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 std::borrow::Cow;
9 : use std::collections::{BTreeSet, HashMap};
10 : use std::fs::File;
11 : use std::path::PathBuf;
12 : use std::process::exit;
13 : use std::str::FromStr;
14 : use std::time::Duration;
15 :
16 : use anyhow::{Context, Result, anyhow, bail};
17 : use clap::Parser;
18 : use compute_api::requests::ComputeClaimsScope;
19 : use compute_api::spec::{ComputeMode, PageserverProtocol};
20 : use control_plane::broker::StorageBroker;
21 : use control_plane::endpoint::{ComputeControlPlane, EndpointTerminateMode};
22 : use control_plane::endpoint_storage::{ENDPOINT_STORAGE_DEFAULT_ADDR, EndpointStorage};
23 : use control_plane::local_env;
24 : use control_plane::local_env::{
25 : EndpointStorageConf, InitForceMode, LocalEnv, NeonBroker, NeonLocalInitConf,
26 : NeonLocalInitPageserverConf, SafekeeperConf,
27 : };
28 : use control_plane::pageserver::PageServerNode;
29 : use control_plane::safekeeper::SafekeeperNode;
30 : use control_plane::storage_controller::{
31 : NeonStorageControllerStartArgs, NeonStorageControllerStopArgs, StorageController,
32 : };
33 : use nix::fcntl::{Flock, FlockArg};
34 : use pageserver_api::config::{
35 : DEFAULT_GRPC_LISTEN_PORT as DEFAULT_PAGESERVER_GRPC_PORT,
36 : DEFAULT_HTTP_LISTEN_PORT as DEFAULT_PAGESERVER_HTTP_PORT,
37 : DEFAULT_PG_LISTEN_PORT as DEFAULT_PAGESERVER_PG_PORT,
38 : };
39 : use pageserver_api::controller_api::{
40 : NodeAvailabilityWrapper, PlacementPolicy, TenantCreateRequest,
41 : };
42 : use pageserver_api::models::{
43 : ShardParameters, TenantConfigRequest, TimelineCreateRequest, TimelineInfo,
44 : };
45 : use pageserver_api::shard::{DEFAULT_STRIPE_SIZE, ShardCount, ShardStripeSize, TenantShardId};
46 : use postgres_backend::AuthType;
47 : use postgres_connection::parse_host_port;
48 : use safekeeper_api::membership::{SafekeeperGeneration, SafekeeperId};
49 : use safekeeper_api::{
50 : DEFAULT_HTTP_LISTEN_PORT as DEFAULT_SAFEKEEPER_HTTP_PORT,
51 : DEFAULT_PG_LISTEN_PORT as DEFAULT_SAFEKEEPER_PG_PORT, PgMajorVersion, PgVersionId,
52 : };
53 : use storage_broker::DEFAULT_LISTEN_ADDR as DEFAULT_BROKER_ADDR;
54 : use tokio::task::JoinSet;
55 : use url::Host;
56 : use utils::auth::{Claims, Scope};
57 : use utils::id::{NodeId, TenantId, TenantTimelineId, TimelineId};
58 : use utils::lsn::Lsn;
59 : use utils::project_git_version;
60 :
61 : // Default id of a safekeeper node, if not specified on the command line.
62 : const DEFAULT_SAFEKEEPER_ID: NodeId = NodeId(1);
63 : const DEFAULT_PAGESERVER_ID: NodeId = NodeId(1);
64 : const DEFAULT_BRANCH_NAME: &str = "main";
65 : project_git_version!(GIT_VERSION);
66 :
67 : #[allow(dead_code)]
68 : const DEFAULT_PG_VERSION: PgMajorVersion = PgMajorVersion::PG17;
69 : const DEFAULT_PG_VERSION_NUM: &str = "17";
70 :
71 : const DEFAULT_PAGESERVER_CONTROL_PLANE_API: &str = "http://127.0.0.1:1234/upcall/v1/";
72 :
73 : #[derive(clap::Parser)]
74 : #[command(version = GIT_VERSION, about, name = "Neon CLI")]
75 : struct Cli {
76 : #[command(subcommand)]
77 : command: NeonLocalCmd,
78 : }
79 :
80 : #[derive(clap::Subcommand)]
81 : enum NeonLocalCmd {
82 : Init(InitCmdArgs),
83 :
84 : #[command(subcommand)]
85 : Tenant(TenantCmd),
86 : #[command(subcommand)]
87 : Timeline(TimelineCmd),
88 : #[command(subcommand)]
89 : Pageserver(PageserverCmd),
90 : #[command(subcommand)]
91 : #[clap(alias = "storage_controller")]
92 : StorageController(StorageControllerCmd),
93 : #[command(subcommand)]
94 : #[clap(alias = "storage_broker")]
95 : StorageBroker(StorageBrokerCmd),
96 : #[command(subcommand)]
97 : Safekeeper(SafekeeperCmd),
98 : #[command(subcommand)]
99 : EndpointStorage(EndpointStorageCmd),
100 : #[command(subcommand)]
101 : Endpoint(EndpointCmd),
102 : #[command(subcommand)]
103 : Mappings(MappingsCmd),
104 :
105 : Start(StartCmdArgs),
106 : Stop(StopCmdArgs),
107 : }
108 :
109 : #[derive(clap::Args)]
110 : #[clap(about = "Initialize a new Neon repository, preparing configs for services to start with")]
111 : struct InitCmdArgs {
112 : #[clap(long, help("How many pageservers to create (default 1)"))]
113 : num_pageservers: Option<u16>,
114 :
115 : #[clap(long)]
116 : config: Option<PathBuf>,
117 :
118 : #[clap(long, help("Force initialization even if the repository is not empty"))]
119 : #[arg(value_parser)]
120 : #[clap(default_value = "must-not-exist")]
121 : force: InitForceMode,
122 : }
123 :
124 : #[derive(clap::Args)]
125 : #[clap(about = "Start pageserver and safekeepers")]
126 : struct StartCmdArgs {
127 : #[clap(long = "start-timeout", default_value = "10s")]
128 : timeout: humantime::Duration,
129 : }
130 :
131 : #[derive(clap::Args)]
132 : #[clap(about = "Stop pageserver and safekeepers")]
133 : struct StopCmdArgs {
134 : #[arg(value_enum)]
135 : #[clap(long, default_value_t = StopMode::Fast)]
136 : mode: StopMode,
137 : }
138 :
139 : #[derive(Clone, Copy, clap::ValueEnum)]
140 : enum StopMode {
141 : Fast,
142 : Immediate,
143 : }
144 :
145 : #[derive(clap::Subcommand)]
146 : #[clap(about = "Manage tenants")]
147 : enum TenantCmd {
148 : List,
149 : Create(TenantCreateCmdArgs),
150 : SetDefault(TenantSetDefaultCmdArgs),
151 : Config(TenantConfigCmdArgs),
152 : Import(TenantImportCmdArgs),
153 : }
154 :
155 : #[derive(clap::Args)]
156 : struct TenantCreateCmdArgs {
157 : #[clap(
158 : long = "tenant-id",
159 : help = "Tenant id. Represented as a hexadecimal string 32 symbols length"
160 : )]
161 : tenant_id: Option<TenantId>,
162 :
163 : #[clap(
164 : long,
165 : help = "Use a specific timeline id when creating a tenant and its initial timeline"
166 : )]
167 : timeline_id: Option<TimelineId>,
168 :
169 : #[clap(short = 'c')]
170 : config: Vec<String>,
171 :
172 : #[arg(default_value = DEFAULT_PG_VERSION_NUM)]
173 : #[clap(long, help = "Postgres version to use for the initial timeline")]
174 : pg_version: PgMajorVersion,
175 :
176 : #[clap(
177 : long,
178 : help = "Use this tenant in future CLI commands where tenant_id is needed, but not specified"
179 : )]
180 : set_default: bool,
181 :
182 : #[clap(long, help = "Number of shards in the new tenant")]
183 : #[arg(default_value_t = 0)]
184 : shard_count: u8,
185 : #[clap(long, help = "Sharding stripe size in pages")]
186 : shard_stripe_size: Option<u32>,
187 :
188 : #[clap(long, help = "Placement policy shards in this tenant")]
189 : #[arg(value_parser = parse_placement_policy)]
190 : placement_policy: Option<PlacementPolicy>,
191 : }
192 :
193 0 : fn parse_placement_policy(s: &str) -> anyhow::Result<PlacementPolicy> {
194 0 : Ok(serde_json::from_str::<PlacementPolicy>(s)?)
195 0 : }
196 :
197 : #[derive(clap::Args)]
198 : #[clap(
199 : about = "Set a particular tenant as default in future CLI commands where tenant_id is needed, but not specified"
200 : )]
201 : struct TenantSetDefaultCmdArgs {
202 : #[clap(
203 : long = "tenant-id",
204 : help = "Tenant id. Represented as a hexadecimal string 32 symbols length"
205 : )]
206 : tenant_id: TenantId,
207 : }
208 :
209 : #[derive(clap::Args)]
210 : struct TenantConfigCmdArgs {
211 : #[clap(
212 : long = "tenant-id",
213 : help = "Tenant id. Represented as a hexadecimal string 32 symbols length"
214 : )]
215 : tenant_id: Option<TenantId>,
216 :
217 : #[clap(short = 'c')]
218 : config: Vec<String>,
219 : }
220 :
221 : #[derive(clap::Args)]
222 : #[clap(
223 : about = "Import a tenant that is present in remote storage, and create branches for its timelines"
224 : )]
225 : struct TenantImportCmdArgs {
226 : #[clap(
227 : long = "tenant-id",
228 : help = "Tenant id. Represented as a hexadecimal string 32 symbols length"
229 : )]
230 : tenant_id: TenantId,
231 : }
232 :
233 : #[derive(clap::Subcommand)]
234 : #[clap(about = "Manage timelines")]
235 : enum TimelineCmd {
236 : List(TimelineListCmdArgs),
237 : Branch(TimelineBranchCmdArgs),
238 : Create(TimelineCreateCmdArgs),
239 : Import(TimelineImportCmdArgs),
240 : }
241 :
242 : #[derive(clap::Args)]
243 : #[clap(about = "List all timelines available to this pageserver")]
244 : struct TimelineListCmdArgs {
245 : #[clap(
246 : long = "tenant-id",
247 : help = "Tenant id. Represented as a hexadecimal string 32 symbols length"
248 : )]
249 : tenant_shard_id: Option<TenantShardId>,
250 : }
251 :
252 : #[derive(clap::Args)]
253 : #[clap(about = "Create a new timeline, branching off from another timeline")]
254 : struct TimelineBranchCmdArgs {
255 : #[clap(
256 : long = "tenant-id",
257 : help = "Tenant id. Represented as a hexadecimal string 32 symbols length"
258 : )]
259 : tenant_id: Option<TenantId>,
260 :
261 : #[clap(long, help = "New timeline's ID")]
262 : timeline_id: Option<TimelineId>,
263 :
264 : #[clap(long, help = "Human-readable alias for the new timeline")]
265 : branch_name: String,
266 :
267 : #[clap(
268 : long,
269 : 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."
270 : )]
271 : ancestor_branch_name: Option<String>,
272 :
273 : #[clap(
274 : long,
275 : help = "When using another timeline as base, use a specific Lsn in it instead of the latest one"
276 : )]
277 : ancestor_start_lsn: Option<Lsn>,
278 : }
279 :
280 : #[derive(clap::Args)]
281 : #[clap(about = "Create a new blank timeline")]
282 : struct TimelineCreateCmdArgs {
283 : #[clap(
284 : long = "tenant-id",
285 : help = "Tenant id. Represented as a hexadecimal string 32 symbols length"
286 : )]
287 : tenant_id: Option<TenantId>,
288 :
289 : #[clap(long, help = "New timeline's ID")]
290 : timeline_id: Option<TimelineId>,
291 :
292 : #[clap(long, help = "Human-readable alias for the new timeline")]
293 : branch_name: String,
294 :
295 : #[arg(default_value = DEFAULT_PG_VERSION_NUM)]
296 : #[clap(long, help = "Postgres version")]
297 : pg_version: PgMajorVersion,
298 : }
299 :
300 : #[derive(clap::Args)]
301 : #[clap(about = "Import timeline from a basebackup directory")]
302 : struct TimelineImportCmdArgs {
303 : #[clap(
304 : long = "tenant-id",
305 : help = "Tenant id. Represented as a hexadecimal string 32 symbols length"
306 : )]
307 : tenant_id: Option<TenantId>,
308 :
309 : #[clap(long, help = "New timeline's ID")]
310 : timeline_id: TimelineId,
311 :
312 : #[clap(long, help = "Human-readable alias for the new timeline")]
313 : branch_name: String,
314 :
315 : #[clap(long, help = "Basebackup tarfile to import")]
316 : base_tarfile: PathBuf,
317 :
318 : #[clap(long, help = "Lsn the basebackup starts at")]
319 : base_lsn: Lsn,
320 :
321 : #[clap(long, help = "Wal to add after base")]
322 : wal_tarfile: Option<PathBuf>,
323 :
324 : #[clap(long, help = "Lsn the basebackup ends at")]
325 : end_lsn: Option<Lsn>,
326 :
327 : #[arg(default_value = DEFAULT_PG_VERSION_NUM)]
328 : #[clap(long, help = "Postgres version of the backup being imported")]
329 : pg_version: PgMajorVersion,
330 : }
331 :
332 : #[derive(clap::Subcommand)]
333 : #[clap(about = "Manage pageservers")]
334 : enum PageserverCmd {
335 : Status(PageserverStatusCmdArgs),
336 : Start(PageserverStartCmdArgs),
337 : Stop(PageserverStopCmdArgs),
338 : Restart(PageserverRestartCmdArgs),
339 : }
340 :
341 : #[derive(clap::Args)]
342 : #[clap(about = "Show status of a local pageserver")]
343 : struct PageserverStatusCmdArgs {
344 : #[clap(long = "id", help = "pageserver id")]
345 : pageserver_id: Option<NodeId>,
346 : }
347 :
348 : #[derive(clap::Args)]
349 : #[clap(about = "Start local pageserver")]
350 : struct PageserverStartCmdArgs {
351 : #[clap(long = "id", help = "pageserver id")]
352 : pageserver_id: Option<NodeId>,
353 :
354 : #[clap(short = 't', long, help = "timeout until we fail the command")]
355 : #[arg(default_value = "10s")]
356 : start_timeout: humantime::Duration,
357 : }
358 :
359 : #[derive(clap::Args)]
360 : #[clap(about = "Stop local pageserver")]
361 : struct PageserverStopCmdArgs {
362 : #[clap(long = "id", help = "pageserver id")]
363 : pageserver_id: Option<NodeId>,
364 :
365 : #[clap(
366 : short = 'm',
367 : help = "If 'immediate', don't flush repository data at shutdown"
368 : )]
369 : #[arg(value_enum, default_value = "fast")]
370 : stop_mode: StopMode,
371 : }
372 :
373 : #[derive(clap::Args)]
374 : #[clap(about = "Restart local pageserver")]
375 : struct PageserverRestartCmdArgs {
376 : #[clap(long = "id", help = "pageserver id")]
377 : pageserver_id: Option<NodeId>,
378 :
379 : #[clap(short = 't', long, help = "timeout until we fail the command")]
380 : #[arg(default_value = "10s")]
381 : start_timeout: humantime::Duration,
382 : }
383 :
384 : #[derive(clap::Subcommand)]
385 : #[clap(about = "Manage storage controller")]
386 : enum StorageControllerCmd {
387 : Start(StorageControllerStartCmdArgs),
388 : Stop(StorageControllerStopCmdArgs),
389 : }
390 :
391 : #[derive(clap::Args)]
392 : #[clap(about = "Start storage controller")]
393 : struct StorageControllerStartCmdArgs {
394 : #[clap(short = 't', long, help = "timeout until we fail the command")]
395 : #[arg(default_value = "10s")]
396 : start_timeout: humantime::Duration,
397 :
398 : #[clap(
399 : long,
400 : help = "Identifier used to distinguish storage controller instances"
401 : )]
402 : #[arg(default_value_t = 1)]
403 : instance_id: u8,
404 :
405 : #[clap(
406 : long,
407 : help = "Base port for the storage controller instance idenfified by instance-id (defaults to pageserver cplane api)"
408 : )]
409 : base_port: Option<u16>,
410 :
411 : #[clap(
412 : long,
413 : help = "Whether the storage controller should handle pageserver-reported local disk loss events."
414 : )]
415 : handle_ps_local_disk_loss: Option<bool>,
416 : }
417 :
418 : #[derive(clap::Args)]
419 : #[clap(about = "Stop storage controller")]
420 : struct StorageControllerStopCmdArgs {
421 : #[clap(
422 : short = 'm',
423 : help = "If 'immediate', don't flush repository data at shutdown"
424 : )]
425 : #[arg(value_enum, default_value = "fast")]
426 : stop_mode: StopMode,
427 :
428 : #[clap(
429 : long,
430 : help = "Identifier used to distinguish storage controller instances"
431 : )]
432 : #[arg(default_value_t = 1)]
433 : instance_id: u8,
434 : }
435 :
436 : #[derive(clap::Subcommand)]
437 : #[clap(about = "Manage storage broker")]
438 : enum StorageBrokerCmd {
439 : Start(StorageBrokerStartCmdArgs),
440 : Stop(StorageBrokerStopCmdArgs),
441 : }
442 :
443 : #[derive(clap::Args)]
444 : #[clap(about = "Start broker")]
445 : struct StorageBrokerStartCmdArgs {
446 : #[clap(short = 't', long, help = "timeout until we fail the command")]
447 : #[arg(default_value = "10s")]
448 : start_timeout: humantime::Duration,
449 : }
450 :
451 : #[derive(clap::Args)]
452 : #[clap(about = "stop broker")]
453 : struct StorageBrokerStopCmdArgs {
454 : #[clap(
455 : short = 'm',
456 : help = "If 'immediate', don't flush repository data at shutdown"
457 : )]
458 : #[arg(value_enum, default_value = "fast")]
459 : stop_mode: StopMode,
460 : }
461 :
462 : #[derive(clap::Subcommand)]
463 : #[clap(about = "Manage safekeepers")]
464 : enum SafekeeperCmd {
465 : Start(SafekeeperStartCmdArgs),
466 : Stop(SafekeeperStopCmdArgs),
467 : Restart(SafekeeperRestartCmdArgs),
468 : }
469 :
470 : #[derive(clap::Subcommand)]
471 : #[clap(about = "Manage object storage")]
472 : enum EndpointStorageCmd {
473 : Start(EndpointStorageStartCmd),
474 : Stop(EndpointStorageStopCmd),
475 : }
476 :
477 : #[derive(clap::Args)]
478 : #[clap(about = "Start object storage")]
479 : struct EndpointStorageStartCmd {
480 : #[clap(short = 't', long, help = "timeout until we fail the command")]
481 : #[arg(default_value = "10s")]
482 : start_timeout: humantime::Duration,
483 : }
484 :
485 : #[derive(clap::Args)]
486 : #[clap(about = "Stop object storage")]
487 : struct EndpointStorageStopCmd {
488 : #[arg(value_enum, default_value = "fast")]
489 : #[clap(
490 : short = 'm',
491 : help = "If 'immediate', don't flush repository data at shutdown"
492 : )]
493 : stop_mode: StopMode,
494 : }
495 :
496 : #[derive(clap::Args)]
497 : #[clap(about = "Start local safekeeper")]
498 : struct SafekeeperStartCmdArgs {
499 : #[clap(help = "safekeeper id")]
500 : #[arg(default_value_t = NodeId(1))]
501 : id: NodeId,
502 :
503 : #[clap(
504 : short = 'e',
505 : long = "safekeeper-extra-opt",
506 : help = "Additional safekeeper invocation options, e.g. -e=--http-auth-public-key-path=foo"
507 : )]
508 : extra_opt: Vec<String>,
509 :
510 : #[clap(short = 't', long, help = "timeout until we fail the command")]
511 : #[arg(default_value = "10s")]
512 : start_timeout: humantime::Duration,
513 : }
514 :
515 : #[derive(clap::Args)]
516 : #[clap(about = "Stop local safekeeper")]
517 : struct SafekeeperStopCmdArgs {
518 : #[clap(help = "safekeeper id")]
519 : #[arg(default_value_t = NodeId(1))]
520 : id: NodeId,
521 :
522 : #[arg(value_enum, default_value = "fast")]
523 : #[clap(
524 : short = 'm',
525 : help = "If 'immediate', don't flush repository data at shutdown"
526 : )]
527 : stop_mode: StopMode,
528 : }
529 :
530 : #[derive(clap::Args)]
531 : #[clap(about = "Restart local safekeeper")]
532 : struct SafekeeperRestartCmdArgs {
533 : #[clap(help = "safekeeper id")]
534 : #[arg(default_value_t = NodeId(1))]
535 : id: NodeId,
536 :
537 : #[arg(value_enum, default_value = "fast")]
538 : #[clap(
539 : short = 'm',
540 : help = "If 'immediate', don't flush repository data at shutdown"
541 : )]
542 : stop_mode: StopMode,
543 :
544 : #[clap(
545 : short = 'e',
546 : long = "safekeeper-extra-opt",
547 : help = "Additional safekeeper invocation options, e.g. -e=--http-auth-public-key-path=foo"
548 : )]
549 : extra_opt: Vec<String>,
550 :
551 : #[clap(short = 't', long, help = "timeout until we fail the command")]
552 : #[arg(default_value = "10s")]
553 : start_timeout: humantime::Duration,
554 : }
555 :
556 : #[derive(clap::Subcommand)]
557 : #[clap(about = "Manage Postgres instances")]
558 : enum EndpointCmd {
559 : List(EndpointListCmdArgs),
560 : Create(EndpointCreateCmdArgs),
561 : Start(EndpointStartCmdArgs),
562 : Reconfigure(EndpointReconfigureCmdArgs),
563 : Stop(EndpointStopCmdArgs),
564 : GenerateJwt(EndpointGenerateJwtCmdArgs),
565 : }
566 :
567 : #[derive(clap::Args)]
568 : #[clap(about = "List endpoints")]
569 : struct EndpointListCmdArgs {
570 : #[clap(
571 : long = "tenant-id",
572 : help = "Tenant id. Represented as a hexadecimal string 32 symbols length"
573 : )]
574 : tenant_shard_id: Option<TenantShardId>,
575 : }
576 :
577 : #[derive(clap::Args)]
578 : #[clap(about = "Create a compute endpoint")]
579 : struct EndpointCreateCmdArgs {
580 : #[clap(
581 : long = "tenant-id",
582 : help = "Tenant id. Represented as a hexadecimal string 32 symbols length"
583 : )]
584 : tenant_id: Option<TenantId>,
585 :
586 : #[clap(help = "Postgres endpoint id")]
587 : endpoint_id: Option<String>,
588 : #[clap(long, help = "Name of the branch the endpoint will run on")]
589 : branch_name: Option<String>,
590 : #[clap(
591 : long,
592 : help = "Specify Lsn on the timeline to start from. By default, end of the timeline would be used"
593 : )]
594 : lsn: Option<Lsn>,
595 : #[clap(long)]
596 : pg_port: Option<u16>,
597 : #[clap(long, alias = "http-port")]
598 : external_http_port: Option<u16>,
599 : #[clap(long)]
600 : internal_http_port: Option<u16>,
601 : #[clap(long = "pageserver-id")]
602 : endpoint_pageserver_id: Option<NodeId>,
603 :
604 : #[clap(
605 : long,
606 : help = "Don't do basebackup, create endpoint directory with only config files",
607 : action = clap::ArgAction::Set,
608 : default_value_t = false
609 : )]
610 : config_only: bool,
611 :
612 : #[arg(default_value = DEFAULT_PG_VERSION_NUM)]
613 : #[clap(long, help = "Postgres version")]
614 : pg_version: PgMajorVersion,
615 :
616 : /// Use gRPC to communicate with Pageservers, by generating grpc:// connstrings.
617 : ///
618 : /// Specified on creation such that it's retained across reconfiguration and restarts.
619 : ///
620 : /// NB: not yet supported by computes.
621 : #[clap(long)]
622 : grpc: bool,
623 :
624 : #[clap(
625 : long,
626 : help = "If set, the node will be a hot replica on the specified timeline",
627 : action = clap::ArgAction::Set,
628 : default_value_t = false
629 : )]
630 : hot_standby: bool,
631 :
632 : #[clap(long, help = "If set, will set up the catalog for neon_superuser")]
633 : update_catalog: bool,
634 :
635 : #[clap(
636 : long,
637 : help = "Allow multiple primary endpoints running on the same branch. Shouldn't be used normally, but useful for tests."
638 : )]
639 : allow_multiple: bool,
640 :
641 : /// Only allow changing it on creation
642 : #[clap(long, help = "Name of the privileged role for the endpoint")]
643 : privileged_role_name: Option<String>,
644 : }
645 :
646 : #[derive(clap::Args)]
647 : #[clap(about = "Start postgres. If the endpoint doesn't exist yet, it is created.")]
648 : struct EndpointStartCmdArgs {
649 : #[clap(help = "Postgres endpoint id")]
650 : endpoint_id: String,
651 : #[clap(long = "pageserver-id")]
652 : endpoint_pageserver_id: Option<NodeId>,
653 :
654 : #[clap(
655 : long,
656 : help = "Safekeepers membership generation to prefix neon.safekeepers with. Normally neon_local sets it on its own, but this option allows to override. Non zero value forces endpoint to use membership configurations."
657 : )]
658 : safekeepers_generation: Option<u32>,
659 : #[clap(
660 : long,
661 : help = "List of safekeepers endpoint will talk to. Normally neon_local chooses them on its own, but this option allows to override."
662 : )]
663 : safekeepers: Option<String>,
664 :
665 : #[clap(
666 : long,
667 : help = "Configure the remote extensions storage proxy gateway URL to request for extensions.",
668 : alias = "remote-ext-config"
669 : )]
670 : remote_ext_base_url: Option<String>,
671 :
672 : #[clap(
673 : long,
674 : help = "If set, will create test user `user` and `neondb` database. Requires `update-catalog = true`"
675 : )]
676 : create_test_user: bool,
677 :
678 : #[clap(
679 : long,
680 : help = "Allow multiple primary endpoints running on the same branch. Shouldn't be used normally, but useful for tests."
681 : )]
682 : allow_multiple: bool,
683 :
684 : #[clap(short = 't', long, value_parser= humantime::parse_duration, help = "timeout until we fail the command")]
685 : #[arg(default_value = "90s")]
686 : start_timeout: Duration,
687 :
688 : #[clap(
689 : long,
690 : help = "Download LFC cache from endpoint storage on endpoint startup",
691 : default_value = "false"
692 : )]
693 : autoprewarm: bool,
694 :
695 : #[clap(long, help = "Upload LFC cache to endpoint storage periodically")]
696 : offload_lfc_interval_seconds: Option<std::num::NonZeroU64>,
697 :
698 : #[clap(
699 : long,
700 : help = "Run in development mode, skipping VM-specific operations like process termination",
701 : action = clap::ArgAction::SetTrue
702 : )]
703 : dev: bool,
704 : }
705 :
706 : #[derive(clap::Args)]
707 : #[clap(about = "Reconfigure an endpoint")]
708 : struct EndpointReconfigureCmdArgs {
709 : #[clap(
710 : long = "tenant-id",
711 : help = "Tenant id. Represented as a hexadecimal string 32 symbols length"
712 : )]
713 : tenant_id: Option<TenantId>,
714 :
715 : #[clap(help = "Postgres endpoint id")]
716 : endpoint_id: String,
717 : #[clap(long = "pageserver-id")]
718 : endpoint_pageserver_id: Option<NodeId>,
719 :
720 : #[clap(long)]
721 : safekeepers: Option<String>,
722 : }
723 :
724 : #[derive(clap::Args)]
725 : #[clap(about = "Stop an endpoint")]
726 : struct EndpointStopCmdArgs {
727 : #[clap(help = "Postgres endpoint id")]
728 : endpoint_id: String,
729 :
730 : #[clap(
731 : long,
732 : help = "Also delete data directory (now optional, should be default in future)"
733 : )]
734 : destroy: bool,
735 :
736 : #[clap(long, help = "Postgres shutdown mode")]
737 : #[clap(default_value = "fast")]
738 : mode: EndpointTerminateMode,
739 : }
740 :
741 : #[derive(clap::Args)]
742 : #[clap(about = "Generate a JWT for an endpoint")]
743 : struct EndpointGenerateJwtCmdArgs {
744 : #[clap(help = "Postgres endpoint id")]
745 : endpoint_id: String,
746 :
747 : #[clap(short = 's', long, help = "Scope to generate the JWT with", value_parser = ComputeClaimsScope::from_str)]
748 : scope: Option<ComputeClaimsScope>,
749 : }
750 :
751 : #[derive(clap::Subcommand)]
752 : #[clap(about = "Manage neon_local branch name mappings")]
753 : enum MappingsCmd {
754 : Map(MappingsMapCmdArgs),
755 : }
756 :
757 : #[derive(clap::Args)]
758 : #[clap(about = "Create new mapping which cannot exist already")]
759 : struct MappingsMapCmdArgs {
760 : #[clap(
761 : long,
762 : help = "Tenant id. Represented as a hexadecimal string 32 symbols length"
763 : )]
764 : tenant_id: TenantId,
765 : #[clap(
766 : long,
767 : help = "Timeline id. Represented as a hexadecimal string 32 symbols length"
768 : )]
769 : timeline_id: TimelineId,
770 : #[clap(long, help = "Branch name to give to the timeline")]
771 : branch_name: String,
772 : }
773 :
774 : ///
775 : /// Timelines tree element used as a value in the HashMap.
776 : ///
777 : struct TimelineTreeEl {
778 : /// `TimelineInfo` received from the `pageserver` via the `timeline_list` http API call.
779 : pub info: TimelineInfo,
780 : /// Name, recovered from neon config mappings
781 : pub name: Option<String>,
782 : /// Holds all direct children of this timeline referenced using `timeline_id`.
783 : pub children: BTreeSet<TimelineId>,
784 : }
785 :
786 : /// A flock-based guard over the neon_local repository directory
787 : struct RepoLock {
788 : _file: Flock<File>,
789 : }
790 :
791 : impl RepoLock {
792 0 : fn new() -> Result<Self> {
793 0 : let repo_dir = File::open(local_env::base_path())?;
794 0 : match Flock::lock(repo_dir, FlockArg::LockExclusive) {
795 0 : Ok(f) => Ok(Self { _file: f }),
796 0 : Err((_, e)) => Err(e).context("flock error"),
797 : }
798 0 : }
799 : }
800 :
801 : // Main entry point for the 'neon_local' CLI utility
802 : //
803 : // This utility helps to manage neon installation. That includes following:
804 : // * Management of local postgres installations running on top of the
805 : // pageserver.
806 : // * Providing CLI api to the pageserver
807 : // * TODO: export/import to/from usual postgres
808 0 : fn main() -> Result<()> {
809 0 : let cli = Cli::parse();
810 :
811 : // Check for 'neon init' command first.
812 0 : let (subcommand_result, _lock) = if let NeonLocalCmd::Init(args) = cli.command {
813 0 : (handle_init(&args).map(|env| Some(Cow::Owned(env))), None)
814 : } else {
815 : // This tool uses a collection of simple files to store its state, and consequently
816 : // it is not generally safe to run multiple commands concurrently. Rather than expect
817 : // all callers to know this, use a lock file to protect against concurrent execution.
818 0 : let _repo_lock = RepoLock::new().unwrap();
819 :
820 : // all other commands need an existing config
821 0 : let env = LocalEnv::load_config(&local_env::base_path()).context("Error loading config")?;
822 0 : let original_env = env.clone();
823 0 : let env = Box::leak(Box::new(env));
824 0 : let rt = tokio::runtime::Builder::new_current_thread()
825 0 : .enable_all()
826 0 : .build()
827 0 : .unwrap();
828 :
829 0 : let subcommand_result = match cli.command {
830 0 : NeonLocalCmd::Init(_) => unreachable!("init was handled earlier already"),
831 0 : NeonLocalCmd::Start(args) => rt.block_on(handle_start_all(&args, env)),
832 0 : NeonLocalCmd::Stop(args) => rt.block_on(handle_stop_all(&args, env)),
833 0 : NeonLocalCmd::Tenant(subcmd) => rt.block_on(handle_tenant(&subcmd, env)),
834 0 : NeonLocalCmd::Timeline(subcmd) => rt.block_on(handle_timeline(&subcmd, env)),
835 0 : NeonLocalCmd::Pageserver(subcmd) => rt.block_on(handle_pageserver(&subcmd, env)),
836 0 : NeonLocalCmd::StorageController(subcmd) => {
837 0 : rt.block_on(handle_storage_controller(&subcmd, env))
838 : }
839 0 : NeonLocalCmd::StorageBroker(subcmd) => rt.block_on(handle_storage_broker(&subcmd, env)),
840 0 : NeonLocalCmd::Safekeeper(subcmd) => rt.block_on(handle_safekeeper(&subcmd, env)),
841 0 : NeonLocalCmd::EndpointStorage(subcmd) => {
842 0 : rt.block_on(handle_endpoint_storage(&subcmd, env))
843 : }
844 0 : NeonLocalCmd::Endpoint(subcmd) => rt.block_on(handle_endpoint(&subcmd, env)),
845 0 : NeonLocalCmd::Mappings(subcmd) => handle_mappings(&subcmd, env),
846 : };
847 :
848 0 : let subcommand_result = if &original_env != env {
849 0 : subcommand_result.map(|()| Some(Cow::Borrowed(env)))
850 : } else {
851 0 : subcommand_result.map(|()| None)
852 : };
853 0 : (subcommand_result, Some(_repo_lock))
854 : };
855 :
856 0 : match subcommand_result {
857 0 : Ok(Some(updated_env)) => updated_env.persist_config()?,
858 0 : Ok(None) => (),
859 0 : Err(e) => {
860 0 : eprintln!("command failed: {e:?}");
861 0 : exit(1);
862 : }
863 : }
864 0 : Ok(())
865 0 : }
866 :
867 : ///
868 : /// Prints timelines list as a tree-like structure.
869 : ///
870 0 : fn print_timelines_tree(
871 0 : timelines: Vec<TimelineInfo>,
872 0 : mut timeline_name_mappings: HashMap<TenantTimelineId, String>,
873 0 : ) -> Result<()> {
874 0 : let mut timelines_hash = timelines
875 0 : .iter()
876 0 : .map(|t| {
877 0 : (
878 0 : t.timeline_id,
879 0 : TimelineTreeEl {
880 0 : info: t.clone(),
881 0 : children: BTreeSet::new(),
882 0 : name: timeline_name_mappings
883 0 : .remove(&TenantTimelineId::new(t.tenant_id.tenant_id, t.timeline_id)),
884 0 : },
885 0 : )
886 0 : })
887 0 : .collect::<HashMap<_, _>>();
888 :
889 : // Memorize all direct children of each timeline.
890 0 : for timeline in timelines.iter() {
891 0 : if let Some(ancestor_timeline_id) = timeline.ancestor_timeline_id {
892 0 : timelines_hash
893 0 : .get_mut(&ancestor_timeline_id)
894 0 : .context("missing timeline info in the HashMap")?
895 : .children
896 0 : .insert(timeline.timeline_id);
897 0 : }
898 : }
899 :
900 0 : for timeline in timelines_hash.values() {
901 : // Start with root local timelines (no ancestors) first.
902 0 : if timeline.info.ancestor_timeline_id.is_none() {
903 0 : print_timeline(0, &Vec::from([true]), timeline, &timelines_hash)?;
904 0 : }
905 : }
906 :
907 0 : Ok(())
908 0 : }
909 :
910 : ///
911 : /// Recursively prints timeline info with all its children.
912 : ///
913 0 : fn print_timeline(
914 0 : nesting_level: usize,
915 0 : is_last: &[bool],
916 0 : timeline: &TimelineTreeEl,
917 0 : timelines: &HashMap<TimelineId, TimelineTreeEl>,
918 0 : ) -> Result<()> {
919 0 : if nesting_level > 0 {
920 0 : let ancestor_lsn = match timeline.info.ancestor_lsn {
921 0 : Some(lsn) => lsn.to_string(),
922 0 : None => "Unknown Lsn".to_string(),
923 : };
924 :
925 0 : let mut br_sym = "┣━";
926 :
927 : // Draw each nesting padding with proper style
928 : // depending on whether its timeline ended or not.
929 0 : if nesting_level > 1 {
930 0 : for l in &is_last[1..is_last.len() - 1] {
931 0 : if *l {
932 0 : print!(" ");
933 0 : } else {
934 0 : print!("┃ ");
935 0 : }
936 : }
937 0 : }
938 :
939 : // We are the last in this sub-timeline
940 0 : if *is_last.last().unwrap() {
941 0 : br_sym = "┗━";
942 0 : }
943 :
944 0 : print!("{br_sym} @{ancestor_lsn}: ");
945 0 : }
946 :
947 : // Finally print a timeline id and name with new line
948 0 : println!(
949 0 : "{} [{}]",
950 0 : timeline.name.as_deref().unwrap_or("_no_name_"),
951 : timeline.info.timeline_id
952 : );
953 :
954 0 : let len = timeline.children.len();
955 0 : let mut i: usize = 0;
956 0 : let mut is_last_new = Vec::from(is_last);
957 0 : is_last_new.push(false);
958 :
959 0 : for child in &timeline.children {
960 0 : i += 1;
961 :
962 : // Mark that the last padding is the end of the timeline
963 0 : if i == len {
964 0 : if let Some(last) = is_last_new.last_mut() {
965 0 : *last = true;
966 0 : }
967 0 : }
968 :
969 0 : print_timeline(
970 0 : nesting_level + 1,
971 0 : &is_last_new,
972 0 : timelines
973 0 : .get(child)
974 0 : .context("missing timeline info in the HashMap")?,
975 0 : timelines,
976 0 : )?;
977 : }
978 :
979 0 : Ok(())
980 0 : }
981 :
982 : /// Helper function to get tenant id from an optional --tenant_id option or from the config file
983 0 : fn get_tenant_id(
984 0 : tenant_id_arg: Option<TenantId>,
985 0 : env: &local_env::LocalEnv,
986 0 : ) -> anyhow::Result<TenantId> {
987 0 : if let Some(tenant_id_from_arguments) = tenant_id_arg {
988 0 : Ok(tenant_id_from_arguments)
989 0 : } else if let Some(default_id) = env.default_tenant_id {
990 0 : Ok(default_id)
991 : } else {
992 0 : anyhow::bail!("No tenant id. Use --tenant-id, or set a default tenant");
993 : }
994 0 : }
995 :
996 : /// Helper function to get tenant-shard ID from an optional --tenant_id option or from the config file,
997 : /// for commands that accept a shard suffix
998 0 : fn get_tenant_shard_id(
999 0 : tenant_shard_id_arg: Option<TenantShardId>,
1000 0 : env: &local_env::LocalEnv,
1001 0 : ) -> anyhow::Result<TenantShardId> {
1002 0 : if let Some(tenant_id_from_arguments) = tenant_shard_id_arg {
1003 0 : Ok(tenant_id_from_arguments)
1004 0 : } else if let Some(default_id) = env.default_tenant_id {
1005 0 : Ok(TenantShardId::unsharded(default_id))
1006 : } else {
1007 0 : anyhow::bail!("No tenant shard id. Use --tenant-id, or set a default tenant");
1008 : }
1009 0 : }
1010 :
1011 0 : fn handle_init(args: &InitCmdArgs) -> anyhow::Result<LocalEnv> {
1012 : // Create the in-memory `LocalEnv` that we'd normally load from disk in `load_config`.
1013 0 : let init_conf: NeonLocalInitConf = if let Some(config_path) = &args.config {
1014 : // User (likely the Python test suite) provided a description of the environment.
1015 0 : if args.num_pageservers.is_some() {
1016 0 : bail!(
1017 0 : "Cannot specify both --num-pageservers and --config, use key `pageservers` in the --config file instead"
1018 : );
1019 0 : }
1020 : // load and parse the file
1021 0 : let contents = std::fs::read_to_string(config_path).with_context(|| {
1022 0 : format!(
1023 0 : "Could not read configuration file '{}'",
1024 0 : config_path.display()
1025 : )
1026 0 : })?;
1027 0 : toml_edit::de::from_str(&contents)?
1028 : } else {
1029 : // User (likely interactive) did not provide a description of the environment, give them the default
1030 : NeonLocalInitConf {
1031 0 : control_plane_api: Some(DEFAULT_PAGESERVER_CONTROL_PLANE_API.parse().unwrap()),
1032 0 : broker: NeonBroker {
1033 0 : listen_addr: Some(DEFAULT_BROKER_ADDR.parse().unwrap()),
1034 0 : listen_https_addr: None,
1035 0 : },
1036 0 : safekeepers: vec![SafekeeperConf {
1037 0 : id: DEFAULT_SAFEKEEPER_ID,
1038 0 : pg_port: DEFAULT_SAFEKEEPER_PG_PORT,
1039 0 : http_port: DEFAULT_SAFEKEEPER_HTTP_PORT,
1040 0 : ..Default::default()
1041 0 : }],
1042 0 : pageservers: (0..args.num_pageservers.unwrap_or(1))
1043 0 : .map(|i| {
1044 0 : let pageserver_id = NodeId(DEFAULT_PAGESERVER_ID.0 + i as u64);
1045 0 : let pg_port = DEFAULT_PAGESERVER_PG_PORT + i;
1046 0 : let http_port = DEFAULT_PAGESERVER_HTTP_PORT + i;
1047 0 : let grpc_port = DEFAULT_PAGESERVER_GRPC_PORT + i;
1048 0 : NeonLocalInitPageserverConf {
1049 0 : id: pageserver_id,
1050 0 : listen_pg_addr: format!("127.0.0.1:{pg_port}"),
1051 0 : listen_http_addr: format!("127.0.0.1:{http_port}"),
1052 0 : listen_https_addr: None,
1053 0 : listen_grpc_addr: Some(format!("127.0.0.1:{grpc_port}")),
1054 0 : pg_auth_type: AuthType::Trust,
1055 0 : http_auth_type: AuthType::Trust,
1056 0 : grpc_auth_type: AuthType::Trust,
1057 0 : other: Default::default(),
1058 0 : // Typical developer machines use disks with slow fsync, and we don't care
1059 0 : // about data integrity: disable disk syncs.
1060 0 : no_sync: true,
1061 0 : }
1062 0 : })
1063 0 : .collect(),
1064 0 : endpoint_storage: EndpointStorageConf {
1065 0 : listen_addr: ENDPOINT_STORAGE_DEFAULT_ADDR,
1066 0 : },
1067 0 : pg_distrib_dir: None,
1068 0 : neon_distrib_dir: None,
1069 0 : default_tenant_id: TenantId::from_array(std::array::from_fn(|_| 0)),
1070 0 : storage_controller: None,
1071 0 : control_plane_hooks_api: None,
1072 : generate_local_ssl_certs: false,
1073 : }
1074 : };
1075 :
1076 0 : LocalEnv::init(init_conf, &args.force)
1077 0 : .context("materialize initial neon_local environment on disk")?;
1078 0 : Ok(LocalEnv::load_config(&local_env::base_path())
1079 0 : .expect("freshly written config should be loadable"))
1080 0 : }
1081 :
1082 : /// The default pageserver is the one where CLI tenant/timeline operations are sent by default.
1083 : /// For typical interactive use, one would just run with a single pageserver. Scenarios with
1084 : /// tenant/timeline placement across multiple pageservers are managed by python test code rather
1085 : /// than this CLI.
1086 0 : fn get_default_pageserver(env: &local_env::LocalEnv) -> PageServerNode {
1087 0 : let ps_conf = env
1088 0 : .pageservers
1089 0 : .first()
1090 0 : .expect("Config is validated to contain at least one pageserver");
1091 0 : PageServerNode::from_env(env, ps_conf)
1092 0 : }
1093 :
1094 0 : async fn handle_tenant(subcmd: &TenantCmd, env: &mut local_env::LocalEnv) -> anyhow::Result<()> {
1095 0 : let pageserver = get_default_pageserver(env);
1096 0 : match subcmd {
1097 : TenantCmd::List => {
1098 0 : for t in pageserver.tenant_list().await? {
1099 0 : println!("{} {:?}", t.id, t.state);
1100 0 : }
1101 : }
1102 0 : TenantCmd::Import(args) => {
1103 0 : let tenant_id = args.tenant_id;
1104 :
1105 0 : let storage_controller = StorageController::from_env(env);
1106 0 : let create_response = storage_controller.tenant_import(tenant_id).await?;
1107 :
1108 0 : let shard_zero = create_response
1109 0 : .shards
1110 0 : .first()
1111 0 : .expect("Import response omitted shards");
1112 :
1113 0 : let attached_pageserver_id = shard_zero.node_id;
1114 0 : let pageserver =
1115 0 : PageServerNode::from_env(env, env.get_pageserver_conf(attached_pageserver_id)?);
1116 :
1117 0 : println!(
1118 0 : "Imported tenant {tenant_id}, attached to pageserver {attached_pageserver_id}"
1119 : );
1120 :
1121 0 : let timelines = pageserver
1122 0 : .http_client
1123 0 : .list_timelines(shard_zero.shard_id)
1124 0 : .await?;
1125 :
1126 : // Pick a 'main' timeline that has no ancestors, the rest will get arbitrary names
1127 0 : let main_timeline = timelines
1128 0 : .iter()
1129 0 : .find(|t| t.ancestor_timeline_id.is_none())
1130 0 : .expect("No timelines found")
1131 : .timeline_id;
1132 :
1133 0 : let mut branch_i = 0;
1134 0 : for timeline in timelines.iter() {
1135 0 : let branch_name = if timeline.timeline_id == main_timeline {
1136 0 : "main".to_string()
1137 : } else {
1138 0 : branch_i += 1;
1139 0 : format!("branch_{branch_i}")
1140 : };
1141 :
1142 0 : println!(
1143 0 : "Importing timeline {tenant_id}/{} as branch {branch_name}",
1144 : timeline.timeline_id
1145 : );
1146 :
1147 0 : env.register_branch_mapping(branch_name, tenant_id, timeline.timeline_id)?;
1148 : }
1149 : }
1150 0 : TenantCmd::Create(args) => {
1151 0 : let tenant_conf: HashMap<_, _> =
1152 0 : args.config.iter().flat_map(|c| c.split_once(':')).collect();
1153 :
1154 0 : let tenant_conf = PageServerNode::parse_config(tenant_conf)?;
1155 :
1156 : // If tenant ID was not specified, generate one
1157 0 : let tenant_id = args.tenant_id.unwrap_or_else(TenantId::generate);
1158 :
1159 : // We must register the tenant with the storage controller, so
1160 : // that when the pageserver restarts, it will be re-attached.
1161 0 : let storage_controller = StorageController::from_env(env);
1162 0 : storage_controller
1163 0 : .tenant_create(TenantCreateRequest {
1164 0 : // Note that ::unsharded here isn't actually because the tenant is unsharded, its because the
1165 0 : // storage controller expects a shard-naive tenant_id in this attribute, and the TenantCreateRequest
1166 0 : // type is used both in the storage controller (for creating tenants) and in the pageserver (for
1167 0 : // creating shards)
1168 0 : new_tenant_id: TenantShardId::unsharded(tenant_id),
1169 0 : generation: None,
1170 0 : shard_parameters: ShardParameters {
1171 0 : count: ShardCount::new(args.shard_count),
1172 0 : stripe_size: args
1173 0 : .shard_stripe_size
1174 0 : .map(ShardStripeSize)
1175 0 : .unwrap_or(DEFAULT_STRIPE_SIZE),
1176 0 : },
1177 0 : placement_policy: args.placement_policy.clone(),
1178 0 : config: tenant_conf,
1179 0 : })
1180 0 : .await?;
1181 0 : println!("tenant {tenant_id} successfully created on the pageserver");
1182 :
1183 : // Create an initial timeline for the new tenant
1184 0 : let new_timeline_id = args.timeline_id.unwrap_or(TimelineId::generate());
1185 :
1186 : // FIXME: passing None for ancestor_start_lsn is not kosher in a sharded world: we can't have
1187 : // different shards picking different start lsns. Maybe we have to teach storage controller
1188 : // to let shard 0 branch first and then propagate the chosen LSN to other shards.
1189 0 : storage_controller
1190 0 : .tenant_timeline_create(
1191 0 : tenant_id,
1192 0 : TimelineCreateRequest {
1193 0 : new_timeline_id,
1194 0 : mode: pageserver_api::models::TimelineCreateRequestMode::Bootstrap {
1195 0 : existing_initdb_timeline_id: None,
1196 0 : pg_version: Some(args.pg_version),
1197 0 : },
1198 0 : },
1199 0 : )
1200 0 : .await?;
1201 :
1202 0 : env.register_branch_mapping(
1203 0 : DEFAULT_BRANCH_NAME.to_string(),
1204 0 : tenant_id,
1205 0 : new_timeline_id,
1206 0 : )?;
1207 :
1208 0 : println!("Created an initial timeline '{new_timeline_id}' for tenant: {tenant_id}",);
1209 :
1210 0 : if args.set_default {
1211 0 : println!("Setting tenant {tenant_id} as a default one");
1212 0 : env.default_tenant_id = Some(tenant_id);
1213 0 : }
1214 : }
1215 0 : TenantCmd::SetDefault(args) => {
1216 0 : println!("Setting tenant {} as a default one", args.tenant_id);
1217 0 : env.default_tenant_id = Some(args.tenant_id);
1218 0 : }
1219 0 : TenantCmd::Config(args) => {
1220 0 : let tenant_id = get_tenant_id(args.tenant_id, env)?;
1221 0 : let tenant_conf: HashMap<_, _> =
1222 0 : args.config.iter().flat_map(|c| c.split_once(':')).collect();
1223 0 : let config = PageServerNode::parse_config(tenant_conf)?;
1224 :
1225 0 : let req = TenantConfigRequest { tenant_id, config };
1226 :
1227 0 : let storage_controller = StorageController::from_env(env);
1228 0 : storage_controller
1229 0 : .set_tenant_config(&req)
1230 0 : .await
1231 0 : .with_context(|| format!("Tenant config failed for tenant with id {tenant_id}"))?;
1232 0 : println!("tenant {tenant_id} successfully configured via storcon");
1233 : }
1234 : }
1235 0 : Ok(())
1236 0 : }
1237 :
1238 0 : async fn handle_timeline(cmd: &TimelineCmd, env: &mut local_env::LocalEnv) -> Result<()> {
1239 0 : let pageserver = get_default_pageserver(env);
1240 :
1241 0 : match cmd {
1242 0 : TimelineCmd::List(args) => {
1243 : // TODO(sharding): this command shouldn't have to specify a shard ID: we should ask the storage controller
1244 : // where shard 0 is attached, and query there.
1245 0 : let tenant_shard_id = get_tenant_shard_id(args.tenant_shard_id, env)?;
1246 0 : let timelines = pageserver.timeline_list(&tenant_shard_id).await?;
1247 0 : print_timelines_tree(timelines, env.timeline_name_mappings())?;
1248 : }
1249 0 : TimelineCmd::Create(args) => {
1250 0 : let tenant_id = get_tenant_id(args.tenant_id, env)?;
1251 0 : let new_branch_name = &args.branch_name;
1252 0 : let new_timeline_id_opt = args.timeline_id;
1253 0 : let new_timeline_id = new_timeline_id_opt.unwrap_or(TimelineId::generate());
1254 :
1255 0 : let storage_controller = StorageController::from_env(env);
1256 0 : let create_req = TimelineCreateRequest {
1257 0 : new_timeline_id,
1258 0 : mode: pageserver_api::models::TimelineCreateRequestMode::Bootstrap {
1259 0 : existing_initdb_timeline_id: None,
1260 0 : pg_version: Some(args.pg_version),
1261 0 : },
1262 0 : };
1263 0 : let timeline_info = storage_controller
1264 0 : .tenant_timeline_create(tenant_id, create_req)
1265 0 : .await?;
1266 :
1267 0 : let last_record_lsn = timeline_info.last_record_lsn;
1268 0 : env.register_branch_mapping(new_branch_name.to_string(), tenant_id, new_timeline_id)?;
1269 :
1270 0 : println!(
1271 0 : "Created timeline '{}' at Lsn {last_record_lsn} for tenant: {tenant_id}",
1272 : timeline_info.timeline_id
1273 : );
1274 : }
1275 : // TODO: rename to import-basebackup-plus-wal
1276 0 : TimelineCmd::Import(args) => {
1277 0 : let tenant_id = get_tenant_id(args.tenant_id, env)?;
1278 0 : let timeline_id = args.timeline_id;
1279 0 : let branch_name = &args.branch_name;
1280 :
1281 : // Parse base inputs
1282 0 : let base = (args.base_lsn, args.base_tarfile.clone());
1283 :
1284 : // Parse pg_wal inputs
1285 0 : let wal_tarfile = args.wal_tarfile.clone();
1286 0 : let end_lsn = args.end_lsn;
1287 : // TODO validate both or none are provided
1288 0 : let pg_wal = end_lsn.zip(wal_tarfile);
1289 :
1290 0 : println!("Importing timeline into pageserver ...");
1291 0 : pageserver
1292 0 : .timeline_import(tenant_id, timeline_id, base, pg_wal, args.pg_version)
1293 0 : .await?;
1294 0 : if env.storage_controller.timelines_onto_safekeepers {
1295 0 : println!("Creating timeline on safekeeper ...");
1296 0 : let timeline_info = pageserver
1297 0 : .timeline_info(
1298 0 : TenantShardId::unsharded(tenant_id),
1299 0 : timeline_id,
1300 0 : pageserver_client::mgmt_api::ForceAwaitLogicalSize::No,
1301 0 : )
1302 0 : .await?;
1303 0 : let default_sk = SafekeeperNode::from_env(env, env.safekeepers.first().unwrap());
1304 0 : let default_host = default_sk
1305 0 : .conf
1306 0 : .listen_addr
1307 0 : .clone()
1308 0 : .unwrap_or_else(|| "localhost".to_string());
1309 0 : let mconf = safekeeper_api::membership::Configuration {
1310 0 : generation: SafekeeperGeneration::new(1),
1311 0 : members: safekeeper_api::membership::MemberSet {
1312 0 : m: vec![SafekeeperId {
1313 0 : host: default_host,
1314 0 : id: default_sk.conf.id,
1315 0 : pg_port: default_sk.conf.pg_port,
1316 0 : }],
1317 0 : },
1318 0 : new_members: None,
1319 0 : };
1320 0 : let pg_version = PgVersionId::from(args.pg_version);
1321 0 : let req = safekeeper_api::models::TimelineCreateRequest {
1322 0 : tenant_id,
1323 0 : timeline_id,
1324 0 : mconf,
1325 0 : pg_version,
1326 0 : system_id: None,
1327 0 : wal_seg_size: None,
1328 0 : start_lsn: timeline_info.last_record_lsn,
1329 0 : commit_lsn: None,
1330 0 : };
1331 0 : default_sk.create_timeline(&req).await?;
1332 0 : }
1333 0 : env.register_branch_mapping(branch_name.to_string(), tenant_id, timeline_id)?;
1334 0 : println!("Done");
1335 : }
1336 0 : TimelineCmd::Branch(args) => {
1337 0 : let tenant_id = get_tenant_id(args.tenant_id, env)?;
1338 0 : let new_timeline_id = args.timeline_id.unwrap_or(TimelineId::generate());
1339 0 : let new_branch_name = &args.branch_name;
1340 0 : let ancestor_branch_name = args
1341 0 : .ancestor_branch_name
1342 0 : .clone()
1343 0 : .unwrap_or(DEFAULT_BRANCH_NAME.to_owned());
1344 0 : let ancestor_timeline_id = env
1345 0 : .get_branch_timeline_id(&ancestor_branch_name, tenant_id)
1346 0 : .ok_or_else(|| {
1347 0 : anyhow!("Found no timeline id for branch name '{ancestor_branch_name}'")
1348 0 : })?;
1349 :
1350 0 : let start_lsn = args.ancestor_start_lsn;
1351 0 : let storage_controller = StorageController::from_env(env);
1352 0 : let create_req = TimelineCreateRequest {
1353 0 : new_timeline_id,
1354 0 : mode: pageserver_api::models::TimelineCreateRequestMode::Branch {
1355 0 : ancestor_timeline_id,
1356 0 : ancestor_start_lsn: start_lsn,
1357 0 : read_only: false,
1358 0 : pg_version: None,
1359 0 : },
1360 0 : };
1361 0 : let timeline_info = storage_controller
1362 0 : .tenant_timeline_create(tenant_id, create_req)
1363 0 : .await?;
1364 :
1365 0 : let last_record_lsn = timeline_info.last_record_lsn;
1366 :
1367 0 : env.register_branch_mapping(new_branch_name.to_string(), tenant_id, new_timeline_id)?;
1368 :
1369 0 : println!(
1370 0 : "Created timeline '{}' at Lsn {last_record_lsn} for tenant: {tenant_id}. Ancestor timeline: '{ancestor_branch_name}'",
1371 : timeline_info.timeline_id
1372 : );
1373 : }
1374 : }
1375 :
1376 0 : Ok(())
1377 0 : }
1378 :
1379 0 : async fn handle_endpoint(subcmd: &EndpointCmd, env: &local_env::LocalEnv) -> Result<()> {
1380 0 : let mut cplane = ComputeControlPlane::load(env.clone())?;
1381 :
1382 0 : match subcmd {
1383 0 : EndpointCmd::List(args) => {
1384 : // TODO(sharding): this command shouldn't have to specify a shard ID: we should ask the storage controller
1385 : // where shard 0 is attached, and query there.
1386 0 : let tenant_shard_id = get_tenant_shard_id(args.tenant_shard_id, env)?;
1387 :
1388 0 : let timeline_name_mappings = env.timeline_name_mappings();
1389 :
1390 0 : let mut table = comfy_table::Table::new();
1391 :
1392 0 : table.load_preset(comfy_table::presets::NOTHING);
1393 :
1394 0 : table.set_header([
1395 0 : "ENDPOINT",
1396 0 : "ADDRESS",
1397 0 : "TIMELINE",
1398 0 : "BRANCH NAME",
1399 0 : "LSN",
1400 0 : "STATUS",
1401 0 : ]);
1402 :
1403 0 : for (endpoint_id, endpoint) in cplane
1404 0 : .endpoints
1405 0 : .iter()
1406 0 : .filter(|(_, endpoint)| endpoint.tenant_id == tenant_shard_id.tenant_id)
1407 : {
1408 0 : let lsn_str = match endpoint.mode {
1409 0 : ComputeMode::Static(lsn) => {
1410 : // -> read-only endpoint
1411 : // Use the node's LSN.
1412 0 : lsn.to_string()
1413 : }
1414 : _ => {
1415 : // As the LSN here refers to the one that the compute is started with,
1416 : // we display nothing as it is a primary/hot standby compute.
1417 0 : "---".to_string()
1418 : }
1419 : };
1420 :
1421 0 : let branch_name = timeline_name_mappings
1422 0 : .get(&TenantTimelineId::new(
1423 0 : tenant_shard_id.tenant_id,
1424 0 : endpoint.timeline_id,
1425 0 : ))
1426 0 : .map(|name| name.as_str())
1427 0 : .unwrap_or("?");
1428 :
1429 0 : table.add_row([
1430 0 : endpoint_id.as_str(),
1431 0 : &endpoint.pg_address.to_string(),
1432 0 : &endpoint.timeline_id.to_string(),
1433 0 : branch_name,
1434 0 : lsn_str.as_str(),
1435 0 : &format!("{}", endpoint.status()),
1436 0 : ]);
1437 : }
1438 :
1439 0 : println!("{table}");
1440 : }
1441 0 : EndpointCmd::Create(args) => {
1442 0 : let tenant_id = get_tenant_id(args.tenant_id, env)?;
1443 0 : let branch_name = args
1444 0 : .branch_name
1445 0 : .clone()
1446 0 : .unwrap_or(DEFAULT_BRANCH_NAME.to_owned());
1447 0 : let endpoint_id = args
1448 0 : .endpoint_id
1449 0 : .clone()
1450 0 : .unwrap_or_else(|| format!("ep-{branch_name}"));
1451 :
1452 0 : let timeline_id = env
1453 0 : .get_branch_timeline_id(&branch_name, tenant_id)
1454 0 : .ok_or_else(|| anyhow!("Found no timeline id for branch name '{branch_name}'"))?;
1455 :
1456 0 : let mode = match (args.lsn, args.hot_standby) {
1457 0 : (Some(lsn), false) => ComputeMode::Static(lsn),
1458 0 : (None, true) => ComputeMode::Replica,
1459 0 : (None, false) => ComputeMode::Primary,
1460 0 : (Some(_), true) => anyhow::bail!("cannot specify both lsn and hot-standby"),
1461 : };
1462 :
1463 0 : match (mode, args.hot_standby) {
1464 : (ComputeMode::Static(_), true) => {
1465 0 : bail!(
1466 0 : "Cannot start a node in hot standby mode when it is already configured as a static replica"
1467 : )
1468 : }
1469 : (ComputeMode::Primary, true) => {
1470 0 : bail!(
1471 0 : "Cannot start a node as a hot standby replica, it is already configured as primary node"
1472 : )
1473 : }
1474 0 : _ => {}
1475 : }
1476 :
1477 0 : if !args.allow_multiple {
1478 0 : cplane.check_conflicting_endpoints(mode, tenant_id, timeline_id)?;
1479 0 : }
1480 :
1481 0 : cplane.new_endpoint(
1482 0 : &endpoint_id,
1483 0 : tenant_id,
1484 0 : timeline_id,
1485 0 : args.pg_port,
1486 0 : args.external_http_port,
1487 0 : args.internal_http_port,
1488 0 : args.pg_version,
1489 0 : mode,
1490 0 : args.grpc,
1491 0 : !args.update_catalog,
1492 : false,
1493 0 : args.privileged_role_name.clone(),
1494 0 : )?;
1495 : }
1496 0 : EndpointCmd::Start(args) => {
1497 0 : let endpoint_id = &args.endpoint_id;
1498 0 : let pageserver_id = args.endpoint_pageserver_id;
1499 0 : let remote_ext_base_url = &args.remote_ext_base_url;
1500 :
1501 0 : let default_generation = env
1502 0 : .storage_controller
1503 0 : .timelines_onto_safekeepers
1504 0 : .then_some(1);
1505 0 : let safekeepers_generation = args
1506 0 : .safekeepers_generation
1507 0 : .or(default_generation)
1508 0 : .map(SafekeeperGeneration::new);
1509 : // If --safekeepers argument is given, use only the listed
1510 : // safekeeper nodes; otherwise all from the env.
1511 0 : let safekeepers = if let Some(safekeepers) = parse_safekeepers(&args.safekeepers)? {
1512 0 : safekeepers
1513 : } else {
1514 0 : env.safekeepers.iter().map(|sk| sk.id).collect()
1515 : };
1516 :
1517 0 : let endpoint = cplane
1518 0 : .endpoints
1519 0 : .get(endpoint_id.as_str())
1520 0 : .ok_or_else(|| anyhow::anyhow!("endpoint {endpoint_id} not found"))?;
1521 :
1522 0 : if !args.allow_multiple {
1523 0 : cplane.check_conflicting_endpoints(
1524 0 : endpoint.mode,
1525 0 : endpoint.tenant_id,
1526 0 : endpoint.timeline_id,
1527 0 : )?;
1528 0 : }
1529 :
1530 0 : let (pageservers, stripe_size) = if let Some(pageserver_id) = pageserver_id {
1531 0 : let conf = env.get_pageserver_conf(pageserver_id).unwrap();
1532 : // Use gRPC if requested.
1533 0 : let pageserver = if endpoint.grpc {
1534 0 : let grpc_addr = conf.listen_grpc_addr.as_ref().expect("bad config");
1535 0 : let (host, port) = parse_host_port(grpc_addr)?;
1536 0 : let port = port.unwrap_or(DEFAULT_PAGESERVER_GRPC_PORT);
1537 0 : (PageserverProtocol::Grpc, host, port)
1538 : } else {
1539 0 : let (host, port) = parse_host_port(&conf.listen_pg_addr)?;
1540 0 : let port = port.unwrap_or(5432);
1541 0 : (PageserverProtocol::Libpq, host, port)
1542 : };
1543 : // If caller is telling us what pageserver to use, this is not a tenant which is
1544 : // fully managed by storage controller, therefore not sharded.
1545 0 : (vec![pageserver], DEFAULT_STRIPE_SIZE)
1546 : } else {
1547 : // Look up the currently attached location of the tenant, and its striping metadata,
1548 : // to pass these on to postgres.
1549 0 : let storage_controller = StorageController::from_env(env);
1550 0 : let locate_result = storage_controller.tenant_locate(endpoint.tenant_id).await?;
1551 0 : let pageservers = futures::future::try_join_all(
1552 0 : locate_result.shards.into_iter().map(|shard| async move {
1553 0 : if let ComputeMode::Static(lsn) = endpoint.mode {
1554 : // Initialize LSN leases for static computes.
1555 0 : let conf = env.get_pageserver_conf(shard.node_id).unwrap();
1556 0 : let pageserver = PageServerNode::from_env(env, conf);
1557 :
1558 0 : pageserver
1559 0 : .http_client
1560 0 : .timeline_init_lsn_lease(shard.shard_id, endpoint.timeline_id, lsn)
1561 0 : .await?;
1562 0 : }
1563 :
1564 0 : let pageserver = if endpoint.grpc {
1565 : (
1566 0 : PageserverProtocol::Grpc,
1567 0 : Host::parse(&shard.listen_grpc_addr.expect("no gRPC address"))?,
1568 0 : shard.listen_grpc_port.expect("no gRPC port"),
1569 : )
1570 : } else {
1571 : (
1572 0 : PageserverProtocol::Libpq,
1573 0 : Host::parse(&shard.listen_pg_addr)?,
1574 0 : shard.listen_pg_port,
1575 : )
1576 : };
1577 0 : anyhow::Ok(pageserver)
1578 0 : }),
1579 : )
1580 0 : .await?;
1581 0 : let stripe_size = locate_result.shard_params.stripe_size;
1582 :
1583 0 : (pageservers, stripe_size)
1584 : };
1585 0 : assert!(!pageservers.is_empty());
1586 :
1587 0 : let ps_conf = env.get_pageserver_conf(DEFAULT_PAGESERVER_ID)?;
1588 0 : let auth_token = if matches!(ps_conf.pg_auth_type, AuthType::NeonJWT) {
1589 0 : let claims = Claims::new(Some(endpoint.tenant_id), Scope::Tenant);
1590 :
1591 0 : Some(env.generate_auth_token(&claims)?)
1592 : } else {
1593 0 : None
1594 : };
1595 :
1596 0 : let exp = (std::time::SystemTime::now().duration_since(std::time::UNIX_EPOCH)?
1597 0 : + Duration::from_secs(86400))
1598 0 : .as_secs();
1599 0 : let claims = endpoint_storage::claims::EndpointStorageClaims {
1600 0 : tenant_id: endpoint.tenant_id,
1601 0 : timeline_id: endpoint.timeline_id,
1602 0 : endpoint_id: endpoint_id.to_string(),
1603 0 : exp,
1604 0 : };
1605 :
1606 0 : let endpoint_storage_token = env.generate_auth_token(&claims)?;
1607 0 : let endpoint_storage_addr = env.endpoint_storage.listen_addr.to_string();
1608 :
1609 0 : let args = control_plane::endpoint::EndpointStartArgs {
1610 0 : auth_token,
1611 0 : endpoint_storage_token,
1612 0 : endpoint_storage_addr,
1613 0 : safekeepers_generation,
1614 0 : safekeepers,
1615 0 : pageservers,
1616 0 : remote_ext_base_url: remote_ext_base_url.clone(),
1617 0 : shard_stripe_size: stripe_size.0 as usize,
1618 0 : create_test_user: args.create_test_user,
1619 0 : start_timeout: args.start_timeout,
1620 0 : autoprewarm: args.autoprewarm,
1621 0 : offload_lfc_interval_seconds: args.offload_lfc_interval_seconds,
1622 0 : dev: args.dev,
1623 0 : };
1624 :
1625 0 : println!("Starting existing endpoint {endpoint_id}...");
1626 0 : endpoint.start(args).await?;
1627 : }
1628 0 : EndpointCmd::Reconfigure(args) => {
1629 0 : let endpoint_id = &args.endpoint_id;
1630 0 : let endpoint = cplane
1631 0 : .endpoints
1632 0 : .get(endpoint_id.as_str())
1633 0 : .with_context(|| format!("postgres endpoint {endpoint_id} is not found"))?;
1634 0 : let pageservers = if let Some(ps_id) = args.endpoint_pageserver_id {
1635 0 : let conf = env.get_pageserver_conf(ps_id)?;
1636 : // Use gRPC if requested.
1637 0 : let pageserver = if endpoint.grpc {
1638 0 : let grpc_addr = conf.listen_grpc_addr.as_ref().expect("bad config");
1639 0 : let (host, port) = parse_host_port(grpc_addr)?;
1640 0 : let port = port.unwrap_or(DEFAULT_PAGESERVER_GRPC_PORT);
1641 0 : (PageserverProtocol::Grpc, host, port)
1642 : } else {
1643 0 : let (host, port) = parse_host_port(&conf.listen_pg_addr)?;
1644 0 : let port = port.unwrap_or(5432);
1645 0 : (PageserverProtocol::Libpq, host, port)
1646 : };
1647 0 : vec![pageserver]
1648 : } else {
1649 0 : let storage_controller = StorageController::from_env(env);
1650 0 : storage_controller
1651 0 : .tenant_locate(endpoint.tenant_id)
1652 0 : .await?
1653 : .shards
1654 0 : .into_iter()
1655 0 : .map(|shard| {
1656 : // Use gRPC if requested.
1657 0 : if endpoint.grpc {
1658 0 : (
1659 0 : PageserverProtocol::Grpc,
1660 0 : Host::parse(&shard.listen_grpc_addr.expect("no gRPC address"))
1661 0 : .expect("bad hostname"),
1662 0 : shard.listen_grpc_port.expect("no gRPC port"),
1663 0 : )
1664 : } else {
1665 0 : (
1666 0 : PageserverProtocol::Libpq,
1667 0 : Host::parse(&shard.listen_pg_addr).expect("bad hostname"),
1668 0 : shard.listen_pg_port,
1669 0 : )
1670 : }
1671 0 : })
1672 0 : .collect::<Vec<_>>()
1673 : };
1674 : // If --safekeepers argument is given, use only the listed
1675 : // safekeeper nodes; otherwise all from the env.
1676 0 : let safekeepers = parse_safekeepers(&args.safekeepers)?;
1677 0 : endpoint
1678 0 : .reconfigure(Some(pageservers), None, safekeepers, None)
1679 0 : .await?;
1680 : }
1681 0 : EndpointCmd::Stop(args) => {
1682 0 : let endpoint_id = &args.endpoint_id;
1683 0 : let endpoint = cplane
1684 0 : .endpoints
1685 0 : .get(endpoint_id)
1686 0 : .with_context(|| format!("postgres endpoint {endpoint_id} is not found"))?;
1687 0 : match endpoint.stop(args.mode, args.destroy).await?.lsn {
1688 0 : Some(lsn) => println!("{lsn}"),
1689 0 : None => println!("null"),
1690 : }
1691 : }
1692 0 : EndpointCmd::GenerateJwt(args) => {
1693 0 : let endpoint = {
1694 0 : let endpoint_id = &args.endpoint_id;
1695 :
1696 0 : cplane
1697 0 : .endpoints
1698 0 : .get(endpoint_id)
1699 0 : .with_context(|| format!("postgres endpoint {endpoint_id} is not found"))?
1700 : };
1701 :
1702 0 : let jwt = endpoint.generate_jwt(args.scope)?;
1703 :
1704 0 : print!("{jwt}");
1705 : }
1706 : }
1707 :
1708 0 : Ok(())
1709 0 : }
1710 :
1711 : /// Parse --safekeepers as list of safekeeper ids.
1712 0 : fn parse_safekeepers(safekeepers_str: &Option<String>) -> Result<Option<Vec<NodeId>>> {
1713 0 : if let Some(safekeepers_str) = safekeepers_str {
1714 0 : let mut safekeepers: Vec<NodeId> = Vec::new();
1715 0 : for sk_id in safekeepers_str.split(',').map(str::trim) {
1716 0 : let sk_id = NodeId(
1717 0 : u64::from_str(sk_id)
1718 0 : .map_err(|_| anyhow!("invalid node ID \"{sk_id}\" in --safekeepers list"))?,
1719 : );
1720 0 : safekeepers.push(sk_id);
1721 : }
1722 0 : Ok(Some(safekeepers))
1723 : } else {
1724 0 : Ok(None)
1725 : }
1726 0 : }
1727 :
1728 0 : fn handle_mappings(subcmd: &MappingsCmd, env: &mut local_env::LocalEnv) -> Result<()> {
1729 0 : match subcmd {
1730 0 : MappingsCmd::Map(args) => {
1731 0 : env.register_branch_mapping(
1732 0 : args.branch_name.to_owned(),
1733 0 : args.tenant_id,
1734 0 : args.timeline_id,
1735 0 : )?;
1736 :
1737 0 : Ok(())
1738 : }
1739 : }
1740 0 : }
1741 :
1742 0 : fn get_pageserver(
1743 0 : env: &local_env::LocalEnv,
1744 0 : pageserver_id_arg: Option<NodeId>,
1745 0 : ) -> Result<PageServerNode> {
1746 0 : let node_id = pageserver_id_arg.unwrap_or(DEFAULT_PAGESERVER_ID);
1747 :
1748 0 : Ok(PageServerNode::from_env(
1749 0 : env,
1750 0 : env.get_pageserver_conf(node_id)?,
1751 : ))
1752 0 : }
1753 :
1754 0 : async fn handle_pageserver(subcmd: &PageserverCmd, env: &local_env::LocalEnv) -> Result<()> {
1755 0 : match subcmd {
1756 0 : PageserverCmd::Start(args) => {
1757 0 : if let Err(e) = get_pageserver(env, args.pageserver_id)?
1758 0 : .start(&args.start_timeout)
1759 0 : .await
1760 : {
1761 0 : eprintln!("pageserver start failed: {e}");
1762 0 : exit(1);
1763 0 : }
1764 : }
1765 :
1766 0 : PageserverCmd::Stop(args) => {
1767 0 : let immediate = match args.stop_mode {
1768 0 : StopMode::Fast => false,
1769 0 : StopMode::Immediate => true,
1770 : };
1771 0 : if let Err(e) = get_pageserver(env, args.pageserver_id)?.stop(immediate) {
1772 0 : eprintln!("pageserver stop failed: {e}");
1773 0 : exit(1);
1774 0 : }
1775 : }
1776 :
1777 0 : PageserverCmd::Restart(args) => {
1778 0 : let pageserver = get_pageserver(env, args.pageserver_id)?;
1779 : //TODO what shutdown strategy should we use here?
1780 0 : if let Err(e) = pageserver.stop(false) {
1781 0 : eprintln!("pageserver stop failed: {e}");
1782 0 : exit(1);
1783 0 : }
1784 :
1785 0 : if let Err(e) = pageserver.start(&args.start_timeout).await {
1786 0 : eprintln!("pageserver start failed: {e}");
1787 0 : exit(1);
1788 0 : }
1789 : }
1790 :
1791 0 : PageserverCmd::Status(args) => {
1792 0 : match get_pageserver(env, args.pageserver_id)?
1793 0 : .check_status()
1794 0 : .await
1795 : {
1796 0 : Ok(_) => println!("Page server is up and running"),
1797 0 : Err(err) => {
1798 0 : eprintln!("Page server is not available: {err}");
1799 0 : exit(1);
1800 : }
1801 : }
1802 : }
1803 : }
1804 0 : Ok(())
1805 0 : }
1806 :
1807 0 : async fn handle_storage_controller(
1808 0 : subcmd: &StorageControllerCmd,
1809 0 : env: &local_env::LocalEnv,
1810 0 : ) -> Result<()> {
1811 0 : let svc = StorageController::from_env(env);
1812 0 : match subcmd {
1813 0 : StorageControllerCmd::Start(args) => {
1814 0 : let start_args = NeonStorageControllerStartArgs {
1815 0 : instance_id: args.instance_id,
1816 0 : base_port: args.base_port,
1817 0 : start_timeout: args.start_timeout,
1818 0 : handle_ps_local_disk_loss: args.handle_ps_local_disk_loss,
1819 0 : };
1820 :
1821 0 : if let Err(e) = svc.start(start_args).await {
1822 0 : eprintln!("start failed: {e}");
1823 0 : exit(1);
1824 0 : }
1825 : }
1826 :
1827 0 : StorageControllerCmd::Stop(args) => {
1828 0 : let stop_args = NeonStorageControllerStopArgs {
1829 0 : instance_id: args.instance_id,
1830 0 : immediate: match args.stop_mode {
1831 0 : StopMode::Fast => false,
1832 0 : StopMode::Immediate => true,
1833 : },
1834 : };
1835 0 : if let Err(e) = svc.stop(stop_args).await {
1836 0 : eprintln!("stop failed: {e}");
1837 0 : exit(1);
1838 0 : }
1839 : }
1840 : }
1841 0 : Ok(())
1842 0 : }
1843 :
1844 0 : fn get_safekeeper(env: &local_env::LocalEnv, id: NodeId) -> Result<SafekeeperNode> {
1845 0 : if let Some(node) = env.safekeepers.iter().find(|node| node.id == id) {
1846 0 : Ok(SafekeeperNode::from_env(env, node))
1847 : } else {
1848 0 : bail!("could not find safekeeper {id}")
1849 : }
1850 0 : }
1851 :
1852 0 : async fn handle_safekeeper(subcmd: &SafekeeperCmd, env: &local_env::LocalEnv) -> Result<()> {
1853 0 : match subcmd {
1854 0 : SafekeeperCmd::Start(args) => {
1855 0 : let safekeeper = get_safekeeper(env, args.id)?;
1856 :
1857 0 : if let Err(e) = safekeeper.start(&args.extra_opt, &args.start_timeout).await {
1858 0 : eprintln!("safekeeper start failed: {e}");
1859 0 : exit(1);
1860 0 : }
1861 : }
1862 :
1863 0 : SafekeeperCmd::Stop(args) => {
1864 0 : let safekeeper = get_safekeeper(env, args.id)?;
1865 0 : let immediate = match args.stop_mode {
1866 0 : StopMode::Fast => false,
1867 0 : StopMode::Immediate => true,
1868 : };
1869 0 : if let Err(e) = safekeeper.stop(immediate) {
1870 0 : eprintln!("safekeeper stop failed: {e}");
1871 0 : exit(1);
1872 0 : }
1873 : }
1874 :
1875 0 : SafekeeperCmd::Restart(args) => {
1876 0 : let safekeeper = get_safekeeper(env, args.id)?;
1877 0 : let immediate = match args.stop_mode {
1878 0 : StopMode::Fast => false,
1879 0 : StopMode::Immediate => true,
1880 : };
1881 :
1882 0 : if let Err(e) = safekeeper.stop(immediate) {
1883 0 : eprintln!("safekeeper stop failed: {e}");
1884 0 : exit(1);
1885 0 : }
1886 :
1887 0 : if let Err(e) = safekeeper.start(&args.extra_opt, &args.start_timeout).await {
1888 0 : eprintln!("safekeeper start failed: {e}");
1889 0 : exit(1);
1890 0 : }
1891 : }
1892 : }
1893 0 : Ok(())
1894 0 : }
1895 :
1896 0 : async fn handle_endpoint_storage(
1897 0 : subcmd: &EndpointStorageCmd,
1898 0 : env: &local_env::LocalEnv,
1899 0 : ) -> Result<()> {
1900 : use EndpointStorageCmd::*;
1901 0 : let storage = EndpointStorage::from_env(env);
1902 :
1903 : // In tests like test_forward_compatibility or test_graceful_cluster_restart
1904 : // old neon binaries (without endpoint_storage) are present
1905 0 : if !storage.bin.exists() {
1906 0 : eprintln!(
1907 0 : "{} binary not found. Ignore if this is a compatibility test",
1908 : storage.bin
1909 : );
1910 0 : return Ok(());
1911 0 : }
1912 :
1913 0 : match subcmd {
1914 0 : Start(EndpointStorageStartCmd { start_timeout }) => {
1915 0 : if let Err(e) = storage.start(start_timeout).await {
1916 0 : eprintln!("endpoint_storage start failed: {e}");
1917 0 : exit(1);
1918 0 : }
1919 : }
1920 0 : Stop(EndpointStorageStopCmd { stop_mode }) => {
1921 0 : let immediate = match stop_mode {
1922 0 : StopMode::Fast => false,
1923 0 : StopMode::Immediate => true,
1924 : };
1925 0 : if let Err(e) = storage.stop(immediate) {
1926 0 : eprintln!("proxy stop failed: {e}");
1927 0 : exit(1);
1928 0 : }
1929 : }
1930 : };
1931 0 : Ok(())
1932 0 : }
1933 :
1934 0 : async fn handle_storage_broker(subcmd: &StorageBrokerCmd, env: &local_env::LocalEnv) -> Result<()> {
1935 0 : match subcmd {
1936 0 : StorageBrokerCmd::Start(args) => {
1937 0 : let storage_broker = StorageBroker::from_env(env);
1938 0 : if let Err(e) = storage_broker.start(&args.start_timeout).await {
1939 0 : eprintln!("broker start failed: {e}");
1940 0 : exit(1);
1941 0 : }
1942 : }
1943 :
1944 0 : StorageBrokerCmd::Stop(_args) => {
1945 : // FIXME: stop_mode unused
1946 0 : let storage_broker = StorageBroker::from_env(env);
1947 0 : if let Err(e) = storage_broker.stop() {
1948 0 : eprintln!("broker stop failed: {e}");
1949 0 : exit(1);
1950 0 : }
1951 : }
1952 : }
1953 0 : Ok(())
1954 0 : }
1955 :
1956 0 : async fn handle_start_all(
1957 0 : args: &StartCmdArgs,
1958 0 : env: &'static local_env::LocalEnv,
1959 0 : ) -> anyhow::Result<()> {
1960 : // FIXME: this was called "retry_timeout", is it right?
1961 0 : let Err(errors) = handle_start_all_impl(env, args.timeout).await else {
1962 0 : neon_start_status_check(env, args.timeout.as_ref())
1963 0 : .await
1964 0 : .context("status check after successful startup of all services")?;
1965 0 : return Ok(());
1966 : };
1967 :
1968 0 : eprintln!("startup failed because one or more services could not be started");
1969 :
1970 0 : for e in errors {
1971 0 : eprintln!("{e}");
1972 0 : let debug_repr = format!("{e:?}");
1973 0 : for line in debug_repr.lines() {
1974 0 : eprintln!(" {line}");
1975 0 : }
1976 : }
1977 :
1978 0 : try_stop_all(env, true).await;
1979 :
1980 0 : exit(2);
1981 0 : }
1982 :
1983 : /// Returns Ok() if and only if all services could be started successfully.
1984 : /// Otherwise, returns the list of errors that occurred during startup.
1985 0 : async fn handle_start_all_impl(
1986 0 : env: &'static local_env::LocalEnv,
1987 0 : retry_timeout: humantime::Duration,
1988 0 : ) -> Result<(), Vec<anyhow::Error>> {
1989 : // Endpoints are not started automatically
1990 :
1991 0 : let mut js = JoinSet::new();
1992 :
1993 : // force infalliblity through closure
1994 : #[allow(clippy::redundant_closure_call)]
1995 0 : (|| {
1996 0 : js.spawn(async move {
1997 0 : let storage_broker = StorageBroker::from_env(env);
1998 0 : storage_broker
1999 0 : .start(&retry_timeout)
2000 0 : .await
2001 0 : .map_err(|e| e.context("start storage_broker"))
2002 0 : });
2003 :
2004 0 : js.spawn(async move {
2005 0 : let storage_controller = StorageController::from_env(env);
2006 0 : storage_controller
2007 0 : .start(NeonStorageControllerStartArgs::with_default_instance_id(
2008 0 : retry_timeout,
2009 0 : ))
2010 0 : .await
2011 0 : .map_err(|e| e.context("start storage_controller"))
2012 0 : });
2013 :
2014 0 : for ps_conf in &env.pageservers {
2015 0 : js.spawn(async move {
2016 0 : let pageserver = PageServerNode::from_env(env, ps_conf);
2017 0 : pageserver
2018 0 : .start(&retry_timeout)
2019 0 : .await
2020 0 : .map_err(|e| e.context(format!("start pageserver {}", ps_conf.id)))
2021 0 : });
2022 : }
2023 :
2024 0 : for node in env.safekeepers.iter() {
2025 0 : js.spawn(async move {
2026 0 : let safekeeper = SafekeeperNode::from_env(env, node);
2027 0 : safekeeper
2028 0 : .start(&[], &retry_timeout)
2029 0 : .await
2030 0 : .map_err(|e| e.context(format!("start safekeeper {}", safekeeper.id)))
2031 0 : });
2032 : }
2033 :
2034 0 : js.spawn(async move {
2035 0 : EndpointStorage::from_env(env)
2036 0 : .start(&retry_timeout)
2037 0 : .await
2038 0 : .map_err(|e| e.context("start endpoint_storage"))
2039 0 : });
2040 : })();
2041 :
2042 0 : let mut errors = Vec::new();
2043 0 : while let Some(result) = js.join_next().await {
2044 0 : let result = result.expect("we don't panic or cancel the tasks");
2045 0 : if let Err(e) = result {
2046 0 : errors.push(e);
2047 0 : }
2048 : }
2049 :
2050 0 : if !errors.is_empty() {
2051 0 : return Err(errors);
2052 0 : }
2053 :
2054 0 : Ok(())
2055 0 : }
2056 :
2057 0 : async fn neon_start_status_check(
2058 0 : env: &local_env::LocalEnv,
2059 0 : retry_timeout: &Duration,
2060 0 : ) -> anyhow::Result<()> {
2061 : const RETRY_INTERVAL: Duration = Duration::from_millis(100);
2062 : const NOTICE_AFTER_RETRIES: Duration = Duration::from_secs(5);
2063 :
2064 0 : let storcon = StorageController::from_env(env);
2065 :
2066 0 : let retries = retry_timeout.as_millis() / RETRY_INTERVAL.as_millis();
2067 0 : let notice_after_retries = retry_timeout.as_millis() / NOTICE_AFTER_RETRIES.as_millis();
2068 :
2069 0 : println!("\nRunning neon status check");
2070 :
2071 0 : for retry in 0..retries {
2072 0 : if retry == notice_after_retries {
2073 0 : println!("\nNeon status check has not passed yet, continuing to wait")
2074 0 : }
2075 :
2076 0 : let mut passed = true;
2077 0 : let mut nodes = storcon.node_list().await?;
2078 0 : let mut pageservers = env.pageservers.clone();
2079 :
2080 0 : if nodes.len() != pageservers.len() {
2081 0 : continue;
2082 0 : }
2083 :
2084 0 : nodes.sort_by_key(|ps| ps.id);
2085 0 : pageservers.sort_by_key(|ps| ps.id);
2086 :
2087 0 : for (idx, pageserver) in pageservers.iter().enumerate() {
2088 0 : let node = &nodes[idx];
2089 0 : if node.id != pageserver.id {
2090 0 : passed = false;
2091 0 : break;
2092 0 : }
2093 :
2094 0 : if !matches!(node.availability, NodeAvailabilityWrapper::Active) {
2095 0 : passed = false;
2096 0 : break;
2097 0 : }
2098 : }
2099 :
2100 0 : if passed {
2101 0 : println!("\nNeon started and passed status check");
2102 0 : return Ok(());
2103 0 : }
2104 :
2105 0 : tokio::time::sleep(RETRY_INTERVAL).await;
2106 : }
2107 :
2108 0 : anyhow::bail!("\nNeon passed status check")
2109 0 : }
2110 :
2111 0 : async fn handle_stop_all(args: &StopCmdArgs, env: &local_env::LocalEnv) -> Result<()> {
2112 0 : let immediate = match args.mode {
2113 0 : StopMode::Fast => false,
2114 0 : StopMode::Immediate => true,
2115 : };
2116 :
2117 0 : try_stop_all(env, immediate).await;
2118 :
2119 0 : Ok(())
2120 0 : }
2121 :
2122 0 : async fn try_stop_all(env: &local_env::LocalEnv, immediate: bool) {
2123 0 : let mode = if immediate {
2124 0 : EndpointTerminateMode::Immediate
2125 : } else {
2126 0 : EndpointTerminateMode::Fast
2127 : };
2128 : // Stop all endpoints
2129 0 : match ComputeControlPlane::load(env.clone()) {
2130 0 : Ok(cplane) => {
2131 0 : for (_k, node) in cplane.endpoints {
2132 0 : if let Err(e) = node.stop(mode, false).await {
2133 0 : eprintln!("postgres stop failed: {e:#}");
2134 0 : }
2135 : }
2136 : }
2137 0 : Err(e) => {
2138 0 : eprintln!("postgres stop failed, could not restore control plane data from env: {e:#}")
2139 : }
2140 : }
2141 :
2142 0 : let storage = EndpointStorage::from_env(env);
2143 0 : if let Err(e) = storage.stop(immediate) {
2144 0 : eprintln!("endpoint_storage stop failed: {e:#}");
2145 0 : }
2146 :
2147 0 : for ps_conf in &env.pageservers {
2148 0 : let pageserver = PageServerNode::from_env(env, ps_conf);
2149 0 : if let Err(e) = pageserver.stop(immediate) {
2150 0 : eprintln!("pageserver {} stop failed: {:#}", ps_conf.id, e);
2151 0 : }
2152 : }
2153 :
2154 0 : for node in env.safekeepers.iter() {
2155 0 : let safekeeper = SafekeeperNode::from_env(env, node);
2156 0 : if let Err(e) = safekeeper.stop(immediate) {
2157 0 : eprintln!("safekeeper {} stop failed: {:#}", safekeeper.id, e);
2158 0 : }
2159 : }
2160 :
2161 0 : let storage_broker = StorageBroker::from_env(env);
2162 0 : if let Err(e) = storage_broker.stop() {
2163 0 : eprintln!("neon broker stop failed: {e:#}");
2164 0 : }
2165 :
2166 : // Stop all storage controller instances. In the most common case there's only one,
2167 : // but iterate though the base data directory in order to discover the instances.
2168 0 : let storcon_instances = env
2169 0 : .storage_controller_instances()
2170 0 : .await
2171 0 : .expect("Must inspect data dir");
2172 0 : for (instance_id, _instance_dir_path) in storcon_instances {
2173 0 : let storage_controller = StorageController::from_env(env);
2174 0 : let stop_args = NeonStorageControllerStopArgs {
2175 0 : instance_id,
2176 0 : immediate,
2177 0 : };
2178 :
2179 0 : if let Err(e) = storage_controller.stop(stop_args).await {
2180 0 : eprintln!("Storage controller instance {instance_id} stop failed: {e:#}");
2181 0 : }
2182 : }
2183 0 : }
|
