That is, slightly above 128 kB.
529 : pub const DEFAULT_MAX_VECTORED_READ_BYTES: usize = 130 * 1024; // 130 KiB
530 :
531 : pub const DEFAULT_IMAGE_COMPRESSION: ImageCompressionAlgorithm =
532 : ImageCompressionAlgorithm::Zstd { level: Some(1) };
533 :
534 : pub const DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB: usize = 0;
535 :
536 : pub const DEFAULT_IO_BUFFER_ALIGNMENT: usize = 512;
537 :
538 : pub const DEFAULT_WAL_RECEIVER_PROTOCOL: utils::postgres_client::PostgresClientProtocol =
539 : utils::postgres_client::PostgresClientProtocol::Vanilla;
540 :
541 : pub const DEFAULT_SSL_KEY_FILE: &str = "server.key";
542 : pub const DEFAULT_SSL_CERT_FILE: &str = "server.crt";
543 : }
544 :
545 : impl Default for ConfigToml {
546 1500 : fn default() -> Self {
547 : use defaults::*;
548 :
549 : Self {
550 1500 : listen_pg_addr: (DEFAULT_PG_LISTEN_ADDR.to_string()),
551 1500 : listen_http_addr: (DEFAULT_HTTP_LISTEN_ADDR.to_string()),
552 1500 : listen_https_addr: (None),
553 1500 : ssl_key_file: Utf8PathBuf::from(DEFAULT_SSL_KEY_FILE),
554 1500 : ssl_cert_file: Utf8PathBuf::from(DEFAULT_SSL_CERT_FILE),
555 1500 : ssl_cert_reload_period: Duration::from_secs(60),
556 1500 : ssl_ca_file: None,
557 1500 : availability_zone: (None),
558 1500 : wait_lsn_timeout: (humantime::parse_duration(DEFAULT_WAIT_LSN_TIMEOUT)
559 1500 : .expect("cannot parse default wait lsn timeout")),
560 1500 : wal_redo_timeout: (humantime::parse_duration(DEFAULT_WAL_REDO_TIMEOUT)
561 1500 : .expect("cannot parse default wal redo timeout")),
562 1500 : superuser: (DEFAULT_SUPERUSER.to_string()),
563 1500 : locale: DEFAULT_LOCALE.to_string(),
564 1500 : page_cache_size: (DEFAULT_PAGE_CACHE_SIZE),
565 1500 : max_file_descriptors: (DEFAULT_MAX_FILE_DESCRIPTORS),
566 1500 : pg_distrib_dir: None, // Utf8PathBuf::from("./pg_install"), // TODO: formely, this was std::env::current_dir()
567 1500 : http_auth_type: (AuthType::Trust),
568 1500 : pg_auth_type: (AuthType::Trust),
569 1500 : auth_validation_public_key_path: (None),
570 1500 : remote_storage: None,
571 1500 : broker_endpoint: (storage_broker::DEFAULT_ENDPOINT
572 1500 : .parse()
573 1500 : .expect("failed to parse default broker endpoint")),
574 1500 : broker_keepalive_interval: (humantime::parse_duration(
575 1500 : storage_broker::DEFAULT_KEEPALIVE_INTERVAL,
576 1500 : )
577 1500 : .expect("cannot parse default keepalive interval")),
578 1500 : log_format: (LogFormat::from_str(DEFAULT_LOG_FORMAT).unwrap()),
579 1500 :
580 1500 : concurrent_tenant_warmup: (NonZeroUsize::new(DEFAULT_CONCURRENT_TENANT_WARMUP)
581 1500 : .expect("Invalid default constant")),
582 1500 : concurrent_tenant_size_logical_size_queries: NonZeroUsize::new(
583 1500 : DEFAULT_CONCURRENT_TENANT_SIZE_LOGICAL_SIZE_QUERIES,
584 1500 : )
585 1500 : .unwrap(),
586 1500 : metric_collection_interval: (humantime::parse_duration(
587 1500 : DEFAULT_METRIC_COLLECTION_INTERVAL,
588 1500 : )
589 1500 : .expect("cannot parse default metric collection interval")),
590 1500 : synthetic_size_calculation_interval: (humantime::parse_duration(
591 1500 : DEFAULT_SYNTHETIC_SIZE_CALCULATION_INTERVAL,
592 1500 : )
593 1500 : .expect("cannot parse default synthetic size calculation interval")),
594 1500 : metric_collection_endpoint: (DEFAULT_METRIC_COLLECTION_ENDPOINT),
595 1500 :
596 1500 : metric_collection_bucket: (None),
597 1500 :
598 1500 : disk_usage_based_eviction: (None),
599 1500 :
600 1500 : test_remote_failures: (0),
601 1500 :
602 1500 : ondemand_download_behavior_treat_error_as_warn: (false),
603 1500 :
604 1500 : background_task_maximum_delay: (humantime::parse_duration(
605 1500 : DEFAULT_BACKGROUND_TASK_MAXIMUM_DELAY,
606 1500 : )
607 1500 : .unwrap()),
608 1500 :
609 1500 : control_plane_api: (None),
610 1500 : control_plane_api_token: (None),
611 1500 : control_plane_emergency_mode: (false),
612 1500 :
613 1500 : import_pgdata_upcall_api: (None),
614 1500 : import_pgdata_upcall_api_token: (None),
615 1500 : import_pgdata_aws_endpoint_url: (None),
616 1500 :
617 1500 : heatmap_upload_concurrency: (DEFAULT_HEATMAP_UPLOAD_CONCURRENCY),
618 1500 : secondary_download_concurrency: (DEFAULT_SECONDARY_DOWNLOAD_CONCURRENCY),
619 1500 :
620 1500 : ingest_batch_size: (DEFAULT_INGEST_BATCH_SIZE),
621 1500 :
622 1500 : virtual_file_io_engine: None,
623 1500 :
624 1500 : max_vectored_read_bytes: (MaxVectoredReadBytes(
625 1500 : NonZeroUsize::new(DEFAULT_MAX_VECTORED_READ_BYTES).unwrap(),
626 1500 : )),
627 1500 : image_compression: (DEFAULT_IMAGE_COMPRESSION),
628 1500 : timeline_offloading: true,
629 1500 : ephemeral_bytes_per_memory_kb: (DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB),
630 1500 : l0_flush: None,
631 1500 : virtual_file_io_mode: None,
632 1500 : tenant_config: TenantConfigToml::default(),
633 1500 : no_sync: None,
634 1500 : wal_receiver_protocol: DEFAULT_WAL_RECEIVER_PROTOCOL,
635 1500 : page_service_pipelining: if !cfg!(test) {
636 1500 : PageServicePipeliningConfig::Serial
637 : } else {
638 : // Do not turn this into the default until scattered reads have been
639 : // validated and rolled-out fully.
640 0 : PageServicePipeliningConfig::Pipelined(PageServicePipeliningConfigPipelined {
641 0 : max_batch_size: NonZeroUsize::new(32).unwrap(),
642 0 : execution: PageServiceProtocolPipelinedExecutionStrategy::ConcurrentFutures,
643 0 : batching: PageServiceProtocolPipelinedBatchingStrategy::ScatteredLsn,
644 0 : })
645 : },
646 1500 : get_vectored_concurrent_io: if !cfg!(test) {
647 1500 : GetVectoredConcurrentIo::Sequential
648 : } else {
649 0 : GetVectoredConcurrentIo::SidecarTask
650 : },
651 1500 : enable_read_path_debugging: if cfg!(test) || cfg!(feature = "testing") {
652 1500 : Some(true)
653 : } else {
654 0 : None
655 : },
656 1500 : validate_wal_contiguity: None,
657 1500 : load_previous_heatmap: None,
658 1500 : generate_unarchival_heatmap: None,
659 1500 : tracing: None,
660 1500 : enable_tls_page_service_api: false,
661 1500 : dev_mode: false,
662 1500 : }
663 1500 : }
664 : }
665 :
666 : pub mod tenant_conf_defaults {
667 :
668 : // FIXME: This current value is very low. I would imagine something like 1 GB or 10 GB
669 : // would be more appropriate. But a low value forces the code to be exercised more,
670 : // which is good for now to trigger bugs.
671 : // This parameter actually determines L0 layer file size.
672 : pub const DEFAULT_CHECKPOINT_DISTANCE: u64 = 256 * 1024 * 1024;
673 : pub const DEFAULT_CHECKPOINT_TIMEOUT: &str = "10 m";
674 :
675 : // FIXME the below configs are only used by legacy algorithm. The new algorithm
676 : // has different parameters.
677 :
678 : // Target file size, when creating image and delta layers.
679 : // This parameter determines L1 layer file size.
680 : pub const DEFAULT_COMPACTION_TARGET_SIZE: u64 = 128 * 1024 * 1024;
681 :
682 : pub const DEFAULT_COMPACTION_PERIOD: &str = "20 s";
683 : pub const DEFAULT_COMPACTION_THRESHOLD: usize = 10;
684 : pub const DEFAULT_COMPACTION_SHARD_ANCESTOR: bool = true;
685 :
686 : // This value needs to be tuned to avoid OOM. We have 3/4*CPUs threads for L0 compaction, that's
687 : // 3/4*8=6 on most of our pageservers. Compacting 10 layers requires a maximum of
688 : // DEFAULT_CHECKPOINT_DISTANCE*10 memory, that's 2560MB. So with this config, we can get a maximum peak
689 : // compaction usage of 15360MB.
690 : pub const DEFAULT_COMPACTION_UPPER_LIMIT: usize = 10;
691 : // Enable L0 compaction pass and semaphore by default. L0 compaction must be responsive to avoid
692 : // read amp.
693 : pub const DEFAULT_COMPACTION_L0_FIRST: bool = true;
694 : pub const DEFAULT_COMPACTION_L0_SEMAPHORE: bool = true;
695 :
696 : pub const DEFAULT_COMPACTION_ALGORITHM: crate::models::CompactionAlgorithm =
697 : crate::models::CompactionAlgorithm::Legacy;
698 :
699 : pub const DEFAULT_GC_HORIZON: u64 = 64 * 1024 * 1024;
700 :
701 : // Large DEFAULT_GC_PERIOD is fine as long as PITR_INTERVAL is larger.
702 : // If there's a need to decrease this value, first make sure that GC
703 : // doesn't hold a layer map write lock for non-trivial operations.
704 : // Relevant: https://github.com/neondatabase/neon/issues/3394
705 : pub const DEFAULT_GC_PERIOD: &str = "1 hr";
706 : pub const DEFAULT_IMAGE_CREATION_THRESHOLD: usize = 3;
707 : // Currently, any value other than 0 will trigger image layer creation preemption immediately with L0 backpressure
708 : // without looking at the exact number of L0 layers.
709 : // It was expected to have the following behavior:
710 : // > If there are more than threshold * compaction_threshold (that is 3 * 10 in the default config) L0 layers, image
711 : // > layer creation will end immediately. Set to 0 to disable.
712 : pub const DEFAULT_IMAGE_CREATION_PREEMPT_THRESHOLD: usize = 3;
713 : pub const DEFAULT_PITR_INTERVAL: &str = "7 days";
714 : pub const DEFAULT_WALRECEIVER_CONNECT_TIMEOUT: &str = "10 seconds";
715 : pub const DEFAULT_WALRECEIVER_LAGGING_WAL_TIMEOUT: &str = "10 seconds";
716 : // The default limit on WAL lag should be set to avoid causing disconnects under high throughput
717 : // scenarios: since the broker stats are updated ~1/s, a value of 1GiB should be sufficient for
718 : // throughputs up to 1GiB/s per timeline.
719 : pub const DEFAULT_MAX_WALRECEIVER_LSN_WAL_LAG: u64 = 1024 * 1024 * 1024;
720 : pub const DEFAULT_EVICTIONS_LOW_RESIDENCE_DURATION_METRIC_THRESHOLD: &str = "24 hour";
721 : // By default ingest enough WAL for two new L0 layers before checking if new image
722 : // image layers should be created.
723 : pub const DEFAULT_IMAGE_LAYER_CREATION_CHECK_THRESHOLD: u8 = 2;
724 : pub const DEFAULT_GC_COMPACTION_ENABLED: bool = false;
725 : pub const DEFAULT_GC_COMPACTION_VERIFICATION: bool = true;
726 : pub const DEFAULT_GC_COMPACTION_INITIAL_THRESHOLD_KB: u64 = 5 * 1024 * 1024; // 5GB
727 : pub const DEFAULT_GC_COMPACTION_RATIO_PERCENT: u64 = 100;
728 : }
729 :
730 : impl Default for TenantConfigToml {
731 1500 : fn default() -> Self {
732 : use tenant_conf_defaults::*;
733 1500 : Self {
734 1500 : checkpoint_distance: DEFAULT_CHECKPOINT_DISTANCE,
735 1500 : checkpoint_timeout: humantime::parse_duration(DEFAULT_CHECKPOINT_TIMEOUT)
736 1500 : .expect("cannot parse default checkpoint timeout"),
737 1500 : compaction_target_size: DEFAULT_COMPACTION_TARGET_SIZE,
738 1500 : compaction_period: humantime::parse_duration(DEFAULT_COMPACTION_PERIOD)
739 1500 : .expect("cannot parse default compaction period"),
740 1500 : compaction_threshold: DEFAULT_COMPACTION_THRESHOLD,
741 1500 : compaction_upper_limit: DEFAULT_COMPACTION_UPPER_LIMIT,
742 1500 : compaction_algorithm: crate::models::CompactionAlgorithmSettings {
743 1500 : kind: DEFAULT_COMPACTION_ALGORITHM,
744 1500 : },
745 1500 : compaction_shard_ancestor: DEFAULT_COMPACTION_SHARD_ANCESTOR,
746 1500 : compaction_l0_first: DEFAULT_COMPACTION_L0_FIRST,
747 1500 : compaction_l0_semaphore: DEFAULT_COMPACTION_L0_SEMAPHORE,
748 1500 : l0_flush_delay_threshold: None,
749 1500 : l0_flush_stall_threshold: None,
750 1500 : gc_horizon: DEFAULT_GC_HORIZON,
751 1500 : gc_period: humantime::parse_duration(DEFAULT_GC_PERIOD)
752 1500 : .expect("cannot parse default gc period"),
753 1500 : image_creation_threshold: DEFAULT_IMAGE_CREATION_THRESHOLD,
754 1500 : pitr_interval: humantime::parse_duration(DEFAULT_PITR_INTERVAL)
755 1500 : .expect("cannot parse default PITR interval"),
756 1500 : walreceiver_connect_timeout: humantime::parse_duration(
757 1500 : DEFAULT_WALRECEIVER_CONNECT_TIMEOUT,
758 1500 : )
759 1500 : .expect("cannot parse default walreceiver connect timeout"),
760 1500 : lagging_wal_timeout: humantime::parse_duration(DEFAULT_WALRECEIVER_LAGGING_WAL_TIMEOUT)
761 1500 : .expect("cannot parse default walreceiver lagging wal timeout"),
762 1500 : max_lsn_wal_lag: NonZeroU64::new(DEFAULT_MAX_WALRECEIVER_LSN_WAL_LAG)
763 1500 : .expect("cannot parse default max walreceiver Lsn wal lag"),
764 1500 : eviction_policy: crate::models::EvictionPolicy::NoEviction,
765 1500 : min_resident_size_override: None,
766 1500 : evictions_low_residence_duration_metric_threshold: humantime::parse_duration(
767 1500 : DEFAULT_EVICTIONS_LOW_RESIDENCE_DURATION_METRIC_THRESHOLD,
768 1500 : )
769 1500 : .expect("cannot parse default evictions_low_residence_duration_metric_threshold"),
770 1500 : heatmap_period: Duration::ZERO,
771 1500 : lazy_slru_download: false,
772 1500 : timeline_get_throttle: crate::models::ThrottleConfig::disabled(),
773 1500 : image_layer_creation_check_threshold: DEFAULT_IMAGE_LAYER_CREATION_CHECK_THRESHOLD,
774 1500 : image_creation_preempt_threshold: DEFAULT_IMAGE_CREATION_PREEMPT_THRESHOLD,
775 1500 : lsn_lease_length: LsnLease::DEFAULT_LENGTH,
776 1500 : lsn_lease_length_for_ts: LsnLease::DEFAULT_LENGTH_FOR_TS,
777 1500 : timeline_offloading: true,
778 1500 : wal_receiver_protocol_override: None,
779 1500 : rel_size_v2_enabled: false,
780 1500 : gc_compaction_enabled: DEFAULT_GC_COMPACTION_ENABLED,
781 1500 : gc_compaction_verification: DEFAULT_GC_COMPACTION_VERIFICATION,
782 1500 : gc_compaction_initial_threshold_kb: DEFAULT_GC_COMPACTION_INITIAL_THRESHOLD_KB,
783 1500 : gc_compaction_ratio_percent: DEFAULT_GC_COMPACTION_RATIO_PERCENT,
784 1500 : sampling_ratio: None,
785 1500 : }
786 1500 : }
787 : }
|
