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