Line data Source code
1 : use anyhow::{anyhow, bail, Result};
2 : use reqwest::StatusCode;
3 : use std::fs::File;
4 : use std::path::Path;
5 : use tokio_postgres::Client;
6 : use tracing::{error, info, instrument, warn};
7 :
8 : use crate::config;
9 : use crate::metrics::{CPlaneRequestRPC, CPLANE_REQUESTS_TOTAL, UNKNOWN_HTTP_STATUS};
10 : use crate::migration::MigrationRunner;
11 : use crate::params::PG_HBA_ALL_MD5;
12 : use crate::pg_helpers::*;
13 :
14 : use compute_api::responses::{ControlPlaneComputeStatus, ControlPlaneSpecResponse};
15 : use compute_api::spec::ComputeSpec;
16 :
17 : // Do control plane request and return response if any. In case of error it
18 : // returns a bool flag indicating whether it makes sense to retry the request
19 : // and a string with error message.
20 0 : fn do_control_plane_request(
21 0 : uri: &str,
22 0 : jwt: &str,
23 0 : ) -> Result<ControlPlaneSpecResponse, (bool, String, String)> {
24 0 : let resp = reqwest::blocking::Client::new()
25 0 : .get(uri)
26 0 : .header("Authorization", format!("Bearer {}", jwt))
27 0 : .send()
28 0 : .map_err(|e| {
29 0 : (
30 0 : true,
31 0 : format!("could not perform spec request to control plane: {:?}", e),
32 0 : UNKNOWN_HTTP_STATUS.to_string(),
33 0 : )
34 0 : })?;
35 :
36 0 : let status = resp.status();
37 0 : match status {
38 0 : StatusCode::OK => match resp.json::<ControlPlaneSpecResponse>() {
39 0 : Ok(spec_resp) => Ok(spec_resp),
40 0 : Err(e) => Err((
41 0 : true,
42 0 : format!("could not deserialize control plane response: {:?}", e),
43 0 : status.to_string(),
44 0 : )),
45 : },
46 0 : StatusCode::SERVICE_UNAVAILABLE => Err((
47 0 : true,
48 0 : "control plane is temporarily unavailable".to_string(),
49 0 : status.to_string(),
50 0 : )),
51 : StatusCode::BAD_GATEWAY => {
52 : // We have a problem with intermittent 502 errors now
53 : // https://github.com/neondatabase/cloud/issues/2353
54 : // It's fine to retry GET request in this case.
55 0 : Err((
56 0 : true,
57 0 : "control plane request failed with 502".to_string(),
58 0 : status.to_string(),
59 0 : ))
60 : }
61 : // Another code, likely 500 or 404, means that compute is unknown to the control plane
62 : // or some internal failure happened. Doesn't make much sense to retry in this case.
63 0 : _ => Err((
64 0 : false,
65 0 : format!("unexpected control plane response status code: {}", status),
66 0 : status.to_string(),
67 0 : )),
68 : }
69 0 : }
70 :
71 : /// Request spec from the control-plane by compute_id. If `NEON_CONTROL_PLANE_TOKEN`
72 : /// env variable is set, it will be used for authorization.
73 0 : pub fn get_spec_from_control_plane(
74 0 : base_uri: &str,
75 0 : compute_id: &str,
76 0 : ) -> Result<Option<ComputeSpec>> {
77 0 : let cp_uri = format!("{base_uri}/compute/api/v2/computes/{compute_id}/spec");
78 0 : let jwt: String = match std::env::var("NEON_CONTROL_PLANE_TOKEN") {
79 0 : Ok(v) => v,
80 0 : Err(_) => "".to_string(),
81 : };
82 0 : let mut attempt = 1;
83 0 : let mut spec: Result<Option<ComputeSpec>> = Ok(None);
84 0 :
85 0 : info!("getting spec from control plane: {}", cp_uri);
86 :
87 : // Do 3 attempts to get spec from the control plane using the following logic:
88 : // - network error -> then retry
89 : // - compute id is unknown or any other error -> bail out
90 : // - no spec for compute yet (Empty state) -> return Ok(None)
91 : // - got spec -> return Ok(Some(spec))
92 0 : while attempt < 4 {
93 0 : spec = match do_control_plane_request(&cp_uri, &jwt) {
94 0 : Ok(spec_resp) => {
95 0 : CPLANE_REQUESTS_TOTAL
96 0 : .with_label_values(&[
97 0 : CPlaneRequestRPC::GetSpec.as_str(),
98 0 : &StatusCode::OK.to_string(),
99 0 : ])
100 0 : .inc();
101 0 : match spec_resp.status {
102 0 : ControlPlaneComputeStatus::Empty => Ok(None),
103 : ControlPlaneComputeStatus::Attached => {
104 0 : if let Some(spec) = spec_resp.spec {
105 0 : Ok(Some(spec))
106 : } else {
107 0 : bail!("compute is attached, but spec is empty")
108 : }
109 : }
110 : }
111 : }
112 0 : Err((retry, msg, status)) => {
113 0 : CPLANE_REQUESTS_TOTAL
114 0 : .with_label_values(&[CPlaneRequestRPC::GetSpec.as_str(), &status])
115 0 : .inc();
116 0 : if retry {
117 0 : Err(anyhow!(msg))
118 : } else {
119 0 : bail!(msg);
120 : }
121 : }
122 : };
123 :
124 0 : if let Err(e) = &spec {
125 0 : error!("attempt {} to get spec failed with: {}", attempt, e);
126 : } else {
127 0 : return spec;
128 : }
129 :
130 0 : attempt += 1;
131 0 : std::thread::sleep(std::time::Duration::from_millis(100));
132 : }
133 :
134 : // All attempts failed, return error.
135 0 : spec
136 0 : }
137 :
138 : /// Check `pg_hba.conf` and update if needed to allow external connections.
139 0 : pub fn update_pg_hba(pgdata_path: &Path) -> Result<()> {
140 0 : // XXX: consider making it a part of spec.json
141 0 : info!("checking pg_hba.conf");
142 0 : let pghba_path = pgdata_path.join("pg_hba.conf");
143 0 :
144 0 : if config::line_in_file(&pghba_path, PG_HBA_ALL_MD5)? {
145 0 : info!("updated pg_hba.conf to allow external connections");
146 : } else {
147 0 : info!("pg_hba.conf is up-to-date");
148 : }
149 :
150 0 : Ok(())
151 0 : }
152 :
153 : /// Create a standby.signal file
154 0 : pub fn add_standby_signal(pgdata_path: &Path) -> Result<()> {
155 0 : // XXX: consider making it a part of spec.json
156 0 : info!("adding standby.signal");
157 0 : let signalfile = pgdata_path.join("standby.signal");
158 0 :
159 0 : if !signalfile.exists() {
160 0 : info!("created standby.signal");
161 0 : File::create(signalfile)?;
162 : } else {
163 0 : info!("reused pre-existing standby.signal");
164 : }
165 0 : Ok(())
166 0 : }
167 :
168 : #[instrument(skip_all)]
169 : pub async fn handle_neon_extension_upgrade(client: &mut Client) -> Result<()> {
170 : info!("handle neon extension upgrade");
171 : let query = "ALTER EXTENSION neon UPDATE";
172 : info!("update neon extension version with query: {}", query);
173 : client.simple_query(query).await?;
174 :
175 : Ok(())
176 : }
177 :
178 : #[instrument(skip_all)]
179 : pub async fn handle_migrations(client: &mut Client) -> Result<()> {
180 : info!("handle migrations");
181 :
182 : // !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
183 : // !BE SURE TO ONLY ADD MIGRATIONS TO THE END OF THIS ARRAY. IF YOU DO NOT, VERY VERY BAD THINGS MAY HAPPEN!
184 : // !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
185 :
186 : // Add new migrations in numerical order.
187 : let migrations = [
188 : include_str!("./migrations/0001-neon_superuser_bypass_rls.sql"),
189 : include_str!("./migrations/0002-alter_roles.sql"),
190 : include_str!("./migrations/0003-grant_pg_create_subscription_to_neon_superuser.sql"),
191 : include_str!("./migrations/0004-grant_pg_monitor_to_neon_superuser.sql"),
192 : include_str!("./migrations/0005-grant_all_on_tables_to_neon_superuser.sql"),
193 : include_str!("./migrations/0006-grant_all_on_sequences_to_neon_superuser.sql"),
194 : include_str!(
195 : "./migrations/0007-grant_all_on_tables_to_neon_superuser_with_grant_option.sql"
196 : ),
197 : include_str!(
198 : "./migrations/0008-grant_all_on_sequences_to_neon_superuser_with_grant_option.sql"
199 : ),
200 : include_str!("./migrations/0009-revoke_replication_for_previously_allowed_roles.sql"),
201 : include_str!(
202 : "./migrations/0010-grant_snapshot_synchronization_funcs_to_neon_superuser.sql"
203 : ),
204 : include_str!(
205 : "./migrations/0011-grant_pg_show_replication_origin_status_to_neon_superuser.sql"
206 : ),
207 : ];
208 :
209 : MigrationRunner::new(client, &migrations)
210 : .run_migrations()
211 : .await?;
212 :
213 : Ok(())
214 : }
215 :
216 : /// Connect to the database as superuser and pre-create anon extension
217 : /// if it is present in shared_preload_libraries
218 : #[instrument(skip_all)]
219 : pub async fn handle_extension_anon(
220 : spec: &ComputeSpec,
221 : db_owner: &str,
222 : db_client: &mut Client,
223 : grants_only: bool,
224 : ) -> Result<()> {
225 : info!("handle extension anon");
226 :
227 : if let Some(libs) = spec.cluster.settings.find("shared_preload_libraries") {
228 : if libs.contains("anon") {
229 : if !grants_only {
230 : // check if extension is already initialized using anon.is_initialized()
231 : let query = "SELECT anon.is_initialized()";
232 : match db_client.query(query, &[]).await {
233 : Ok(rows) => {
234 : if !rows.is_empty() {
235 : let is_initialized: bool = rows[0].get(0);
236 : if is_initialized {
237 : info!("anon extension is already initialized");
238 : return Ok(());
239 : }
240 : }
241 : }
242 : Err(e) => {
243 : warn!(
244 : "anon extension is_installed check failed with expected error: {}",
245 : e
246 : );
247 : }
248 : };
249 :
250 : // Create anon extension if this compute needs it
251 : // Users cannot create it themselves, because superuser is required.
252 : let mut query = "CREATE EXTENSION IF NOT EXISTS anon CASCADE";
253 : info!("creating anon extension with query: {}", query);
254 : match db_client.query(query, &[]).await {
255 : Ok(_) => {}
256 : Err(e) => {
257 : error!("anon extension creation failed with error: {}", e);
258 : return Ok(());
259 : }
260 : }
261 :
262 : // check that extension is installed
263 : query = "SELECT extname FROM pg_extension WHERE extname = 'anon'";
264 : let rows = db_client.query(query, &[]).await?;
265 : if rows.is_empty() {
266 : error!("anon extension is not installed");
267 : return Ok(());
268 : }
269 :
270 : // Initialize anon extension
271 : // This also requires superuser privileges, so users cannot do it themselves.
272 : query = "SELECT anon.init()";
273 : match db_client.query(query, &[]).await {
274 : Ok(_) => {}
275 : Err(e) => {
276 : error!("anon.init() failed with error: {}", e);
277 : return Ok(());
278 : }
279 : }
280 : }
281 :
282 : // check that extension is installed, if not bail early
283 : let query = "SELECT extname FROM pg_extension WHERE extname = 'anon'";
284 : match db_client.query(query, &[]).await {
285 : Ok(rows) => {
286 : if rows.is_empty() {
287 : error!("anon extension is not installed");
288 : return Ok(());
289 : }
290 : }
291 : Err(e) => {
292 : error!("anon extension check failed with error: {}", e);
293 : return Ok(());
294 : }
295 : };
296 :
297 : let query = format!("GRANT ALL ON SCHEMA anon TO {}", db_owner);
298 : info!("granting anon extension permissions with query: {}", query);
299 : db_client.simple_query(&query).await?;
300 :
301 : // Grant permissions to db_owner to use anon extension functions
302 : let query = format!("GRANT ALL ON ALL FUNCTIONS IN SCHEMA anon TO {}", db_owner);
303 : info!("granting anon extension permissions with query: {}", query);
304 : db_client.simple_query(&query).await?;
305 :
306 : // This is needed, because some functions are defined as SECURITY DEFINER.
307 : // In Postgres SECURITY DEFINER functions are executed with the privileges
308 : // of the owner.
309 : // In anon extension this it is needed to access some GUCs, which are only accessible to
310 : // superuser. But we've patched postgres to allow db_owner to access them as well.
311 : // So we need to change owner of these functions to db_owner.
312 : let query = format!("
313 : SELECT 'ALTER FUNCTION '||nsp.nspname||'.'||p.proname||'('||pg_get_function_identity_arguments(p.oid)||') OWNER TO {};'
314 : from pg_proc p
315 : join pg_namespace nsp ON p.pronamespace = nsp.oid
316 : where nsp.nspname = 'anon';", db_owner);
317 :
318 : info!("change anon extension functions owner to db owner");
319 : db_client.simple_query(&query).await?;
320 :
321 : // affects views as well
322 : let query = format!("GRANT ALL ON ALL TABLES IN SCHEMA anon TO {}", db_owner);
323 : info!("granting anon extension permissions with query: {}", query);
324 : db_client.simple_query(&query).await?;
325 :
326 : let query = format!("GRANT ALL ON ALL SEQUENCES IN SCHEMA anon TO {}", db_owner);
327 : info!("granting anon extension permissions with query: {}", query);
328 : db_client.simple_query(&query).await?;
329 : }
330 : }
331 :
332 : Ok(())
333 : }
|
