Line data Source code
1 : //! Code to manage safekeepers
2 : //!
3 : //! In the local test environment, the data for each safekeeper is stored in
4 : //!
5 : //! ```text
6 : //! .neon/safekeepers/<safekeeper id>
7 : //! ```
8 : use std::io::Write;
9 : use std::path::PathBuf;
10 : use std::time::Duration;
11 : use std::{io, result};
12 :
13 : use anyhow::Context;
14 : use camino::Utf8PathBuf;
15 : use postgres_connection::PgConnectionConfig;
16 : use reqwest::{IntoUrl, Method};
17 : use thiserror::Error;
18 : use utils::auth::{Claims, Scope};
19 : use utils::{http::error::HttpErrorBody, id::NodeId};
20 :
21 : use crate::{
22 : background_process,
23 : local_env::{LocalEnv, SafekeeperConf},
24 : };
25 :
26 0 : #[derive(Error, Debug)]
27 : pub enum SafekeeperHttpError {
28 : #[error("Reqwest error: {0}")]
29 : Transport(#[from] reqwest::Error),
30 :
31 : #[error("Error: {0}")]
32 : Response(String),
33 : }
34 :
35 : type Result<T> = result::Result<T, SafekeeperHttpError>;
36 :
37 : #[async_trait::async_trait]
38 : pub trait ResponseErrorMessageExt: Sized {
39 : async fn error_from_body(self) -> Result<Self>;
40 : }
41 :
42 : #[async_trait::async_trait]
43 : impl ResponseErrorMessageExt for reqwest::Response {
44 0 : async fn error_from_body(self) -> Result<Self> {
45 0 : let status = self.status();
46 0 : if !(status.is_client_error() || status.is_server_error()) {
47 0 : return Ok(self);
48 0 : }
49 0 :
50 0 : // reqwest does not export its error construction utility functions, so let's craft the message ourselves
51 0 : let url = self.url().to_owned();
52 0 : Err(SafekeeperHttpError::Response(
53 0 : match self.json::<HttpErrorBody>().await {
54 0 : Ok(err_body) => format!("Error: {}", err_body.msg),
55 0 : Err(_) => format!("Http error ({}) at {}.", status.as_u16(), url),
56 0 : },
57 0 : ))
58 0 : }
59 : }
60 :
61 : //
62 : // Control routines for safekeeper.
63 : //
64 : // Used in CLI and tests.
65 : //
66 : #[derive(Debug)]
67 : pub struct SafekeeperNode {
68 : pub id: NodeId,
69 :
70 : pub conf: SafekeeperConf,
71 :
72 : pub pg_connection_config: PgConnectionConfig,
73 : pub env: LocalEnv,
74 : pub http_client: reqwest::Client,
75 : pub listen_addr: String,
76 : pub http_base_url: String,
77 : }
78 :
79 : impl SafekeeperNode {
80 0 : pub fn from_env(env: &LocalEnv, conf: &SafekeeperConf) -> SafekeeperNode {
81 0 : let listen_addr = if let Some(ref listen_addr) = conf.listen_addr {
82 0 : listen_addr.clone()
83 : } else {
84 0 : "127.0.0.1".to_string()
85 : };
86 0 : SafekeeperNode {
87 0 : id: conf.id,
88 0 : conf: conf.clone(),
89 0 : pg_connection_config: Self::safekeeper_connection_config(&listen_addr, conf.pg_port),
90 0 : env: env.clone(),
91 0 : http_client: reqwest::Client::new(),
92 0 : http_base_url: format!("http://{}:{}/v1", listen_addr, conf.http_port),
93 0 : listen_addr,
94 0 : }
95 0 : }
96 :
97 : /// Construct libpq connection string for connecting to this safekeeper.
98 0 : fn safekeeper_connection_config(addr: &str, port: u16) -> PgConnectionConfig {
99 0 : PgConnectionConfig::new_host_port(url::Host::parse(addr).unwrap(), port)
100 0 : }
101 :
102 0 : pub fn datadir_path_by_id(env: &LocalEnv, sk_id: NodeId) -> PathBuf {
103 0 : env.safekeeper_data_dir(&format!("sk{sk_id}"))
104 0 : }
105 :
106 0 : pub fn datadir_path(&self) -> PathBuf {
107 0 : SafekeeperNode::datadir_path_by_id(&self.env, self.id)
108 0 : }
109 :
110 0 : pub fn pid_file(&self) -> Utf8PathBuf {
111 0 : Utf8PathBuf::from_path_buf(self.datadir_path().join("safekeeper.pid"))
112 0 : .expect("non-Unicode path")
113 0 : }
114 :
115 0 : pub async fn start(
116 0 : &self,
117 0 : extra_opts: Vec<String>,
118 0 : retry_timeout: &Duration,
119 0 : ) -> anyhow::Result<()> {
120 0 : print!(
121 0 : "Starting safekeeper at '{}' in '{}', retrying for {:?}",
122 0 : self.pg_connection_config.raw_address(),
123 0 : self.datadir_path().display(),
124 0 : retry_timeout,
125 0 : );
126 0 : io::stdout().flush().unwrap();
127 0 :
128 0 : let listen_pg = format!("{}:{}", self.listen_addr, self.conf.pg_port);
129 0 : let listen_http = format!("{}:{}", self.listen_addr, self.conf.http_port);
130 0 : let id = self.id;
131 0 : let datadir = self.datadir_path();
132 0 :
133 0 : let id_string = id.to_string();
134 0 : // TODO: add availability_zone to the config.
135 0 : // Right now we just specify any value here and use it to check metrics in tests.
136 0 : let availability_zone = format!("sk-{}", id_string);
137 :
138 0 : let mut args = vec![
139 0 : "-D".to_owned(),
140 0 : datadir
141 0 : .to_str()
142 0 : .with_context(|| {
143 0 : format!("Datadir path {datadir:?} cannot be represented as a unicode string")
144 0 : })?
145 0 : .to_owned(),
146 0 : "--id".to_owned(),
147 0 : id_string,
148 0 : "--listen-pg".to_owned(),
149 0 : listen_pg,
150 0 : "--listen-http".to_owned(),
151 0 : listen_http,
152 0 : "--availability-zone".to_owned(),
153 0 : availability_zone,
154 : ];
155 0 : if let Some(pg_tenant_only_port) = self.conf.pg_tenant_only_port {
156 0 : let listen_pg_tenant_only = format!("{}:{}", self.listen_addr, pg_tenant_only_port);
157 0 : args.extend(["--listen-pg-tenant-only".to_owned(), listen_pg_tenant_only]);
158 0 : }
159 0 : if !self.conf.sync {
160 0 : args.push("--no-sync".to_owned());
161 0 : }
162 :
163 0 : let broker_endpoint = format!("{}", self.env.broker.client_url());
164 0 : args.extend(["--broker-endpoint".to_owned(), broker_endpoint]);
165 0 :
166 0 : let mut backup_threads = String::new();
167 0 : if let Some(threads) = self.conf.backup_threads {
168 0 : backup_threads = threads.to_string();
169 0 : args.extend(["--backup-threads".to_owned(), backup_threads]);
170 0 : } else {
171 0 : drop(backup_threads);
172 0 : }
173 :
174 0 : if let Some(ref remote_storage) = self.conf.remote_storage {
175 0 : args.extend(["--remote-storage".to_owned(), remote_storage.clone()]);
176 0 : }
177 :
178 0 : let key_path = self.env.base_data_dir.join("auth_public_key.pem");
179 0 : if self.conf.auth_enabled {
180 0 : let key_path_string = key_path
181 0 : .to_str()
182 0 : .with_context(|| {
183 0 : format!("Key path {key_path:?} cannot be represented as a unicode string")
184 0 : })?
185 0 : .to_owned();
186 0 : args.extend([
187 0 : "--pg-auth-public-key-path".to_owned(),
188 0 : key_path_string.clone(),
189 0 : ]);
190 0 : args.extend([
191 0 : "--pg-tenant-only-auth-public-key-path".to_owned(),
192 0 : key_path_string.clone(),
193 0 : ]);
194 0 : args.extend([
195 0 : "--http-auth-public-key-path".to_owned(),
196 0 : key_path_string.clone(),
197 0 : ]);
198 0 : }
199 :
200 0 : args.extend(extra_opts);
201 0 :
202 0 : background_process::start_process(
203 0 : &format!("safekeeper-{id}"),
204 0 : &datadir,
205 0 : &self.env.safekeeper_bin(),
206 0 : &args,
207 0 : self.safekeeper_env_variables()?,
208 0 : background_process::InitialPidFile::Expect(self.pid_file()),
209 0 : retry_timeout,
210 0 : || async {
211 0 : match self.check_status().await {
212 0 : Ok(()) => Ok(true),
213 0 : Err(SafekeeperHttpError::Transport(_)) => Ok(false),
214 0 : Err(e) => Err(anyhow::anyhow!("Failed to check node status: {e}")),
215 0 : }
216 0 : },
217 : )
218 0 : .await
219 0 : }
220 :
221 0 : fn safekeeper_env_variables(&self) -> anyhow::Result<Vec<(String, String)>> {
222 0 : // Generate a token to connect from safekeeper to peers
223 0 : if self.conf.auth_enabled {
224 0 : let token = self
225 0 : .env
226 0 : .generate_auth_token(&Claims::new(None, Scope::SafekeeperData))?;
227 0 : Ok(vec![("SAFEKEEPER_AUTH_TOKEN".to_owned(), token)])
228 : } else {
229 0 : Ok(Vec::new())
230 : }
231 0 : }
232 :
233 : ///
234 : /// Stop the server.
235 : ///
236 : /// If 'immediate' is true, we use SIGQUIT, killing the process immediately.
237 : /// Otherwise we use SIGTERM, triggering a clean shutdown
238 : ///
239 : /// If the server is not running, returns success
240 : ///
241 0 : pub fn stop(&self, immediate: bool) -> anyhow::Result<()> {
242 0 : background_process::stop_process(
243 0 : immediate,
244 0 : &format!("safekeeper {}", self.id),
245 0 : &self.pid_file(),
246 0 : )
247 0 : }
248 :
249 0 : fn http_request<U: IntoUrl>(&self, method: Method, url: U) -> reqwest::RequestBuilder {
250 0 : // TODO: authentication
251 0 : //if self.env.auth_type == AuthType::NeonJWT {
252 0 : // builder = builder.bearer_auth(&self.env.safekeeper_auth_token)
253 0 : //}
254 0 : self.http_client.request(method, url)
255 0 : }
256 :
257 0 : pub async fn check_status(&self) -> Result<()> {
258 0 : self.http_request(Method::GET, format!("{}/{}", self.http_base_url, "status"))
259 0 : .send()
260 0 : .await?
261 0 : .error_from_body()
262 0 : .await?;
263 0 : Ok(())
264 0 : }
265 : }
|
