Line data Source code
1 : //! The ComputeSpec contains all the information needed to start up
2 : //! the right version of PostgreSQL, and connect it to the storage nodes.
3 : //! It can be passed as part of the `config.json`, or the control plane can
4 : //! provide it by calling the compute_ctl's `/compute_ctl` endpoint, or
5 : //! compute_ctl can fetch it by calling the control plane's API.
6 : use std::collections::HashMap;
7 :
8 : use indexmap::IndexMap;
9 : use regex::Regex;
10 : use remote_storage::RemotePath;
11 : use serde::{Deserialize, Serialize};
12 : use utils::id::{TenantId, TimelineId};
13 : use utils::lsn::Lsn;
14 :
15 : use crate::responses::TlsConfig;
16 :
17 : /// String type alias representing Postgres identifier and
18 : /// intended to be used for DB / role names.
19 : pub type PgIdent = String;
20 :
21 : /// String type alias representing Postgres extension version
22 : pub type ExtVersion = String;
23 :
24 6 : fn default_reconfigure_concurrency() -> usize {
25 6 : 1
26 6 : }
27 :
28 : /// Cluster spec or configuration represented as an optional number of
29 : /// delta operations + final cluster state description.
30 45 : #[derive(Clone, Debug, Default, Deserialize, Serialize)]
31 : pub struct ComputeSpec {
32 : pub format_version: f32,
33 :
34 : // The control plane also includes a 'timestamp' field in the JSON document,
35 : // but we don't use it for anything. Serde will ignore missing fields when
36 : // deserializing it.
37 : pub operation_uuid: Option<String>,
38 :
39 : /// Compute features to enable. These feature flags are provided, when we
40 : /// know all the details about client's compute, so they cannot be used
41 : /// to change `Empty` compute behavior.
42 : #[serde(default)]
43 : pub features: Vec<ComputeFeature>,
44 :
45 : /// If compute_ctl was passed `--resize-swap-on-bind`, a value of `Some(_)` instructs
46 : /// compute_ctl to `/neonvm/bin/resize-swap` with the given size, when the spec is first
47 : /// received.
48 : ///
49 : /// Both this field and `--resize-swap-on-bind` are required, so that the control plane's
50 : /// spec generation doesn't need to be aware of the actual compute it's running on, while
51 : /// guaranteeing gradual rollout of swap. Otherwise, without `--resize-swap-on-bind`, we could
52 : /// end up trying to resize swap in VMs without it -- or end up *not* resizing swap, thus
53 : /// giving every VM much more swap than it should have (32GiB).
54 : ///
55 : /// Eventually we may remove `--resize-swap-on-bind` and exclusively use `swap_size_bytes` for
56 : /// enabling the swap resizing behavior once rollout is complete.
57 : ///
58 : /// See neondatabase/cloud#12047 for more.
59 : #[serde(default)]
60 : pub swap_size_bytes: Option<u64>,
61 :
62 : /// If compute_ctl was passed `--set-disk-quota-for-fs`, a value of `Some(_)` instructs
63 : /// compute_ctl to run `/neonvm/bin/set-disk-quota` with the given size and fs, when the
64 : /// spec is first received.
65 : ///
66 : /// Both this field and `--set-disk-quota-for-fs` are required, so that the control plane's
67 : /// spec generation doesn't need to be aware of the actual compute it's running on, while
68 : /// guaranteeing gradual rollout of disk quota.
69 : #[serde(default)]
70 : pub disk_quota_bytes: Option<u64>,
71 :
72 : /// Disables the vm-monitor behavior that resizes LFC on upscale/downscale, instead relying on
73 : /// the initial size of LFC.
74 : ///
75 : /// This is intended for use when the LFC size is being overridden from the default but
76 : /// autoscaling is still enabled, and we don't want the vm-monitor to interfere with the custom
77 : /// LFC sizing.
78 : #[serde(default)]
79 : pub disable_lfc_resizing: Option<bool>,
80 :
81 : /// Expected cluster state at the end of transition process.
82 : pub cluster: Cluster,
83 : pub delta_operations: Option<Vec<DeltaOp>>,
84 :
85 : /// An optional hint that can be passed to speed up startup time if we know
86 : /// that no pg catalog mutations (like role creation, database creation,
87 : /// extension creation) need to be done on the actual database to start.
88 : #[serde(default)] // Default false
89 : pub skip_pg_catalog_updates: bool,
90 :
91 : // Information needed to connect to the storage layer.
92 : //
93 : // `tenant_id`, `timeline_id` and `pageserver_connstring` are always needed.
94 : //
95 : // Depending on `mode`, this can be a primary read-write node, a read-only
96 : // replica, or a read-only node pinned at an older LSN.
97 : // `safekeeper_connstrings` must be set for a primary.
98 : //
99 : // For backwards compatibility, the control plane may leave out all of
100 : // these, and instead set the "neon.tenant_id", "neon.timeline_id",
101 : // etc. GUCs in cluster.settings. TODO: Once the control plane has been
102 : // updated to fill these fields, we can make these non optional.
103 : pub tenant_id: Option<TenantId>,
104 : pub timeline_id: Option<TimelineId>,
105 : pub pageserver_connstring: Option<String>,
106 :
107 : // More neon ids that we expose to the compute_ctl
108 : // and to postgres as neon extension GUCs.
109 : pub project_id: Option<String>,
110 : pub branch_id: Option<String>,
111 : pub endpoint_id: Option<String>,
112 :
113 : /// Safekeeper membership config generation. It is put in
114 : /// neon.safekeepers GUC and serves two purposes:
115 : /// 1) Non zero value forces walproposer to use membership configurations.
116 : /// 2) If walproposer wants to update list of safekeepers to connect to
117 : /// taking them from some safekeeper mconf, it should check what value
118 : /// is newer by comparing the generation.
119 : ///
120 : /// Note: it could be SafekeeperGeneration, but this needs linking
121 : /// compute_ctl with postgres_ffi.
122 : #[serde(default)]
123 : pub safekeepers_generation: Option<u32>,
124 : #[serde(default)]
125 : pub safekeeper_connstrings: Vec<String>,
126 :
127 : #[serde(default)]
128 : pub mode: ComputeMode,
129 :
130 : /// If set, 'storage_auth_token' is used as the password to authenticate to
131 : /// the pageserver and safekeepers.
132 : pub storage_auth_token: Option<String>,
133 :
134 : // information about available remote extensions
135 : pub remote_extensions: Option<RemoteExtSpec>,
136 :
137 : pub pgbouncer_settings: Option<IndexMap<String, String>>,
138 :
139 : // Stripe size for pageserver sharding, in pages
140 : #[serde(default)]
141 : pub shard_stripe_size: Option<usize>,
142 :
143 : /// Local Proxy configuration used for JWT authentication
144 : #[serde(default)]
145 : pub local_proxy_config: Option<LocalProxySpec>,
146 :
147 : /// Number of concurrent connections during the parallel RunInEachDatabase
148 : /// phase of the apply config process.
149 : ///
150 : /// We need a higher concurrency during reconfiguration in case of many DBs,
151 : /// but instance is already running and used by client. We can easily get out of
152 : /// `max_connections` limit, and the current code won't handle that.
153 : ///
154 : /// Default is 1, but also allow control plane to override this value for specific
155 : /// projects. It's also recommended to bump `superuser_reserved_connections` +=
156 : /// `reconfigure_concurrency` for such projects to ensure that we always have
157 : /// enough spare connections for reconfiguration process to succeed.
158 : #[serde(default = "default_reconfigure_concurrency")]
159 : pub reconfigure_concurrency: usize,
160 :
161 : /// If set to true, the compute_ctl will drop all subscriptions before starting the
162 : /// compute. This is needed when we start an endpoint on a branch, so that child
163 : /// would not compete with parent branch subscriptions
164 : /// over the same replication content from publisher.
165 : #[serde(default)] // Default false
166 : pub drop_subscriptions_before_start: bool,
167 :
168 : /// Log level for compute audit logging
169 : #[serde(default)]
170 : pub audit_log_level: ComputeAudit,
171 :
172 : /// Hostname and the port of the otel collector. Leave empty to disable Postgres logs forwarding.
173 : /// Example: config-shy-breeze-123-collector-monitoring.neon-telemetry.svc.cluster.local:10514
174 : pub logs_export_host: Option<String>,
175 : }
176 :
177 : /// Feature flag to signal `compute_ctl` to enable certain experimental functionality.
178 3 : #[derive(Serialize, Clone, Copy, Debug, Deserialize, PartialEq, Eq)]
179 : #[serde(rename_all = "snake_case")]
180 : pub enum ComputeFeature {
181 : // XXX: Add more feature flags here.
182 : /// Enable the experimental activity monitor logic, which uses `pg_stat_database` to
183 : /// track short-lived connections as user activity.
184 : ActivityMonitorExperimental,
185 :
186 : /// This is a special feature flag that is used to represent unknown feature flags.
187 : /// Basically all unknown to enum flags are represented as this one. See unit test
188 : /// `parse_unknown_features()` for more details.
189 : #[serde(other)]
190 : UnknownFeature,
191 : }
192 :
193 44 : #[derive(Clone, Debug, Default, Deserialize, Serialize)]
194 : pub struct RemoteExtSpec {
195 : pub public_extensions: Option<Vec<String>>,
196 : pub custom_extensions: Option<Vec<String>>,
197 : pub library_index: HashMap<String, String>,
198 : pub extension_data: HashMap<String, ExtensionData>,
199 : }
200 :
201 18 : #[derive(Clone, Debug, Serialize, Deserialize)]
202 : pub struct ExtensionData {
203 : pub control_data: HashMap<String, String>,
204 : pub archive_path: String,
205 : }
206 :
207 : impl RemoteExtSpec {
208 6 : pub fn get_ext(
209 6 : &self,
210 6 : ext_name: &str,
211 6 : is_library: bool,
212 6 : build_tag: &str,
213 6 : pg_major_version: &str,
214 6 : ) -> anyhow::Result<(String, RemotePath)> {
215 6 : let mut real_ext_name = ext_name;
216 6 : if is_library {
217 : // sometimes library names might have a suffix like
218 : // library.so or library.so.3. We strip this off
219 : // because library_index is based on the name without the file extension
220 1 : let strip_lib_suffix = Regex::new(r"\.so.*").unwrap();
221 1 : let lib_raw_name = strip_lib_suffix.replace(real_ext_name, "").to_string();
222 1 :
223 1 : real_ext_name = self
224 1 : .library_index
225 1 : .get(&lib_raw_name)
226 1 : .ok_or(anyhow::anyhow!("library {} is not found", lib_raw_name))?;
227 5 : }
228 :
229 : // Check if extension is present in public or custom.
230 : // If not, then it is not allowed to be used by this compute.
231 6 : if !self
232 6 : .public_extensions
233 6 : .as_ref()
234 6 : .is_some_and(|exts| exts.iter().any(|e| e == real_ext_name))
235 4 : && !self
236 4 : .custom_extensions
237 4 : .as_ref()
238 4 : .is_some_and(|exts| exts.iter().any(|e| e == real_ext_name))
239 : {
240 3 : return Err(anyhow::anyhow!("extension {} is not found", real_ext_name));
241 3 : }
242 3 :
243 3 : match self.extension_data.get(real_ext_name) {
244 3 : Some(_ext_data) => {
245 3 : // Construct the path to the extension archive
246 3 : // BUILD_TAG/PG_MAJOR_VERSION/extensions/EXTENSION_NAME.tar.zst
247 3 : //
248 3 : // Keep it in sync with path generation in
249 3 : // https://github.com/neondatabase/build-custom-extensions/tree/main
250 3 : let archive_path_str =
251 3 : format!("{build_tag}/{pg_major_version}/extensions/{real_ext_name}.tar.zst");
252 3 : Ok((
253 3 : real_ext_name.to_string(),
254 3 : RemotePath::from_string(&archive_path_str)?,
255 : ))
256 : }
257 0 : None => Err(anyhow::anyhow!(
258 0 : "real_ext_name {} is not found",
259 0 : real_ext_name
260 0 : )),
261 : }
262 6 : }
263 : }
264 :
265 0 : #[derive(Clone, Copy, Debug, Default, Eq, PartialEq, Deserialize, Serialize)]
266 : pub enum ComputeMode {
267 : /// A read-write node
268 : #[default]
269 : Primary,
270 : /// A read-only node, pinned at a particular LSN
271 : Static(Lsn),
272 : /// A read-only node that follows the tip of the branch in hot standby mode
273 : ///
274 : /// Future versions may want to distinguish between replicas with hot standby
275 : /// feedback and other kinds of replication configurations.
276 : Replica,
277 : }
278 :
279 : impl ComputeMode {
280 : /// Convert the compute mode to a string that can be used to identify the type of compute,
281 : /// which means that if it's a static compute, the LSN will not be included.
282 0 : pub fn to_type_str(&self) -> &'static str {
283 0 : match self {
284 0 : ComputeMode::Primary => "primary",
285 0 : ComputeMode::Static(_) => "static",
286 0 : ComputeMode::Replica => "replica",
287 : }
288 0 : }
289 : }
290 :
291 : /// Log level for audit logging
292 0 : #[derive(Clone, Debug, Default, Eq, PartialEq, Deserialize, Serialize)]
293 : pub enum ComputeAudit {
294 : #[default]
295 : Disabled,
296 : // Deprecated, use Base instead
297 : Log,
298 : // (pgaudit.log = 'ddl', pgaudit.log_parameter='off')
299 : // logged to the standard postgresql log stream
300 : Base,
301 : // Deprecated, use Full or Extended instead
302 : Hipaa,
303 : // (pgaudit.log = 'all, -misc', pgaudit.log_parameter='off')
304 : // logged to separate files collected by rsyslog
305 : // into dedicated log storage with strict access
306 : Extended,
307 : // (pgaudit.log='all', pgaudit.log_parameter='on'),
308 : // logged to separate files collected by rsyslog
309 : // into dedicated log storage with strict access.
310 : Full,
311 : }
312 :
313 36 : #[derive(Clone, Debug, Default, Deserialize, Serialize, PartialEq, Eq)]
314 : pub struct Cluster {
315 : pub cluster_id: Option<String>,
316 : pub name: Option<String>,
317 : pub state: Option<String>,
318 : pub roles: Vec<Role>,
319 : pub databases: Vec<Database>,
320 :
321 : /// Desired contents of 'postgresql.conf' file. (The 'compute_ctl'
322 : /// tool may add additional settings to the final file.)
323 : pub postgresql_conf: Option<String>,
324 :
325 : /// Additional settings that will be appended to the 'postgresql.conf' file.
326 : pub settings: GenericOptions,
327 : }
328 :
329 : /// Single cluster state changing operation that could not be represented as
330 : /// a static `Cluster` structure. For example:
331 : /// - DROP DATABASE
332 : /// - DROP ROLE
333 : /// - ALTER ROLE name RENAME TO new_name
334 : /// - ALTER DATABASE name RENAME TO new_name
335 60 : #[derive(Clone, Debug, Deserialize, Serialize)]
336 : pub struct DeltaOp {
337 : pub action: String,
338 : pub name: PgIdent,
339 : pub new_name: Option<PgIdent>,
340 : }
341 :
342 : /// Rust representation of Postgres role info with only those fields
343 : /// that matter for us.
344 90 : #[derive(Clone, Debug, Deserialize, Serialize, PartialEq, Eq)]
345 : pub struct Role {
346 : pub name: PgIdent,
347 : pub encrypted_password: Option<String>,
348 : pub options: GenericOptions,
349 : }
350 :
351 : /// Rust representation of Postgres database info with only those fields
352 : /// that matter for us.
353 42 : #[derive(Clone, Debug, Deserialize, Serialize, PartialEq, Eq)]
354 : pub struct Database {
355 : pub name: PgIdent,
356 : pub owner: PgIdent,
357 : pub options: GenericOptions,
358 : // These are derived flags, not present in the spec file.
359 : // They are never set by the control plane.
360 : #[serde(skip_deserializing, default)]
361 : pub restrict_conn: bool,
362 : #[serde(skip_deserializing, default)]
363 : pub invalid: bool,
364 : }
365 :
366 : /// Common type representing both SQL statement params with or without value,
367 : /// like `LOGIN` or `OWNER username` in the `CREATE/ALTER ROLE`, and config
368 : /// options like `wal_level = logical`.
369 468 : #[derive(Clone, Debug, Deserialize, Serialize, PartialEq, Eq)]
370 : pub struct GenericOption {
371 : pub name: String,
372 : pub value: Option<String>,
373 : pub vartype: String,
374 : }
375 :
376 : /// Optional collection of `GenericOption`'s. Type alias allows us to
377 : /// declare a `trait` on it.
378 : pub type GenericOptions = Option<Vec<GenericOption>>;
379 :
380 : /// Configured the local_proxy application with the relevant JWKS and roles it should
381 : /// use for authorizing connect requests using JWT.
382 0 : #[derive(Clone, Debug, Deserialize, Serialize)]
383 : pub struct LocalProxySpec {
384 : #[serde(default)]
385 : #[serde(skip_serializing_if = "Option::is_none")]
386 : pub jwks: Option<Vec<JwksSettings>>,
387 : #[serde(default)]
388 : #[serde(skip_serializing_if = "Option::is_none")]
389 : pub tls: Option<TlsConfig>,
390 : }
391 :
392 0 : #[derive(Clone, Debug, Deserialize, Serialize)]
393 : pub struct JwksSettings {
394 : pub id: String,
395 : pub role_names: Vec<String>,
396 : pub jwks_url: String,
397 : pub provider_name: String,
398 : pub jwt_audience: Option<String>,
399 : }
400 :
401 : #[cfg(test)]
402 : mod tests {
403 : use std::fs::File;
404 :
405 : use super::*;
406 :
407 : #[test]
408 1 : fn allow_installing_remote_extensions() {
409 1 : let rspec: RemoteExtSpec = serde_json::from_value(serde_json::json!({
410 1 : "public_extensions": null,
411 1 : "custom_extensions": null,
412 1 : "library_index": {},
413 1 : "extension_data": {},
414 1 : }))
415 1 : .unwrap();
416 1 :
417 1 : rspec
418 1 : .get_ext("ext", false, "latest", "v17")
419 1 : .expect_err("Extension should not be found");
420 1 :
421 1 : let rspec: RemoteExtSpec = serde_json::from_value(serde_json::json!({
422 1 : "public_extensions": [],
423 1 : "custom_extensions": null,
424 1 : "library_index": {},
425 1 : "extension_data": {},
426 1 : }))
427 1 : .unwrap();
428 1 :
429 1 : rspec
430 1 : .get_ext("ext", false, "latest", "v17")
431 1 : .expect_err("Extension should not be found");
432 1 :
433 1 : let rspec: RemoteExtSpec = serde_json::from_value(serde_json::json!({
434 1 : "public_extensions": [],
435 1 : "custom_extensions": [],
436 1 : "library_index": {
437 1 : "ext": "ext"
438 1 : },
439 1 : "extension_data": {
440 1 : "ext": {
441 1 : "control_data": {
442 1 : "ext.control": ""
443 1 : },
444 1 : "archive_path": ""
445 1 : }
446 1 : },
447 1 : }))
448 1 : .unwrap();
449 1 :
450 1 : rspec
451 1 : .get_ext("ext", false, "latest", "v17")
452 1 : .expect_err("Extension should not be found");
453 1 :
454 1 : let rspec: RemoteExtSpec = serde_json::from_value(serde_json::json!({
455 1 : "public_extensions": [],
456 1 : "custom_extensions": ["ext"],
457 1 : "library_index": {
458 1 : "ext": "ext"
459 1 : },
460 1 : "extension_data": {
461 1 : "ext": {
462 1 : "control_data": {
463 1 : "ext.control": ""
464 1 : },
465 1 : "archive_path": ""
466 1 : }
467 1 : },
468 1 : }))
469 1 : .unwrap();
470 1 :
471 1 : rspec
472 1 : .get_ext("ext", false, "latest", "v17")
473 1 : .expect("Extension should be found");
474 1 :
475 1 : let rspec: RemoteExtSpec = serde_json::from_value(serde_json::json!({
476 1 : "public_extensions": ["ext"],
477 1 : "custom_extensions": [],
478 1 : "library_index": {
479 1 : "extlib": "ext",
480 1 : },
481 1 : "extension_data": {
482 1 : "ext": {
483 1 : "control_data": {
484 1 : "ext.control": ""
485 1 : },
486 1 : "archive_path": ""
487 1 : }
488 1 : },
489 1 : }))
490 1 : .unwrap();
491 1 :
492 1 : rspec
493 1 : .get_ext("ext", false, "latest", "v17")
494 1 : .expect("Extension should be found");
495 1 :
496 1 : // test library index for the case when library name
497 1 : // doesn't match the extension name
498 1 : rspec
499 1 : .get_ext("extlib", true, "latest", "v17")
500 1 : .expect("Library should be found");
501 1 : }
502 :
503 : #[test]
504 1 : fn parse_spec_file() {
505 1 : let file = File::open("tests/cluster_spec.json").unwrap();
506 1 : let spec: ComputeSpec = serde_json::from_reader(file).unwrap();
507 1 :
508 1 : // Features list defaults to empty vector.
509 1 : assert!(spec.features.is_empty());
510 :
511 : // Reconfigure concurrency defaults to 1.
512 1 : assert_eq!(spec.reconfigure_concurrency, 1);
513 1 : }
514 :
515 : #[test]
516 1 : fn parse_unknown_fields() {
517 1 : // Forward compatibility test
518 1 : let file = File::open("tests/cluster_spec.json").unwrap();
519 1 : let mut json: serde_json::Value = serde_json::from_reader(file).unwrap();
520 1 : let ob = json.as_object_mut().unwrap();
521 1 : ob.insert("unknown_field_123123123".into(), "hello".into());
522 1 : let _spec: ComputeSpec = serde_json::from_value(json).unwrap();
523 1 : }
524 :
525 : #[test]
526 1 : fn parse_unknown_features() {
527 1 : // Test that unknown feature flags do not cause any errors.
528 1 : let file = File::open("tests/cluster_spec.json").unwrap();
529 1 : let mut json: serde_json::Value = serde_json::from_reader(file).unwrap();
530 1 : let ob = json.as_object_mut().unwrap();
531 1 :
532 1 : // Add unknown feature flags.
533 1 : let features = vec!["foo_bar_feature", "baz_feature"];
534 1 : ob.insert("features".into(), features.into());
535 1 :
536 1 : let spec: ComputeSpec = serde_json::from_value(json).unwrap();
537 1 :
538 1 : assert!(spec.features.len() == 2);
539 1 : assert!(spec.features.contains(&ComputeFeature::UnknownFeature));
540 1 : assert_eq!(spec.features, vec![ComputeFeature::UnknownFeature; 2]);
541 1 : }
542 :
543 : #[test]
544 1 : fn parse_known_features() {
545 1 : // Test that we can properly parse known feature flags.
546 1 : let file = File::open("tests/cluster_spec.json").unwrap();
547 1 : let mut json: serde_json::Value = serde_json::from_reader(file).unwrap();
548 1 : let ob = json.as_object_mut().unwrap();
549 1 :
550 1 : // Add known feature flags.
551 1 : let features = vec!["activity_monitor_experimental"];
552 1 : ob.insert("features".into(), features.into());
553 1 :
554 1 : let spec: ComputeSpec = serde_json::from_value(json).unwrap();
555 1 :
556 1 : assert_eq!(
557 1 : spec.features,
558 1 : vec![ComputeFeature::ActivityMonitorExperimental]
559 1 : );
560 1 : }
561 : }
|
