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 : /// Cluster spec or configuration represented as an optional number of
20 : /// delta operations + final cluster state description.
21 31705 : #[derive(Clone, Debug, Default, Deserialize, Serialize)]
22 : pub struct ComputeSpec {
23 : pub format_version: f32,
24 :
25 : // The control plane also includes a 'timestamp' field in the JSON document,
26 : // but we don't use it for anything. Serde will ignore missing fields when
27 : // deserializing it.
28 : pub operation_uuid: Option<String>,
29 :
30 : /// Compute features to enable. These feature flags are provided, when we
31 : /// know all the details about client's compute, so they cannot be used
32 : /// to change `Empty` compute behavior.
33 : #[serde(default)]
34 : pub features: Vec<ComputeFeature>,
35 :
36 : /// Expected cluster state at the end of transition process.
37 : pub cluster: Cluster,
38 : pub delta_operations: Option<Vec<DeltaOp>>,
39 :
40 : /// An optinal hint that can be passed to speed up startup time if we know
41 : /// that no pg catalog mutations (like role creation, database creation,
42 : /// extension creation) need to be done on the actual database to start.
43 : #[serde(default)] // Default false
44 : pub skip_pg_catalog_updates: bool,
45 :
46 : // Information needed to connect to the storage layer.
47 : //
48 : // `tenant_id`, `timeline_id` and `pageserver_connstring` are always needed.
49 : //
50 : // Depending on `mode`, this can be a primary read-write node, a read-only
51 : // replica, or a read-only node pinned at an older LSN.
52 : // `safekeeper_connstrings` must be set for a primary.
53 : //
54 : // For backwards compatibility, the control plane may leave out all of
55 : // these, and instead set the "neon.tenant_id", "neon.timeline_id",
56 : // etc. GUCs in cluster.settings. TODO: Once the control plane has been
57 : // updated to fill these fields, we can make these non optional.
58 : pub tenant_id: Option<TenantId>,
59 :
60 : pub timeline_id: Option<TimelineId>,
61 :
62 : pub pageserver_connstring: Option<String>,
63 :
64 : #[serde(default)]
65 : pub safekeeper_connstrings: Vec<String>,
66 :
67 : #[serde(default)]
68 : pub mode: ComputeMode,
69 :
70 : /// If set, 'storage_auth_token' is used as the password to authenticate to
71 : /// the pageserver and safekeepers.
72 : pub storage_auth_token: Option<String>,
73 :
74 : // information about available remote extensions
75 : pub remote_extensions: Option<RemoteExtSpec>,
76 :
77 : pub pgbouncer_settings: Option<HashMap<String, String>>,
78 :
79 : // Stripe size for pageserver sharding, in pages
80 : #[serde(default)]
81 : pub shard_stripe_size: Option<usize>,
82 : }
83 :
84 : /// Feature flag to signal `compute_ctl` to enable certain experimental functionality.
85 42 : #[derive(Serialize, Clone, Copy, Debug, Deserialize, PartialEq, Eq)]
86 : #[serde(rename_all = "snake_case")]
87 : pub enum ComputeFeature {
88 : // XXX: Add more feature flags here.
89 : /// Enable the experimental activity monitor logic, which uses `pg_stat_database` to
90 : /// track short-lived connections as user activity.
91 : ActivityMonitorExperimental,
92 :
93 : /// Enable running migrations
94 : Migrations,
95 :
96 : /// This is a special feature flag that is used to represent unknown feature flags.
97 : /// Basically all unknown to enum flags are represented as this one. See unit test
98 : /// `parse_unknown_features()` for more details.
99 : #[serde(other)]
100 : UnknownFeature,
101 : }
102 :
103 122 : #[derive(Clone, Debug, Default, Deserialize, Serialize)]
104 : pub struct RemoteExtSpec {
105 : pub public_extensions: Option<Vec<String>>,
106 : pub custom_extensions: Option<Vec<String>>,
107 : pub library_index: HashMap<String, String>,
108 : pub extension_data: HashMap<String, ExtensionData>,
109 : }
110 :
111 130 : #[derive(Clone, Debug, Serialize, Deserialize)]
112 : pub struct ExtensionData {
113 : pub control_data: HashMap<String, String>,
114 : pub archive_path: String,
115 : }
116 :
117 : impl RemoteExtSpec {
118 2 : pub fn get_ext(
119 2 : &self,
120 2 : ext_name: &str,
121 2 : is_library: bool,
122 2 : build_tag: &str,
123 2 : pg_major_version: &str,
124 2 : ) -> anyhow::Result<(String, RemotePath)> {
125 2 : let mut real_ext_name = ext_name;
126 2 : if is_library {
127 : // sometimes library names might have a suffix like
128 : // library.so or library.so.3. We strip this off
129 : // because library_index is based on the name without the file extension
130 0 : let strip_lib_suffix = Regex::new(r"\.so.*").unwrap();
131 0 : let lib_raw_name = strip_lib_suffix.replace(real_ext_name, "").to_string();
132 0 :
133 0 : real_ext_name = self
134 0 : .library_index
135 0 : .get(&lib_raw_name)
136 0 : .ok_or(anyhow::anyhow!("library {} is not found", lib_raw_name))?;
137 2 : }
138 :
139 : // Check if extension is present in public or custom.
140 : // If not, then it is not allowed to be used by this compute.
141 2 : if let Some(public_extensions) = &self.public_extensions {
142 0 : if !public_extensions.contains(&real_ext_name.to_string()) {
143 0 : if let Some(custom_extensions) = &self.custom_extensions {
144 0 : if !custom_extensions.contains(&real_ext_name.to_string()) {
145 0 : return Err(anyhow::anyhow!("extension {} is not found", real_ext_name));
146 0 : }
147 0 : }
148 0 : }
149 2 : }
150 :
151 2 : match self.extension_data.get(real_ext_name) {
152 2 : Some(_ext_data) => {
153 2 : // Construct the path to the extension archive
154 2 : // BUILD_TAG/PG_MAJOR_VERSION/extensions/EXTENSION_NAME.tar.zst
155 2 : //
156 2 : // Keep it in sync with path generation in
157 2 : // https://github.com/neondatabase/build-custom-extensions/tree/main
158 2 : let archive_path_str =
159 2 : format!("{build_tag}/{pg_major_version}/extensions/{real_ext_name}.tar.zst");
160 2 : Ok((
161 2 : real_ext_name.to_string(),
162 2 : RemotePath::from_string(&archive_path_str)?,
163 : ))
164 : }
165 0 : None => Err(anyhow::anyhow!(
166 0 : "real_ext_name {} is not found",
167 0 : real_ext_name
168 0 : )),
169 : }
170 2 : }
171 : }
172 :
173 9826 : #[derive(Clone, Copy, Debug, Default, Eq, PartialEq, Deserialize, Serialize)]
174 : pub enum ComputeMode {
175 : /// A read-write node
176 : #[default]
177 : Primary,
178 : /// A read-only node, pinned at a particular LSN
179 : Static(Lsn),
180 : /// A read-only node that follows the tip of the branch in hot standby mode
181 : ///
182 : /// Future versions may want to distinguish between replicas with hot standby
183 : /// feedback and other kinds of replication configurations.
184 : Replica,
185 : }
186 :
187 15411 : #[derive(Clone, Debug, Default, Deserialize, Serialize)]
188 : pub struct Cluster {
189 : pub cluster_id: Option<String>,
190 : pub name: Option<String>,
191 : pub state: Option<String>,
192 : pub roles: Vec<Role>,
193 : pub databases: Vec<Database>,
194 :
195 : /// Desired contents of 'postgresql.conf' file. (The 'compute_ctl'
196 : /// tool may add additional settings to the final file.)
197 : pub postgresql_conf: Option<String>,
198 :
199 : /// Additional settings that will be appended to the 'postgresql.conf' file.
200 : pub settings: GenericOptions,
201 : }
202 :
203 : /// Single cluster state changing operation that could not be represented as
204 : /// a static `Cluster` structure. For example:
205 : /// - DROP DATABASE
206 : /// - DROP ROLE
207 : /// - ALTER ROLE name RENAME TO new_name
208 : /// - ALTER DATABASE name RENAME TO new_name
209 288 : #[derive(Clone, Debug, Deserialize, Serialize)]
210 : pub struct DeltaOp {
211 : pub action: String,
212 : pub name: PgIdent,
213 : pub new_name: Option<PgIdent>,
214 : }
215 :
216 : /// Rust representation of Postgres role info with only those fields
217 : /// that matter for us.
218 432 : #[derive(Clone, Debug, Deserialize, Serialize)]
219 : pub struct Role {
220 : pub name: PgIdent,
221 : pub encrypted_password: Option<String>,
222 : pub options: GenericOptions,
223 : }
224 :
225 : /// Rust representation of Postgres database info with only those fields
226 : /// that matter for us.
227 2749 : #[derive(Clone, Debug, Deserialize, Serialize)]
228 : pub struct Database {
229 : pub name: PgIdent,
230 : pub owner: PgIdent,
231 : pub options: GenericOptions,
232 : // These are derived flags, not present in the spec file.
233 : // They are never set by the control plane.
234 : #[serde(skip_deserializing, default)]
235 : pub restrict_conn: bool,
236 : #[serde(skip_deserializing, default)]
237 : pub invalid: bool,
238 : }
239 :
240 : /// Common type representing both SQL statement params with or without value,
241 : /// like `LOGIN` or `OWNER username` in the `CREATE/ALTER ROLE`, and config
242 : /// options like `wal_level = logical`.
243 2184 : #[derive(Clone, Debug, Deserialize, Serialize)]
244 : pub struct GenericOption {
245 : pub name: String,
246 : pub value: Option<String>,
247 : pub vartype: String,
248 : }
249 :
250 : /// Optional collection of `GenericOption`'s. Type alias allows us to
251 : /// declare a `trait` on it.
252 : pub type GenericOptions = Option<Vec<GenericOption>>;
253 :
254 : #[cfg(test)]
255 : mod tests {
256 : use super::*;
257 : use std::fs::File;
258 :
259 2 : #[test]
260 2 : fn parse_spec_file() {
261 2 : let file = File::open("tests/cluster_spec.json").unwrap();
262 2 : let spec: ComputeSpec = serde_json::from_reader(file).unwrap();
263 :
264 : // Features list defaults to empty vector.
265 2 : assert!(spec.features.is_empty());
266 2 : }
267 :
268 2 : #[test]
269 2 : fn parse_unknown_fields() {
270 2 : // Forward compatibility test
271 2 : let file = File::open("tests/cluster_spec.json").unwrap();
272 2 : let mut json: serde_json::Value = serde_json::from_reader(file).unwrap();
273 2 : let ob = json.as_object_mut().unwrap();
274 2 : ob.insert("unknown_field_123123123".into(), "hello".into());
275 2 : let _spec: ComputeSpec = serde_json::from_value(json).unwrap();
276 2 : }
277 :
278 2 : #[test]
279 2 : fn parse_unknown_features() {
280 2 : // Test that unknown feature flags do not cause any errors.
281 2 : let file = File::open("tests/cluster_spec.json").unwrap();
282 2 : let mut json: serde_json::Value = serde_json::from_reader(file).unwrap();
283 2 : let ob = json.as_object_mut().unwrap();
284 2 :
285 2 : // Add unknown feature flags.
286 2 : let features = vec!["foo_bar_feature", "baz_feature"];
287 2 : ob.insert("features".into(), features.into());
288 2 :
289 2 : let spec: ComputeSpec = serde_json::from_value(json).unwrap();
290 :
291 2 : assert!(spec.features.len() == 2);
292 2 : assert!(spec.features.contains(&ComputeFeature::UnknownFeature));
293 2 : assert_eq!(spec.features, vec![ComputeFeature::UnknownFeature; 2]);
294 2 : }
295 :
296 2 : #[test]
297 2 : fn parse_known_features() {
298 2 : // Test that we can properly parse known feature flags.
299 2 : let file = File::open("tests/cluster_spec.json").unwrap();
300 2 : let mut json: serde_json::Value = serde_json::from_reader(file).unwrap();
301 2 : let ob = json.as_object_mut().unwrap();
302 2 :
303 2 : // Add known feature flags.
304 2 : let features = vec!["activity_monitor_experimental"];
305 2 : ob.insert("features".into(), features.into());
306 2 :
307 2 : let spec: ComputeSpec = serde_json::from_value(json).unwrap();
308 2 :
309 2 : assert_eq!(
310 2 : spec.features,
311 2 : vec![ComputeFeature::ActivityMonitorExperimental]
312 2 : );
313 2 : }
314 : }
|
