Line data Source code
1 : //!
2 : //! Postgres wrapper (`compute_ctl`) is intended to be run as a Docker entrypoint or as a `systemd`
3 : //! `ExecStart` option. It will handle all the `Neon` specifics during compute node
4 : //! initialization:
5 : //! - `compute_ctl` accepts cluster (compute node) specification as a JSON file.
6 : //! - Every start is a fresh start, so the data directory is removed and
7 : //! initialized again on each run.
8 : //! - If remote_extension_config is provided, it will be used to fetch extensions list
9 : //! and download `shared_preload_libraries` from the remote storage.
10 : //! - Next it will put configuration files into the `PGDATA` directory.
11 : //! - Sync safekeepers and get commit LSN.
12 : //! - Get `basebackup` from pageserver using the returned on the previous step LSN.
13 : //! - Try to start `postgres` and wait until it is ready to accept connections.
14 : //! - Check and alter/drop/create roles and databases.
15 : //! - Hang waiting on the `postmaster` process to exit.
16 : //!
17 : //! Also `compute_ctl` spawns two separate service threads:
18 : //! - `compute-monitor` checks the last Postgres activity timestamp and saves it
19 : //! into the shared `ComputeNode`;
20 : //! - `http-endpoint` runs a Hyper HTTP API server, which serves readiness and the
21 : //! last activity requests.
22 : //!
23 : //! If `AUTOSCALING` environment variable is set, `compute_ctl` will start the
24 : //! `vm-monitor` located in [`neon/libs/vm_monitor`]. For VM compute nodes,
25 : //! `vm-monitor` communicates with the VM autoscaling system. It coordinates
26 : //! downscaling and requests immediate upscaling under resource pressure.
27 : //!
28 : //! Usage example:
29 : //! ```sh
30 : //! compute_ctl -D /var/db/postgres/compute \
31 : //! -C 'postgresql://cloud_admin@localhost/postgres' \
32 : //! -c /var/db/postgres/configs/config.json \
33 : //! -b /usr/local/bin/postgres \
34 : //! -r http://pg-ext-s3-gateway \
35 : //! ```
36 : use std::ffi::OsString;
37 : use std::fs::File;
38 : use std::process::exit;
39 : use std::sync::Arc;
40 : use std::sync::atomic::AtomicU64;
41 : use std::sync::mpsc;
42 : use std::thread;
43 : use std::time::Duration;
44 :
45 : use anyhow::{Context, Result, bail};
46 : use clap::Parser;
47 : use compute_api::responses::ComputeConfig;
48 : use compute_tools::compute::{
49 : BUILD_TAG, ComputeNode, ComputeNodeParams, forward_termination_signal,
50 : };
51 : use compute_tools::extension_server::get_pg_version_string;
52 : use compute_tools::logger::*;
53 : use compute_tools::params::*;
54 : use compute_tools::spec::*;
55 : use rlimit::{Resource, setrlimit};
56 : use signal_hook::consts::{SIGINT, SIGQUIT, SIGTERM};
57 : use signal_hook::iterator::Signals;
58 : use tracing::{error, info};
59 : use url::Url;
60 : use utils::failpoint_support;
61 :
62 : #[derive(Debug, Parser)]
63 : #[command(rename_all = "kebab-case")]
64 : struct Cli {
65 : #[arg(short = 'b', long, default_value = "postgres", env = "POSTGRES_PATH")]
66 : pub pgbin: String,
67 :
68 : /// The base URL for the remote extension storage proxy gateway.
69 : #[arg(short = 'r', long, value_parser = Self::parse_remote_ext_base_url)]
70 : pub remote_ext_base_url: Option<Url>,
71 :
72 : /// The port to bind the external listening HTTP server to. Clients running
73 : /// outside the compute will talk to the compute through this port. Keep
74 : /// the previous name for this argument around for a smoother release
75 : /// with the control plane.
76 : #[arg(long, default_value_t = 3080)]
77 : pub external_http_port: u16,
78 :
79 : /// The port to bind the internal listening HTTP server to. Clients include
80 : /// the neon extension (for installing remote extensions) and local_proxy.
81 : #[arg(long, default_value_t = 3081)]
82 : pub internal_http_port: u16,
83 :
84 : #[arg(short = 'D', long, value_name = "DATADIR")]
85 : pub pgdata: String,
86 :
87 : #[arg(short = 'C', long, value_name = "DATABASE_URL")]
88 : pub connstr: String,
89 :
90 : #[cfg(target_os = "linux")]
91 : #[arg(long, default_value = "neon-postgres")]
92 : pub cgroup: String,
93 :
94 : #[cfg(target_os = "linux")]
95 : #[arg(
96 : long,
97 : default_value = "host=localhost port=5432 dbname=postgres user=cloud_admin sslmode=disable application_name=vm-monitor"
98 : )]
99 : pub filecache_connstr: String,
100 :
101 : #[cfg(target_os = "linux")]
102 : #[arg(long, default_value = "0.0.0.0:10301")]
103 : pub vm_monitor_addr: String,
104 :
105 : #[arg(long, action = clap::ArgAction::SetTrue)]
106 : pub resize_swap_on_bind: bool,
107 :
108 : #[arg(long)]
109 : pub set_disk_quota_for_fs: Option<String>,
110 :
111 : #[arg(short = 'c', long)]
112 : pub config: Option<OsString>,
113 :
114 : #[arg(short = 'i', long, group = "compute-id")]
115 : pub compute_id: String,
116 :
117 : #[arg(
118 : short = 'p',
119 : long,
120 : conflicts_with = "config",
121 : value_name = "CONTROL_PLANE_API_BASE_URL",
122 : requires = "compute-id"
123 : )]
124 : pub control_plane_uri: Option<String>,
125 :
126 : /// Interval in seconds for collecting installed extensions statistics
127 : #[arg(long, default_value = "3600")]
128 : pub installed_extensions_collection_interval: u64,
129 :
130 : /// Run in development mode, skipping VM-specific operations like process termination
131 : #[arg(long, action = clap::ArgAction::SetTrue)]
132 : pub dev: bool,
133 : }
134 :
135 : impl Cli {
136 : /// Parse a URL from an argument. By default, this isn't necessary, but we
137 : /// want to do some sanity checking.
138 3 : fn parse_remote_ext_base_url(value: &str) -> Result<Url> {
139 : // Remove extra trailing slashes, and add one. We use Url::join() later
140 : // when downloading remote extensions. If the base URL is something like
141 : // http://example.com/pg-ext-s3-gateway, and join() is called with
142 : // something like "xyz", the resulting URL is http://example.com/xyz.
143 3 : let value = value.trim_end_matches('/').to_owned() + "/";
144 3 : let url = Url::parse(&value)?;
145 :
146 3 : if url.query_pairs().count() != 0 {
147 1 : bail!("parameters detected in remote extensions base URL")
148 2 : }
149 :
150 2 : Ok(url)
151 3 : }
152 : }
153 :
154 0 : fn main() -> Result<()> {
155 0 : let cli = Cli::parse();
156 :
157 0 : let scenario = failpoint_support::init();
158 :
159 : // For historical reasons, the main thread that processes the config and launches postgres
160 : // is synchronous, but we always have this tokio runtime available and we "enter" it so
161 : // that you can use tokio::spawn() and tokio::runtime::Handle::current().block_on(...)
162 : // from all parts of compute_ctl.
163 0 : let runtime = tokio::runtime::Builder::new_multi_thread()
164 0 : .enable_all()
165 0 : .build()?;
166 0 : let _rt_guard = runtime.enter();
167 :
168 0 : runtime.block_on(init(cli.dev))?;
169 :
170 : // enable core dumping for all child processes
171 0 : setrlimit(Resource::CORE, rlimit::INFINITY, rlimit::INFINITY)?;
172 :
173 0 : let connstr = Url::parse(&cli.connstr).context("cannot parse connstr as a URL")?;
174 :
175 0 : let config = get_config(&cli)?;
176 :
177 0 : let compute_node = ComputeNode::new(
178 0 : ComputeNodeParams {
179 0 : compute_id: cli.compute_id,
180 0 : connstr,
181 0 : pgdata: cli.pgdata.clone(),
182 0 : pgbin: cli.pgbin.clone(),
183 0 : pgversion: get_pg_version_string(&cli.pgbin),
184 0 : external_http_port: cli.external_http_port,
185 0 : internal_http_port: cli.internal_http_port,
186 0 : remote_ext_base_url: cli.remote_ext_base_url.clone(),
187 0 : resize_swap_on_bind: cli.resize_swap_on_bind,
188 0 : set_disk_quota_for_fs: cli.set_disk_quota_for_fs,
189 0 : #[cfg(target_os = "linux")]
190 0 : filecache_connstr: cli.filecache_connstr,
191 0 : #[cfg(target_os = "linux")]
192 0 : cgroup: cli.cgroup,
193 0 : #[cfg(target_os = "linux")]
194 0 : vm_monitor_addr: cli.vm_monitor_addr,
195 0 : installed_extensions_collection_interval: Arc::new(AtomicU64::new(
196 0 : cli.installed_extensions_collection_interval,
197 0 : )),
198 0 : },
199 0 : config,
200 0 : )?;
201 :
202 0 : let exit_code = compute_node.run()?;
203 :
204 0 : scenario.teardown();
205 :
206 0 : deinit_and_exit(exit_code);
207 0 : }
208 :
209 0 : async fn init(dev_mode: bool) -> Result<()> {
210 0 : init_tracing_and_logging(DEFAULT_LOG_LEVEL).await?;
211 :
212 0 : let mut signals = Signals::new([SIGINT, SIGTERM, SIGQUIT])?;
213 0 : thread::spawn(move || {
214 0 : for sig in signals.forever() {
215 0 : handle_exit_signal(sig, dev_mode);
216 0 : }
217 0 : });
218 :
219 0 : info!("compute build_tag: {}", &BUILD_TAG.to_string());
220 :
221 0 : Ok(())
222 0 : }
223 :
224 0 : fn get_config(cli: &Cli) -> Result<ComputeConfig> {
225 : // First, read the config from the path if provided
226 0 : if let Some(ref config) = cli.config {
227 0 : let file = File::open(config)?;
228 0 : return Ok(serde_json::from_reader(&file)?);
229 0 : }
230 :
231 : // If the config wasn't provided in the CLI arguments, then retrieve it from
232 : // the control plane
233 0 : match get_config_from_control_plane(cli.control_plane_uri.as_ref().unwrap(), &cli.compute_id) {
234 0 : Ok(config) => Ok(config),
235 0 : Err(e) => {
236 0 : error!(
237 0 : "cannot get response from control plane: {}\n\
238 0 : neither spec nor confirmation that compute is in the Empty state was received",
239 : e
240 : );
241 0 : Err(e)
242 : }
243 : }
244 0 : }
245 :
246 0 : fn deinit_and_exit(exit_code: Option<i32>) -> ! {
247 : // Shutdown trace pipeline gracefully, so that it has a chance to send any
248 : // pending traces before we exit. Shutting down OTEL tracing provider may
249 : // hang for quite some time, see, for example:
250 : // - https://github.com/open-telemetry/opentelemetry-rust/issues/868
251 : // - and our problems with staging https://github.com/neondatabase/cloud/issues/3707#issuecomment-1493983636
252 : //
253 : // Yet, we want computes to shut down fast enough, as we may need a new one
254 : // for the same timeline ASAP. So wait no longer than 2s for the shutdown to
255 : // complete, then just error out and exit the main thread.
256 0 : info!("shutting down tracing");
257 0 : let (sender, receiver) = mpsc::channel();
258 0 : let _ = thread::spawn(move || {
259 0 : tracing_utils::shutdown_tracing();
260 0 : sender.send(()).ok()
261 0 : });
262 0 : let shutdown_res = receiver.recv_timeout(Duration::from_millis(2000));
263 0 : if shutdown_res.is_err() {
264 0 : error!("timed out while shutting down tracing, exiting anyway");
265 0 : }
266 :
267 0 : info!("shutting down");
268 0 : exit(exit_code.unwrap_or(1))
269 : }
270 :
271 : /// When compute_ctl is killed, send also termination signal to sync-safekeepers
272 : /// to prevent leakage. TODO: it is better to convert compute_ctl to async and
273 : /// wait for termination which would be easy then.
274 0 : fn handle_exit_signal(sig: i32, dev_mode: bool) {
275 0 : info!("received {sig} termination signal");
276 0 : forward_termination_signal(dev_mode);
277 0 : exit(1);
278 : }
279 :
280 : #[cfg(test)]
281 : mod test {
282 : use clap::{CommandFactory, Parser};
283 : use url::Url;
284 :
285 : use super::Cli;
286 :
287 : #[test]
288 1 : fn verify_cli() {
289 1 : Cli::command().debug_assert()
290 1 : }
291 :
292 : #[test]
293 1 : fn verify_remote_ext_base_url() {
294 1 : let cli = Cli::parse_from([
295 1 : "compute_ctl",
296 1 : "--pgdata=test",
297 1 : "--connstr=test",
298 1 : "--compute-id=test",
299 1 : "--remote-ext-base-url",
300 1 : "https://example.com/subpath",
301 1 : ]);
302 1 : assert_eq!(
303 1 : cli.remote_ext_base_url.unwrap(),
304 1 : Url::parse("https://example.com/subpath/").unwrap()
305 : );
306 :
307 1 : let cli = Cli::parse_from([
308 1 : "compute_ctl",
309 1 : "--pgdata=test",
310 1 : "--connstr=test",
311 1 : "--compute-id=test",
312 1 : "--remote-ext-base-url",
313 1 : "https://example.com//",
314 1 : ]);
315 1 : assert_eq!(
316 1 : cli.remote_ext_base_url.unwrap(),
317 1 : Url::parse("https://example.com").unwrap()
318 : );
319 :
320 1 : Cli::try_parse_from([
321 1 : "compute_ctl",
322 1 : "--pgdata=test",
323 1 : "--connstr=test",
324 1 : "--compute-id=test",
325 1 : "--remote-ext-base-url",
326 1 : "https://example.com?hello=world",
327 1 : ])
328 1 : .expect_err("URL parameters are not allowed");
329 1 : }
330 : }
|
