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 : #[serde_as]
55 20 : #[derive(Clone, Debug, serde::Deserialize, serde::Serialize)]
56 : #[serde(default, deny_unknown_fields)]
57 : pub struct ConfigToml {
58 : // types mapped 1:1 into the runtime PageServerConfig type
59 : pub listen_pg_addr: String,
60 : pub listen_http_addr: String,
61 : pub listen_https_addr: Option<String>,
62 : pub ssl_key_file: Utf8PathBuf,
63 : pub ssl_cert_file: Utf8PathBuf,
64 : #[serde(with = "humantime_serde")]
65 : pub ssl_cert_reload_period: Duration,
66 : pub ssl_ca_file: Option<Utf8PathBuf>,
67 : pub availability_zone: Option<String>,
68 : #[serde(with = "humantime_serde")]
69 : pub wait_lsn_timeout: Duration,
70 : #[serde(with = "humantime_serde")]
71 : pub wal_redo_timeout: Duration,
72 : pub superuser: String,
73 : pub locale: String,
74 : pub page_cache_size: usize,
75 : pub max_file_descriptors: usize,
76 : pub pg_distrib_dir: Option<Utf8PathBuf>,
77 : #[serde_as(as = "serde_with::DisplayFromStr")]
78 : pub http_auth_type: AuthType,
79 : #[serde_as(as = "serde_with::DisplayFromStr")]
80 : pub pg_auth_type: AuthType,
81 : pub auth_validation_public_key_path: Option<Utf8PathBuf>,
82 : pub remote_storage: Option<RemoteStorageConfig>,
83 : pub tenant_config: TenantConfigToml,
84 : #[serde_as(as = "serde_with::DisplayFromStr")]
85 : pub broker_endpoint: storage_broker::Uri,
86 : #[serde(with = "humantime_serde")]
87 : pub broker_keepalive_interval: Duration,
88 : #[serde_as(as = "serde_with::DisplayFromStr")]
89 : pub log_format: LogFormat,
90 : pub concurrent_tenant_warmup: NonZeroUsize,
91 : pub concurrent_tenant_size_logical_size_queries: NonZeroUsize,
92 : #[serde(with = "humantime_serde")]
93 : pub metric_collection_interval: Duration,
94 : pub metric_collection_endpoint: Option<reqwest::Url>,
95 : pub metric_collection_bucket: Option<RemoteStorageConfig>,
96 : #[serde(with = "humantime_serde")]
97 : pub synthetic_size_calculation_interval: Duration,
98 : pub disk_usage_based_eviction: Option<DiskUsageEvictionTaskConfig>,
99 : pub test_remote_failures: u64,
100 : pub ondemand_download_behavior_treat_error_as_warn: bool,
101 : #[serde(with = "humantime_serde")]
102 : pub background_task_maximum_delay: Duration,
103 : pub control_plane_api: Option<reqwest::Url>,
104 : pub control_plane_api_token: Option<String>,
105 : pub control_plane_emergency_mode: bool,
106 : /// Unstable feature: subject to change or removal without notice.
107 : /// See <https://github.com/neondatabase/neon/pull/9218>.
108 : pub import_pgdata_upcall_api: Option<reqwest::Url>,
109 : /// Unstable feature: subject to change or removal without notice.
110 : /// See <https://github.com/neondatabase/neon/pull/9218>.
111 : pub import_pgdata_upcall_api_token: Option<String>,
112 : /// Unstable feature: subject to change or removal without notice.
113 : /// See <https://github.com/neondatabase/neon/pull/9218>.
114 : pub import_pgdata_aws_endpoint_url: Option<reqwest::Url>,
115 : pub heatmap_upload_concurrency: usize,
116 : pub secondary_download_concurrency: usize,
117 : pub virtual_file_io_engine: Option<crate::models::virtual_file::IoEngineKind>,
118 : pub ingest_batch_size: u64,
119 : pub max_vectored_read_bytes: MaxVectoredReadBytes,
120 : pub image_compression: ImageCompressionAlgorithm,
121 : pub timeline_offloading: bool,
122 : pub ephemeral_bytes_per_memory_kb: usize,
123 : pub l0_flush: Option<crate::models::L0FlushConfig>,
124 : pub virtual_file_io_mode: Option<crate::models::virtual_file::IoMode>,
125 : #[serde(skip_serializing_if = "Option::is_none")]
126 : pub no_sync: Option<bool>,
127 : pub wal_receiver_protocol: PostgresClientProtocol,
128 : pub page_service_pipelining: PageServicePipeliningConfig,
129 : pub get_vectored_concurrent_io: GetVectoredConcurrentIo,
130 : pub enable_read_path_debugging: Option<bool>,
131 : #[serde(skip_serializing_if = "Option::is_none")]
132 : pub validate_wal_contiguity: Option<bool>,
133 : #[serde(skip_serializing_if = "Option::is_none")]
134 : pub load_previous_heatmap: Option<bool>,
135 : #[serde(skip_serializing_if = "Option::is_none")]
136 : pub generate_unarchival_heatmap: Option<bool>,
137 : }
138 :
139 4 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
140 : #[serde(deny_unknown_fields)]
141 : pub struct DiskUsageEvictionTaskConfig {
142 : pub max_usage_pct: utils::serde_percent::Percent,
143 : pub min_avail_bytes: u64,
144 : #[serde(with = "humantime_serde")]
145 : pub period: Duration,
146 : #[cfg(feature = "testing")]
147 : pub mock_statvfs: Option<statvfs::mock::Behavior>,
148 : /// Select sorting for evicted layers
149 : #[serde(default)]
150 : pub eviction_order: EvictionOrder,
151 : }
152 :
153 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
154 : #[serde(tag = "mode", rename_all = "kebab-case")]
155 : #[serde(deny_unknown_fields)]
156 : pub enum PageServicePipeliningConfig {
157 : Serial,
158 : Pipelined(PageServicePipeliningConfigPipelined),
159 : }
160 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
161 : #[serde(deny_unknown_fields)]
162 : pub struct PageServicePipeliningConfigPipelined {
163 : /// Causes runtime errors if larger than max get_vectored batch size.
164 : pub max_batch_size: NonZeroUsize,
165 : pub execution: PageServiceProtocolPipelinedExecutionStrategy,
166 : }
167 :
168 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
169 : #[serde(rename_all = "kebab-case")]
170 : pub enum PageServiceProtocolPipelinedExecutionStrategy {
171 : ConcurrentFutures,
172 : Tasks,
173 : }
174 :
175 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
176 : #[serde(tag = "mode", rename_all = "kebab-case")]
177 : #[serde(deny_unknown_fields)]
178 : pub enum GetVectoredConcurrentIo {
179 : /// The read path is fully sequential: layers are visited
180 : /// one after the other and IOs are issued and waited upon
181 : /// from the same task that traverses the layers.
182 : Sequential,
183 : /// The read path still traverses layers sequentially, and
184 : /// index blocks will be read into the PS PageCache from
185 : /// that task, with waiting.
186 : /// But data IOs are dispatched and waited upon from a sidecar
187 : /// task so that the traversing task can continue to traverse
188 : /// layers while the IOs are in flight.
189 : /// If the PS PageCache miss rate is low, this improves
190 : /// throughput dramatically.
191 : SidecarTask,
192 : }
193 :
194 : pub mod statvfs {
195 : pub mod mock {
196 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
197 : #[serde(tag = "type")]
198 : pub enum Behavior {
199 : Success {
200 : blocksize: u64,
201 : total_blocks: u64,
202 : name_filter: Option<utils::serde_regex::Regex>,
203 : },
204 : #[cfg(feature = "testing")]
205 : Failure { mocked_error: MockedError },
206 : }
207 :
208 : #[cfg(feature = "testing")]
209 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
210 : #[allow(clippy::upper_case_acronyms)]
211 : pub enum MockedError {
212 : EIO,
213 : }
214 :
215 : #[cfg(feature = "testing")]
216 : impl From<MockedError> for nix::Error {
217 0 : fn from(e: MockedError) -> Self {
218 0 : match e {
219 0 : MockedError::EIO => nix::Error::EIO,
220 0 : }
221 0 : }
222 : }
223 : }
224 : }
225 :
226 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
227 : #[serde(tag = "type", content = "args")]
228 : pub enum EvictionOrder {
229 : RelativeAccessed {
230 : highest_layer_count_loses_first: bool,
231 : },
232 : }
233 :
234 : impl Default for EvictionOrder {
235 4 : fn default() -> Self {
236 4 : Self::RelativeAccessed {
237 4 : highest_layer_count_loses_first: true,
238 4 : }
239 4 : }
240 : }
241 :
242 0 : #[derive(Copy, Clone, Debug, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
243 : #[serde(transparent)]
244 : pub struct MaxVectoredReadBytes(pub NonZeroUsize);
245 :
246 : /// Tenant-level configuration values, used for various purposes.
247 4 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
248 : #[serde(deny_unknown_fields, default)]
249 : pub struct TenantConfigToml {
250 : // Flush out an inmemory layer, if it's holding WAL older than this
251 : // This puts a backstop on how much WAL needs to be re-digested if the
252 : // page server crashes.
253 : // This parameter actually determines L0 layer file size.
254 : pub checkpoint_distance: u64,
255 : // Inmemory layer is also flushed at least once in checkpoint_timeout to
256 : // eventually upload WAL after activity is stopped.
257 : #[serde(with = "humantime_serde")]
258 : pub checkpoint_timeout: Duration,
259 : // Target file size, when creating image and delta layers.
260 : // This parameter determines L1 layer file size.
261 : pub compaction_target_size: u64,
262 : // How often to check if there's compaction work to be done.
263 : // Duration::ZERO means automatic compaction is disabled.
264 : #[serde(with = "humantime_serde")]
265 : pub compaction_period: Duration,
266 : /// Level0 delta layer threshold for compaction.
267 : pub compaction_threshold: usize,
268 : /// Controls the amount of L0 included in a single compaction iteration.
269 : /// The unit is `checkpoint_distance`, i.e., a size.
270 : /// We add L0s to the set of layers to compact until their cumulative
271 : /// size exceeds `compaction_upper_limit * checkpoint_distance`.
272 : pub compaction_upper_limit: usize,
273 : pub compaction_algorithm: crate::models::CompactionAlgorithmSettings,
274 : /// If true, compact down L0 across all tenant timelines before doing regular compaction. L0
275 : /// compaction must be responsive to avoid read amp during heavy ingestion. Defaults to true.
276 : pub compaction_l0_first: bool,
277 : /// If true, use a separate semaphore (i.e. concurrency limit) for the L0 compaction pass. Only
278 : /// has an effect if `compaction_l0_first` is true. Defaults to true.
279 : pub compaction_l0_semaphore: bool,
280 : /// Level0 delta layer threshold at which to delay layer flushes such that they take 2x as long,
281 : /// and block on layer flushes during ephemeral layer rolls, for compaction backpressure. This
282 : /// helps compaction keep up with WAL ingestion, and avoids read amplification blowing up.
283 : /// Should be >compaction_threshold. 0 to disable. Defaults to 3x compaction_threshold.
284 : pub l0_flush_delay_threshold: Option<usize>,
285 : /// Level0 delta layer threshold at which to stall layer flushes. Must be >compaction_threshold
286 : /// to avoid deadlock. 0 to disable. Disabled by default.
287 : pub l0_flush_stall_threshold: Option<usize>,
288 : // Determines how much history is retained, to allow
289 : // branching and read replicas at an older point in time.
290 : // The unit is #of bytes of WAL.
291 : // Page versions older than this are garbage collected away.
292 : pub gc_horizon: u64,
293 : // Interval at which garbage collection is triggered.
294 : // Duration::ZERO means automatic GC is disabled
295 : #[serde(with = "humantime_serde")]
296 : pub gc_period: Duration,
297 : // Delta layer churn threshold to create L1 image layers.
298 : pub image_creation_threshold: usize,
299 : // Determines how much history is retained, to allow
300 : // branching and read replicas at an older point in time.
301 : // The unit is time.
302 : // Page versions older than this are garbage collected away.
303 : #[serde(with = "humantime_serde")]
304 : pub pitr_interval: Duration,
305 : /// Maximum amount of time to wait while opening a connection to receive wal, before erroring.
306 : #[serde(with = "humantime_serde")]
307 : pub walreceiver_connect_timeout: Duration,
308 : /// Considers safekeepers stalled after no WAL updates were received longer than this threshold.
309 : /// A stalled safekeeper will be changed to a newer one when it appears.
310 : #[serde(with = "humantime_serde")]
311 : pub lagging_wal_timeout: Duration,
312 : /// Considers safekeepers lagging when their WAL is behind another safekeeper for more than this threshold.
313 : /// A lagging safekeeper will be changed after `lagging_wal_timeout` time elapses since the last WAL update,
314 : /// to avoid eager reconnects.
315 : pub max_lsn_wal_lag: NonZeroU64,
316 : pub eviction_policy: crate::models::EvictionPolicy,
317 : pub min_resident_size_override: Option<u64>,
318 : // See the corresponding metric's help string.
319 : #[serde(with = "humantime_serde")]
320 : pub evictions_low_residence_duration_metric_threshold: Duration,
321 :
322 : /// If non-zero, the period between uploads of a heatmap from attached tenants. This
323 : /// may be disabled if a Tenant will not have secondary locations: only secondary
324 : /// locations will use the heatmap uploaded by attached locations.
325 : #[serde(with = "humantime_serde")]
326 : pub heatmap_period: Duration,
327 :
328 : /// If true then SLRU segments are dowloaded on demand, if false SLRU segments are included in basebackup
329 : pub lazy_slru_download: bool,
330 :
331 : pub timeline_get_throttle: crate::models::ThrottleConfig,
332 :
333 : // How much WAL must be ingested before checking again whether a new image layer is required.
334 : // Expresed in multiples of checkpoint distance.
335 : pub image_layer_creation_check_threshold: u8,
336 :
337 : // How many multiples of L0 `compaction_threshold` will preempt image layer creation and do L0 compaction.
338 : // Set to 0 to disable preemption.
339 : pub image_creation_preempt_threshold: usize,
340 :
341 : /// The length for an explicit LSN lease request.
342 : /// Layers needed to reconstruct pages at LSN will not be GC-ed during this interval.
343 : #[serde(with = "humantime_serde")]
344 : pub lsn_lease_length: Duration,
345 :
346 : /// The length for an implicit LSN lease granted as part of `get_lsn_by_timestamp` request.
347 : /// Layers needed to reconstruct pages at LSN will not be GC-ed during this interval.
348 : #[serde(with = "humantime_serde")]
349 : pub lsn_lease_length_for_ts: Duration,
350 :
351 : /// Enable auto-offloading of timelines.
352 : /// (either this flag or the pageserver-global one need to be set)
353 : pub timeline_offloading: bool,
354 :
355 : pub wal_receiver_protocol_override: Option<PostgresClientProtocol>,
356 :
357 : /// Enable rel_size_v2 for this tenant. Once enabled, the tenant will persist this information into
358 : /// `index_part.json`, and it cannot be reversed.
359 : pub rel_size_v2_enabled: bool,
360 :
361 : // gc-compaction related configs
362 : /// Enable automatic gc-compaction trigger on this tenant.
363 : pub gc_compaction_enabled: bool,
364 : /// The initial threshold for gc-compaction in KB. Once the total size of layers below the gc-horizon is above this threshold,
365 : /// gc-compaction will be triggered.
366 : pub gc_compaction_initial_threshold_kb: u64,
367 : /// 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)
368 : /// is above this ratio, gc-compaction will be triggered.
369 : pub gc_compaction_ratio_percent: u64,
370 : }
371 :
372 : pub mod defaults {
373 : pub use storage_broker::DEFAULT_ENDPOINT as BROKER_DEFAULT_ENDPOINT;
374 :
375 : use crate::models::ImageCompressionAlgorithm;
376 :
377 : pub const DEFAULT_WAIT_LSN_TIMEOUT: &str = "300 s";
378 : pub const DEFAULT_WAL_REDO_TIMEOUT: &str = "60 s";
379 :
380 : pub const DEFAULT_SUPERUSER: &str = "cloud_admin";
381 : pub const DEFAULT_LOCALE: &str = if cfg!(target_os = "macos") {
382 : "C"
383 : } else {
384 : "C.UTF-8"
385 : };
386 :
387 : pub const DEFAULT_PAGE_CACHE_SIZE: usize = 8192;
388 : pub const DEFAULT_MAX_FILE_DESCRIPTORS: usize = 100;
389 :
390 : pub const DEFAULT_LOG_FORMAT: &str = "plain";
391 :
392 : pub const DEFAULT_CONCURRENT_TENANT_WARMUP: usize = 8;
393 :
394 : pub const DEFAULT_CONCURRENT_TENANT_SIZE_LOGICAL_SIZE_QUERIES: usize = 1;
395 :
396 : pub const DEFAULT_METRIC_COLLECTION_INTERVAL: &str = "10 min";
397 : pub const DEFAULT_METRIC_COLLECTION_ENDPOINT: Option<reqwest::Url> = None;
398 : pub const DEFAULT_SYNTHETIC_SIZE_CALCULATION_INTERVAL: &str = "10 min";
399 : pub const DEFAULT_BACKGROUND_TASK_MAXIMUM_DELAY: &str = "10s";
400 :
401 : pub const DEFAULT_HEATMAP_UPLOAD_CONCURRENCY: usize = 8;
402 : pub const DEFAULT_SECONDARY_DOWNLOAD_CONCURRENCY: usize = 1;
403 :
404 : pub const DEFAULT_INGEST_BATCH_SIZE: u64 = 100;
405 :
406 : /// Soft limit for the maximum size of a vectored read.
407 : ///
408 : /// This is determined by the largest NeonWalRecord that can exist (minus dbdir and reldir keys
409 : /// which are bounded by the blob io limits only). As of this writing, that is a `NeonWalRecord::ClogSetCommitted` record,
410 : /// with 32k xids. That's the max number of XIDS on a single CLOG page. The size of such a record
411 : /// is `sizeof(Transactionid) * 32768 + (some fixed overhead from 'timestamp`, the Vec length and whatever extra serde serialization adds)`.
412 : /// That is, slightly above 128 kB.
413 : pub const DEFAULT_MAX_VECTORED_READ_BYTES: usize = 130 * 1024; // 130 KiB
414 :
415 : pub const DEFAULT_IMAGE_COMPRESSION: ImageCompressionAlgorithm =
416 : ImageCompressionAlgorithm::Zstd { level: Some(1) };
417 :
418 : pub const DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB: usize = 0;
419 :
420 : pub const DEFAULT_IO_BUFFER_ALIGNMENT: usize = 512;
421 :
422 : pub const DEFAULT_WAL_RECEIVER_PROTOCOL: utils::postgres_client::PostgresClientProtocol =
423 : utils::postgres_client::PostgresClientProtocol::Vanilla;
424 :
425 : pub const DEFAULT_SSL_KEY_FILE: &str = "server.key";
426 : pub const DEFAULT_SSL_CERT_FILE: &str = "server.crt";
427 : }
428 :
429 : impl Default for ConfigToml {
430 488 : fn default() -> Self {
431 : use defaults::*;
432 :
433 : Self {
434 488 : listen_pg_addr: (DEFAULT_PG_LISTEN_ADDR.to_string()),
435 488 : listen_http_addr: (DEFAULT_HTTP_LISTEN_ADDR.to_string()),
436 488 : listen_https_addr: (None),
437 488 : ssl_key_file: Utf8PathBuf::from(DEFAULT_SSL_KEY_FILE),
438 488 : ssl_cert_file: Utf8PathBuf::from(DEFAULT_SSL_CERT_FILE),
439 488 : ssl_cert_reload_period: Duration::from_secs(60),
440 488 : ssl_ca_file: None,
441 488 : availability_zone: (None),
442 488 : wait_lsn_timeout: (humantime::parse_duration(DEFAULT_WAIT_LSN_TIMEOUT)
443 488 : .expect("cannot parse default wait lsn timeout")),
444 488 : wal_redo_timeout: (humantime::parse_duration(DEFAULT_WAL_REDO_TIMEOUT)
445 488 : .expect("cannot parse default wal redo timeout")),
446 488 : superuser: (DEFAULT_SUPERUSER.to_string()),
447 488 : locale: DEFAULT_LOCALE.to_string(),
448 488 : page_cache_size: (DEFAULT_PAGE_CACHE_SIZE),
449 488 : max_file_descriptors: (DEFAULT_MAX_FILE_DESCRIPTORS),
450 488 : pg_distrib_dir: None, // Utf8PathBuf::from("./pg_install"), // TODO: formely, this was std::env::current_dir()
451 488 : http_auth_type: (AuthType::Trust),
452 488 : pg_auth_type: (AuthType::Trust),
453 488 : auth_validation_public_key_path: (None),
454 488 : remote_storage: None,
455 488 : broker_endpoint: (storage_broker::DEFAULT_ENDPOINT
456 488 : .parse()
457 488 : .expect("failed to parse default broker endpoint")),
458 488 : broker_keepalive_interval: (humantime::parse_duration(
459 488 : storage_broker::DEFAULT_KEEPALIVE_INTERVAL,
460 488 : )
461 488 : .expect("cannot parse default keepalive interval")),
462 488 : log_format: (LogFormat::from_str(DEFAULT_LOG_FORMAT).unwrap()),
463 488 :
464 488 : concurrent_tenant_warmup: (NonZeroUsize::new(DEFAULT_CONCURRENT_TENANT_WARMUP)
465 488 : .expect("Invalid default constant")),
466 488 : concurrent_tenant_size_logical_size_queries: NonZeroUsize::new(
467 488 : DEFAULT_CONCURRENT_TENANT_SIZE_LOGICAL_SIZE_QUERIES,
468 488 : )
469 488 : .unwrap(),
470 488 : metric_collection_interval: (humantime::parse_duration(
471 488 : DEFAULT_METRIC_COLLECTION_INTERVAL,
472 488 : )
473 488 : .expect("cannot parse default metric collection interval")),
474 488 : synthetic_size_calculation_interval: (humantime::parse_duration(
475 488 : DEFAULT_SYNTHETIC_SIZE_CALCULATION_INTERVAL,
476 488 : )
477 488 : .expect("cannot parse default synthetic size calculation interval")),
478 488 : metric_collection_endpoint: (DEFAULT_METRIC_COLLECTION_ENDPOINT),
479 488 :
480 488 : metric_collection_bucket: (None),
481 488 :
482 488 : disk_usage_based_eviction: (None),
483 488 :
484 488 : test_remote_failures: (0),
485 488 :
486 488 : ondemand_download_behavior_treat_error_as_warn: (false),
487 488 :
488 488 : background_task_maximum_delay: (humantime::parse_duration(
489 488 : DEFAULT_BACKGROUND_TASK_MAXIMUM_DELAY,
490 488 : )
491 488 : .unwrap()),
492 488 :
493 488 : control_plane_api: (None),
494 488 : control_plane_api_token: (None),
495 488 : control_plane_emergency_mode: (false),
496 488 :
497 488 : import_pgdata_upcall_api: (None),
498 488 : import_pgdata_upcall_api_token: (None),
499 488 : import_pgdata_aws_endpoint_url: (None),
500 488 :
501 488 : heatmap_upload_concurrency: (DEFAULT_HEATMAP_UPLOAD_CONCURRENCY),
502 488 : secondary_download_concurrency: (DEFAULT_SECONDARY_DOWNLOAD_CONCURRENCY),
503 488 :
504 488 : ingest_batch_size: (DEFAULT_INGEST_BATCH_SIZE),
505 488 :
506 488 : virtual_file_io_engine: None,
507 488 :
508 488 : max_vectored_read_bytes: (MaxVectoredReadBytes(
509 488 : NonZeroUsize::new(DEFAULT_MAX_VECTORED_READ_BYTES).unwrap(),
510 488 : )),
511 488 : image_compression: (DEFAULT_IMAGE_COMPRESSION),
512 488 : timeline_offloading: true,
513 488 : ephemeral_bytes_per_memory_kb: (DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB),
514 488 : l0_flush: None,
515 488 : virtual_file_io_mode: None,
516 488 : tenant_config: TenantConfigToml::default(),
517 488 : no_sync: None,
518 488 : wal_receiver_protocol: DEFAULT_WAL_RECEIVER_PROTOCOL,
519 488 : page_service_pipelining: if !cfg!(test) {
520 488 : PageServicePipeliningConfig::Serial
521 : } else {
522 0 : PageServicePipeliningConfig::Pipelined(PageServicePipeliningConfigPipelined {
523 0 : max_batch_size: NonZeroUsize::new(32).unwrap(),
524 0 : execution: PageServiceProtocolPipelinedExecutionStrategy::ConcurrentFutures,
525 0 : })
526 : },
527 488 : get_vectored_concurrent_io: if !cfg!(test) {
528 488 : GetVectoredConcurrentIo::Sequential
529 : } else {
530 0 : GetVectoredConcurrentIo::SidecarTask
531 : },
532 488 : enable_read_path_debugging: if cfg!(test) || cfg!(feature = "testing") {
533 488 : Some(true)
534 : } else {
535 0 : None
536 : },
537 488 : validate_wal_contiguity: None,
538 488 : load_previous_heatmap: None,
539 488 : generate_unarchival_heatmap: None,
540 488 : }
541 488 : }
542 : }
543 :
544 : pub mod tenant_conf_defaults {
545 :
546 : // FIXME: This current value is very low. I would imagine something like 1 GB or 10 GB
547 : // would be more appropriate. But a low value forces the code to be exercised more,
548 : // which is good for now to trigger bugs.
549 : // This parameter actually determines L0 layer file size.
550 : pub const DEFAULT_CHECKPOINT_DISTANCE: u64 = 256 * 1024 * 1024;
551 : pub const DEFAULT_CHECKPOINT_TIMEOUT: &str = "10 m";
552 :
553 : // FIXME the below configs are only used by legacy algorithm. The new algorithm
554 : // has different parameters.
555 :
556 : // Target file size, when creating image and delta layers.
557 : // This parameter determines L1 layer file size.
558 : pub const DEFAULT_COMPACTION_TARGET_SIZE: u64 = 128 * 1024 * 1024;
559 :
560 : pub const DEFAULT_COMPACTION_PERIOD: &str = "20 s";
561 : pub const DEFAULT_COMPACTION_THRESHOLD: usize = 10;
562 :
563 : // This value needs to be tuned to avoid OOM. We have 3/4*CPUs threads for L0 compaction, that's
564 : // 3/4*16=9 on most of our pageservers. Compacting 20 layers requires about 1 GB memory (could
565 : // be reduced later by optimizing L0 hole calculation to avoid loading all keys into memory). So
566 : // with this config, we can get a maximum peak compaction usage of 9 GB.
567 : pub const DEFAULT_COMPACTION_UPPER_LIMIT: usize = 20;
568 : // Enable L0 compaction pass and semaphore by default. L0 compaction must be responsive to avoid
569 : // read amp.
570 : pub const DEFAULT_COMPACTION_L0_FIRST: bool = true;
571 : pub const DEFAULT_COMPACTION_L0_SEMAPHORE: bool = true;
572 :
573 : pub const DEFAULT_COMPACTION_ALGORITHM: crate::models::CompactionAlgorithm =
574 : crate::models::CompactionAlgorithm::Legacy;
575 :
576 : pub const DEFAULT_GC_HORIZON: u64 = 64 * 1024 * 1024;
577 :
578 : // Large DEFAULT_GC_PERIOD is fine as long as PITR_INTERVAL is larger.
579 : // If there's a need to decrease this value, first make sure that GC
580 : // doesn't hold a layer map write lock for non-trivial operations.
581 : // Relevant: https://github.com/neondatabase/neon/issues/3394
582 : pub const DEFAULT_GC_PERIOD: &str = "1 hr";
583 : pub const DEFAULT_IMAGE_CREATION_THRESHOLD: usize = 3;
584 : // If there are more than threshold * compaction_threshold (that is 3 * 10 in the default config) L0 layers, image
585 : // layer creation will end immediately. Set to 0 to disable.
586 : pub const DEFAULT_IMAGE_CREATION_PREEMPT_THRESHOLD: usize = 3;
587 : pub const DEFAULT_PITR_INTERVAL: &str = "7 days";
588 : pub const DEFAULT_WALRECEIVER_CONNECT_TIMEOUT: &str = "10 seconds";
589 : pub const DEFAULT_WALRECEIVER_LAGGING_WAL_TIMEOUT: &str = "10 seconds";
590 : // The default limit on WAL lag should be set to avoid causing disconnects under high throughput
591 : // scenarios: since the broker stats are updated ~1/s, a value of 1GiB should be sufficient for
592 : // throughputs up to 1GiB/s per timeline.
593 : pub const DEFAULT_MAX_WALRECEIVER_LSN_WAL_LAG: u64 = 1024 * 1024 * 1024;
594 : pub const DEFAULT_EVICTIONS_LOW_RESIDENCE_DURATION_METRIC_THRESHOLD: &str = "24 hour";
595 : // By default ingest enough WAL for two new L0 layers before checking if new image
596 : // image layers should be created.
597 : pub const DEFAULT_IMAGE_LAYER_CREATION_CHECK_THRESHOLD: u8 = 2;
598 : pub const DEFAULT_GC_COMPACTION_ENABLED: bool = false;
599 : pub const DEFAULT_GC_COMPACTION_INITIAL_THRESHOLD_KB: u64 = 5 * 1024 * 1024; // 5GB
600 : pub const DEFAULT_GC_COMPACTION_RATIO_PERCENT: u64 = 100;
601 : }
602 :
603 : impl Default for TenantConfigToml {
604 488 : fn default() -> Self {
605 : use tenant_conf_defaults::*;
606 488 : Self {
607 488 : checkpoint_distance: DEFAULT_CHECKPOINT_DISTANCE,
608 488 : checkpoint_timeout: humantime::parse_duration(DEFAULT_CHECKPOINT_TIMEOUT)
609 488 : .expect("cannot parse default checkpoint timeout"),
610 488 : compaction_target_size: DEFAULT_COMPACTION_TARGET_SIZE,
611 488 : compaction_period: humantime::parse_duration(DEFAULT_COMPACTION_PERIOD)
612 488 : .expect("cannot parse default compaction period"),
613 488 : compaction_threshold: DEFAULT_COMPACTION_THRESHOLD,
614 488 : compaction_upper_limit: DEFAULT_COMPACTION_UPPER_LIMIT,
615 488 : compaction_algorithm: crate::models::CompactionAlgorithmSettings {
616 488 : kind: DEFAULT_COMPACTION_ALGORITHM,
617 488 : },
618 488 : compaction_l0_first: DEFAULT_COMPACTION_L0_FIRST,
619 488 : compaction_l0_semaphore: DEFAULT_COMPACTION_L0_SEMAPHORE,
620 488 : l0_flush_delay_threshold: None,
621 488 : l0_flush_stall_threshold: None,
622 488 : gc_horizon: DEFAULT_GC_HORIZON,
623 488 : gc_period: humantime::parse_duration(DEFAULT_GC_PERIOD)
624 488 : .expect("cannot parse default gc period"),
625 488 : image_creation_threshold: DEFAULT_IMAGE_CREATION_THRESHOLD,
626 488 : pitr_interval: humantime::parse_duration(DEFAULT_PITR_INTERVAL)
627 488 : .expect("cannot parse default PITR interval"),
628 488 : walreceiver_connect_timeout: humantime::parse_duration(
629 488 : DEFAULT_WALRECEIVER_CONNECT_TIMEOUT,
630 488 : )
631 488 : .expect("cannot parse default walreceiver connect timeout"),
632 488 : lagging_wal_timeout: humantime::parse_duration(DEFAULT_WALRECEIVER_LAGGING_WAL_TIMEOUT)
633 488 : .expect("cannot parse default walreceiver lagging wal timeout"),
634 488 : max_lsn_wal_lag: NonZeroU64::new(DEFAULT_MAX_WALRECEIVER_LSN_WAL_LAG)
635 488 : .expect("cannot parse default max walreceiver Lsn wal lag"),
636 488 : eviction_policy: crate::models::EvictionPolicy::NoEviction,
637 488 : min_resident_size_override: None,
638 488 : evictions_low_residence_duration_metric_threshold: humantime::parse_duration(
639 488 : DEFAULT_EVICTIONS_LOW_RESIDENCE_DURATION_METRIC_THRESHOLD,
640 488 : )
641 488 : .expect("cannot parse default evictions_low_residence_duration_metric_threshold"),
642 488 : heatmap_period: Duration::ZERO,
643 488 : lazy_slru_download: false,
644 488 : timeline_get_throttle: crate::models::ThrottleConfig::disabled(),
645 488 : image_layer_creation_check_threshold: DEFAULT_IMAGE_LAYER_CREATION_CHECK_THRESHOLD,
646 488 : image_creation_preempt_threshold: DEFAULT_IMAGE_CREATION_PREEMPT_THRESHOLD,
647 488 : lsn_lease_length: LsnLease::DEFAULT_LENGTH,
648 488 : lsn_lease_length_for_ts: LsnLease::DEFAULT_LENGTH_FOR_TS,
649 488 : timeline_offloading: true,
650 488 : wal_receiver_protocol_override: None,
651 488 : rel_size_v2_enabled: false,
652 488 : gc_compaction_enabled: DEFAULT_GC_COMPACTION_ENABLED,
653 488 : gc_compaction_initial_threshold_kb: DEFAULT_GC_COMPACTION_INITIAL_THRESHOLD_KB,
654 488 : gc_compaction_ratio_percent: DEFAULT_GC_COMPACTION_RATIO_PERCENT,
655 488 : }
656 488 : }
657 : }
|
