Line data Source code
1 : //!
2 : //! `neon_local` is an executable that can be used to create a local
3 : //! Neon environment, for testing purposes. The local environment is
4 : //! quite different from the cloud environment with Kubernetes, but it
5 : //! easier to work with locally. The python tests in `test_runner`
6 : //! rely on `neon_local` to set up the environment for each test.
7 : //!
8 : use anyhow::{anyhow, bail, Context, Result};
9 : use clap::{value_parser, Arg, ArgAction, ArgMatches, Command, ValueEnum};
10 : use compute_api::spec::ComputeMode;
11 : use control_plane::endpoint::ComputeControlPlane;
12 : use control_plane::local_env::{InitForceMode, LocalEnv};
13 : use control_plane::pageserver::{PageServerNode, PAGESERVER_REMOTE_STORAGE_DIR};
14 : use control_plane::safekeeper::SafekeeperNode;
15 : use control_plane::storage_controller::StorageController;
16 : use control_plane::{broker, local_env};
17 : use pageserver_api::controller_api::PlacementPolicy;
18 : use pageserver_api::models::{
19 : ShardParameters, TenantCreateRequest, TimelineCreateRequest, TimelineInfo,
20 : };
21 : use pageserver_api::shard::{ShardCount, ShardStripeSize, TenantShardId};
22 : use pageserver_api::{
23 : DEFAULT_HTTP_LISTEN_PORT as DEFAULT_PAGESERVER_HTTP_PORT,
24 : DEFAULT_PG_LISTEN_PORT as DEFAULT_PAGESERVER_PG_PORT,
25 : };
26 : use postgres_backend::AuthType;
27 : use postgres_connection::parse_host_port;
28 : use safekeeper_api::{
29 : DEFAULT_HTTP_LISTEN_PORT as DEFAULT_SAFEKEEPER_HTTP_PORT,
30 : DEFAULT_PG_LISTEN_PORT as DEFAULT_SAFEKEEPER_PG_PORT,
31 : };
32 : use std::collections::{BTreeSet, HashMap};
33 : use std::path::PathBuf;
34 : use std::process::exit;
35 : use std::str::FromStr;
36 : use storage_broker::DEFAULT_LISTEN_ADDR as DEFAULT_BROKER_ADDR;
37 : use url::Host;
38 : use utils::{
39 : auth::{Claims, Scope},
40 : id::{NodeId, TenantId, TenantTimelineId, TimelineId},
41 : lsn::Lsn,
42 : project_git_version,
43 : };
44 :
45 : // Default id of a safekeeper node, if not specified on the command line.
46 : const DEFAULT_SAFEKEEPER_ID: NodeId = NodeId(1);
47 : const DEFAULT_PAGESERVER_ID: NodeId = NodeId(1);
48 : const DEFAULT_BRANCH_NAME: &str = "main";
49 : project_git_version!(GIT_VERSION);
50 :
51 : const DEFAULT_PG_VERSION: &str = "15";
52 :
53 : const DEFAULT_PAGESERVER_CONTROL_PLANE_API: &str = "http://127.0.0.1:1234/upcall/v1/";
54 :
55 0 : fn default_conf(num_pageservers: u16) -> String {
56 0 : let mut template = format!(
57 0 : r#"
58 0 : # Default built-in configuration, defined in main.rs
59 0 : control_plane_api = '{DEFAULT_PAGESERVER_CONTROL_PLANE_API}'
60 0 :
61 0 : [broker]
62 0 : listen_addr = '{DEFAULT_BROKER_ADDR}'
63 0 :
64 0 : [[safekeepers]]
65 0 : id = {DEFAULT_SAFEKEEPER_ID}
66 0 : pg_port = {DEFAULT_SAFEKEEPER_PG_PORT}
67 0 : http_port = {DEFAULT_SAFEKEEPER_HTTP_PORT}
68 0 :
69 0 : "#,
70 0 : );
71 :
72 0 : for i in 0..num_pageservers {
73 0 : let pageserver_id = NodeId(DEFAULT_PAGESERVER_ID.0 + i as u64);
74 0 : let pg_port = DEFAULT_PAGESERVER_PG_PORT + i;
75 0 : let http_port = DEFAULT_PAGESERVER_HTTP_PORT + i;
76 0 :
77 0 : template += &format!(
78 0 : r#"
79 0 : [[pageservers]]
80 0 : id = {pageserver_id}
81 0 : listen_pg_addr = '127.0.0.1:{pg_port}'
82 0 : listen_http_addr = '127.0.0.1:{http_port}'
83 0 : pg_auth_type = '{trust_auth}'
84 0 : http_auth_type = '{trust_auth}'
85 0 : "#,
86 0 : trust_auth = AuthType::Trust,
87 0 : )
88 : }
89 :
90 0 : template
91 0 : }
92 :
93 : ///
94 : /// Timelines tree element used as a value in the HashMap.
95 : ///
96 : struct TimelineTreeEl {
97 : /// `TimelineInfo` received from the `pageserver` via the `timeline_list` http API call.
98 : pub info: TimelineInfo,
99 : /// Name, recovered from neon config mappings
100 : pub name: Option<String>,
101 : /// Holds all direct children of this timeline referenced using `timeline_id`.
102 : pub children: BTreeSet<TimelineId>,
103 : }
104 :
105 : // Main entry point for the 'neon_local' CLI utility
106 : //
107 : // This utility helps to manage neon installation. That includes following:
108 : // * Management of local postgres installations running on top of the
109 : // pageserver.
110 : // * Providing CLI api to the pageserver
111 : // * TODO: export/import to/from usual postgres
112 0 : fn main() -> Result<()> {
113 0 : let matches = cli().get_matches();
114 :
115 0 : let (sub_name, sub_args) = match matches.subcommand() {
116 0 : Some(subcommand_data) => subcommand_data,
117 0 : None => bail!("no subcommand provided"),
118 : };
119 :
120 : // Check for 'neon init' command first.
121 0 : let subcommand_result = if sub_name == "init" {
122 0 : handle_init(sub_args).map(Some)
123 : } else {
124 : // all other commands need an existing config
125 0 : let mut env = LocalEnv::load_config().context("Error loading config")?;
126 0 : let original_env = env.clone();
127 0 :
128 0 : let rt = tokio::runtime::Builder::new_current_thread()
129 0 : .enable_all()
130 0 : .build()
131 0 : .unwrap();
132 :
133 0 : let subcommand_result = match sub_name {
134 0 : "tenant" => rt.block_on(handle_tenant(sub_args, &mut env)),
135 0 : "timeline" => rt.block_on(handle_timeline(sub_args, &mut env)),
136 0 : "start" => rt.block_on(handle_start_all(sub_args, &env)),
137 0 : "stop" => rt.block_on(handle_stop_all(sub_args, &env)),
138 0 : "pageserver" => rt.block_on(handle_pageserver(sub_args, &env)),
139 0 : "storage_controller" => rt.block_on(handle_storage_controller(sub_args, &env)),
140 0 : "safekeeper" => rt.block_on(handle_safekeeper(sub_args, &env)),
141 0 : "endpoint" => rt.block_on(handle_endpoint(sub_args, &env)),
142 0 : "mappings" => handle_mappings(sub_args, &mut env),
143 0 : "pg" => bail!("'pg' subcommand has been renamed to 'endpoint'"),
144 0 : _ => bail!("unexpected subcommand {sub_name}"),
145 : };
146 :
147 0 : if original_env != env {
148 0 : subcommand_result.map(|()| Some(env))
149 : } else {
150 0 : subcommand_result.map(|()| None)
151 : }
152 : };
153 :
154 0 : match subcommand_result {
155 0 : Ok(Some(updated_env)) => updated_env.persist_config(&updated_env.base_data_dir)?,
156 0 : Ok(None) => (),
157 0 : Err(e) => {
158 0 : eprintln!("command failed: {e:?}");
159 0 : exit(1);
160 : }
161 : }
162 0 : Ok(())
163 0 : }
164 :
165 : ///
166 : /// Prints timelines list as a tree-like structure.
167 : ///
168 0 : fn print_timelines_tree(
169 0 : timelines: Vec<TimelineInfo>,
170 0 : mut timeline_name_mappings: HashMap<TenantTimelineId, String>,
171 0 : ) -> Result<()> {
172 0 : let mut timelines_hash = timelines
173 0 : .iter()
174 0 : .map(|t| {
175 0 : (
176 0 : t.timeline_id,
177 0 : TimelineTreeEl {
178 0 : info: t.clone(),
179 0 : children: BTreeSet::new(),
180 0 : name: timeline_name_mappings
181 0 : .remove(&TenantTimelineId::new(t.tenant_id.tenant_id, t.timeline_id)),
182 0 : },
183 0 : )
184 0 : })
185 0 : .collect::<HashMap<_, _>>();
186 :
187 : // Memorize all direct children of each timeline.
188 0 : for timeline in timelines.iter() {
189 0 : if let Some(ancestor_timeline_id) = timeline.ancestor_timeline_id {
190 0 : timelines_hash
191 0 : .get_mut(&ancestor_timeline_id)
192 0 : .context("missing timeline info in the HashMap")?
193 : .children
194 0 : .insert(timeline.timeline_id);
195 0 : }
196 : }
197 :
198 0 : for timeline in timelines_hash.values() {
199 : // Start with root local timelines (no ancestors) first.
200 0 : if timeline.info.ancestor_timeline_id.is_none() {
201 0 : print_timeline(0, &Vec::from([true]), timeline, &timelines_hash)?;
202 0 : }
203 : }
204 :
205 0 : Ok(())
206 0 : }
207 :
208 : ///
209 : /// Recursively prints timeline info with all its children.
210 : ///
211 0 : fn print_timeline(
212 0 : nesting_level: usize,
213 0 : is_last: &[bool],
214 0 : timeline: &TimelineTreeEl,
215 0 : timelines: &HashMap<TimelineId, TimelineTreeEl>,
216 0 : ) -> Result<()> {
217 0 : if nesting_level > 0 {
218 0 : let ancestor_lsn = match timeline.info.ancestor_lsn {
219 0 : Some(lsn) => lsn.to_string(),
220 0 : None => "Unknown Lsn".to_string(),
221 : };
222 :
223 0 : let mut br_sym = "┣━";
224 0 :
225 0 : // Draw each nesting padding with proper style
226 0 : // depending on whether its timeline ended or not.
227 0 : if nesting_level > 1 {
228 0 : for l in &is_last[1..is_last.len() - 1] {
229 0 : if *l {
230 0 : print!(" ");
231 0 : } else {
232 0 : print!("┃ ");
233 0 : }
234 : }
235 0 : }
236 :
237 : // We are the last in this sub-timeline
238 0 : if *is_last.last().unwrap() {
239 0 : br_sym = "┗━";
240 0 : }
241 :
242 0 : print!("{} @{}: ", br_sym, ancestor_lsn);
243 0 : }
244 :
245 : // Finally print a timeline id and name with new line
246 0 : println!(
247 0 : "{} [{}]",
248 0 : timeline.name.as_deref().unwrap_or("_no_name_"),
249 0 : timeline.info.timeline_id
250 0 : );
251 0 :
252 0 : let len = timeline.children.len();
253 0 : let mut i: usize = 0;
254 0 : let mut is_last_new = Vec::from(is_last);
255 0 : is_last_new.push(false);
256 :
257 0 : for child in &timeline.children {
258 0 : i += 1;
259 0 :
260 0 : // Mark that the last padding is the end of the timeline
261 0 : if i == len {
262 0 : if let Some(last) = is_last_new.last_mut() {
263 0 : *last = true;
264 0 : }
265 0 : }
266 :
267 : print_timeline(
268 0 : nesting_level + 1,
269 0 : &is_last_new,
270 0 : timelines
271 0 : .get(child)
272 0 : .context("missing timeline info in the HashMap")?,
273 0 : timelines,
274 0 : )?;
275 : }
276 :
277 0 : Ok(())
278 0 : }
279 :
280 : /// Returns a map of timeline IDs to timeline_id@lsn strings.
281 : /// Connects to the pageserver to query this information.
282 0 : async fn get_timeline_infos(
283 0 : env: &local_env::LocalEnv,
284 0 : tenant_shard_id: &TenantShardId,
285 0 : ) -> Result<HashMap<TimelineId, TimelineInfo>> {
286 0 : Ok(get_default_pageserver(env)
287 0 : .timeline_list(tenant_shard_id)
288 0 : .await?
289 0 : .into_iter()
290 0 : .map(|timeline_info| (timeline_info.timeline_id, timeline_info))
291 0 : .collect())
292 0 : }
293 :
294 : // Helper function to parse --tenant_id option, or get the default from config file
295 0 : fn get_tenant_id(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> anyhow::Result<TenantId> {
296 0 : if let Some(tenant_id_from_arguments) = parse_tenant_id(sub_match).transpose() {
297 0 : tenant_id_from_arguments
298 0 : } else if let Some(default_id) = env.default_tenant_id {
299 0 : Ok(default_id)
300 : } else {
301 0 : anyhow::bail!("No tenant id. Use --tenant-id, or set a default tenant");
302 : }
303 0 : }
304 :
305 : // Helper function to parse --tenant_id option, for commands that accept a shard suffix
306 0 : fn get_tenant_shard_id(
307 0 : sub_match: &ArgMatches,
308 0 : env: &local_env::LocalEnv,
309 0 : ) -> anyhow::Result<TenantShardId> {
310 0 : if let Some(tenant_id_from_arguments) = parse_tenant_shard_id(sub_match).transpose() {
311 0 : tenant_id_from_arguments
312 0 : } else if let Some(default_id) = env.default_tenant_id {
313 0 : Ok(TenantShardId::unsharded(default_id))
314 : } else {
315 0 : anyhow::bail!("No tenant shard id. Use --tenant-id, or set a default tenant");
316 : }
317 0 : }
318 :
319 0 : fn parse_tenant_id(sub_match: &ArgMatches) -> anyhow::Result<Option<TenantId>> {
320 0 : sub_match
321 0 : .get_one::<String>("tenant-id")
322 0 : .map(|tenant_id| TenantId::from_str(tenant_id))
323 0 : .transpose()
324 0 : .context("Failed to parse tenant id from the argument string")
325 0 : }
326 :
327 0 : fn parse_tenant_shard_id(sub_match: &ArgMatches) -> anyhow::Result<Option<TenantShardId>> {
328 0 : sub_match
329 0 : .get_one::<String>("tenant-id")
330 0 : .map(|id_str| TenantShardId::from_str(id_str))
331 0 : .transpose()
332 0 : .context("Failed to parse tenant shard id from the argument string")
333 0 : }
334 :
335 0 : fn parse_timeline_id(sub_match: &ArgMatches) -> anyhow::Result<Option<TimelineId>> {
336 0 : sub_match
337 0 : .get_one::<String>("timeline-id")
338 0 : .map(|timeline_id| TimelineId::from_str(timeline_id))
339 0 : .transpose()
340 0 : .context("Failed to parse timeline id from the argument string")
341 0 : }
342 :
343 0 : fn handle_init(init_match: &ArgMatches) -> anyhow::Result<LocalEnv> {
344 0 : let num_pageservers = init_match
345 0 : .get_one::<u16>("num-pageservers")
346 0 : .expect("num-pageservers arg has a default");
347 : // Create config file
348 0 : let toml_file: String = if let Some(config_path) = init_match.get_one::<PathBuf>("config") {
349 : // load and parse the file
350 0 : std::fs::read_to_string(config_path).with_context(|| {
351 0 : format!(
352 0 : "Could not read configuration file '{}'",
353 0 : config_path.display()
354 0 : )
355 0 : })?
356 : } else {
357 : // Built-in default config
358 0 : default_conf(*num_pageservers)
359 : };
360 :
361 0 : let pg_version = init_match
362 0 : .get_one::<u32>("pg-version")
363 0 : .copied()
364 0 : .context("Failed to parse postgres version from the argument string")?;
365 :
366 0 : let mut env =
367 0 : LocalEnv::parse_config(&toml_file).context("Failed to create neon configuration")?;
368 0 : let force = init_match.get_one("force").expect("we set a default value");
369 0 : env.init(pg_version, force)
370 0 : .context("Failed to initialize neon repository")?;
371 :
372 : // Create remote storage location for default LocalFs remote storage
373 0 : std::fs::create_dir_all(env.base_data_dir.join(PAGESERVER_REMOTE_STORAGE_DIR))?;
374 :
375 : // Initialize pageserver, create initial tenant and timeline.
376 0 : for ps_conf in &env.pageservers {
377 0 : PageServerNode::from_env(&env, ps_conf)
378 0 : .initialize(&pageserver_config_overrides(init_match))
379 0 : .unwrap_or_else(|e| {
380 0 : eprintln!("pageserver init failed: {e:?}");
381 0 : exit(1);
382 0 : });
383 0 : }
384 :
385 0 : Ok(env)
386 0 : }
387 :
388 : /// The default pageserver is the one where CLI tenant/timeline operations are sent by default.
389 : /// For typical interactive use, one would just run with a single pageserver. Scenarios with
390 : /// tenant/timeline placement across multiple pageservers are managed by python test code rather
391 : /// than this CLI.
392 0 : fn get_default_pageserver(env: &local_env::LocalEnv) -> PageServerNode {
393 0 : let ps_conf = env
394 0 : .pageservers
395 0 : .first()
396 0 : .expect("Config is validated to contain at least one pageserver");
397 0 : PageServerNode::from_env(env, ps_conf)
398 0 : }
399 :
400 0 : fn pageserver_config_overrides(init_match: &ArgMatches) -> Vec<&str> {
401 0 : init_match
402 0 : .get_many::<String>("pageserver-config-override")
403 0 : .into_iter()
404 0 : .flatten()
405 0 : .map(String::as_str)
406 0 : .collect()
407 0 : }
408 :
409 0 : async fn handle_tenant(
410 0 : tenant_match: &ArgMatches,
411 0 : env: &mut local_env::LocalEnv,
412 0 : ) -> anyhow::Result<()> {
413 0 : let pageserver = get_default_pageserver(env);
414 0 : match tenant_match.subcommand() {
415 0 : Some(("list", _)) => {
416 0 : for t in pageserver.tenant_list().await? {
417 0 : println!("{} {:?}", t.id, t.state);
418 0 : }
419 : }
420 0 : Some(("create", create_match)) => {
421 0 : let tenant_conf: HashMap<_, _> = create_match
422 0 : .get_many::<String>("config")
423 0 : .map(|vals: clap::parser::ValuesRef<'_, String>| {
424 0 : vals.flat_map(|c| c.split_once(':')).collect()
425 0 : })
426 0 : .unwrap_or_default();
427 0 :
428 0 : let shard_count: u8 = create_match
429 0 : .get_one::<u8>("shard-count")
430 0 : .cloned()
431 0 : .unwrap_or(0);
432 0 :
433 0 : let shard_stripe_size: Option<u32> =
434 0 : create_match.get_one::<u32>("shard-stripe-size").cloned();
435 :
436 0 : let placement_policy = match create_match.get_one::<String>("placement-policy") {
437 0 : Some(s) if !s.is_empty() => serde_json::from_str::<PlacementPolicy>(s)?,
438 0 : _ => PlacementPolicy::Attached(0),
439 : };
440 :
441 0 : let tenant_conf = PageServerNode::parse_config(tenant_conf)?;
442 :
443 : // If tenant ID was not specified, generate one
444 0 : let tenant_id = parse_tenant_id(create_match)?.unwrap_or_else(TenantId::generate);
445 0 :
446 0 : // We must register the tenant with the storage controller, so
447 0 : // that when the pageserver restarts, it will be re-attached.
448 0 : let storage_controller = StorageController::from_env(env);
449 0 : storage_controller
450 0 : .tenant_create(TenantCreateRequest {
451 0 : // Note that ::unsharded here isn't actually because the tenant is unsharded, its because the
452 0 : // storage controller expecfs a shard-naive tenant_id in this attribute, and the TenantCreateRequest
453 0 : // type is used both in storage controller (for creating tenants) and in pageserver (for creating shards)
454 0 : new_tenant_id: TenantShardId::unsharded(tenant_id),
455 0 : generation: None,
456 0 : shard_parameters: ShardParameters {
457 0 : count: ShardCount::new(shard_count),
458 0 : stripe_size: shard_stripe_size
459 0 : .map(ShardStripeSize)
460 0 : .unwrap_or(ShardParameters::DEFAULT_STRIPE_SIZE),
461 0 : },
462 0 : placement_policy: Some(placement_policy),
463 0 : config: tenant_conf,
464 0 : })
465 0 : .await?;
466 0 : println!("tenant {tenant_id} successfully created on the pageserver");
467 :
468 : // Create an initial timeline for the new tenant
469 0 : let new_timeline_id =
470 0 : parse_timeline_id(create_match)?.unwrap_or(TimelineId::generate());
471 0 : let pg_version = create_match
472 0 : .get_one::<u32>("pg-version")
473 0 : .copied()
474 0 : .context("Failed to parse postgres version from the argument string")?;
475 :
476 : // FIXME: passing None for ancestor_start_lsn is not kosher in a sharded world: we can't have
477 : // different shards picking different start lsns. Maybe we have to teach storage controller
478 : // to let shard 0 branch first and then propagate the chosen LSN to other shards.
479 0 : storage_controller
480 0 : .tenant_timeline_create(
481 0 : tenant_id,
482 0 : TimelineCreateRequest {
483 0 : new_timeline_id,
484 0 : ancestor_timeline_id: None,
485 0 : ancestor_start_lsn: None,
486 0 : existing_initdb_timeline_id: None,
487 0 : pg_version: Some(pg_version),
488 0 : },
489 0 : )
490 0 : .await?;
491 :
492 0 : env.register_branch_mapping(
493 0 : DEFAULT_BRANCH_NAME.to_string(),
494 0 : tenant_id,
495 0 : new_timeline_id,
496 0 : )?;
497 :
498 0 : println!("Created an initial timeline '{new_timeline_id}' for tenant: {tenant_id}",);
499 0 :
500 0 : if create_match.get_flag("set-default") {
501 0 : println!("Setting tenant {tenant_id} as a default one");
502 0 : env.default_tenant_id = Some(tenant_id);
503 0 : }
504 : }
505 0 : Some(("set-default", set_default_match)) => {
506 0 : let tenant_id =
507 0 : parse_tenant_id(set_default_match)?.context("No tenant id specified")?;
508 0 : println!("Setting tenant {tenant_id} as a default one");
509 0 : env.default_tenant_id = Some(tenant_id);
510 : }
511 0 : Some(("config", create_match)) => {
512 0 : let tenant_id = get_tenant_id(create_match, env)?;
513 0 : let tenant_conf: HashMap<_, _> = create_match
514 0 : .get_many::<String>("config")
515 0 : .map(|vals| vals.flat_map(|c| c.split_once(':')).collect())
516 0 : .unwrap_or_default();
517 0 :
518 0 : pageserver
519 0 : .tenant_config(tenant_id, tenant_conf)
520 0 : .await
521 0 : .with_context(|| format!("Tenant config failed for tenant with id {tenant_id}"))?;
522 0 : println!("tenant {tenant_id} successfully configured on the pageserver");
523 : }
524 :
525 0 : Some((sub_name, _)) => bail!("Unexpected tenant subcommand '{}'", sub_name),
526 0 : None => bail!("no tenant subcommand provided"),
527 : }
528 0 : Ok(())
529 0 : }
530 :
531 0 : async fn handle_timeline(timeline_match: &ArgMatches, env: &mut local_env::LocalEnv) -> Result<()> {
532 0 : let pageserver = get_default_pageserver(env);
533 0 :
534 0 : match timeline_match.subcommand() {
535 0 : Some(("list", list_match)) => {
536 : // TODO(sharding): this command shouldn't have to specify a shard ID: we should ask the storage controller
537 : // where shard 0 is attached, and query there.
538 0 : let tenant_shard_id = get_tenant_shard_id(list_match, env)?;
539 0 : let timelines = pageserver.timeline_list(&tenant_shard_id).await?;
540 0 : print_timelines_tree(timelines, env.timeline_name_mappings())?;
541 : }
542 0 : Some(("create", create_match)) => {
543 0 : let tenant_id = get_tenant_id(create_match, env)?;
544 0 : let new_branch_name = create_match
545 0 : .get_one::<String>("branch-name")
546 0 : .ok_or_else(|| anyhow!("No branch name provided"))?;
547 :
548 0 : let pg_version = create_match
549 0 : .get_one::<u32>("pg-version")
550 0 : .copied()
551 0 : .context("Failed to parse postgres version from the argument string")?;
552 :
553 0 : let new_timeline_id_opt = parse_timeline_id(create_match)?;
554 0 : let new_timeline_id = new_timeline_id_opt.unwrap_or(TimelineId::generate());
555 0 :
556 0 : let storage_controller = StorageController::from_env(env);
557 0 : let create_req = TimelineCreateRequest {
558 0 : new_timeline_id,
559 0 : ancestor_timeline_id: None,
560 0 : existing_initdb_timeline_id: None,
561 0 : ancestor_start_lsn: None,
562 0 : pg_version: Some(pg_version),
563 0 : };
564 0 : let timeline_info = storage_controller
565 0 : .tenant_timeline_create(tenant_id, create_req)
566 0 : .await?;
567 :
568 0 : let last_record_lsn = timeline_info.last_record_lsn;
569 0 : env.register_branch_mapping(new_branch_name.to_string(), tenant_id, new_timeline_id)?;
570 :
571 0 : println!(
572 0 : "Created timeline '{}' at Lsn {last_record_lsn} for tenant: {tenant_id}",
573 0 : timeline_info.timeline_id
574 0 : );
575 : }
576 0 : Some(("import", import_match)) => {
577 0 : let tenant_id = get_tenant_id(import_match, env)?;
578 0 : let timeline_id = parse_timeline_id(import_match)?.expect("No timeline id provided");
579 0 : let name = import_match
580 0 : .get_one::<String>("node-name")
581 0 : .ok_or_else(|| anyhow!("No node name provided"))?;
582 0 : let update_catalog = import_match
583 0 : .get_one::<bool>("update-catalog")
584 0 : .cloned()
585 0 : .unwrap_or_default();
586 :
587 : // Parse base inputs
588 0 : let base_tarfile = import_match
589 0 : .get_one::<PathBuf>("base-tarfile")
590 0 : .ok_or_else(|| anyhow!("No base-tarfile provided"))?
591 0 : .to_owned();
592 0 : let base_lsn = Lsn::from_str(
593 0 : import_match
594 0 : .get_one::<String>("base-lsn")
595 0 : .ok_or_else(|| anyhow!("No base-lsn provided"))?,
596 0 : )?;
597 0 : let base = (base_lsn, base_tarfile);
598 0 :
599 0 : // Parse pg_wal inputs
600 0 : let wal_tarfile = import_match.get_one::<PathBuf>("wal-tarfile").cloned();
601 0 : let end_lsn = import_match
602 0 : .get_one::<String>("end-lsn")
603 0 : .map(|s| Lsn::from_str(s).unwrap());
604 0 : // TODO validate both or none are provided
605 0 : let pg_wal = end_lsn.zip(wal_tarfile);
606 :
607 0 : let pg_version = import_match
608 0 : .get_one::<u32>("pg-version")
609 0 : .copied()
610 0 : .context("Failed to parse postgres version from the argument string")?;
611 :
612 0 : let mut cplane = ComputeControlPlane::load(env.clone())?;
613 0 : println!("Importing timeline into pageserver ...");
614 0 : pageserver
615 0 : .timeline_import(tenant_id, timeline_id, base, pg_wal, pg_version)
616 0 : .await?;
617 0 : env.register_branch_mapping(name.to_string(), tenant_id, timeline_id)?;
618 :
619 0 : println!("Creating endpoint for imported timeline ...");
620 0 : cplane.new_endpoint(
621 0 : name,
622 0 : tenant_id,
623 0 : timeline_id,
624 0 : None,
625 0 : None,
626 0 : pg_version,
627 0 : ComputeMode::Primary,
628 0 : !update_catalog,
629 0 : )?;
630 0 : println!("Done");
631 : }
632 0 : Some(("branch", branch_match)) => {
633 0 : let tenant_id = get_tenant_id(branch_match, env)?;
634 0 : let new_branch_name = branch_match
635 0 : .get_one::<String>("branch-name")
636 0 : .ok_or_else(|| anyhow!("No branch name provided"))?;
637 0 : let ancestor_branch_name = branch_match
638 0 : .get_one::<String>("ancestor-branch-name")
639 0 : .map(|s| s.as_str())
640 0 : .unwrap_or(DEFAULT_BRANCH_NAME);
641 0 : let ancestor_timeline_id = env
642 0 : .get_branch_timeline_id(ancestor_branch_name, tenant_id)
643 0 : .ok_or_else(|| {
644 0 : anyhow!("Found no timeline id for branch name '{ancestor_branch_name}'")
645 0 : })?;
646 :
647 0 : let start_lsn = branch_match
648 0 : .get_one::<String>("ancestor-start-lsn")
649 0 : .map(|lsn_str| Lsn::from_str(lsn_str))
650 0 : .transpose()
651 0 : .context("Failed to parse ancestor start Lsn from the request")?;
652 0 : let new_timeline_id = TimelineId::generate();
653 0 : let storage_controller = StorageController::from_env(env);
654 0 : let create_req = TimelineCreateRequest {
655 0 : new_timeline_id,
656 0 : ancestor_timeline_id: Some(ancestor_timeline_id),
657 0 : existing_initdb_timeline_id: None,
658 0 : ancestor_start_lsn: start_lsn,
659 0 : pg_version: None,
660 0 : };
661 0 : let timeline_info = storage_controller
662 0 : .tenant_timeline_create(tenant_id, create_req)
663 0 : .await?;
664 :
665 0 : let last_record_lsn = timeline_info.last_record_lsn;
666 0 :
667 0 : env.register_branch_mapping(new_branch_name.to_string(), tenant_id, new_timeline_id)?;
668 :
669 0 : println!(
670 0 : "Created timeline '{}' at Lsn {last_record_lsn} for tenant: {tenant_id}. Ancestor timeline: '{ancestor_branch_name}'",
671 0 : timeline_info.timeline_id
672 0 : );
673 : }
674 0 : Some((sub_name, _)) => bail!("Unexpected tenant subcommand '{sub_name}'"),
675 0 : None => bail!("no tenant subcommand provided"),
676 : }
677 :
678 0 : Ok(())
679 0 : }
680 :
681 0 : async fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
682 0 : let (sub_name, sub_args) = match ep_match.subcommand() {
683 0 : Some(ep_subcommand_data) => ep_subcommand_data,
684 0 : None => bail!("no endpoint subcommand provided"),
685 : };
686 0 : let mut cplane = ComputeControlPlane::load(env.clone())?;
687 :
688 0 : match sub_name {
689 0 : "list" => {
690 : // TODO(sharding): this command shouldn't have to specify a shard ID: we should ask the storage controller
691 : // where shard 0 is attached, and query there.
692 0 : let tenant_shard_id = get_tenant_shard_id(sub_args, env)?;
693 0 : let timeline_infos = get_timeline_infos(env, &tenant_shard_id)
694 0 : .await
695 0 : .unwrap_or_else(|e| {
696 0 : eprintln!("Failed to load timeline info: {}", e);
697 0 : HashMap::new()
698 0 : });
699 0 :
700 0 : let timeline_name_mappings = env.timeline_name_mappings();
701 0 :
702 0 : let mut table = comfy_table::Table::new();
703 0 :
704 0 : table.load_preset(comfy_table::presets::NOTHING);
705 0 :
706 0 : table.set_header([
707 0 : "ENDPOINT",
708 0 : "ADDRESS",
709 0 : "TIMELINE",
710 0 : "BRANCH NAME",
711 0 : "LSN",
712 0 : "STATUS",
713 0 : ]);
714 :
715 0 : for (endpoint_id, endpoint) in cplane
716 0 : .endpoints
717 0 : .iter()
718 0 : .filter(|(_, endpoint)| endpoint.tenant_id == tenant_shard_id.tenant_id)
719 0 : {
720 0 : let lsn_str = match endpoint.mode {
721 0 : ComputeMode::Static(lsn) => {
722 0 : // -> read-only endpoint
723 0 : // Use the node's LSN.
724 0 : lsn.to_string()
725 : }
726 : _ => {
727 : // -> primary endpoint or hot replica
728 : // Use the LSN at the end of the timeline.
729 0 : timeline_infos
730 0 : .get(&endpoint.timeline_id)
731 0 : .map(|bi| bi.last_record_lsn.to_string())
732 0 : .unwrap_or_else(|| "?".to_string())
733 : }
734 : };
735 :
736 0 : let branch_name = timeline_name_mappings
737 0 : .get(&TenantTimelineId::new(
738 0 : tenant_shard_id.tenant_id,
739 0 : endpoint.timeline_id,
740 0 : ))
741 0 : .map(|name| name.as_str())
742 0 : .unwrap_or("?");
743 0 :
744 0 : table.add_row([
745 0 : endpoint_id.as_str(),
746 0 : &endpoint.pg_address.to_string(),
747 0 : &endpoint.timeline_id.to_string(),
748 0 : branch_name,
749 0 : lsn_str.as_str(),
750 0 : &format!("{}", endpoint.status()),
751 0 : ]);
752 : }
753 :
754 0 : println!("{table}");
755 : }
756 0 : "create" => {
757 0 : let tenant_id = get_tenant_id(sub_args, env)?;
758 0 : let branch_name = sub_args
759 0 : .get_one::<String>("branch-name")
760 0 : .map(|s| s.as_str())
761 0 : .unwrap_or(DEFAULT_BRANCH_NAME);
762 0 : let endpoint_id = sub_args
763 0 : .get_one::<String>("endpoint_id")
764 0 : .map(String::to_string)
765 0 : .unwrap_or_else(|| format!("ep-{branch_name}"));
766 0 : let update_catalog = sub_args
767 0 : .get_one::<bool>("update-catalog")
768 0 : .cloned()
769 0 : .unwrap_or_default();
770 :
771 0 : let lsn = sub_args
772 0 : .get_one::<String>("lsn")
773 0 : .map(|lsn_str| Lsn::from_str(lsn_str))
774 0 : .transpose()
775 0 : .context("Failed to parse Lsn from the request")?;
776 0 : let timeline_id = env
777 0 : .get_branch_timeline_id(branch_name, tenant_id)
778 0 : .ok_or_else(|| anyhow!("Found no timeline id for branch name '{branch_name}'"))?;
779 :
780 0 : let pg_port: Option<u16> = sub_args.get_one::<u16>("pg-port").copied();
781 0 : let http_port: Option<u16> = sub_args.get_one::<u16>("http-port").copied();
782 0 : let pg_version = sub_args
783 0 : .get_one::<u32>("pg-version")
784 0 : .copied()
785 0 : .context("Failed to parse postgres version from the argument string")?;
786 :
787 0 : let hot_standby = sub_args
788 0 : .get_one::<bool>("hot-standby")
789 0 : .copied()
790 0 : .unwrap_or(false);
791 :
792 0 : let mode = match (lsn, hot_standby) {
793 0 : (Some(lsn), false) => ComputeMode::Static(lsn),
794 0 : (None, true) => ComputeMode::Replica,
795 0 : (None, false) => ComputeMode::Primary,
796 0 : (Some(_), true) => anyhow::bail!("cannot specify both lsn and hot-standby"),
797 : };
798 :
799 0 : match (mode, hot_standby) {
800 : (ComputeMode::Static(_), true) => {
801 0 : bail!("Cannot start a node in hot standby mode when it is already configured as a static replica")
802 : }
803 : (ComputeMode::Primary, true) => {
804 0 : bail!("Cannot start a node as a hot standby replica, it is already configured as primary node")
805 : }
806 0 : _ => {}
807 0 : }
808 0 :
809 0 : cplane.check_conflicting_endpoints(mode, tenant_id, timeline_id)?;
810 :
811 0 : cplane.new_endpoint(
812 0 : &endpoint_id,
813 0 : tenant_id,
814 0 : timeline_id,
815 0 : pg_port,
816 0 : http_port,
817 0 : pg_version,
818 0 : mode,
819 0 : !update_catalog,
820 0 : )?;
821 : }
822 0 : "start" => {
823 0 : let endpoint_id = sub_args
824 0 : .get_one::<String>("endpoint_id")
825 0 : .ok_or_else(|| anyhow!("No endpoint ID was provided to start"))?;
826 :
827 0 : let pageserver_id =
828 0 : if let Some(id_str) = sub_args.get_one::<String>("endpoint-pageserver-id") {
829 : Some(NodeId(
830 0 : id_str.parse().context("while parsing pageserver id")?,
831 : ))
832 : } else {
833 0 : None
834 : };
835 :
836 0 : let remote_ext_config = sub_args.get_one::<String>("remote-ext-config");
837 :
838 : // If --safekeepers argument is given, use only the listed safekeeper nodes.
839 0 : let safekeepers =
840 0 : if let Some(safekeepers_str) = sub_args.get_one::<String>("safekeepers") {
841 0 : let mut safekeepers: Vec<NodeId> = Vec::new();
842 0 : for sk_id in safekeepers_str.split(',').map(str::trim) {
843 0 : let sk_id = NodeId(u64::from_str(sk_id).map_err(|_| {
844 0 : anyhow!("invalid node ID \"{sk_id}\" in --safekeepers list")
845 0 : })?);
846 0 : safekeepers.push(sk_id);
847 : }
848 0 : safekeepers
849 : } else {
850 0 : env.safekeepers.iter().map(|sk| sk.id).collect()
851 : };
852 :
853 0 : let endpoint = cplane
854 0 : .endpoints
855 0 : .get(endpoint_id.as_str())
856 0 : .ok_or_else(|| anyhow::anyhow!("endpoint {endpoint_id} not found"))?;
857 :
858 0 : let create_test_user = sub_args
859 0 : .get_one::<bool>("create-test-user")
860 0 : .cloned()
861 0 : .unwrap_or_default();
862 0 :
863 0 : cplane.check_conflicting_endpoints(
864 0 : endpoint.mode,
865 0 : endpoint.tenant_id,
866 0 : endpoint.timeline_id,
867 0 : )?;
868 :
869 0 : let (pageservers, stripe_size) = if let Some(pageserver_id) = pageserver_id {
870 0 : let conf = env.get_pageserver_conf(pageserver_id).unwrap();
871 0 : let parsed = parse_host_port(&conf.listen_pg_addr).expect("Bad config");
872 0 : (
873 0 : vec![(parsed.0, parsed.1.unwrap_or(5432))],
874 0 : // If caller is telling us what pageserver to use, this is not a tenant which is
875 0 : // full managed by storage controller, therefore not sharded.
876 0 : ShardParameters::DEFAULT_STRIPE_SIZE,
877 0 : )
878 : } else {
879 : // Look up the currently attached location of the tenant, and its striping metadata,
880 : // to pass these on to postgres.
881 0 : let storage_controller = StorageController::from_env(env);
882 0 : let locate_result = storage_controller.tenant_locate(endpoint.tenant_id).await?;
883 0 : let pageservers = locate_result
884 0 : .shards
885 0 : .into_iter()
886 0 : .map(|shard| {
887 0 : (
888 0 : Host::parse(&shard.listen_pg_addr)
889 0 : .expect("Storage controller reported bad hostname"),
890 0 : shard.listen_pg_port,
891 0 : )
892 0 : })
893 0 : .collect::<Vec<_>>();
894 0 : let stripe_size = locate_result.shard_params.stripe_size;
895 0 :
896 0 : (pageservers, stripe_size)
897 : };
898 0 : assert!(!pageservers.is_empty());
899 :
900 0 : let ps_conf = env.get_pageserver_conf(DEFAULT_PAGESERVER_ID)?;
901 0 : let auth_token = if matches!(ps_conf.pg_auth_type, AuthType::NeonJWT) {
902 0 : let claims = Claims::new(Some(endpoint.tenant_id), Scope::Tenant);
903 0 :
904 0 : Some(env.generate_auth_token(&claims)?)
905 : } else {
906 0 : None
907 : };
908 :
909 0 : println!("Starting existing endpoint {endpoint_id}...");
910 0 : endpoint
911 0 : .start(
912 0 : &auth_token,
913 0 : safekeepers,
914 0 : pageservers,
915 0 : remote_ext_config,
916 0 : stripe_size.0 as usize,
917 0 : create_test_user,
918 0 : )
919 0 : .await?;
920 : }
921 0 : "reconfigure" => {
922 0 : let endpoint_id = sub_args
923 0 : .get_one::<String>("endpoint_id")
924 0 : .ok_or_else(|| anyhow!("No endpoint ID provided to reconfigure"))?;
925 0 : let endpoint = cplane
926 0 : .endpoints
927 0 : .get(endpoint_id.as_str())
928 0 : .with_context(|| format!("postgres endpoint {endpoint_id} is not found"))?;
929 0 : let pageservers =
930 0 : if let Some(id_str) = sub_args.get_one::<String>("endpoint-pageserver-id") {
931 0 : let ps_id = NodeId(id_str.parse().context("while parsing pageserver id")?);
932 0 : let pageserver = PageServerNode::from_env(env, env.get_pageserver_conf(ps_id)?);
933 0 : vec![(
934 0 : pageserver.pg_connection_config.host().clone(),
935 0 : pageserver.pg_connection_config.port(),
936 0 : )]
937 : } else {
938 0 : let storage_controller = StorageController::from_env(env);
939 0 : storage_controller
940 0 : .tenant_locate(endpoint.tenant_id)
941 0 : .await?
942 : .shards
943 0 : .into_iter()
944 0 : .map(|shard| {
945 0 : (
946 0 : Host::parse(&shard.listen_pg_addr)
947 0 : .expect("Storage controller reported malformed host"),
948 0 : shard.listen_pg_port,
949 0 : )
950 0 : })
951 0 : .collect::<Vec<_>>()
952 : };
953 0 : endpoint.reconfigure(pageservers, None).await?;
954 : }
955 0 : "stop" => {
956 0 : let endpoint_id = sub_args
957 0 : .get_one::<String>("endpoint_id")
958 0 : .ok_or_else(|| anyhow!("No endpoint ID was provided to stop"))?;
959 0 : let destroy = sub_args.get_flag("destroy");
960 0 : let mode = sub_args.get_one::<String>("mode").expect("has a default");
961 :
962 0 : let endpoint = cplane
963 0 : .endpoints
964 0 : .get(endpoint_id.as_str())
965 0 : .with_context(|| format!("postgres endpoint {endpoint_id} is not found"))?;
966 0 : endpoint.stop(mode, destroy)?;
967 : }
968 :
969 0 : _ => bail!("Unexpected endpoint subcommand '{sub_name}'"),
970 : }
971 :
972 0 : Ok(())
973 0 : }
974 :
975 0 : fn handle_mappings(sub_match: &ArgMatches, env: &mut local_env::LocalEnv) -> Result<()> {
976 0 : let (sub_name, sub_args) = match sub_match.subcommand() {
977 0 : Some(ep_subcommand_data) => ep_subcommand_data,
978 0 : None => bail!("no mappings subcommand provided"),
979 : };
980 :
981 0 : match sub_name {
982 0 : "map" => {
983 0 : let branch_name = sub_args
984 0 : .get_one::<String>("branch-name")
985 0 : .expect("branch-name argument missing");
986 0 :
987 0 : let tenant_id = sub_args
988 0 : .get_one::<String>("tenant-id")
989 0 : .map(|x| TenantId::from_str(x))
990 0 : .expect("tenant-id argument missing")
991 0 : .expect("malformed tenant-id arg");
992 0 :
993 0 : let timeline_id = sub_args
994 0 : .get_one::<String>("timeline-id")
995 0 : .map(|x| TimelineId::from_str(x))
996 0 : .expect("timeline-id argument missing")
997 0 : .expect("malformed timeline-id arg");
998 0 :
999 0 : env.register_branch_mapping(branch_name.to_owned(), tenant_id, timeline_id)?;
1000 :
1001 0 : Ok(())
1002 : }
1003 0 : other => unimplemented!("mappings subcommand {other}"),
1004 : }
1005 0 : }
1006 :
1007 0 : fn get_pageserver(env: &local_env::LocalEnv, args: &ArgMatches) -> Result<PageServerNode> {
1008 0 : let node_id = if let Some(id_str) = args.get_one::<String>("pageserver-id") {
1009 0 : NodeId(id_str.parse().context("while parsing pageserver id")?)
1010 : } else {
1011 0 : DEFAULT_PAGESERVER_ID
1012 : };
1013 :
1014 : Ok(PageServerNode::from_env(
1015 0 : env,
1016 0 : env.get_pageserver_conf(node_id)?,
1017 : ))
1018 0 : }
1019 :
1020 0 : async fn handle_pageserver(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
1021 0 : match sub_match.subcommand() {
1022 0 : Some(("start", subcommand_args)) => {
1023 0 : if let Err(e) = get_pageserver(env, subcommand_args)?
1024 0 : .start(&pageserver_config_overrides(subcommand_args))
1025 0 : .await
1026 : {
1027 0 : eprintln!("pageserver start failed: {e}");
1028 0 : exit(1);
1029 0 : }
1030 : }
1031 :
1032 0 : Some(("stop", subcommand_args)) => {
1033 0 : let immediate = subcommand_args
1034 0 : .get_one::<String>("stop-mode")
1035 0 : .map(|s| s.as_str())
1036 0 : == Some("immediate");
1037 :
1038 0 : if let Err(e) = get_pageserver(env, subcommand_args)?.stop(immediate) {
1039 0 : eprintln!("pageserver stop failed: {}", e);
1040 0 : exit(1);
1041 0 : }
1042 : }
1043 :
1044 0 : Some(("restart", subcommand_args)) => {
1045 0 : let pageserver = get_pageserver(env, subcommand_args)?;
1046 : //TODO what shutdown strategy should we use here?
1047 0 : if let Err(e) = pageserver.stop(false) {
1048 0 : eprintln!("pageserver stop failed: {}", e);
1049 0 : exit(1);
1050 0 : }
1051 :
1052 0 : if let Err(e) = pageserver
1053 0 : .start(&pageserver_config_overrides(subcommand_args))
1054 0 : .await
1055 : {
1056 0 : eprintln!("pageserver start failed: {e}");
1057 0 : exit(1);
1058 0 : }
1059 : }
1060 :
1061 0 : Some(("status", subcommand_args)) => {
1062 0 : match get_pageserver(env, subcommand_args)?.check_status().await {
1063 0 : Ok(_) => println!("Page server is up and running"),
1064 0 : Err(err) => {
1065 0 : eprintln!("Page server is not available: {}", err);
1066 0 : exit(1);
1067 : }
1068 : }
1069 : }
1070 :
1071 0 : Some((sub_name, _)) => bail!("Unexpected pageserver subcommand '{}'", sub_name),
1072 0 : None => bail!("no pageserver subcommand provided"),
1073 : }
1074 0 : Ok(())
1075 0 : }
1076 :
1077 0 : async fn handle_storage_controller(
1078 0 : sub_match: &ArgMatches,
1079 0 : env: &local_env::LocalEnv,
1080 0 : ) -> Result<()> {
1081 0 : let svc = StorageController::from_env(env);
1082 0 : match sub_match.subcommand() {
1083 0 : Some(("start", _start_match)) => {
1084 0 : if let Err(e) = svc.start().await {
1085 0 : eprintln!("start failed: {e}");
1086 0 : exit(1);
1087 0 : }
1088 : }
1089 :
1090 0 : Some(("stop", stop_match)) => {
1091 0 : let immediate = stop_match
1092 0 : .get_one::<String>("stop-mode")
1093 0 : .map(|s| s.as_str())
1094 0 : == Some("immediate");
1095 :
1096 0 : if let Err(e) = svc.stop(immediate).await {
1097 0 : eprintln!("stop failed: {}", e);
1098 0 : exit(1);
1099 0 : }
1100 : }
1101 0 : Some((sub_name, _)) => bail!("Unexpected storage_controller subcommand '{}'", sub_name),
1102 0 : None => bail!("no storage_controller subcommand provided"),
1103 : }
1104 0 : Ok(())
1105 0 : }
1106 :
1107 0 : fn get_safekeeper(env: &local_env::LocalEnv, id: NodeId) -> Result<SafekeeperNode> {
1108 0 : if let Some(node) = env.safekeepers.iter().find(|node| node.id == id) {
1109 0 : Ok(SafekeeperNode::from_env(env, node))
1110 : } else {
1111 0 : bail!("could not find safekeeper {id}")
1112 : }
1113 0 : }
1114 :
1115 : // Get list of options to append to safekeeper command invocation.
1116 0 : fn safekeeper_extra_opts(init_match: &ArgMatches) -> Vec<String> {
1117 0 : init_match
1118 0 : .get_many::<String>("safekeeper-extra-opt")
1119 0 : .into_iter()
1120 0 : .flatten()
1121 0 : .map(|s| s.to_owned())
1122 0 : .collect()
1123 0 : }
1124 :
1125 0 : async fn handle_safekeeper(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
1126 0 : let (sub_name, sub_args) = match sub_match.subcommand() {
1127 0 : Some(safekeeper_command_data) => safekeeper_command_data,
1128 0 : None => bail!("no safekeeper subcommand provided"),
1129 : };
1130 :
1131 : // All the commands take an optional safekeeper name argument
1132 0 : let sk_id = if let Some(id_str) = sub_args.get_one::<String>("id") {
1133 0 : NodeId(id_str.parse().context("while parsing safekeeper id")?)
1134 : } else {
1135 0 : DEFAULT_SAFEKEEPER_ID
1136 : };
1137 0 : let safekeeper = get_safekeeper(env, sk_id)?;
1138 :
1139 0 : match sub_name {
1140 0 : "start" => {
1141 0 : let extra_opts = safekeeper_extra_opts(sub_args);
1142 :
1143 0 : if let Err(e) = safekeeper.start(extra_opts).await {
1144 0 : eprintln!("safekeeper start failed: {}", e);
1145 0 : exit(1);
1146 0 : }
1147 : }
1148 :
1149 0 : "stop" => {
1150 0 : let immediate =
1151 0 : sub_args.get_one::<String>("stop-mode").map(|s| s.as_str()) == Some("immediate");
1152 :
1153 0 : if let Err(e) = safekeeper.stop(immediate) {
1154 0 : eprintln!("safekeeper stop failed: {}", e);
1155 0 : exit(1);
1156 0 : }
1157 : }
1158 :
1159 0 : "restart" => {
1160 0 : let immediate =
1161 0 : sub_args.get_one::<String>("stop-mode").map(|s| s.as_str()) == Some("immediate");
1162 :
1163 0 : if let Err(e) = safekeeper.stop(immediate) {
1164 0 : eprintln!("safekeeper stop failed: {}", e);
1165 0 : exit(1);
1166 0 : }
1167 0 :
1168 0 : let extra_opts = safekeeper_extra_opts(sub_args);
1169 0 : if let Err(e) = safekeeper.start(extra_opts).await {
1170 0 : eprintln!("safekeeper start failed: {}", e);
1171 0 : exit(1);
1172 0 : }
1173 : }
1174 :
1175 : _ => {
1176 0 : bail!("Unexpected safekeeper subcommand '{}'", sub_name)
1177 : }
1178 : }
1179 0 : Ok(())
1180 0 : }
1181 :
1182 0 : async fn handle_start_all(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> anyhow::Result<()> {
1183 0 : // Endpoints are not started automatically
1184 0 :
1185 0 : broker::start_broker_process(env).await?;
1186 :
1187 : // Only start the storage controller if the pageserver is configured to need it
1188 0 : if env.control_plane_api.is_some() {
1189 0 : let storage_controller = StorageController::from_env(env);
1190 0 : if let Err(e) = storage_controller.start().await {
1191 0 : eprintln!("storage_controller start failed: {:#}", e);
1192 0 : try_stop_all(env, true).await;
1193 0 : exit(1);
1194 0 : }
1195 0 : }
1196 :
1197 0 : for ps_conf in &env.pageservers {
1198 0 : let pageserver = PageServerNode::from_env(env, ps_conf);
1199 0 : if let Err(e) = pageserver
1200 0 : .start(&pageserver_config_overrides(sub_match))
1201 0 : .await
1202 : {
1203 0 : eprintln!("pageserver {} start failed: {:#}", ps_conf.id, e);
1204 0 : try_stop_all(env, true).await;
1205 0 : exit(1);
1206 0 : }
1207 : }
1208 :
1209 0 : for node in env.safekeepers.iter() {
1210 0 : let safekeeper = SafekeeperNode::from_env(env, node);
1211 0 : if let Err(e) = safekeeper.start(vec![]).await {
1212 0 : eprintln!("safekeeper {} start failed: {:#}", safekeeper.id, e);
1213 0 : try_stop_all(env, false).await;
1214 0 : exit(1);
1215 0 : }
1216 : }
1217 0 : Ok(())
1218 0 : }
1219 :
1220 0 : async fn handle_stop_all(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
1221 0 : let immediate =
1222 0 : sub_match.get_one::<String>("stop-mode").map(|s| s.as_str()) == Some("immediate");
1223 0 :
1224 0 : try_stop_all(env, immediate).await;
1225 :
1226 0 : Ok(())
1227 0 : }
1228 :
1229 0 : async fn try_stop_all(env: &local_env::LocalEnv, immediate: bool) {
1230 0 : // Stop all endpoints
1231 0 : match ComputeControlPlane::load(env.clone()) {
1232 0 : Ok(cplane) => {
1233 0 : for (_k, node) in cplane.endpoints {
1234 0 : if let Err(e) = node.stop(if immediate { "immediate" } else { "fast" }, false) {
1235 0 : eprintln!("postgres stop failed: {e:#}");
1236 0 : }
1237 : }
1238 : }
1239 0 : Err(e) => {
1240 0 : eprintln!("postgres stop failed, could not restore control plane data from env: {e:#}")
1241 : }
1242 : }
1243 :
1244 0 : for ps_conf in &env.pageservers {
1245 0 : let pageserver = PageServerNode::from_env(env, ps_conf);
1246 0 : if let Err(e) = pageserver.stop(immediate) {
1247 0 : eprintln!("pageserver {} stop failed: {:#}", ps_conf.id, e);
1248 0 : }
1249 : }
1250 :
1251 0 : for node in env.safekeepers.iter() {
1252 0 : let safekeeper = SafekeeperNode::from_env(env, node);
1253 0 : if let Err(e) = safekeeper.stop(immediate) {
1254 0 : eprintln!("safekeeper {} stop failed: {:#}", safekeeper.id, e);
1255 0 : }
1256 : }
1257 :
1258 0 : if let Err(e) = broker::stop_broker_process(env) {
1259 0 : eprintln!("neon broker stop failed: {e:#}");
1260 0 : }
1261 :
1262 0 : if env.control_plane_api.is_some() {
1263 0 : let storage_controller = StorageController::from_env(env);
1264 0 : if let Err(e) = storage_controller.stop(immediate).await {
1265 0 : eprintln!("storage controller stop failed: {e:#}");
1266 0 : }
1267 0 : }
1268 0 : }
1269 :
1270 2 : fn cli() -> Command {
1271 2 : let branch_name_arg = Arg::new("branch-name")
1272 2 : .long("branch-name")
1273 2 : .help("Name of the branch to be created or used as an alias for other services")
1274 2 : .required(false);
1275 2 :
1276 2 : let endpoint_id_arg = Arg::new("endpoint_id")
1277 2 : .help("Postgres endpoint id")
1278 2 : .required(false);
1279 2 :
1280 2 : let safekeeper_id_arg = Arg::new("id").help("safekeeper id").required(false);
1281 2 :
1282 2 : // --id, when using a pageserver command
1283 2 : let pageserver_id_arg = Arg::new("pageserver-id")
1284 2 : .long("id")
1285 2 : .global(true)
1286 2 : .help("pageserver id")
1287 2 : .required(false);
1288 2 : // --pageserver-id when using a non-pageserver command
1289 2 : let endpoint_pageserver_id_arg = Arg::new("endpoint-pageserver-id")
1290 2 : .long("pageserver-id")
1291 2 : .required(false);
1292 2 :
1293 2 : let safekeeper_extra_opt_arg = Arg::new("safekeeper-extra-opt")
1294 2 : .short('e')
1295 2 : .long("safekeeper-extra-opt")
1296 2 : .num_args(1)
1297 2 : .action(ArgAction::Append)
1298 2 : .help("Additional safekeeper invocation options, e.g. -e=--http-auth-public-key-path=foo")
1299 2 : .required(false);
1300 2 :
1301 2 : let tenant_id_arg = Arg::new("tenant-id")
1302 2 : .long("tenant-id")
1303 2 : .help("Tenant id. Represented as a hexadecimal string 32 symbols length")
1304 2 : .required(false);
1305 2 :
1306 2 : let timeline_id_arg = Arg::new("timeline-id")
1307 2 : .long("timeline-id")
1308 2 : .help("Timeline id. Represented as a hexadecimal string 32 symbols length")
1309 2 : .required(false);
1310 2 :
1311 2 : let pg_version_arg = Arg::new("pg-version")
1312 2 : .long("pg-version")
1313 2 : .help("Postgres version to use for the initial tenant")
1314 2 : .required(false)
1315 2 : .value_parser(value_parser!(u32))
1316 2 : .default_value(DEFAULT_PG_VERSION);
1317 2 :
1318 2 : let pg_port_arg = Arg::new("pg-port")
1319 2 : .long("pg-port")
1320 2 : .required(false)
1321 2 : .value_parser(value_parser!(u16))
1322 2 : .value_name("pg-port");
1323 2 :
1324 2 : let http_port_arg = Arg::new("http-port")
1325 2 : .long("http-port")
1326 2 : .required(false)
1327 2 : .value_parser(value_parser!(u16))
1328 2 : .value_name("http-port");
1329 2 :
1330 2 : let safekeepers_arg = Arg::new("safekeepers")
1331 2 : .long("safekeepers")
1332 2 : .required(false)
1333 2 : .value_name("safekeepers");
1334 2 :
1335 2 : let stop_mode_arg = Arg::new("stop-mode")
1336 2 : .short('m')
1337 2 : .value_parser(["fast", "immediate"])
1338 2 : .default_value("fast")
1339 2 : .help("If 'immediate', don't flush repository data at shutdown")
1340 2 : .required(false)
1341 2 : .value_name("stop-mode");
1342 2 :
1343 2 : let pageserver_config_args = Arg::new("pageserver-config-override")
1344 2 : .long("pageserver-config-override")
1345 2 : .num_args(1)
1346 2 : .action(ArgAction::Append)
1347 2 : .help("Additional pageserver's configuration options or overrides, refer to pageserver's 'config-override' CLI parameter docs for more")
1348 2 : .required(false);
1349 2 :
1350 2 : let remote_ext_config_args = Arg::new("remote-ext-config")
1351 2 : .long("remote-ext-config")
1352 2 : .num_args(1)
1353 2 : .help("Configure the remote extensions storage proxy gateway to request for extensions.")
1354 2 : .required(false);
1355 2 :
1356 2 : let lsn_arg = Arg::new("lsn")
1357 2 : .long("lsn")
1358 2 : .help("Specify Lsn on the timeline to start from. By default, end of the timeline would be used.")
1359 2 : .required(false);
1360 2 :
1361 2 : let hot_standby_arg = Arg::new("hot-standby")
1362 2 : .value_parser(value_parser!(bool))
1363 2 : .long("hot-standby")
1364 2 : .help("If set, the node will be a hot replica on the specified timeline")
1365 2 : .required(false);
1366 2 :
1367 2 : let force_arg = Arg::new("force")
1368 2 : .value_parser(value_parser!(InitForceMode))
1369 2 : .long("force")
1370 2 : .default_value(
1371 2 : InitForceMode::MustNotExist
1372 2 : .to_possible_value()
1373 2 : .unwrap()
1374 2 : .get_name()
1375 2 : .to_owned(),
1376 2 : )
1377 2 : .help("Force initialization even if the repository is not empty")
1378 2 : .required(false);
1379 2 :
1380 2 : let num_pageservers_arg = Arg::new("num-pageservers")
1381 2 : .value_parser(value_parser!(u16))
1382 2 : .long("num-pageservers")
1383 2 : .help("How many pageservers to create (default 1)")
1384 2 : .required(false)
1385 2 : .default_value("1");
1386 2 :
1387 2 : let update_catalog = Arg::new("update-catalog")
1388 2 : .value_parser(value_parser!(bool))
1389 2 : .long("update-catalog")
1390 2 : .help("If set, will set up the catalog for neon_superuser")
1391 2 : .required(false);
1392 2 :
1393 2 : let create_test_user = Arg::new("create-test-user")
1394 2 : .value_parser(value_parser!(bool))
1395 2 : .long("create-test-user")
1396 2 : .help("If set, will create test user `user` and `neondb` database. Requires `update-catalog = true`")
1397 2 : .required(false);
1398 2 :
1399 2 : Command::new("Neon CLI")
1400 2 : .arg_required_else_help(true)
1401 2 : .version(GIT_VERSION)
1402 2 : .subcommand(
1403 2 : Command::new("init")
1404 2 : .about("Initialize a new Neon repository, preparing configs for services to start with")
1405 2 : .arg(pageserver_config_args.clone())
1406 2 : .arg(num_pageservers_arg.clone())
1407 2 : .arg(
1408 2 : Arg::new("config")
1409 2 : .long("config")
1410 2 : .required(false)
1411 2 : .value_parser(value_parser!(PathBuf))
1412 2 : .value_name("config"),
1413 2 : )
1414 2 : .arg(pg_version_arg.clone())
1415 2 : .arg(force_arg)
1416 2 : )
1417 2 : .subcommand(
1418 2 : Command::new("timeline")
1419 2 : .about("Manage timelines")
1420 2 : .arg_required_else_help(true)
1421 2 : .subcommand(Command::new("list")
1422 2 : .about("List all timelines, available to this pageserver")
1423 2 : .arg(tenant_id_arg.clone()))
1424 2 : .subcommand(Command::new("branch")
1425 2 : .about("Create a new timeline, using another timeline as a base, copying its data")
1426 2 : .arg(tenant_id_arg.clone())
1427 2 : .arg(branch_name_arg.clone())
1428 2 : .arg(Arg::new("ancestor-branch-name").long("ancestor-branch-name")
1429 2 : .help("Use last Lsn of another timeline (and its data) as base when creating the new timeline. The timeline gets resolved by its branch name.").required(false))
1430 2 : .arg(Arg::new("ancestor-start-lsn").long("ancestor-start-lsn")
1431 2 : .help("When using another timeline as base, use a specific Lsn in it instead of the latest one").required(false)))
1432 2 : .subcommand(Command::new("create")
1433 2 : .about("Create a new blank timeline")
1434 2 : .arg(tenant_id_arg.clone())
1435 2 : .arg(timeline_id_arg.clone())
1436 2 : .arg(branch_name_arg.clone())
1437 2 : .arg(pg_version_arg.clone())
1438 2 : )
1439 2 : .subcommand(Command::new("import")
1440 2 : .about("Import timeline from basebackup directory")
1441 2 : .arg(tenant_id_arg.clone())
1442 2 : .arg(timeline_id_arg.clone())
1443 2 : .arg(Arg::new("node-name").long("node-name")
1444 2 : .help("Name to assign to the imported timeline"))
1445 2 : .arg(Arg::new("base-tarfile")
1446 2 : .long("base-tarfile")
1447 2 : .value_parser(value_parser!(PathBuf))
1448 2 : .help("Basebackup tarfile to import")
1449 2 : )
1450 2 : .arg(Arg::new("base-lsn").long("base-lsn")
1451 2 : .help("Lsn the basebackup starts at"))
1452 2 : .arg(Arg::new("wal-tarfile")
1453 2 : .long("wal-tarfile")
1454 2 : .value_parser(value_parser!(PathBuf))
1455 2 : .help("Wal to add after base")
1456 2 : )
1457 2 : .arg(Arg::new("end-lsn").long("end-lsn")
1458 2 : .help("Lsn the basebackup ends at"))
1459 2 : .arg(pg_version_arg.clone())
1460 2 : .arg(update_catalog.clone())
1461 2 : )
1462 2 : ).subcommand(
1463 2 : Command::new("tenant")
1464 2 : .arg_required_else_help(true)
1465 2 : .about("Manage tenants")
1466 2 : .subcommand(Command::new("list"))
1467 2 : .subcommand(Command::new("create")
1468 2 : .arg(tenant_id_arg.clone())
1469 2 : .arg(timeline_id_arg.clone().help("Use a specific timeline id when creating a tenant and its initial timeline"))
1470 2 : .arg(Arg::new("config").short('c').num_args(1).action(ArgAction::Append).required(false))
1471 2 : .arg(pg_version_arg.clone())
1472 2 : .arg(Arg::new("set-default").long("set-default").action(ArgAction::SetTrue).required(false)
1473 2 : .help("Use this tenant in future CLI commands where tenant_id is needed, but not specified"))
1474 2 : .arg(Arg::new("shard-count").value_parser(value_parser!(u8)).long("shard-count").action(ArgAction::Set).help("Number of shards in the new tenant (default 1)"))
1475 2 : .arg(Arg::new("shard-stripe-size").value_parser(value_parser!(u32)).long("shard-stripe-size").action(ArgAction::Set).help("Sharding stripe size in pages"))
1476 2 : .arg(Arg::new("placement-policy").value_parser(value_parser!(String)).long("placement-policy").action(ArgAction::Set).help("Placement policy shards in this tenant"))
1477 2 : )
1478 2 : .subcommand(Command::new("set-default").arg(tenant_id_arg.clone().required(true))
1479 2 : .about("Set a particular tenant as default in future CLI commands where tenant_id is needed, but not specified"))
1480 2 : .subcommand(Command::new("config")
1481 2 : .arg(tenant_id_arg.clone())
1482 2 : .arg(Arg::new("config").short('c').num_args(1).action(ArgAction::Append).required(false)))
1483 2 : )
1484 2 : .subcommand(
1485 2 : Command::new("pageserver")
1486 2 : .arg_required_else_help(true)
1487 2 : .about("Manage pageserver")
1488 2 : .arg(pageserver_id_arg)
1489 2 : .subcommand(Command::new("status"))
1490 2 : .subcommand(Command::new("start")
1491 2 : .about("Start local pageserver")
1492 2 : .arg(pageserver_config_args.clone())
1493 2 : )
1494 2 : .subcommand(Command::new("stop")
1495 2 : .about("Stop local pageserver")
1496 2 : .arg(stop_mode_arg.clone())
1497 2 : )
1498 2 : .subcommand(Command::new("restart")
1499 2 : .about("Restart local pageserver")
1500 2 : .arg(pageserver_config_args.clone())
1501 2 : )
1502 2 : )
1503 2 : .subcommand(
1504 2 : Command::new("storage_controller")
1505 2 : .arg_required_else_help(true)
1506 2 : .about("Manage storage_controller")
1507 2 : .subcommand(Command::new("start").about("Start local pageserver").arg(pageserver_config_args.clone()))
1508 2 : .subcommand(Command::new("stop").about("Stop local pageserver")
1509 2 : .arg(stop_mode_arg.clone()))
1510 2 : )
1511 2 : .subcommand(
1512 2 : Command::new("safekeeper")
1513 2 : .arg_required_else_help(true)
1514 2 : .about("Manage safekeepers")
1515 2 : .subcommand(Command::new("start")
1516 2 : .about("Start local safekeeper")
1517 2 : .arg(safekeeper_id_arg.clone())
1518 2 : .arg(safekeeper_extra_opt_arg.clone())
1519 2 : )
1520 2 : .subcommand(Command::new("stop")
1521 2 : .about("Stop local safekeeper")
1522 2 : .arg(safekeeper_id_arg.clone())
1523 2 : .arg(stop_mode_arg.clone())
1524 2 : )
1525 2 : .subcommand(Command::new("restart")
1526 2 : .about("Restart local safekeeper")
1527 2 : .arg(safekeeper_id_arg)
1528 2 : .arg(stop_mode_arg.clone())
1529 2 : .arg(safekeeper_extra_opt_arg)
1530 2 : )
1531 2 : )
1532 2 : .subcommand(
1533 2 : Command::new("endpoint")
1534 2 : .arg_required_else_help(true)
1535 2 : .about("Manage postgres instances")
1536 2 : .subcommand(Command::new("list").arg(tenant_id_arg.clone()))
1537 2 : .subcommand(Command::new("create")
1538 2 : .about("Create a compute endpoint")
1539 2 : .arg(endpoint_id_arg.clone())
1540 2 : .arg(branch_name_arg.clone())
1541 2 : .arg(tenant_id_arg.clone())
1542 2 : .arg(lsn_arg.clone())
1543 2 : .arg(pg_port_arg.clone())
1544 2 : .arg(http_port_arg.clone())
1545 2 : .arg(endpoint_pageserver_id_arg.clone())
1546 2 : .arg(
1547 2 : Arg::new("config-only")
1548 2 : .help("Don't do basebackup, create endpoint directory with only config files")
1549 2 : .long("config-only")
1550 2 : .required(false))
1551 2 : .arg(pg_version_arg.clone())
1552 2 : .arg(hot_standby_arg.clone())
1553 2 : .arg(update_catalog)
1554 2 : )
1555 2 : .subcommand(Command::new("start")
1556 2 : .about("Start postgres.\n If the endpoint doesn't exist yet, it is created.")
1557 2 : .arg(endpoint_id_arg.clone())
1558 2 : .arg(endpoint_pageserver_id_arg.clone())
1559 2 : .arg(safekeepers_arg)
1560 2 : .arg(remote_ext_config_args)
1561 2 : .arg(create_test_user)
1562 2 : )
1563 2 : .subcommand(Command::new("reconfigure")
1564 2 : .about("Reconfigure the endpoint")
1565 2 : .arg(endpoint_pageserver_id_arg)
1566 2 : .arg(endpoint_id_arg.clone())
1567 2 : .arg(tenant_id_arg.clone())
1568 2 : )
1569 2 : .subcommand(
1570 2 : Command::new("stop")
1571 2 : .arg(endpoint_id_arg)
1572 2 : .arg(
1573 2 : Arg::new("destroy")
1574 2 : .help("Also delete data directory (now optional, should be default in future)")
1575 2 : .long("destroy")
1576 2 : .action(ArgAction::SetTrue)
1577 2 : .required(false)
1578 2 : )
1579 2 : .arg(
1580 2 : Arg::new("mode")
1581 2 : .help("Postgres shutdown mode, passed to \"pg_ctl -m <mode>\"")
1582 2 : .long("mode")
1583 2 : .action(ArgAction::Set)
1584 2 : .required(false)
1585 2 : .value_parser(["smart", "fast", "immediate"])
1586 2 : .default_value("fast")
1587 2 : )
1588 2 : )
1589 2 :
1590 2 : )
1591 2 : .subcommand(
1592 2 : Command::new("mappings")
1593 2 : .arg_required_else_help(true)
1594 2 : .about("Manage neon_local branch name mappings")
1595 2 : .subcommand(
1596 2 : Command::new("map")
1597 2 : .about("Create new mapping which cannot exist already")
1598 2 : .arg(branch_name_arg.clone())
1599 2 : .arg(tenant_id_arg.clone())
1600 2 : .arg(timeline_id_arg.clone())
1601 2 : )
1602 2 : )
1603 2 : // Obsolete old name for 'endpoint'. We now just print an error if it's used.
1604 2 : .subcommand(
1605 2 : Command::new("pg")
1606 2 : .hide(true)
1607 2 : .arg(Arg::new("ignore-rest").allow_hyphen_values(true).num_args(0..).required(false))
1608 2 : .trailing_var_arg(true)
1609 2 : )
1610 2 : .subcommand(
1611 2 : Command::new("start")
1612 2 : .about("Start page server and safekeepers")
1613 2 : .arg(pageserver_config_args)
1614 2 : )
1615 2 : .subcommand(
1616 2 : Command::new("stop")
1617 2 : .about("Stop page server and safekeepers")
1618 2 : .arg(stop_mode_arg)
1619 2 : )
1620 2 : }
1621 :
1622 : #[test]
1623 2 : fn verify_cli() {
1624 2 : cli().debug_assert();
1625 2 : }
|
