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 51 : #[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 : /// Expected cluster state at the end of transition process.
71 : pub cluster: Cluster,
72 : pub delta_operations: Option<Vec<DeltaOp>>,
73 :
74 : /// An optional hint that can be passed to speed up startup time if we know
75 : /// that no pg catalog mutations (like role creation, database creation,
76 : /// extension creation) need to be done on the actual database to start.
77 : #[serde(default)] // Default false
78 : pub skip_pg_catalog_updates: bool,
79 :
80 : // Information needed to connect to the storage layer.
81 : //
82 : // `tenant_id`, `timeline_id` and `pageserver_connstring` are always needed.
83 : //
84 : // Depending on `mode`, this can be a primary read-write node, a read-only
85 : // replica, or a read-only node pinned at an older LSN.
86 : // `safekeeper_connstrings` must be set for a primary.
87 : //
88 : // For backwards compatibility, the control plane may leave out all of
89 : // these, and instead set the "neon.tenant_id", "neon.timeline_id",
90 : // etc. GUCs in cluster.settings. TODO: Once the control plane has been
91 : // updated to fill these fields, we can make these non optional.
92 : pub tenant_id: Option<TenantId>,
93 : pub timeline_id: Option<TimelineId>,
94 : pub pageserver_connstring: Option<String>,
95 :
96 : #[serde(default)]
97 : pub safekeeper_connstrings: Vec<String>,
98 :
99 : #[serde(default)]
100 : pub mode: ComputeMode,
101 :
102 : /// If set, 'storage_auth_token' is used as the password to authenticate to
103 : /// the pageserver and safekeepers.
104 : pub storage_auth_token: Option<String>,
105 :
106 : // information about available remote extensions
107 : pub remote_extensions: Option<RemoteExtSpec>,
108 :
109 : pub pgbouncer_settings: Option<HashMap<String, String>>,
110 :
111 : // Stripe size for pageserver sharding, in pages
112 : #[serde(default)]
113 : pub shard_stripe_size: Option<usize>,
114 :
115 : /// Local Proxy configuration used for JWT authentication
116 : #[serde(default)]
117 : pub local_proxy_config: Option<LocalProxySpec>,
118 :
119 : /// Number of concurrent connections during the parallel RunInEachDatabase
120 : /// phase of the apply config process.
121 : ///
122 : /// We need a higher concurrency during reconfiguration in case of many DBs,
123 : /// but instance is already running and used by client. We can easily get out of
124 : /// `max_connections` limit, and the current code won't handle that.
125 : ///
126 : /// Default is 1, but also allow control plane to override this value for specific
127 : /// projects. It's also recommended to bump `superuser_reserved_connections` +=
128 : /// `reconfigure_concurrency` for such projects to ensure that we always have
129 : /// enough spare connections for reconfiguration process to succeed.
130 : #[serde(default = "default_reconfigure_concurrency")]
131 : pub reconfigure_concurrency: usize,
132 : }
133 :
134 : /// Feature flag to signal `compute_ctl` to enable certain experimental functionality.
135 6 : #[derive(Serialize, Clone, Copy, Debug, Deserialize, PartialEq, Eq)]
136 : #[serde(rename_all = "snake_case")]
137 : pub enum ComputeFeature {
138 : // XXX: Add more feature flags here.
139 : /// Enable the experimental activity monitor logic, which uses `pg_stat_database` to
140 : /// track short-lived connections as user activity.
141 : ActivityMonitorExperimental,
142 :
143 : /// Pre-install and initialize anon extension for every database in the cluster
144 : AnonExtension,
145 :
146 : /// This is a special feature flag that is used to represent unknown feature flags.
147 : /// Basically all unknown to enum flags are represented as this one. See unit test
148 : /// `parse_unknown_features()` for more details.
149 : #[serde(other)]
150 : UnknownFeature,
151 : }
152 :
153 30 : #[derive(Clone, Debug, Default, Deserialize, Serialize)]
154 : pub struct RemoteExtSpec {
155 : pub public_extensions: Option<Vec<String>>,
156 : pub custom_extensions: Option<Vec<String>>,
157 : pub library_index: HashMap<String, String>,
158 : pub extension_data: HashMap<String, ExtensionData>,
159 : }
160 :
161 36 : #[derive(Clone, Debug, Serialize, Deserialize)]
162 : pub struct ExtensionData {
163 : pub control_data: HashMap<String, String>,
164 : pub archive_path: String,
165 : }
166 :
167 : impl RemoteExtSpec {
168 0 : pub fn get_ext(
169 0 : &self,
170 0 : ext_name: &str,
171 0 : is_library: bool,
172 0 : build_tag: &str,
173 0 : pg_major_version: &str,
174 0 : ) -> anyhow::Result<(String, RemotePath)> {
175 0 : let mut real_ext_name = ext_name;
176 0 : if is_library {
177 : // sometimes library names might have a suffix like
178 : // library.so or library.so.3. We strip this off
179 : // because library_index is based on the name without the file extension
180 0 : let strip_lib_suffix = Regex::new(r"\.so.*").unwrap();
181 0 : let lib_raw_name = strip_lib_suffix.replace(real_ext_name, "").to_string();
182 0 :
183 0 : real_ext_name = self
184 0 : .library_index
185 0 : .get(&lib_raw_name)
186 0 : .ok_or(anyhow::anyhow!("library {} is not found", lib_raw_name))?;
187 0 : }
188 :
189 : // Check if extension is present in public or custom.
190 : // If not, then it is not allowed to be used by this compute.
191 0 : if let Some(public_extensions) = &self.public_extensions {
192 0 : if !public_extensions.contains(&real_ext_name.to_string()) {
193 0 : if let Some(custom_extensions) = &self.custom_extensions {
194 0 : if !custom_extensions.contains(&real_ext_name.to_string()) {
195 0 : return Err(anyhow::anyhow!("extension {} is not found", real_ext_name));
196 0 : }
197 0 : }
198 0 : }
199 0 : }
200 :
201 0 : match self.extension_data.get(real_ext_name) {
202 0 : Some(_ext_data) => {
203 0 : // Construct the path to the extension archive
204 0 : // BUILD_TAG/PG_MAJOR_VERSION/extensions/EXTENSION_NAME.tar.zst
205 0 : //
206 0 : // Keep it in sync with path generation in
207 0 : // https://github.com/neondatabase/build-custom-extensions/tree/main
208 0 : let archive_path_str =
209 0 : format!("{build_tag}/{pg_major_version}/extensions/{real_ext_name}.tar.zst");
210 0 : Ok((
211 0 : real_ext_name.to_string(),
212 0 : RemotePath::from_string(&archive_path_str)?,
213 : ))
214 : }
215 0 : None => Err(anyhow::anyhow!(
216 0 : "real_ext_name {} is not found",
217 0 : real_ext_name
218 0 : )),
219 : }
220 0 : }
221 : }
222 :
223 0 : #[derive(Clone, Copy, Debug, Default, Eq, PartialEq, Deserialize, Serialize)]
224 : pub enum ComputeMode {
225 : /// A read-write node
226 : #[default]
227 : Primary,
228 : /// A read-only node, pinned at a particular LSN
229 : Static(Lsn),
230 : /// A read-only node that follows the tip of the branch in hot standby mode
231 : ///
232 : /// Future versions may want to distinguish between replicas with hot standby
233 : /// feedback and other kinds of replication configurations.
234 : Replica,
235 : }
236 :
237 42 : #[derive(Clone, Debug, Default, Deserialize, Serialize)]
238 : pub struct Cluster {
239 : pub cluster_id: Option<String>,
240 : pub name: Option<String>,
241 : pub state: Option<String>,
242 : pub roles: Vec<Role>,
243 : pub databases: Vec<Database>,
244 :
245 : /// Desired contents of 'postgresql.conf' file. (The 'compute_ctl'
246 : /// tool may add additional settings to the final file.)
247 : pub postgresql_conf: Option<String>,
248 :
249 : /// Additional settings that will be appended to the 'postgresql.conf' file.
250 : pub settings: GenericOptions,
251 : }
252 :
253 : /// Single cluster state changing operation that could not be represented as
254 : /// a static `Cluster` structure. For example:
255 : /// - DROP DATABASE
256 : /// - DROP ROLE
257 : /// - ALTER ROLE name RENAME TO new_name
258 : /// - ALTER DATABASE name RENAME TO new_name
259 84 : #[derive(Clone, Debug, Deserialize, Serialize)]
260 : pub struct DeltaOp {
261 : pub action: String,
262 : pub name: PgIdent,
263 : pub new_name: Option<PgIdent>,
264 : }
265 :
266 : /// Rust representation of Postgres role info with only those fields
267 : /// that matter for us.
268 126 : #[derive(Clone, Debug, Deserialize, Serialize)]
269 : pub struct Role {
270 : pub name: PgIdent,
271 : pub encrypted_password: Option<String>,
272 : pub options: GenericOptions,
273 : }
274 :
275 : /// Rust representation of Postgres database info with only those fields
276 : /// that matter for us.
277 60 : #[derive(Clone, Debug, Deserialize, Serialize)]
278 : pub struct Database {
279 : pub name: PgIdent,
280 : pub owner: PgIdent,
281 : pub options: GenericOptions,
282 : // These are derived flags, not present in the spec file.
283 : // They are never set by the control plane.
284 : #[serde(skip_deserializing, default)]
285 : pub restrict_conn: bool,
286 : #[serde(skip_deserializing, default)]
287 : pub invalid: bool,
288 : }
289 :
290 : /// Common type representing both SQL statement params with or without value,
291 : /// like `LOGIN` or `OWNER username` in the `CREATE/ALTER ROLE`, and config
292 : /// options like `wal_level = logical`.
293 624 : #[derive(Clone, Debug, Deserialize, Serialize)]
294 : pub struct GenericOption {
295 : pub name: String,
296 : pub value: Option<String>,
297 : pub vartype: String,
298 : }
299 :
300 : /// Optional collection of `GenericOption`'s. Type alias allows us to
301 : /// declare a `trait` on it.
302 : pub type GenericOptions = Option<Vec<GenericOption>>;
303 :
304 : /// Configured the local_proxy application with the relevant JWKS and roles it should
305 : /// use for authorizing connect requests using JWT.
306 0 : #[derive(Clone, Debug, Deserialize, Serialize)]
307 : pub struct LocalProxySpec {
308 : #[serde(default)]
309 : #[serde(skip_serializing_if = "Option::is_none")]
310 : pub jwks: Option<Vec<JwksSettings>>,
311 : }
312 :
313 0 : #[derive(Clone, Debug, Deserialize, Serialize)]
314 : pub struct JwksSettings {
315 : pub id: String,
316 : pub role_names: Vec<String>,
317 : pub jwks_url: String,
318 : pub provider_name: String,
319 : pub jwt_audience: Option<String>,
320 : }
321 :
322 : #[cfg(test)]
323 : mod tests {
324 : use super::*;
325 : use std::fs::File;
326 :
327 : #[test]
328 1 : fn parse_spec_file() {
329 1 : let file = File::open("tests/cluster_spec.json").unwrap();
330 1 : let spec: ComputeSpec = serde_json::from_reader(file).unwrap();
331 1 :
332 1 : // Features list defaults to empty vector.
333 1 : assert!(spec.features.is_empty());
334 :
335 : // Reconfigure concurrency defaults to 1.
336 1 : assert_eq!(spec.reconfigure_concurrency, 1);
337 1 : }
338 :
339 : #[test]
340 1 : fn parse_unknown_fields() {
341 1 : // Forward compatibility test
342 1 : let file = File::open("tests/cluster_spec.json").unwrap();
343 1 : let mut json: serde_json::Value = serde_json::from_reader(file).unwrap();
344 1 : let ob = json.as_object_mut().unwrap();
345 1 : ob.insert("unknown_field_123123123".into(), "hello".into());
346 1 : let _spec: ComputeSpec = serde_json::from_value(json).unwrap();
347 1 : }
348 :
349 : #[test]
350 1 : fn parse_unknown_features() {
351 1 : // Test that unknown feature flags do not cause any errors.
352 1 : let file = File::open("tests/cluster_spec.json").unwrap();
353 1 : let mut json: serde_json::Value = serde_json::from_reader(file).unwrap();
354 1 : let ob = json.as_object_mut().unwrap();
355 1 :
356 1 : // Add unknown feature flags.
357 1 : let features = vec!["foo_bar_feature", "baz_feature"];
358 1 : ob.insert("features".into(), features.into());
359 1 :
360 1 : let spec: ComputeSpec = serde_json::from_value(json).unwrap();
361 1 :
362 1 : assert!(spec.features.len() == 2);
363 1 : assert!(spec.features.contains(&ComputeFeature::UnknownFeature));
364 1 : assert_eq!(spec.features, vec![ComputeFeature::UnknownFeature; 2]);
365 1 : }
366 :
367 : #[test]
368 1 : fn parse_known_features() {
369 1 : // Test that we can properly parse known feature flags.
370 1 : let file = File::open("tests/cluster_spec.json").unwrap();
371 1 : let mut json: serde_json::Value = serde_json::from_reader(file).unwrap();
372 1 : let ob = json.as_object_mut().unwrap();
373 1 :
374 1 : // Add known feature flags.
375 1 : let features = vec!["activity_monitor_experimental"];
376 1 : ob.insert("features".into(), features.into());
377 1 :
378 1 : let spec: ComputeSpec = serde_json::from_value(json).unwrap();
379 1 :
380 1 : assert_eq!(
381 1 : spec.features,
382 1 : vec![ComputeFeature::ActivityMonitorExperimental]
383 1 : );
384 1 : }
385 : }
|
