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 4 : #[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 :
39 : // Deployment tools may write fields to the metadata file beyond what we
40 : // use in this type: this type intentionally only names fields that require.
41 : #[serde(flatten)]
42 : pub other: HashMap<String, serde_json::Value>,
43 : }
44 :
45 : /// `pageserver.toml`
46 : ///
47 : /// We use serde derive with `#[serde(default)]` to generate a deserializer
48 : /// that fills in the default values for each config field.
49 : ///
50 : /// If there cannot be a static default value because we need to make runtime
51 : /// checks to determine the default, make it an `Option` (which defaults to None).
52 : /// The runtime check should be done in the consuming crate, i.e., `pageserver`.
53 : #[serde_as]
54 20 : #[derive(Clone, Debug, serde::Deserialize, serde::Serialize)]
55 : #[serde(default, deny_unknown_fields)]
56 : pub struct ConfigToml {
57 : // types mapped 1:1 into the runtime PageServerConfig type
58 : pub listen_pg_addr: String,
59 : pub listen_http_addr: String,
60 : pub availability_zone: Option<String>,
61 : #[serde(with = "humantime_serde")]
62 : pub wait_lsn_timeout: Duration,
63 : #[serde(with = "humantime_serde")]
64 : pub wal_redo_timeout: Duration,
65 : pub superuser: String,
66 : pub locale: String,
67 : pub page_cache_size: usize,
68 : pub max_file_descriptors: usize,
69 : pub pg_distrib_dir: Option<Utf8PathBuf>,
70 : #[serde_as(as = "serde_with::DisplayFromStr")]
71 : pub http_auth_type: AuthType,
72 : #[serde_as(as = "serde_with::DisplayFromStr")]
73 : pub pg_auth_type: AuthType,
74 : pub auth_validation_public_key_path: Option<Utf8PathBuf>,
75 : pub remote_storage: Option<RemoteStorageConfig>,
76 : pub tenant_config: TenantConfigToml,
77 : #[serde_as(as = "serde_with::DisplayFromStr")]
78 : pub broker_endpoint: storage_broker::Uri,
79 : #[serde(with = "humantime_serde")]
80 : pub broker_keepalive_interval: Duration,
81 : #[serde_as(as = "serde_with::DisplayFromStr")]
82 : pub log_format: LogFormat,
83 : pub concurrent_tenant_warmup: NonZeroUsize,
84 : pub concurrent_tenant_size_logical_size_queries: NonZeroUsize,
85 : #[serde(with = "humantime_serde")]
86 : pub metric_collection_interval: Duration,
87 : pub metric_collection_endpoint: Option<reqwest::Url>,
88 : pub metric_collection_bucket: Option<RemoteStorageConfig>,
89 : #[serde(with = "humantime_serde")]
90 : pub synthetic_size_calculation_interval: Duration,
91 : pub disk_usage_based_eviction: Option<DiskUsageEvictionTaskConfig>,
92 : pub test_remote_failures: u64,
93 : pub ondemand_download_behavior_treat_error_as_warn: bool,
94 : #[serde(with = "humantime_serde")]
95 : pub background_task_maximum_delay: Duration,
96 : pub control_plane_api: Option<reqwest::Url>,
97 : pub control_plane_api_token: Option<String>,
98 : pub control_plane_emergency_mode: bool,
99 : /// Unstable feature: subject to change or removal without notice.
100 : /// See <https://github.com/neondatabase/neon/pull/9218>.
101 : pub import_pgdata_upcall_api: Option<reqwest::Url>,
102 : /// Unstable feature: subject to change or removal without notice.
103 : /// See <https://github.com/neondatabase/neon/pull/9218>.
104 : pub import_pgdata_upcall_api_token: Option<String>,
105 : /// Unstable feature: subject to change or removal without notice.
106 : /// See <https://github.com/neondatabase/neon/pull/9218>.
107 : pub import_pgdata_aws_endpoint_url: Option<reqwest::Url>,
108 : pub heatmap_upload_concurrency: usize,
109 : pub secondary_download_concurrency: usize,
110 : pub virtual_file_io_engine: Option<crate::models::virtual_file::IoEngineKind>,
111 : pub ingest_batch_size: u64,
112 : pub max_vectored_read_bytes: MaxVectoredReadBytes,
113 : pub image_compression: ImageCompressionAlgorithm,
114 : pub timeline_offloading: bool,
115 : pub ephemeral_bytes_per_memory_kb: usize,
116 : pub l0_flush: Option<crate::models::L0FlushConfig>,
117 : pub virtual_file_io_mode: Option<crate::models::virtual_file::IoMode>,
118 : #[serde(skip_serializing_if = "Option::is_none")]
119 : pub no_sync: Option<bool>,
120 : pub wal_receiver_protocol: PostgresClientProtocol,
121 : pub page_service_pipelining: PageServicePipeliningConfig,
122 : pub get_vectored_concurrent_io: GetVectoredConcurrentIo,
123 : pub enable_read_path_debugging: Option<bool>,
124 : #[serde(skip_serializing_if = "Option::is_none")]
125 : pub validate_wal_contiguity: Option<bool>,
126 : }
127 :
128 4 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
129 : #[serde(deny_unknown_fields)]
130 : pub struct DiskUsageEvictionTaskConfig {
131 : pub max_usage_pct: utils::serde_percent::Percent,
132 : pub min_avail_bytes: u64,
133 : #[serde(with = "humantime_serde")]
134 : pub period: Duration,
135 : #[cfg(feature = "testing")]
136 : pub mock_statvfs: Option<statvfs::mock::Behavior>,
137 : /// Select sorting for evicted layers
138 : #[serde(default)]
139 : pub eviction_order: EvictionOrder,
140 : }
141 :
142 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
143 : #[serde(tag = "mode", rename_all = "kebab-case")]
144 : #[serde(deny_unknown_fields)]
145 : pub enum PageServicePipeliningConfig {
146 : Serial,
147 : Pipelined(PageServicePipeliningConfigPipelined),
148 : }
149 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
150 : #[serde(deny_unknown_fields)]
151 : pub struct PageServicePipeliningConfigPipelined {
152 : /// Causes runtime errors if larger than max get_vectored batch size.
153 : pub max_batch_size: NonZeroUsize,
154 : pub execution: PageServiceProtocolPipelinedExecutionStrategy,
155 : }
156 :
157 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
158 : #[serde(rename_all = "kebab-case")]
159 : pub enum PageServiceProtocolPipelinedExecutionStrategy {
160 : ConcurrentFutures,
161 : Tasks,
162 : }
163 :
164 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
165 : #[serde(tag = "mode", rename_all = "kebab-case")]
166 : #[serde(deny_unknown_fields)]
167 : pub enum GetVectoredConcurrentIo {
168 : /// The read path is fully sequential: layers are visited
169 : /// one after the other and IOs are issued and waited upon
170 : /// from the same task that traverses the layers.
171 : Sequential,
172 : /// The read path still traverses layers sequentially, and
173 : /// index blocks will be read into the PS PageCache from
174 : /// that task, with waiting.
175 : /// But data IOs are dispatched and waited upon from a sidecar
176 : /// task so that the traversing task can continue to traverse
177 : /// layers while the IOs are in flight.
178 : /// If the PS PageCache miss rate is low, this improves
179 : /// throughput dramatically.
180 : SidecarTask,
181 : }
182 :
183 : pub mod statvfs {
184 : pub mod mock {
185 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
186 : #[serde(tag = "type")]
187 : pub enum Behavior {
188 : Success {
189 : blocksize: u64,
190 : total_blocks: u64,
191 : name_filter: Option<utils::serde_regex::Regex>,
192 : },
193 : #[cfg(feature = "testing")]
194 : Failure { mocked_error: MockedError },
195 : }
196 :
197 : #[cfg(feature = "testing")]
198 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
199 : #[allow(clippy::upper_case_acronyms)]
200 : pub enum MockedError {
201 : EIO,
202 : }
203 :
204 : #[cfg(feature = "testing")]
205 : impl From<MockedError> for nix::Error {
206 0 : fn from(e: MockedError) -> Self {
207 0 : match e {
208 0 : MockedError::EIO => nix::Error::EIO,
209 0 : }
210 0 : }
211 : }
212 : }
213 : }
214 :
215 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
216 : #[serde(tag = "type", content = "args")]
217 : pub enum EvictionOrder {
218 : RelativeAccessed {
219 : highest_layer_count_loses_first: bool,
220 : },
221 : }
222 :
223 : impl Default for EvictionOrder {
224 4 : fn default() -> Self {
225 4 : Self::RelativeAccessed {
226 4 : highest_layer_count_loses_first: true,
227 4 : }
228 4 : }
229 : }
230 :
231 0 : #[derive(Copy, Clone, Debug, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
232 : #[serde(transparent)]
233 : pub struct MaxVectoredReadBytes(pub NonZeroUsize);
234 :
235 : /// A tenant's calcuated configuration, which is the result of merging a
236 : /// tenant's TenantConfOpt with the global TenantConf from PageServerConf.
237 : ///
238 : /// For storing and transmitting individual tenant's configuration, see
239 : /// TenantConfOpt.
240 4 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
241 : #[serde(deny_unknown_fields, default)]
242 : pub struct TenantConfigToml {
243 : // Flush out an inmemory layer, if it's holding WAL older than this
244 : // This puts a backstop on how much WAL needs to be re-digested if the
245 : // page server crashes.
246 : // This parameter actually determines L0 layer file size.
247 : pub checkpoint_distance: u64,
248 : // Inmemory layer is also flushed at least once in checkpoint_timeout to
249 : // eventually upload WAL after activity is stopped.
250 : #[serde(with = "humantime_serde")]
251 : pub checkpoint_timeout: Duration,
252 : // Target file size, when creating image and delta layers.
253 : // This parameter determines L1 layer file size.
254 : pub compaction_target_size: u64,
255 : // How often to check if there's compaction work to be done.
256 : // Duration::ZERO means automatic compaction is disabled.
257 : #[serde(with = "humantime_serde")]
258 : pub compaction_period: Duration,
259 : /// Level0 delta layer threshold for compaction.
260 : pub compaction_threshold: usize,
261 : /// Controls the amount of L0 included in a single compaction iteration.
262 : /// The unit is `checkpoint_distance`, i.e., a size.
263 : /// We add L0s to the set of layers to compact until their cumulative
264 : /// size exceeds `compaction_upper_limit * checkpoint_distance`.
265 : pub compaction_upper_limit: usize,
266 : pub compaction_algorithm: crate::models::CompactionAlgorithmSettings,
267 : /// If true, compact down L0 across all tenant timelines before doing regular compaction.
268 : pub compaction_l0_first: bool,
269 : /// If true, use a separate semaphore (i.e. concurrency limit) for the L0 compaction pass. Only
270 : /// has an effect if `compaction_l0_first` is `true`.
271 : pub compaction_l0_semaphore: bool,
272 : /// Level0 delta layer threshold at which to delay layer flushes for compaction backpressure,
273 : /// such that they take 2x as long, and start waiting for layer flushes during ephemeral layer
274 : /// rolls. This helps compaction keep up with WAL ingestion, and avoids read amplification
275 : /// blowing up. Should be >compaction_threshold. 0 to disable. Disabled by default.
276 : pub l0_flush_delay_threshold: Option<usize>,
277 : /// Level0 delta layer threshold at which to stall layer flushes. Must be >compaction_threshold
278 : /// to avoid deadlock. 0 to disable. Disabled by default.
279 : pub l0_flush_stall_threshold: Option<usize>,
280 : /// If true, Level0 delta layer flushes will wait for S3 upload before flushing the next
281 : /// layer. This is a temporary backpressure mechanism which should be removed once
282 : /// l0_flush_{delay,stall}_threshold is fully enabled.
283 : pub l0_flush_wait_upload: bool,
284 : // Determines how much history is retained, to allow
285 : // branching and read replicas at an older point in time.
286 : // The unit is #of bytes of WAL.
287 : // Page versions older than this are garbage collected away.
288 : pub gc_horizon: u64,
289 : // Interval at which garbage collection is triggered.
290 : // Duration::ZERO means automatic GC is disabled
291 : #[serde(with = "humantime_serde")]
292 : pub gc_period: Duration,
293 : // Delta layer churn threshold to create L1 image layers.
294 : pub image_creation_threshold: usize,
295 : // Determines how much history is retained, to allow
296 : // branching and read replicas at an older point in time.
297 : // The unit is time.
298 : // Page versions older than this are garbage collected away.
299 : #[serde(with = "humantime_serde")]
300 : pub pitr_interval: Duration,
301 : /// Maximum amount of time to wait while opening a connection to receive wal, before erroring.
302 : #[serde(with = "humantime_serde")]
303 : pub walreceiver_connect_timeout: Duration,
304 : /// Considers safekeepers stalled after no WAL updates were received longer than this threshold.
305 : /// A stalled safekeeper will be changed to a newer one when it appears.
306 : #[serde(with = "humantime_serde")]
307 : pub lagging_wal_timeout: Duration,
308 : /// Considers safekeepers lagging when their WAL is behind another safekeeper for more than this threshold.
309 : /// A lagging safekeeper will be changed after `lagging_wal_timeout` time elapses since the last WAL update,
310 : /// to avoid eager reconnects.
311 : pub max_lsn_wal_lag: NonZeroU64,
312 : pub eviction_policy: crate::models::EvictionPolicy,
313 : pub min_resident_size_override: Option<u64>,
314 : // See the corresponding metric's help string.
315 : #[serde(with = "humantime_serde")]
316 : pub evictions_low_residence_duration_metric_threshold: Duration,
317 :
318 : /// If non-zero, the period between uploads of a heatmap from attached tenants. This
319 : /// may be disabled if a Tenant will not have secondary locations: only secondary
320 : /// locations will use the heatmap uploaded by attached locations.
321 : #[serde(with = "humantime_serde")]
322 : pub heatmap_period: Duration,
323 :
324 : /// If true then SLRU segments are dowloaded on demand, if false SLRU segments are included in basebackup
325 : pub lazy_slru_download: bool,
326 :
327 : pub timeline_get_throttle: crate::models::ThrottleConfig,
328 :
329 : // How much WAL must be ingested before checking again whether a new image layer is required.
330 : // Expresed in multiples of checkpoint distance.
331 : pub image_layer_creation_check_threshold: u8,
332 :
333 : // How many multiples of L0 `compaction_threshold` will preempt image layer creation and do L0 compaction.
334 : // Set to 0 to disable preemption.
335 : pub image_creation_preempt_threshold: usize,
336 :
337 : /// The length for an explicit LSN lease request.
338 : /// Layers needed to reconstruct pages at LSN will not be GC-ed during this interval.
339 : #[serde(with = "humantime_serde")]
340 : pub lsn_lease_length: Duration,
341 :
342 : /// The length for an implicit LSN lease granted as part of `get_lsn_by_timestamp` request.
343 : /// Layers needed to reconstruct pages at LSN will not be GC-ed during this interval.
344 : #[serde(with = "humantime_serde")]
345 : pub lsn_lease_length_for_ts: Duration,
346 :
347 : /// Enable auto-offloading of timelines.
348 : /// (either this flag or the pageserver-global one need to be set)
349 : pub timeline_offloading: bool,
350 :
351 : pub wal_receiver_protocol_override: Option<PostgresClientProtocol>,
352 :
353 : /// Enable rel_size_v2 for this tenant. Once enabled, the tenant will persist this information into
354 : /// `index_part.json`, and it cannot be reversed.
355 : pub rel_size_v2_enabled: bool,
356 :
357 : // gc-compaction related configs
358 : /// Enable automatic gc-compaction trigger on this tenant.
359 : pub gc_compaction_enabled: bool,
360 : /// The initial threshold for gc-compaction in KB. Once the total size of layers below the gc-horizon is above this threshold,
361 : /// gc-compaction will be triggered.
362 : pub gc_compaction_initial_threshold_kb: u64,
363 : /// 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)
364 : /// is above this ratio, gc-compaction will be triggered.
365 : pub gc_compaction_ratio_percent: u64,
366 : }
367 :
368 : pub mod defaults {
369 : pub use storage_broker::DEFAULT_ENDPOINT as BROKER_DEFAULT_ENDPOINT;
370 :
371 : use crate::models::ImageCompressionAlgorithm;
372 :
373 : pub const DEFAULT_WAIT_LSN_TIMEOUT: &str = "300 s";
374 : pub const DEFAULT_WAL_REDO_TIMEOUT: &str = "60 s";
375 :
376 : pub const DEFAULT_SUPERUSER: &str = "cloud_admin";
377 : pub const DEFAULT_LOCALE: &str = if cfg!(target_os = "macos") {
378 : "C"
379 : } else {
380 : "C.UTF-8"
381 : };
382 :
383 : pub const DEFAULT_PAGE_CACHE_SIZE: usize = 8192;
384 : pub const DEFAULT_MAX_FILE_DESCRIPTORS: usize = 100;
385 :
386 : pub const DEFAULT_LOG_FORMAT: &str = "plain";
387 :
388 : pub const DEFAULT_CONCURRENT_TENANT_WARMUP: usize = 8;
389 :
390 : pub const DEFAULT_CONCURRENT_TENANT_SIZE_LOGICAL_SIZE_QUERIES: usize = 1;
391 :
392 : pub const DEFAULT_METRIC_COLLECTION_INTERVAL: &str = "10 min";
393 : pub const DEFAULT_METRIC_COLLECTION_ENDPOINT: Option<reqwest::Url> = None;
394 : pub const DEFAULT_SYNTHETIC_SIZE_CALCULATION_INTERVAL: &str = "10 min";
395 : pub const DEFAULT_BACKGROUND_TASK_MAXIMUM_DELAY: &str = "10s";
396 :
397 : pub const DEFAULT_HEATMAP_UPLOAD_CONCURRENCY: usize = 8;
398 : pub const DEFAULT_SECONDARY_DOWNLOAD_CONCURRENCY: usize = 1;
399 :
400 : pub const DEFAULT_INGEST_BATCH_SIZE: u64 = 100;
401 :
402 : /// Soft limit for the maximum size of a vectored read.
403 : ///
404 : /// This is determined by the largest NeonWalRecord that can exist (minus dbdir and reldir keys
405 : /// which are bounded by the blob io limits only). As of this writing, that is a `NeonWalRecord::ClogSetCommitted` record,
406 : /// with 32k xids. That's the max number of XIDS on a single CLOG page. The size of such a record
407 : /// is `sizeof(Transactionid) * 32768 + (some fixed overhead from 'timestamp`, the Vec length and whatever extra serde serialization adds)`.
408 : /// That is, slightly above 128 kB.
409 : pub const DEFAULT_MAX_VECTORED_READ_BYTES: usize = 130 * 1024; // 130 KiB
410 :
411 : pub const DEFAULT_IMAGE_COMPRESSION: ImageCompressionAlgorithm =
412 : ImageCompressionAlgorithm::Zstd { level: Some(1) };
413 :
414 : pub const DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB: usize = 0;
415 :
416 : pub const DEFAULT_IO_BUFFER_ALIGNMENT: usize = 512;
417 :
418 : pub const DEFAULT_WAL_RECEIVER_PROTOCOL: utils::postgres_client::PostgresClientProtocol =
419 : utils::postgres_client::PostgresClientProtocol::Vanilla;
420 : }
421 :
422 : impl Default for ConfigToml {
423 488 : fn default() -> Self {
424 : use defaults::*;
425 :
426 : Self {
427 488 : listen_pg_addr: (DEFAULT_PG_LISTEN_ADDR.to_string()),
428 488 : listen_http_addr: (DEFAULT_HTTP_LISTEN_ADDR.to_string()),
429 488 : availability_zone: (None),
430 488 : wait_lsn_timeout: (humantime::parse_duration(DEFAULT_WAIT_LSN_TIMEOUT)
431 488 : .expect("cannot parse default wait lsn timeout")),
432 488 : wal_redo_timeout: (humantime::parse_duration(DEFAULT_WAL_REDO_TIMEOUT)
433 488 : .expect("cannot parse default wal redo timeout")),
434 488 : superuser: (DEFAULT_SUPERUSER.to_string()),
435 488 : locale: DEFAULT_LOCALE.to_string(),
436 488 : page_cache_size: (DEFAULT_PAGE_CACHE_SIZE),
437 488 : max_file_descriptors: (DEFAULT_MAX_FILE_DESCRIPTORS),
438 488 : pg_distrib_dir: None, // Utf8PathBuf::from("./pg_install"), // TODO: formely, this was std::env::current_dir()
439 488 : http_auth_type: (AuthType::Trust),
440 488 : pg_auth_type: (AuthType::Trust),
441 488 : auth_validation_public_key_path: (None),
442 488 : remote_storage: None,
443 488 : broker_endpoint: (storage_broker::DEFAULT_ENDPOINT
444 488 : .parse()
445 488 : .expect("failed to parse default broker endpoint")),
446 488 : broker_keepalive_interval: (humantime::parse_duration(
447 488 : storage_broker::DEFAULT_KEEPALIVE_INTERVAL,
448 488 : )
449 488 : .expect("cannot parse default keepalive interval")),
450 488 : log_format: (LogFormat::from_str(DEFAULT_LOG_FORMAT).unwrap()),
451 488 :
452 488 : concurrent_tenant_warmup: (NonZeroUsize::new(DEFAULT_CONCURRENT_TENANT_WARMUP)
453 488 : .expect("Invalid default constant")),
454 488 : concurrent_tenant_size_logical_size_queries: NonZeroUsize::new(
455 488 : DEFAULT_CONCURRENT_TENANT_SIZE_LOGICAL_SIZE_QUERIES,
456 488 : )
457 488 : .unwrap(),
458 488 : metric_collection_interval: (humantime::parse_duration(
459 488 : DEFAULT_METRIC_COLLECTION_INTERVAL,
460 488 : )
461 488 : .expect("cannot parse default metric collection interval")),
462 488 : synthetic_size_calculation_interval: (humantime::parse_duration(
463 488 : DEFAULT_SYNTHETIC_SIZE_CALCULATION_INTERVAL,
464 488 : )
465 488 : .expect("cannot parse default synthetic size calculation interval")),
466 488 : metric_collection_endpoint: (DEFAULT_METRIC_COLLECTION_ENDPOINT),
467 488 :
468 488 : metric_collection_bucket: (None),
469 488 :
470 488 : disk_usage_based_eviction: (None),
471 488 :
472 488 : test_remote_failures: (0),
473 488 :
474 488 : ondemand_download_behavior_treat_error_as_warn: (false),
475 488 :
476 488 : background_task_maximum_delay: (humantime::parse_duration(
477 488 : DEFAULT_BACKGROUND_TASK_MAXIMUM_DELAY,
478 488 : )
479 488 : .unwrap()),
480 488 :
481 488 : control_plane_api: (None),
482 488 : control_plane_api_token: (None),
483 488 : control_plane_emergency_mode: (false),
484 488 :
485 488 : import_pgdata_upcall_api: (None),
486 488 : import_pgdata_upcall_api_token: (None),
487 488 : import_pgdata_aws_endpoint_url: (None),
488 488 :
489 488 : heatmap_upload_concurrency: (DEFAULT_HEATMAP_UPLOAD_CONCURRENCY),
490 488 : secondary_download_concurrency: (DEFAULT_SECONDARY_DOWNLOAD_CONCURRENCY),
491 488 :
492 488 : ingest_batch_size: (DEFAULT_INGEST_BATCH_SIZE),
493 488 :
494 488 : virtual_file_io_engine: None,
495 488 :
496 488 : max_vectored_read_bytes: (MaxVectoredReadBytes(
497 488 : NonZeroUsize::new(DEFAULT_MAX_VECTORED_READ_BYTES).unwrap(),
498 488 : )),
499 488 : image_compression: (DEFAULT_IMAGE_COMPRESSION),
500 488 : timeline_offloading: true,
501 488 : ephemeral_bytes_per_memory_kb: (DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB),
502 488 : l0_flush: None,
503 488 : virtual_file_io_mode: None,
504 488 : tenant_config: TenantConfigToml::default(),
505 488 : no_sync: None,
506 488 : wal_receiver_protocol: DEFAULT_WAL_RECEIVER_PROTOCOL,
507 488 : page_service_pipelining: if !cfg!(test) {
508 488 : PageServicePipeliningConfig::Serial
509 : } else {
510 0 : PageServicePipeliningConfig::Pipelined(PageServicePipeliningConfigPipelined {
511 0 : max_batch_size: NonZeroUsize::new(32).unwrap(),
512 0 : execution: PageServiceProtocolPipelinedExecutionStrategy::ConcurrentFutures,
513 0 : })
514 : },
515 488 : get_vectored_concurrent_io: if !cfg!(test) {
516 488 : GetVectoredConcurrentIo::Sequential
517 : } else {
518 0 : GetVectoredConcurrentIo::SidecarTask
519 : },
520 488 : enable_read_path_debugging: if cfg!(test) || cfg!(feature = "testing") {
521 488 : Some(true)
522 : } else {
523 0 : None
524 : },
525 488 : validate_wal_contiguity: None,
526 488 : }
527 488 : }
528 : }
529 :
530 : pub mod tenant_conf_defaults {
531 :
532 : // FIXME: This current value is very low. I would imagine something like 1 GB or 10 GB
533 : // would be more appropriate. But a low value forces the code to be exercised more,
534 : // which is good for now to trigger bugs.
535 : // This parameter actually determines L0 layer file size.
536 : pub const DEFAULT_CHECKPOINT_DISTANCE: u64 = 256 * 1024 * 1024;
537 : pub const DEFAULT_CHECKPOINT_TIMEOUT: &str = "10 m";
538 :
539 : // FIXME the below configs are only used by legacy algorithm. The new algorithm
540 : // has different parameters.
541 :
542 : // Target file size, when creating image and delta layers.
543 : // This parameter determines L1 layer file size.
544 : pub const DEFAULT_COMPACTION_TARGET_SIZE: u64 = 128 * 1024 * 1024;
545 :
546 : pub const DEFAULT_COMPACTION_PERIOD: &str = "20 s";
547 : pub const DEFAULT_COMPACTION_THRESHOLD: usize = 10;
548 :
549 : // This value needs to be tuned to avoid OOM. We have 3/4*CPUs threads for L0 compaction, that's
550 : // 3/4*16=9 on most of our pageservers. Compacting 20 layers requires about 1 GB memory (could
551 : // be reduced later by optimizing L0 hole calculation to avoid loading all keys into memory). So
552 : // with this config, we can get a maximum peak compaction usage of 9 GB.
553 : pub const DEFAULT_COMPACTION_UPPER_LIMIT: usize = 20;
554 : pub const DEFAULT_COMPACTION_L0_FIRST: bool = false;
555 : pub const DEFAULT_COMPACTION_L0_SEMAPHORE: bool = true;
556 :
557 : pub const DEFAULT_COMPACTION_ALGORITHM: crate::models::CompactionAlgorithm =
558 : crate::models::CompactionAlgorithm::Legacy;
559 :
560 : pub const DEFAULT_L0_FLUSH_WAIT_UPLOAD: bool = true;
561 :
562 : pub const DEFAULT_GC_HORIZON: u64 = 64 * 1024 * 1024;
563 :
564 : // Large DEFAULT_GC_PERIOD is fine as long as PITR_INTERVAL is larger.
565 : // If there's a need to decrease this value, first make sure that GC
566 : // doesn't hold a layer map write lock for non-trivial operations.
567 : // Relevant: https://github.com/neondatabase/neon/issues/3394
568 : pub const DEFAULT_GC_PERIOD: &str = "1 hr";
569 : pub const DEFAULT_IMAGE_CREATION_THRESHOLD: usize = 3;
570 : // If there are more than threshold * compaction_threshold (that is 3 * 10 in the default config) L0 layers, image
571 : // layer creation will end immediately. Set to 0 to disable. The target default will be 3 once we
572 : // want to enable this feature.
573 : pub const DEFAULT_IMAGE_CREATION_PREEMPT_THRESHOLD: usize = 0;
574 : pub const DEFAULT_PITR_INTERVAL: &str = "7 days";
575 : pub const DEFAULT_WALRECEIVER_CONNECT_TIMEOUT: &str = "10 seconds";
576 : pub const DEFAULT_WALRECEIVER_LAGGING_WAL_TIMEOUT: &str = "10 seconds";
577 : // The default limit on WAL lag should be set to avoid causing disconnects under high throughput
578 : // scenarios: since the broker stats are updated ~1/s, a value of 1GiB should be sufficient for
579 : // throughputs up to 1GiB/s per timeline.
580 : pub const DEFAULT_MAX_WALRECEIVER_LSN_WAL_LAG: u64 = 1024 * 1024 * 1024;
581 : pub const DEFAULT_EVICTIONS_LOW_RESIDENCE_DURATION_METRIC_THRESHOLD: &str = "24 hour";
582 : // By default ingest enough WAL for two new L0 layers before checking if new image
583 : // image layers should be created.
584 : pub const DEFAULT_IMAGE_LAYER_CREATION_CHECK_THRESHOLD: u8 = 2;
585 : pub const DEFAULT_GC_COMPACTION_ENABLED: bool = false;
586 : pub const DEFAULT_GC_COMPACTION_INITIAL_THRESHOLD_KB: u64 = 5 * 1024 * 1024; // 5GB
587 : pub const DEFAULT_GC_COMPACTION_RATIO_PERCENT: u64 = 100;
588 : }
589 :
590 : impl Default for TenantConfigToml {
591 924 : fn default() -> Self {
592 : use tenant_conf_defaults::*;
593 924 : Self {
594 924 : checkpoint_distance: DEFAULT_CHECKPOINT_DISTANCE,
595 924 : checkpoint_timeout: humantime::parse_duration(DEFAULT_CHECKPOINT_TIMEOUT)
596 924 : .expect("cannot parse default checkpoint timeout"),
597 924 : compaction_target_size: DEFAULT_COMPACTION_TARGET_SIZE,
598 924 : compaction_period: humantime::parse_duration(DEFAULT_COMPACTION_PERIOD)
599 924 : .expect("cannot parse default compaction period"),
600 924 : compaction_threshold: DEFAULT_COMPACTION_THRESHOLD,
601 924 : compaction_upper_limit: DEFAULT_COMPACTION_UPPER_LIMIT,
602 924 : compaction_algorithm: crate::models::CompactionAlgorithmSettings {
603 924 : kind: DEFAULT_COMPACTION_ALGORITHM,
604 924 : },
605 924 : compaction_l0_first: DEFAULT_COMPACTION_L0_FIRST,
606 924 : compaction_l0_semaphore: DEFAULT_COMPACTION_L0_SEMAPHORE,
607 924 : l0_flush_delay_threshold: None,
608 924 : l0_flush_stall_threshold: None,
609 924 : l0_flush_wait_upload: DEFAULT_L0_FLUSH_WAIT_UPLOAD,
610 924 : gc_horizon: DEFAULT_GC_HORIZON,
611 924 : gc_period: humantime::parse_duration(DEFAULT_GC_PERIOD)
612 924 : .expect("cannot parse default gc period"),
613 924 : image_creation_threshold: DEFAULT_IMAGE_CREATION_THRESHOLD,
614 924 : pitr_interval: humantime::parse_duration(DEFAULT_PITR_INTERVAL)
615 924 : .expect("cannot parse default PITR interval"),
616 924 : walreceiver_connect_timeout: humantime::parse_duration(
617 924 : DEFAULT_WALRECEIVER_CONNECT_TIMEOUT,
618 924 : )
619 924 : .expect("cannot parse default walreceiver connect timeout"),
620 924 : lagging_wal_timeout: humantime::parse_duration(DEFAULT_WALRECEIVER_LAGGING_WAL_TIMEOUT)
621 924 : .expect("cannot parse default walreceiver lagging wal timeout"),
622 924 : max_lsn_wal_lag: NonZeroU64::new(DEFAULT_MAX_WALRECEIVER_LSN_WAL_LAG)
623 924 : .expect("cannot parse default max walreceiver Lsn wal lag"),
624 924 : eviction_policy: crate::models::EvictionPolicy::NoEviction,
625 924 : min_resident_size_override: None,
626 924 : evictions_low_residence_duration_metric_threshold: humantime::parse_duration(
627 924 : DEFAULT_EVICTIONS_LOW_RESIDENCE_DURATION_METRIC_THRESHOLD,
628 924 : )
629 924 : .expect("cannot parse default evictions_low_residence_duration_metric_threshold"),
630 924 : heatmap_period: Duration::ZERO,
631 924 : lazy_slru_download: false,
632 924 : timeline_get_throttle: crate::models::ThrottleConfig::disabled(),
633 924 : image_layer_creation_check_threshold: DEFAULT_IMAGE_LAYER_CREATION_CHECK_THRESHOLD,
634 924 : image_creation_preempt_threshold: DEFAULT_IMAGE_CREATION_PREEMPT_THRESHOLD,
635 924 : lsn_lease_length: LsnLease::DEFAULT_LENGTH,
636 924 : lsn_lease_length_for_ts: LsnLease::DEFAULT_LENGTH_FOR_TS,
637 924 : timeline_offloading: true,
638 924 : wal_receiver_protocol_override: None,
639 924 : rel_size_v2_enabled: false,
640 924 : gc_compaction_enabled: DEFAULT_GC_COMPACTION_ENABLED,
641 924 : gc_compaction_initial_threshold_kb: DEFAULT_GC_COMPACTION_INITIAL_THRESHOLD_KB,
642 924 : gc_compaction_ratio_percent: DEFAULT_GC_COMPACTION_RATIO_PERCENT,
643 924 : }
644 924 : }
645 : }
|
