Line data Source code
1 : //! This program dumps a remote Postgres database into a local Postgres database
2 : //! and uploads the resulting PGDATA into object storage for import into a Timeline.
3 : //!
4 : //! # Context, Architecture, Design
5 : //!
6 : //! See cloud.git Fast Imports RFC (<https://github.com/neondatabase/cloud/pull/19799>)
7 : //! for the full picture.
8 : //! The RFC describing the storage pieces of importing the PGDATA dump into a Timeline
9 : //! is publicly accessible at <https://github.com/neondatabase/neon/pull/9538>.
10 : //!
11 : //! # This is a Prototype!
12 : //!
13 : //! This program is part of a prototype feature and not yet used in production.
14 : //!
15 : //! The cloud.git RFC contains lots of suggestions for improving e2e throughput
16 : //! of this step of the timeline import process.
17 : //!
18 : //! # Local Testing
19 : //!
20 : //! - Comment out most of the pgxns in The Dockerfile.compute-tools to speed up the build.
21 : //! - Build the image with the following command:
22 : //!
23 : //! ```bash
24 : //! docker buildx build --platform linux/amd64 --build-arg DEBIAN_VERSION=bullseye --build-arg GIT_VERSION=local --build-arg PG_VERSION=v14 --build-arg BUILD_TAG="$(date --iso-8601=s -u)" -t localhost:3030/localregistry/compute-node-v14:latest -f compute/compute-node.Dockerfile .
25 : //! docker push localhost:3030/localregistry/compute-node-v14:latest
26 : //! ```
27 :
28 : use anyhow::Context;
29 : use aws_config::BehaviorVersion;
30 : use camino::{Utf8Path, Utf8PathBuf};
31 : use clap::Parser;
32 : use compute_tools::extension_server::{get_pg_version, PostgresMajorVersion};
33 : use nix::unistd::Pid;
34 : use tracing::{info, info_span, warn, Instrument};
35 : use utils::fs_ext::is_directory_empty;
36 :
37 : #[path = "fast_import/aws_s3_sync.rs"]
38 : mod aws_s3_sync;
39 : #[path = "fast_import/child_stdio_to_log.rs"]
40 : mod child_stdio_to_log;
41 : #[path = "fast_import/s3_uri.rs"]
42 : mod s3_uri;
43 :
44 : #[derive(clap::Parser)]
45 : struct Args {
46 : #[clap(long)]
47 0 : working_directory: Utf8PathBuf,
48 : #[clap(long, env = "NEON_IMPORTER_S3_PREFIX")]
49 0 : s3_prefix: s3_uri::S3Uri,
50 : #[clap(long)]
51 0 : pg_bin_dir: Utf8PathBuf,
52 : #[clap(long)]
53 0 : pg_lib_dir: Utf8PathBuf,
54 : }
55 :
56 : #[serde_with::serde_as]
57 0 : #[derive(serde::Deserialize)]
58 : struct Spec {
59 : encryption_secret: EncryptionSecret,
60 : #[serde_as(as = "serde_with::base64::Base64")]
61 : source_connstring_ciphertext_base64: Vec<u8>,
62 : }
63 :
64 0 : #[derive(serde::Deserialize)]
65 : enum EncryptionSecret {
66 : #[allow(clippy::upper_case_acronyms)]
67 : KMS { key_id: String },
68 : }
69 :
70 : #[tokio::main]
71 0 : pub(crate) async fn main() -> anyhow::Result<()> {
72 0 : utils::logging::init(
73 0 : utils::logging::LogFormat::Plain,
74 0 : utils::logging::TracingErrorLayerEnablement::EnableWithRustLogFilter,
75 0 : utils::logging::Output::Stdout,
76 0 : )?;
77 0 :
78 0 : info!("starting");
79 0 :
80 0 : let Args {
81 0 : working_directory,
82 0 : s3_prefix,
83 0 : pg_bin_dir,
84 0 : pg_lib_dir,
85 0 : } = Args::parse();
86 0 :
87 0 : let aws_config = aws_config::load_defaults(BehaviorVersion::v2024_03_28()).await;
88 0 :
89 0 : let spec: Spec = {
90 0 : let spec_key = s3_prefix.append("/spec.json");
91 0 : let s3_client = aws_sdk_s3::Client::new(&aws_config);
92 0 : let object = s3_client
93 0 : .get_object()
94 0 : .bucket(&spec_key.bucket)
95 0 : .key(spec_key.key)
96 0 : .send()
97 0 : .await
98 0 : .context("get spec from s3")?
99 0 : .body
100 0 : .collect()
101 0 : .await
102 0 : .context("download spec body")?;
103 0 : serde_json::from_slice(&object.into_bytes()).context("parse spec as json")?
104 0 : };
105 0 :
106 0 : match tokio::fs::create_dir(&working_directory).await {
107 0 : Ok(()) => {}
108 0 : Err(e) if e.kind() == std::io::ErrorKind::AlreadyExists => {
109 0 : if !is_directory_empty(&working_directory)
110 0 : .await
111 0 : .context("check if working directory is empty")?
112 0 : {
113 0 : anyhow::bail!("working directory is not empty");
114 0 : } else {
115 0 : // ok
116 0 : }
117 0 : }
118 0 : Err(e) => return Err(anyhow::Error::new(e).context("create working directory")),
119 0 : }
120 0 :
121 0 : let pgdata_dir = working_directory.join("pgdata");
122 0 : tokio::fs::create_dir(&pgdata_dir)
123 0 : .await
124 0 : .context("create pgdata directory")?;
125 0 :
126 0 : //
127 0 : // Setup clients
128 0 : //
129 0 : let aws_config = aws_config::load_defaults(BehaviorVersion::v2024_03_28()).await;
130 0 : let kms_client = aws_sdk_kms::Client::new(&aws_config);
131 0 :
132 0 : //
133 0 : // Initialize pgdata
134 0 : //
135 0 : let pgbin = pg_bin_dir.join("postgres");
136 0 : let pg_version = match get_pg_version(pgbin.as_ref()) {
137 0 : PostgresMajorVersion::V14 => 14,
138 0 : PostgresMajorVersion::V15 => 15,
139 0 : PostgresMajorVersion::V16 => 16,
140 0 : PostgresMajorVersion::V17 => 17,
141 0 : };
142 0 : let superuser = "cloud_admin"; // XXX: this shouldn't be hard-coded
143 0 : postgres_initdb::do_run_initdb(postgres_initdb::RunInitdbArgs {
144 0 : superuser,
145 0 : locale: "en_US.UTF-8", // XXX: this shouldn't be hard-coded,
146 0 : pg_version,
147 0 : initdb_bin: pg_bin_dir.join("initdb").as_ref(),
148 0 : library_search_path: &pg_lib_dir, // TODO: is this right? Prob works in compute image, not sure about neon_local.
149 0 : pgdata: &pgdata_dir,
150 0 : })
151 0 : .await
152 0 : .context("initdb")?;
153 0 :
154 0 : let nproc = num_cpus::get();
155 0 :
156 0 : //
157 0 : // Launch postgres process
158 0 : //
159 0 : let mut postgres_proc = tokio::process::Command::new(pgbin)
160 0 : .arg("-D")
161 0 : .arg(&pgdata_dir)
162 0 : .args(["-c", "wal_level=minimal"])
163 0 : .args(["-c", "shared_buffers=10GB"])
164 0 : .args(["-c", "max_wal_senders=0"])
165 0 : .args(["-c", "fsync=off"])
166 0 : .args(["-c", "full_page_writes=off"])
167 0 : .args(["-c", "synchronous_commit=off"])
168 0 : .args(["-c", "maintenance_work_mem=8388608"])
169 0 : .args(["-c", &format!("max_parallel_maintenance_workers={nproc}")])
170 0 : .args(["-c", &format!("max_parallel_workers={nproc}")])
171 0 : .args(["-c", &format!("max_parallel_workers_per_gather={nproc}")])
172 0 : .args(["-c", &format!("max_worker_processes={nproc}")])
173 0 : .args(["-c", "effective_io_concurrency=100"])
174 0 : .env_clear()
175 0 : .stdout(std::process::Stdio::piped())
176 0 : .stderr(std::process::Stdio::piped())
177 0 : .spawn()
178 0 : .context("spawn postgres")?;
179 0 :
180 0 : info!("spawned postgres, waiting for it to become ready");
181 0 : tokio::spawn(
182 0 : child_stdio_to_log::relay_process_output(
183 0 : postgres_proc.stdout.take(),
184 0 : postgres_proc.stderr.take(),
185 0 : )
186 0 : .instrument(info_span!("postgres")),
187 0 : );
188 0 : let restore_pg_connstring =
189 0 : format!("host=localhost port=5432 user={superuser} dbname=postgres");
190 0 : loop {
191 0 : let res = tokio_postgres::connect(&restore_pg_connstring, tokio_postgres::NoTls).await;
192 0 : if res.is_ok() {
193 0 : info!("postgres is ready, could connect to it");
194 0 : break;
195 0 : }
196 0 : }
197 0 :
198 0 : //
199 0 : // Decrypt connection string
200 0 : //
201 0 : let source_connection_string = {
202 0 : match spec.encryption_secret {
203 0 : EncryptionSecret::KMS { key_id } => {
204 0 : let mut output = kms_client
205 0 : .decrypt()
206 0 : .key_id(key_id)
207 0 : .ciphertext_blob(aws_sdk_s3::primitives::Blob::new(
208 0 : spec.source_connstring_ciphertext_base64,
209 0 : ))
210 0 : .send()
211 0 : .await
212 0 : .context("decrypt source connection string")?;
213 0 : let plaintext = output
214 0 : .plaintext
215 0 : .take()
216 0 : .context("get plaintext source connection string")?;
217 0 : String::from_utf8(plaintext.into_inner())
218 0 : .context("parse source connection string as utf8")?
219 0 : }
220 0 : }
221 0 : };
222 0 :
223 0 : //
224 0 : // Start the work
225 0 : //
226 0 :
227 0 : let dumpdir = working_directory.join("dumpdir");
228 0 :
229 0 : let common_args = [
230 0 : // schema mapping (prob suffices to specify them on one side)
231 0 : "--no-owner".to_string(),
232 0 : "--no-privileges".to_string(),
233 0 : "--no-publications".to_string(),
234 0 : "--no-security-labels".to_string(),
235 0 : "--no-subscriptions".to_string(),
236 0 : "--no-tablespaces".to_string(),
237 0 : // format
238 0 : "--format".to_string(),
239 0 : "directory".to_string(),
240 0 : // concurrency
241 0 : "--jobs".to_string(),
242 0 : num_cpus::get().to_string(),
243 0 : // progress updates
244 0 : "--verbose".to_string(),
245 0 : ];
246 0 :
247 0 : info!("dump into the working directory");
248 0 : {
249 0 : let mut pg_dump = tokio::process::Command::new(pg_bin_dir.join("pg_dump"))
250 0 : .args(&common_args)
251 0 : .arg("-f")
252 0 : .arg(&dumpdir)
253 0 : .arg("--no-sync")
254 0 : // POSITIONAL args
255 0 : // source db (db name included in connection string)
256 0 : .arg(&source_connection_string)
257 0 : // how we run it
258 0 : .env_clear()
259 0 : .kill_on_drop(true)
260 0 : .stdout(std::process::Stdio::piped())
261 0 : .stderr(std::process::Stdio::piped())
262 0 : .spawn()
263 0 : .context("spawn pg_dump")?;
264 0 :
265 0 : info!(pid=%pg_dump.id().unwrap(), "spawned pg_dump");
266 0 :
267 0 : tokio::spawn(
268 0 : child_stdio_to_log::relay_process_output(pg_dump.stdout.take(), pg_dump.stderr.take())
269 0 : .instrument(info_span!("pg_dump")),
270 0 : );
271 0 :
272 0 : let st = pg_dump.wait().await.context("wait for pg_dump")?;
273 0 : info!(status=?st, "pg_dump exited");
274 0 : if !st.success() {
275 0 : warn!(status=%st, "pg_dump failed, restore will likely fail as well");
276 0 : }
277 0 : }
278 0 :
279 0 : // TODO: do it in a streaming way, plenty of internal research done on this already
280 0 : // TODO: do the unlogged table trick
281 0 :
282 0 : info!("restore from working directory into vanilla postgres");
283 0 : {
284 0 : let mut pg_restore = tokio::process::Command::new(pg_bin_dir.join("pg_restore"))
285 0 : .args(&common_args)
286 0 : .arg("-d")
287 0 : .arg(&restore_pg_connstring)
288 0 : // POSITIONAL args
289 0 : .arg(&dumpdir)
290 0 : // how we run it
291 0 : .env_clear()
292 0 : .kill_on_drop(true)
293 0 : .stdout(std::process::Stdio::piped())
294 0 : .stderr(std::process::Stdio::piped())
295 0 : .spawn()
296 0 : .context("spawn pg_restore")?;
297 0 :
298 0 : info!(pid=%pg_restore.id().unwrap(), "spawned pg_restore");
299 0 : tokio::spawn(
300 0 : child_stdio_to_log::relay_process_output(
301 0 : pg_restore.stdout.take(),
302 0 : pg_restore.stderr.take(),
303 0 : )
304 0 : .instrument(info_span!("pg_restore")),
305 0 : );
306 0 : let st = pg_restore.wait().await.context("wait for pg_restore")?;
307 0 : info!(status=?st, "pg_restore exited");
308 0 : if !st.success() {
309 0 : warn!(status=%st, "pg_restore failed, restore will likely fail as well");
310 0 : }
311 0 : }
312 0 :
313 0 : info!("shutdown postgres");
314 0 : {
315 0 : nix::sys::signal::kill(
316 0 : Pid::from_raw(
317 0 : i32::try_from(postgres_proc.id().unwrap()).expect("convert child pid to i32"),
318 0 : ),
319 0 : nix::sys::signal::SIGTERM,
320 0 : )
321 0 : .context("signal postgres to shut down")?;
322 0 : postgres_proc
323 0 : .wait()
324 0 : .await
325 0 : .context("wait for postgres to shut down")?;
326 0 : }
327 0 :
328 0 : info!("upload pgdata");
329 0 : aws_s3_sync::sync(Utf8Path::new(&pgdata_dir), &s3_prefix.append("/pgdata/"))
330 0 : .await
331 0 : .context("sync dump directory to destination")?;
332 0 :
333 0 : info!("write status");
334 0 : {
335 0 : let status_dir = working_directory.join("status");
336 0 : std::fs::create_dir(&status_dir).context("create status directory")?;
337 0 : let status_file = status_dir.join("pgdata");
338 0 : std::fs::write(&status_file, serde_json::json!({"done": true}).to_string())
339 0 : .context("write status file")?;
340 0 : aws_s3_sync::sync(&status_dir, &s3_prefix.append("/status/"))
341 0 : .await
342 0 : .context("sync status directory to destination")?;
343 0 : }
344 0 :
345 0 : Ok(())
346 0 : }
|
