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 : pub availability_zone: Option<String>,
110 : #[serde(with = "humantime_serde")]
111 : pub wait_lsn_timeout: Duration,
112 : #[serde(with = "humantime_serde")]
113 : pub wal_redo_timeout: Duration,
114 : pub superuser: String,
115 : pub locale: String,
116 : pub page_cache_size: usize,
117 : pub max_file_descriptors: usize,
118 : pub pg_distrib_dir: Option<Utf8PathBuf>,
119 : #[serde_as(as = "serde_with::DisplayFromStr")]
120 : pub http_auth_type: AuthType,
121 : #[serde_as(as = "serde_with::DisplayFromStr")]
122 : pub pg_auth_type: AuthType,
123 : pub auth_validation_public_key_path: Option<Utf8PathBuf>,
124 : pub remote_storage: Option<RemoteStorageConfig>,
125 : pub tenant_config: TenantConfigToml,
126 : #[serde_as(as = "serde_with::DisplayFromStr")]
127 : pub broker_endpoint: storage_broker::Uri,
128 : #[serde(with = "humantime_serde")]
129 : pub broker_keepalive_interval: Duration,
130 : #[serde_as(as = "serde_with::DisplayFromStr")]
131 : pub log_format: LogFormat,
132 : pub concurrent_tenant_warmup: NonZeroUsize,
133 : pub concurrent_tenant_size_logical_size_queries: NonZeroUsize,
134 : #[serde(with = "humantime_serde")]
135 : pub metric_collection_interval: Duration,
136 : pub metric_collection_endpoint: Option<reqwest::Url>,
137 : pub metric_collection_bucket: Option<RemoteStorageConfig>,
138 : #[serde(with = "humantime_serde")]
139 : pub synthetic_size_calculation_interval: Duration,
140 : pub disk_usage_based_eviction: Option<DiskUsageEvictionTaskConfig>,
141 : pub test_remote_failures: u64,
142 : pub ondemand_download_behavior_treat_error_as_warn: bool,
143 : #[serde(with = "humantime_serde")]
144 : pub background_task_maximum_delay: Duration,
145 : pub control_plane_api: Option<reqwest::Url>,
146 : pub control_plane_api_token: Option<String>,
147 : pub control_plane_emergency_mode: bool,
148 : /// Unstable feature: subject to change or removal without notice.
149 : /// See <https://github.com/neondatabase/neon/pull/9218>.
150 : pub import_pgdata_upcall_api: Option<reqwest::Url>,
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_token: Option<String>,
154 : /// Unstable feature: subject to change or removal without notice.
155 : /// See <https://github.com/neondatabase/neon/pull/9218>.
156 : pub import_pgdata_aws_endpoint_url: Option<reqwest::Url>,
157 : pub heatmap_upload_concurrency: usize,
158 : pub secondary_download_concurrency: usize,
159 : pub virtual_file_io_engine: Option<crate::models::virtual_file::IoEngineKind>,
160 : pub ingest_batch_size: u64,
161 : pub max_vectored_read_bytes: MaxVectoredReadBytes,
162 : pub image_compression: ImageCompressionAlgorithm,
163 : pub timeline_offloading: bool,
164 : pub ephemeral_bytes_per_memory_kb: usize,
165 : pub l0_flush: Option<crate::models::L0FlushConfig>,
166 : pub virtual_file_io_mode: Option<crate::models::virtual_file::IoMode>,
167 : #[serde(skip_serializing_if = "Option::is_none")]
168 : pub no_sync: Option<bool>,
169 : pub wal_receiver_protocol: PostgresClientProtocol,
170 : pub page_service_pipelining: PageServicePipeliningConfig,
171 : pub get_vectored_concurrent_io: GetVectoredConcurrentIo,
172 : pub enable_read_path_debugging: Option<bool>,
173 : #[serde(skip_serializing_if = "Option::is_none")]
174 : pub validate_wal_contiguity: Option<bool>,
175 : #[serde(skip_serializing_if = "Option::is_none")]
176 : pub load_previous_heatmap: Option<bool>,
177 : #[serde(skip_serializing_if = "Option::is_none")]
178 : pub generate_unarchival_heatmap: Option<bool>,
179 : }
180 :
181 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
182 : pub struct DiskUsageEvictionTaskConfig {
183 : pub max_usage_pct: utils::serde_percent::Percent,
184 : pub min_avail_bytes: u64,
185 : #[serde(with = "humantime_serde")]
186 : pub period: Duration,
187 : #[cfg(feature = "testing")]
188 : pub mock_statvfs: Option<statvfs::mock::Behavior>,
189 : /// Select sorting for evicted layers
190 : #[serde(default)]
191 : pub eviction_order: EvictionOrder,
192 : }
193 :
194 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
195 : #[serde(tag = "mode", rename_all = "kebab-case")]
196 : pub enum PageServicePipeliningConfig {
197 : Serial,
198 : Pipelined(PageServicePipeliningConfigPipelined),
199 : }
200 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
201 : pub struct PageServicePipeliningConfigPipelined {
202 : /// Causes runtime errors if larger than max get_vectored batch size.
203 : pub max_batch_size: NonZeroUsize,
204 : pub execution: PageServiceProtocolPipelinedExecutionStrategy,
205 : }
206 :
207 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
208 : #[serde(rename_all = "kebab-case")]
209 : pub enum PageServiceProtocolPipelinedExecutionStrategy {
210 : ConcurrentFutures,
211 : Tasks,
212 : }
213 :
214 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
215 : #[serde(tag = "mode", rename_all = "kebab-case")]
216 : pub enum GetVectoredConcurrentIo {
217 : /// The read path is fully sequential: layers are visited
218 : /// one after the other and IOs are issued and waited upon
219 : /// from the same task that traverses the layers.
220 : Sequential,
221 : /// The read path still traverses layers sequentially, and
222 : /// index blocks will be read into the PS PageCache from
223 : /// that task, with waiting.
224 : /// But data IOs are dispatched and waited upon from a sidecar
225 : /// task so that the traversing task can continue to traverse
226 : /// layers while the IOs are in flight.
227 : /// If the PS PageCache miss rate is low, this improves
228 : /// throughput dramatically.
229 : SidecarTask,
230 : }
231 :
232 : pub mod statvfs {
233 : pub mod mock {
234 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
235 : #[serde(tag = "type")]
236 : pub enum Behavior {
237 : Success {
238 : blocksize: u64,
239 : total_blocks: u64,
240 : name_filter: Option<utils::serde_regex::Regex>,
241 : },
242 : #[cfg(feature = "testing")]
243 : Failure { mocked_error: MockedError },
244 : }
245 :
246 : #[cfg(feature = "testing")]
247 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
248 : #[allow(clippy::upper_case_acronyms)]
249 : pub enum MockedError {
250 : EIO,
251 : }
252 :
253 : #[cfg(feature = "testing")]
254 : impl From<MockedError> for nix::Error {
255 0 : fn from(e: MockedError) -> Self {
256 0 : match e {
257 0 : MockedError::EIO => nix::Error::EIO,
258 0 : }
259 0 : }
260 : }
261 : }
262 : }
263 :
264 0 : #[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
265 : #[serde(tag = "type", content = "args")]
266 : pub enum EvictionOrder {
267 : RelativeAccessed {
268 : highest_layer_count_loses_first: bool,
269 : },
270 : }
271 :
272 : impl Default for EvictionOrder {
273 4 : fn default() -> Self {
274 4 : Self::RelativeAccessed {
275 4 : highest_layer_count_loses_first: true,
276 4 : }
277 4 : }
278 : }
279 :
280 0 : #[derive(Copy, Clone, Debug, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
281 : #[serde(transparent)]
282 : pub struct MaxVectoredReadBytes(pub NonZeroUsize);
283 :
284 : /// Tenant-level configuration values, used for various purposes.
285 0 : #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
286 : #[serde(default)]
287 : pub struct TenantConfigToml {
288 : // Flush out an inmemory layer, if it's holding WAL older than this
289 : // This puts a backstop on how much WAL needs to be re-digested if the
290 : // page server crashes.
291 : // This parameter actually determines L0 layer file size.
292 : pub checkpoint_distance: u64,
293 : // Inmemory layer is also flushed at least once in checkpoint_timeout to
294 : // eventually upload WAL after activity is stopped.
295 : #[serde(with = "humantime_serde")]
296 : pub checkpoint_timeout: Duration,
297 : // Target file size, when creating image and delta layers.
298 : // This parameter determines L1 layer file size.
299 : pub compaction_target_size: u64,
300 : // How often to check if there's compaction work to be done.
301 : // Duration::ZERO means automatic compaction is disabled.
302 : #[serde(with = "humantime_serde")]
303 : pub compaction_period: Duration,
304 : /// Level0 delta layer threshold for compaction.
305 : pub compaction_threshold: usize,
306 : /// Controls the amount of L0 included in a single compaction iteration.
307 : /// The unit is `checkpoint_distance`, i.e., a size.
308 : /// We add L0s to the set of layers to compact until their cumulative
309 : /// size exceeds `compaction_upper_limit * checkpoint_distance`.
310 : pub compaction_upper_limit: usize,
311 : pub compaction_algorithm: crate::models::CompactionAlgorithmSettings,
312 : /// If true, compact down L0 across all tenant timelines before doing regular compaction. L0
313 : /// compaction must be responsive to avoid read amp during heavy ingestion. Defaults to true.
314 : pub compaction_l0_first: bool,
315 : /// If true, use a separate semaphore (i.e. concurrency limit) for the L0 compaction pass. Only
316 : /// has an effect if `compaction_l0_first` is true. Defaults to true.
317 : pub compaction_l0_semaphore: bool,
318 : /// Level0 delta layer threshold at which to delay layer flushes such that they take 2x as long,
319 : /// and block on layer flushes during ephemeral layer rolls, for compaction backpressure. This
320 : /// helps compaction keep up with WAL ingestion, and avoids read amplification blowing up.
321 : /// Should be >compaction_threshold. 0 to disable. Defaults to 3x compaction_threshold.
322 : pub l0_flush_delay_threshold: Option<usize>,
323 : /// Level0 delta layer threshold at which to stall layer flushes. Must be >compaction_threshold
324 : /// to avoid deadlock. 0 to disable. Disabled by default.
325 : pub l0_flush_stall_threshold: Option<usize>,
326 : /// If true, Level0 delta layer flushes will wait for S3 upload before flushing the next
327 : /// layer. This is a temporary backpressure mechanism which should be removed once
328 : /// l0_flush_{delay,stall}_threshold is fully enabled.
329 : ///
330 : /// TODO: this is no longer enabled, remove it when the config option is no longer set.
331 : pub l0_flush_wait_upload: bool,
332 : // Determines how much history is retained, to allow
333 : // branching and read replicas at an older point in time.
334 : // The unit is #of bytes of WAL.
335 : // Page versions older than this are garbage collected away.
336 : pub gc_horizon: u64,
337 : // Interval at which garbage collection is triggered.
338 : // Duration::ZERO means automatic GC is disabled
339 : #[serde(with = "humantime_serde")]
340 : pub gc_period: Duration,
341 : // Delta layer churn threshold to create L1 image layers.
342 : pub image_creation_threshold: usize,
343 : // Determines how much history is retained, to allow
344 : // branching and read replicas at an older point in time.
345 : // The unit is time.
346 : // Page versions older than this are garbage collected away.
347 : #[serde(with = "humantime_serde")]
348 : pub pitr_interval: Duration,
349 : /// Maximum amount of time to wait while opening a connection to receive wal, before erroring.
350 : #[serde(with = "humantime_serde")]
351 : pub walreceiver_connect_timeout: Duration,
352 : /// Considers safekeepers stalled after no WAL updates were received longer than this threshold.
353 : /// A stalled safekeeper will be changed to a newer one when it appears.
354 : #[serde(with = "humantime_serde")]
355 : pub lagging_wal_timeout: Duration,
356 : /// Considers safekeepers lagging when their WAL is behind another safekeeper for more than this threshold.
357 : /// A lagging safekeeper will be changed after `lagging_wal_timeout` time elapses since the last WAL update,
358 : /// to avoid eager reconnects.
359 : pub max_lsn_wal_lag: NonZeroU64,
360 : pub eviction_policy: crate::models::EvictionPolicy,
361 : pub min_resident_size_override: Option<u64>,
362 : // See the corresponding metric's help string.
363 : #[serde(with = "humantime_serde")]
364 : pub evictions_low_residence_duration_metric_threshold: Duration,
365 :
366 : /// If non-zero, the period between uploads of a heatmap from attached tenants. This
367 : /// may be disabled if a Tenant will not have secondary locations: only secondary
368 : /// locations will use the heatmap uploaded by attached locations.
369 : #[serde(with = "humantime_serde")]
370 : pub heatmap_period: Duration,
371 :
372 : /// If true then SLRU segments are dowloaded on demand, if false SLRU segments are included in basebackup
373 : pub lazy_slru_download: bool,
374 :
375 : pub timeline_get_throttle: crate::models::ThrottleConfig,
376 :
377 : // How much WAL must be ingested before checking again whether a new image layer is required.
378 : // Expresed in multiples of checkpoint distance.
379 : pub image_layer_creation_check_threshold: u8,
380 :
381 : // How many multiples of L0 `compaction_threshold` will preempt image layer creation and do L0 compaction.
382 : // Set to 0 to disable preemption.
383 : pub image_creation_preempt_threshold: usize,
384 :
385 : /// The length for an explicit LSN lease request.
386 : /// Layers needed to reconstruct pages at LSN will not be GC-ed during this interval.
387 : #[serde(with = "humantime_serde")]
388 : pub lsn_lease_length: Duration,
389 :
390 : /// The length for an implicit LSN lease granted as part of `get_lsn_by_timestamp` request.
391 : /// Layers needed to reconstruct pages at LSN will not be GC-ed during this interval.
392 : #[serde(with = "humantime_serde")]
393 : pub lsn_lease_length_for_ts: Duration,
394 :
395 : /// Enable auto-offloading of timelines.
396 : /// (either this flag or the pageserver-global one need to be set)
397 : pub timeline_offloading: bool,
398 :
399 : pub wal_receiver_protocol_override: Option<PostgresClientProtocol>,
400 :
401 : /// Enable rel_size_v2 for this tenant. Once enabled, the tenant will persist this information into
402 : /// `index_part.json`, and it cannot be reversed.
403 : pub rel_size_v2_enabled: bool,
404 :
405 : // gc-compaction related configs
406 : /// Enable automatic gc-compaction trigger on this tenant.
407 : pub gc_compaction_enabled: bool,
408 : /// The initial threshold for gc-compaction in KB. Once the total size of layers below the gc-horizon is above this threshold,
409 : /// gc-compaction will be triggered.
410 : pub gc_compaction_initial_threshold_kb: u64,
411 : /// 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)
412 : /// is above this ratio, gc-compaction will be triggered.
413 : pub gc_compaction_ratio_percent: u64,
414 : }
415 :
416 : pub mod defaults {
417 : pub use storage_broker::DEFAULT_ENDPOINT as BROKER_DEFAULT_ENDPOINT;
418 :
419 : use crate::models::ImageCompressionAlgorithm;
420 :
421 : pub const DEFAULT_WAIT_LSN_TIMEOUT: &str = "300 s";
422 : pub const DEFAULT_WAL_REDO_TIMEOUT: &str = "60 s";
423 :
424 : pub const DEFAULT_SUPERUSER: &str = "cloud_admin";
425 : pub const DEFAULT_LOCALE: &str = if cfg!(target_os = "macos") {
426 : "C"
427 : } else {
428 : "C.UTF-8"
429 : };
430 :
431 : pub const DEFAULT_PAGE_CACHE_SIZE: usize = 8192;
432 : pub const DEFAULT_MAX_FILE_DESCRIPTORS: usize = 100;
433 :
434 : pub const DEFAULT_LOG_FORMAT: &str = "plain";
435 :
436 : pub const DEFAULT_CONCURRENT_TENANT_WARMUP: usize = 8;
437 :
438 : pub const DEFAULT_CONCURRENT_TENANT_SIZE_LOGICAL_SIZE_QUERIES: usize = 1;
439 :
440 : pub const DEFAULT_METRIC_COLLECTION_INTERVAL: &str = "10 min";
441 : pub const DEFAULT_METRIC_COLLECTION_ENDPOINT: Option<reqwest::Url> = None;
442 : pub const DEFAULT_SYNTHETIC_SIZE_CALCULATION_INTERVAL: &str = "10 min";
443 : pub const DEFAULT_BACKGROUND_TASK_MAXIMUM_DELAY: &str = "10s";
444 :
445 : pub const DEFAULT_HEATMAP_UPLOAD_CONCURRENCY: usize = 8;
446 : pub const DEFAULT_SECONDARY_DOWNLOAD_CONCURRENCY: usize = 1;
447 :
448 : pub const DEFAULT_INGEST_BATCH_SIZE: u64 = 100;
449 :
450 : /// Soft limit for the maximum size of a vectored read.
451 : ///
452 : /// This is determined by the largest NeonWalRecord that can exist (minus dbdir and reldir keys
453 : /// which are bounded by the blob io limits only). As of this writing, that is a `NeonWalRecord::ClogSetCommitted` record,
454 : /// with 32k xids. That's the max number of XIDS on a single CLOG page. The size of such a record
455 : /// is `sizeof(Transactionid) * 32768 + (some fixed overhead from 'timestamp`, the Vec length and whatever extra serde serialization adds)`.
456 : /// That is, slightly above 128 kB.
457 : pub const DEFAULT_MAX_VECTORED_READ_BYTES: usize = 130 * 1024; // 130 KiB
458 :
459 : pub const DEFAULT_IMAGE_COMPRESSION: ImageCompressionAlgorithm =
460 : ImageCompressionAlgorithm::Zstd { level: Some(1) };
461 :
462 : pub const DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB: usize = 0;
463 :
464 : pub const DEFAULT_IO_BUFFER_ALIGNMENT: usize = 512;
465 :
466 : pub const DEFAULT_WAL_RECEIVER_PROTOCOL: utils::postgres_client::PostgresClientProtocol =
467 : utils::postgres_client::PostgresClientProtocol::Vanilla;
468 :
469 : pub const DEFAULT_SSL_KEY_FILE: &str = "server.key";
470 : pub const DEFAULT_SSL_CERT_FILE: &str = "server.crt";
471 : }
472 :
473 : impl Default for ConfigToml {
474 488 : fn default() -> Self {
475 : use defaults::*;
476 :
477 : Self {
478 488 : listen_pg_addr: (DEFAULT_PG_LISTEN_ADDR.to_string()),
479 488 : listen_http_addr: (DEFAULT_HTTP_LISTEN_ADDR.to_string()),
480 488 : listen_https_addr: (None),
481 488 : ssl_key_file: Utf8PathBuf::from(DEFAULT_SSL_KEY_FILE),
482 488 : ssl_cert_file: Utf8PathBuf::from(DEFAULT_SSL_CERT_FILE),
483 488 : availability_zone: (None),
484 488 : wait_lsn_timeout: (humantime::parse_duration(DEFAULT_WAIT_LSN_TIMEOUT)
485 488 : .expect("cannot parse default wait lsn timeout")),
486 488 : wal_redo_timeout: (humantime::parse_duration(DEFAULT_WAL_REDO_TIMEOUT)
487 488 : .expect("cannot parse default wal redo timeout")),
488 488 : superuser: (DEFAULT_SUPERUSER.to_string()),
489 488 : locale: DEFAULT_LOCALE.to_string(),
490 488 : page_cache_size: (DEFAULT_PAGE_CACHE_SIZE),
491 488 : max_file_descriptors: (DEFAULT_MAX_FILE_DESCRIPTORS),
492 488 : pg_distrib_dir: None, // Utf8PathBuf::from("./pg_install"), // TODO: formely, this was std::env::current_dir()
493 488 : http_auth_type: (AuthType::Trust),
494 488 : pg_auth_type: (AuthType::Trust),
495 488 : auth_validation_public_key_path: (None),
496 488 : remote_storage: None,
497 488 : broker_endpoint: (storage_broker::DEFAULT_ENDPOINT
498 488 : .parse()
499 488 : .expect("failed to parse default broker endpoint")),
500 488 : broker_keepalive_interval: (humantime::parse_duration(
501 488 : storage_broker::DEFAULT_KEEPALIVE_INTERVAL,
502 488 : )
503 488 : .expect("cannot parse default keepalive interval")),
504 488 : log_format: (LogFormat::from_str(DEFAULT_LOG_FORMAT).unwrap()),
505 488 :
506 488 : concurrent_tenant_warmup: (NonZeroUsize::new(DEFAULT_CONCURRENT_TENANT_WARMUP)
507 488 : .expect("Invalid default constant")),
508 488 : concurrent_tenant_size_logical_size_queries: NonZeroUsize::new(
509 488 : DEFAULT_CONCURRENT_TENANT_SIZE_LOGICAL_SIZE_QUERIES,
510 488 : )
511 488 : .unwrap(),
512 488 : metric_collection_interval: (humantime::parse_duration(
513 488 : DEFAULT_METRIC_COLLECTION_INTERVAL,
514 488 : )
515 488 : .expect("cannot parse default metric collection interval")),
516 488 : synthetic_size_calculation_interval: (humantime::parse_duration(
517 488 : DEFAULT_SYNTHETIC_SIZE_CALCULATION_INTERVAL,
518 488 : )
519 488 : .expect("cannot parse default synthetic size calculation interval")),
520 488 : metric_collection_endpoint: (DEFAULT_METRIC_COLLECTION_ENDPOINT),
521 488 :
522 488 : metric_collection_bucket: (None),
523 488 :
524 488 : disk_usage_based_eviction: (None),
525 488 :
526 488 : test_remote_failures: (0),
527 488 :
528 488 : ondemand_download_behavior_treat_error_as_warn: (false),
529 488 :
530 488 : background_task_maximum_delay: (humantime::parse_duration(
531 488 : DEFAULT_BACKGROUND_TASK_MAXIMUM_DELAY,
532 488 : )
533 488 : .unwrap()),
534 488 :
535 488 : control_plane_api: (None),
536 488 : control_plane_api_token: (None),
537 488 : control_plane_emergency_mode: (false),
538 488 :
539 488 : import_pgdata_upcall_api: (None),
540 488 : import_pgdata_upcall_api_token: (None),
541 488 : import_pgdata_aws_endpoint_url: (None),
542 488 :
543 488 : heatmap_upload_concurrency: (DEFAULT_HEATMAP_UPLOAD_CONCURRENCY),
544 488 : secondary_download_concurrency: (DEFAULT_SECONDARY_DOWNLOAD_CONCURRENCY),
545 488 :
546 488 : ingest_batch_size: (DEFAULT_INGEST_BATCH_SIZE),
547 488 :
548 488 : virtual_file_io_engine: None,
549 488 :
550 488 : max_vectored_read_bytes: (MaxVectoredReadBytes(
551 488 : NonZeroUsize::new(DEFAULT_MAX_VECTORED_READ_BYTES).unwrap(),
552 488 : )),
553 488 : image_compression: (DEFAULT_IMAGE_COMPRESSION),
554 488 : timeline_offloading: true,
555 488 : ephemeral_bytes_per_memory_kb: (DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB),
556 488 : l0_flush: None,
557 488 : virtual_file_io_mode: None,
558 488 : tenant_config: TenantConfigToml::default(),
559 488 : no_sync: None,
560 488 : wal_receiver_protocol: DEFAULT_WAL_RECEIVER_PROTOCOL,
561 488 : page_service_pipelining: if !cfg!(test) {
562 488 : PageServicePipeliningConfig::Serial
563 : } else {
564 0 : PageServicePipeliningConfig::Pipelined(PageServicePipeliningConfigPipelined {
565 0 : max_batch_size: NonZeroUsize::new(32).unwrap(),
566 0 : execution: PageServiceProtocolPipelinedExecutionStrategy::ConcurrentFutures,
567 0 : })
568 : },
569 488 : get_vectored_concurrent_io: if !cfg!(test) {
570 488 : GetVectoredConcurrentIo::Sequential
571 : } else {
572 0 : GetVectoredConcurrentIo::SidecarTask
573 : },
574 488 : enable_read_path_debugging: if cfg!(test) || cfg!(feature = "testing") {
575 488 : Some(true)
576 : } else {
577 0 : None
578 : },
579 488 : validate_wal_contiguity: None,
580 488 : load_previous_heatmap: None,
581 488 : generate_unarchival_heatmap: None,
582 488 : }
583 488 : }
584 : }
585 :
586 : pub mod tenant_conf_defaults {
587 :
588 : // FIXME: This current value is very low. I would imagine something like 1 GB or 10 GB
589 : // would be more appropriate. But a low value forces the code to be exercised more,
590 : // which is good for now to trigger bugs.
591 : // This parameter actually determines L0 layer file size.
592 : pub const DEFAULT_CHECKPOINT_DISTANCE: u64 = 256 * 1024 * 1024;
593 : pub const DEFAULT_CHECKPOINT_TIMEOUT: &str = "10 m";
594 :
595 : // FIXME the below configs are only used by legacy algorithm. The new algorithm
596 : // has different parameters.
597 :
598 : // Target file size, when creating image and delta layers.
599 : // This parameter determines L1 layer file size.
600 : pub const DEFAULT_COMPACTION_TARGET_SIZE: u64 = 128 * 1024 * 1024;
601 :
602 : pub const DEFAULT_COMPACTION_PERIOD: &str = "20 s";
603 : pub const DEFAULT_COMPACTION_THRESHOLD: usize = 10;
604 :
605 : // This value needs to be tuned to avoid OOM. We have 3/4*CPUs threads for L0 compaction, that's
606 : // 3/4*16=9 on most of our pageservers. Compacting 20 layers requires about 1 GB memory (could
607 : // be reduced later by optimizing L0 hole calculation to avoid loading all keys into memory). So
608 : // with this config, we can get a maximum peak compaction usage of 9 GB.
609 : pub const DEFAULT_COMPACTION_UPPER_LIMIT: usize = 20;
610 : // Enable L0 compaction pass and semaphore by default. L0 compaction must be responsive to avoid
611 : // read amp.
612 : pub const DEFAULT_COMPACTION_L0_FIRST: bool = true;
613 : pub const DEFAULT_COMPACTION_L0_SEMAPHORE: bool = true;
614 :
615 : pub const DEFAULT_COMPACTION_ALGORITHM: crate::models::CompactionAlgorithm =
616 : crate::models::CompactionAlgorithm::Legacy;
617 :
618 : pub const DEFAULT_L0_FLUSH_WAIT_UPLOAD: bool = false;
619 :
620 : pub const DEFAULT_GC_HORIZON: u64 = 64 * 1024 * 1024;
621 :
622 : // Large DEFAULT_GC_PERIOD is fine as long as PITR_INTERVAL is larger.
623 : // If there's a need to decrease this value, first make sure that GC
624 : // doesn't hold a layer map write lock for non-trivial operations.
625 : // Relevant: https://github.com/neondatabase/neon/issues/3394
626 : pub const DEFAULT_GC_PERIOD: &str = "1 hr";
627 : pub const DEFAULT_IMAGE_CREATION_THRESHOLD: usize = 3;
628 : // If there are more than threshold * compaction_threshold (that is 3 * 10 in the default config) L0 layers, image
629 : // layer creation will end immediately. Set to 0 to disable.
630 : pub const DEFAULT_IMAGE_CREATION_PREEMPT_THRESHOLD: usize = 3;
631 : pub const DEFAULT_PITR_INTERVAL: &str = "7 days";
632 : pub const DEFAULT_WALRECEIVER_CONNECT_TIMEOUT: &str = "10 seconds";
633 : pub const DEFAULT_WALRECEIVER_LAGGING_WAL_TIMEOUT: &str = "10 seconds";
634 : // The default limit on WAL lag should be set to avoid causing disconnects under high throughput
635 : // scenarios: since the broker stats are updated ~1/s, a value of 1GiB should be sufficient for
636 : // throughputs up to 1GiB/s per timeline.
637 : pub const DEFAULT_MAX_WALRECEIVER_LSN_WAL_LAG: u64 = 1024 * 1024 * 1024;
638 : pub const DEFAULT_EVICTIONS_LOW_RESIDENCE_DURATION_METRIC_THRESHOLD: &str = "24 hour";
639 : // By default ingest enough WAL for two new L0 layers before checking if new image
640 : // image layers should be created.
641 : pub const DEFAULT_IMAGE_LAYER_CREATION_CHECK_THRESHOLD: u8 = 2;
642 : pub const DEFAULT_GC_COMPACTION_ENABLED: bool = false;
643 : pub const DEFAULT_GC_COMPACTION_INITIAL_THRESHOLD_KB: u64 = 5 * 1024 * 1024; // 5GB
644 : pub const DEFAULT_GC_COMPACTION_RATIO_PERCENT: u64 = 100;
645 : }
646 :
647 : impl Default for TenantConfigToml {
648 488 : fn default() -> Self {
649 : use tenant_conf_defaults::*;
650 488 : Self {
651 488 : checkpoint_distance: DEFAULT_CHECKPOINT_DISTANCE,
652 488 : checkpoint_timeout: humantime::parse_duration(DEFAULT_CHECKPOINT_TIMEOUT)
653 488 : .expect("cannot parse default checkpoint timeout"),
654 488 : compaction_target_size: DEFAULT_COMPACTION_TARGET_SIZE,
655 488 : compaction_period: humantime::parse_duration(DEFAULT_COMPACTION_PERIOD)
656 488 : .expect("cannot parse default compaction period"),
657 488 : compaction_threshold: DEFAULT_COMPACTION_THRESHOLD,
658 488 : compaction_upper_limit: DEFAULT_COMPACTION_UPPER_LIMIT,
659 488 : compaction_algorithm: crate::models::CompactionAlgorithmSettings {
660 488 : kind: DEFAULT_COMPACTION_ALGORITHM,
661 488 : },
662 488 : compaction_l0_first: DEFAULT_COMPACTION_L0_FIRST,
663 488 : compaction_l0_semaphore: DEFAULT_COMPACTION_L0_SEMAPHORE,
664 488 : l0_flush_delay_threshold: None,
665 488 : l0_flush_stall_threshold: None,
666 488 : l0_flush_wait_upload: DEFAULT_L0_FLUSH_WAIT_UPLOAD,
667 488 : gc_horizon: DEFAULT_GC_HORIZON,
668 488 : gc_period: humantime::parse_duration(DEFAULT_GC_PERIOD)
669 488 : .expect("cannot parse default gc period"),
670 488 : image_creation_threshold: DEFAULT_IMAGE_CREATION_THRESHOLD,
671 488 : pitr_interval: humantime::parse_duration(DEFAULT_PITR_INTERVAL)
672 488 : .expect("cannot parse default PITR interval"),
673 488 : walreceiver_connect_timeout: humantime::parse_duration(
674 488 : DEFAULT_WALRECEIVER_CONNECT_TIMEOUT,
675 488 : )
676 488 : .expect("cannot parse default walreceiver connect timeout"),
677 488 : lagging_wal_timeout: humantime::parse_duration(DEFAULT_WALRECEIVER_LAGGING_WAL_TIMEOUT)
678 488 : .expect("cannot parse default walreceiver lagging wal timeout"),
679 488 : max_lsn_wal_lag: NonZeroU64::new(DEFAULT_MAX_WALRECEIVER_LSN_WAL_LAG)
680 488 : .expect("cannot parse default max walreceiver Lsn wal lag"),
681 488 : eviction_policy: crate::models::EvictionPolicy::NoEviction,
682 488 : min_resident_size_override: None,
683 488 : evictions_low_residence_duration_metric_threshold: humantime::parse_duration(
684 488 : DEFAULT_EVICTIONS_LOW_RESIDENCE_DURATION_METRIC_THRESHOLD,
685 488 : )
686 488 : .expect("cannot parse default evictions_low_residence_duration_metric_threshold"),
687 488 : heatmap_period: Duration::ZERO,
688 488 : lazy_slru_download: false,
689 488 : timeline_get_throttle: crate::models::ThrottleConfig::disabled(),
690 488 : image_layer_creation_check_threshold: DEFAULT_IMAGE_LAYER_CREATION_CHECK_THRESHOLD,
691 488 : image_creation_preempt_threshold: DEFAULT_IMAGE_CREATION_PREEMPT_THRESHOLD,
692 488 : lsn_lease_length: LsnLease::DEFAULT_LENGTH,
693 488 : lsn_lease_length_for_ts: LsnLease::DEFAULT_LENGTH_FOR_TS,
694 488 : timeline_offloading: true,
695 488 : wal_receiver_protocol_override: None,
696 488 : rel_size_v2_enabled: false,
697 488 : gc_compaction_enabled: DEFAULT_GC_COMPACTION_ENABLED,
698 488 : gc_compaction_initial_threshold_kb: DEFAULT_GC_COMPACTION_INITIAL_THRESHOLD_KB,
699 488 : gc_compaction_ratio_percent: DEFAULT_GC_COMPACTION_RATIO_PERCENT,
700 488 : }
701 488 : }
702 : }
|
