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 : RefreshConfiguration(EndpointRefreshConfigurationArgs),
564 : Stop(EndpointStopCmdArgs),
565 : UpdatePageservers(EndpointUpdatePageserversCmdArgs),
566 : GenerateJwt(EndpointGenerateJwtCmdArgs),
567 : }
568 :
569 : #[derive(clap::Args)]
570 : #[clap(about = "List endpoints")]
571 : struct EndpointListCmdArgs {
572 : #[clap(
573 : long = "tenant-id",
574 : help = "Tenant id. Represented as a hexadecimal string 32 symbols length"
575 : )]
576 : tenant_shard_id: Option<TenantShardId>,
577 : }
578 :
579 : #[derive(clap::Args)]
580 : #[clap(about = "Create a compute endpoint")]
581 : struct EndpointCreateCmdArgs {
582 : #[clap(
583 : long = "tenant-id",
584 : help = "Tenant id. Represented as a hexadecimal string 32 symbols length"
585 : )]
586 : tenant_id: Option<TenantId>,
587 :
588 : #[clap(help = "Postgres endpoint id")]
589 : endpoint_id: Option<String>,
590 : #[clap(long, help = "Name of the branch the endpoint will run on")]
591 : branch_name: Option<String>,
592 : #[clap(
593 : long,
594 : help = "Specify Lsn on the timeline to start from. By default, end of the timeline would be used"
595 : )]
596 : lsn: Option<Lsn>,
597 : #[clap(long)]
598 : pg_port: Option<u16>,
599 : #[clap(long, alias = "http-port")]
600 : external_http_port: Option<u16>,
601 : #[clap(long)]
602 : internal_http_port: Option<u16>,
603 : #[clap(long = "pageserver-id")]
604 : endpoint_pageserver_id: Option<NodeId>,
605 :
606 : #[clap(
607 : long,
608 : help = "Don't do basebackup, create endpoint directory with only config files",
609 : action = clap::ArgAction::Set,
610 : default_value_t = false
611 : )]
612 : config_only: bool,
613 :
614 : #[arg(default_value = DEFAULT_PG_VERSION_NUM)]
615 : #[clap(long, help = "Postgres version")]
616 : pg_version: PgMajorVersion,
617 :
618 : /// Use gRPC to communicate with Pageservers, by generating grpc:// connstrings.
619 : ///
620 : /// Specified on creation such that it's retained across reconfiguration and restarts.
621 : ///
622 : /// NB: not yet supported by computes.
623 : #[clap(long)]
624 : grpc: bool,
625 :
626 : #[clap(
627 : long,
628 : help = "If set, the node will be a hot replica on the specified timeline",
629 : action = clap::ArgAction::Set,
630 : default_value_t = false
631 : )]
632 : hot_standby: bool,
633 :
634 : #[clap(long, help = "If set, will set up the catalog for neon_superuser")]
635 : update_catalog: bool,
636 :
637 : #[clap(
638 : long,
639 : help = "Allow multiple primary endpoints running on the same branch. Shouldn't be used normally, but useful for tests."
640 : )]
641 : allow_multiple: bool,
642 :
643 : /// Only allow changing it on creation
644 : #[clap(long, help = "Name of the privileged role for the endpoint")]
645 : privileged_role_name: Option<String>,
646 : }
647 :
648 : #[derive(clap::Args)]
649 : #[clap(about = "Start postgres. If the endpoint doesn't exist yet, it is created.")]
650 : struct EndpointStartCmdArgs {
651 : #[clap(help = "Postgres endpoint id")]
652 : endpoint_id: String,
653 : #[clap(long = "pageserver-id")]
654 : endpoint_pageserver_id: Option<NodeId>,
655 :
656 : #[clap(
657 : long,
658 : 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."
659 : )]
660 : safekeepers_generation: Option<u32>,
661 : #[clap(
662 : long,
663 : help = "List of safekeepers endpoint will talk to. Normally neon_local chooses them on its own, but this option allows to override."
664 : )]
665 : safekeepers: Option<String>,
666 :
667 : #[clap(
668 : long,
669 : help = "Configure the remote extensions storage proxy gateway URL to request for extensions.",
670 : alias = "remote-ext-config"
671 : )]
672 : remote_ext_base_url: Option<String>,
673 :
674 : #[clap(
675 : long,
676 : help = "If set, will create test user `user` and `neondb` database. Requires `update-catalog = true`"
677 : )]
678 : create_test_user: bool,
679 :
680 : #[clap(
681 : long,
682 : help = "Allow multiple primary endpoints running on the same branch. Shouldn't be used normally, but useful for tests."
683 : )]
684 : allow_multiple: bool,
685 :
686 : #[clap(short = 't', long, value_parser= humantime::parse_duration, help = "timeout until we fail the command")]
687 : #[arg(default_value = "90s")]
688 : start_timeout: Duration,
689 :
690 : #[clap(
691 : long,
692 : help = "Download LFC cache from endpoint storage on endpoint startup",
693 : default_value = "false"
694 : )]
695 : autoprewarm: bool,
696 :
697 : #[clap(long, help = "Upload LFC cache to endpoint storage periodically")]
698 : offload_lfc_interval_seconds: Option<std::num::NonZeroU64>,
699 :
700 : #[clap(
701 : long,
702 : help = "Run in development mode, skipping VM-specific operations like process termination",
703 : action = clap::ArgAction::SetTrue
704 : )]
705 : dev: bool,
706 : }
707 :
708 : #[derive(clap::Args)]
709 : #[clap(about = "Reconfigure an endpoint")]
710 : struct EndpointReconfigureCmdArgs {
711 : #[clap(
712 : long = "tenant-id",
713 : help = "Tenant id. Represented as a hexadecimal string 32 symbols length"
714 : )]
715 : tenant_id: Option<TenantId>,
716 :
717 : #[clap(help = "Postgres endpoint id")]
718 : endpoint_id: String,
719 : #[clap(long = "pageserver-id")]
720 : endpoint_pageserver_id: Option<NodeId>,
721 :
722 : #[clap(long)]
723 : safekeepers: Option<String>,
724 : }
725 :
726 : #[derive(clap::Args)]
727 : #[clap(about = "Refresh the endpoint's configuration by forcing it reload it's spec")]
728 : struct EndpointRefreshConfigurationArgs {
729 : #[clap(help = "Postgres endpoint id")]
730 : endpoint_id: String,
731 : }
732 :
733 : #[derive(clap::Args)]
734 : #[clap(about = "Stop an endpoint")]
735 : struct EndpointStopCmdArgs {
736 : #[clap(help = "Postgres endpoint id")]
737 : endpoint_id: String,
738 :
739 : #[clap(
740 : long,
741 : help = "Also delete data directory (now optional, should be default in future)"
742 : )]
743 : destroy: bool,
744 :
745 : #[clap(long, help = "Postgres shutdown mode")]
746 : #[clap(default_value = "fast")]
747 : mode: EndpointTerminateMode,
748 : }
749 :
750 : #[derive(clap::Args)]
751 : #[clap(about = "Update the pageservers in the spec file of the compute endpoint")]
752 : struct EndpointUpdatePageserversCmdArgs {
753 : #[clap(help = "Postgres endpoint id")]
754 : endpoint_id: String,
755 :
756 : #[clap(short = 'p', long, help = "Specified pageserver id")]
757 : pageserver_id: Option<NodeId>,
758 : }
759 :
760 : #[derive(clap::Args)]
761 : #[clap(about = "Generate a JWT for an endpoint")]
762 : struct EndpointGenerateJwtCmdArgs {
763 : #[clap(help = "Postgres endpoint id")]
764 : endpoint_id: String,
765 :
766 : #[clap(short = 's', long, help = "Scope to generate the JWT with", value_parser = ComputeClaimsScope::from_str)]
767 : scope: Option<ComputeClaimsScope>,
768 : }
769 :
770 : #[derive(clap::Subcommand)]
771 : #[clap(about = "Manage neon_local branch name mappings")]
772 : enum MappingsCmd {
773 : Map(MappingsMapCmdArgs),
774 : }
775 :
776 : #[derive(clap::Args)]
777 : #[clap(about = "Create new mapping which cannot exist already")]
778 : struct MappingsMapCmdArgs {
779 : #[clap(
780 : long,
781 : help = "Tenant id. Represented as a hexadecimal string 32 symbols length"
782 : )]
783 : tenant_id: TenantId,
784 : #[clap(
785 : long,
786 : help = "Timeline id. Represented as a hexadecimal string 32 symbols length"
787 : )]
788 : timeline_id: TimelineId,
789 : #[clap(long, help = "Branch name to give to the timeline")]
790 : branch_name: String,
791 : }
792 :
793 : ///
794 : /// Timelines tree element used as a value in the HashMap.
795 : ///
796 : struct TimelineTreeEl {
797 : /// `TimelineInfo` received from the `pageserver` via the `timeline_list` http API call.
798 : pub info: TimelineInfo,
799 : /// Name, recovered from neon config mappings
800 : pub name: Option<String>,
801 : /// Holds all direct children of this timeline referenced using `timeline_id`.
802 : pub children: BTreeSet<TimelineId>,
803 : }
804 :
805 : /// A flock-based guard over the neon_local repository directory
806 : struct RepoLock {
807 : _file: Flock<File>,
808 : }
809 :
810 : impl RepoLock {
811 0 : fn new() -> Result<Self> {
812 0 : let repo_dir = File::open(local_env::base_path())?;
813 0 : match Flock::lock(repo_dir, FlockArg::LockExclusive) {
814 0 : Ok(f) => Ok(Self { _file: f }),
815 0 : Err((_, e)) => Err(e).context("flock error"),
816 : }
817 0 : }
818 : }
819 :
820 : // Main entry point for the 'neon_local' CLI utility
821 : //
822 : // This utility helps to manage neon installation. That includes following:
823 : // * Management of local postgres installations running on top of the
824 : // pageserver.
825 : // * Providing CLI api to the pageserver
826 : // * TODO: export/import to/from usual postgres
827 0 : fn main() -> Result<()> {
828 0 : let cli = Cli::parse();
829 :
830 : // Check for 'neon init' command first.
831 0 : let (subcommand_result, _lock) = if let NeonLocalCmd::Init(args) = cli.command {
832 0 : (handle_init(&args).map(|env| Some(Cow::Owned(env))), None)
833 : } else {
834 : // This tool uses a collection of simple files to store its state, and consequently
835 : // it is not generally safe to run multiple commands concurrently. Rather than expect
836 : // all callers to know this, use a lock file to protect against concurrent execution.
837 0 : let _repo_lock = RepoLock::new().unwrap();
838 :
839 : // all other commands need an existing config
840 0 : let env = LocalEnv::load_config(&local_env::base_path()).context("Error loading config")?;
841 0 : let original_env = env.clone();
842 0 : let env = Box::leak(Box::new(env));
843 0 : let rt = tokio::runtime::Builder::new_current_thread()
844 0 : .enable_all()
845 0 : .build()
846 0 : .unwrap();
847 :
848 0 : let subcommand_result = match cli.command {
849 0 : NeonLocalCmd::Init(_) => unreachable!("init was handled earlier already"),
850 0 : NeonLocalCmd::Start(args) => rt.block_on(handle_start_all(&args, env)),
851 0 : NeonLocalCmd::Stop(args) => rt.block_on(handle_stop_all(&args, env)),
852 0 : NeonLocalCmd::Tenant(subcmd) => rt.block_on(handle_tenant(&subcmd, env)),
853 0 : NeonLocalCmd::Timeline(subcmd) => rt.block_on(handle_timeline(&subcmd, env)),
854 0 : NeonLocalCmd::Pageserver(subcmd) => rt.block_on(handle_pageserver(&subcmd, env)),
855 0 : NeonLocalCmd::StorageController(subcmd) => {
856 0 : rt.block_on(handle_storage_controller(&subcmd, env))
857 : }
858 0 : NeonLocalCmd::StorageBroker(subcmd) => rt.block_on(handle_storage_broker(&subcmd, env)),
859 0 : NeonLocalCmd::Safekeeper(subcmd) => rt.block_on(handle_safekeeper(&subcmd, env)),
860 0 : NeonLocalCmd::EndpointStorage(subcmd) => {
861 0 : rt.block_on(handle_endpoint_storage(&subcmd, env))
862 : }
863 0 : NeonLocalCmd::Endpoint(subcmd) => rt.block_on(handle_endpoint(&subcmd, env)),
864 0 : NeonLocalCmd::Mappings(subcmd) => handle_mappings(&subcmd, env),
865 : };
866 :
867 0 : let subcommand_result = if &original_env != env {
868 0 : subcommand_result.map(|()| Some(Cow::Borrowed(env)))
869 : } else {
870 0 : subcommand_result.map(|()| None)
871 : };
872 0 : (subcommand_result, Some(_repo_lock))
873 : };
874 :
875 0 : match subcommand_result {
876 0 : Ok(Some(updated_env)) => updated_env.persist_config()?,
877 0 : Ok(None) => (),
878 0 : Err(e) => {
879 0 : eprintln!("command failed: {e:?}");
880 0 : exit(1);
881 : }
882 : }
883 0 : Ok(())
884 0 : }
885 :
886 : ///
887 : /// Prints timelines list as a tree-like structure.
888 : ///
889 0 : fn print_timelines_tree(
890 0 : timelines: Vec<TimelineInfo>,
891 0 : mut timeline_name_mappings: HashMap<TenantTimelineId, String>,
892 0 : ) -> Result<()> {
893 0 : let mut timelines_hash = timelines
894 0 : .iter()
895 0 : .map(|t| {
896 0 : (
897 0 : t.timeline_id,
898 0 : TimelineTreeEl {
899 0 : info: t.clone(),
900 0 : children: BTreeSet::new(),
901 0 : name: timeline_name_mappings
902 0 : .remove(&TenantTimelineId::new(t.tenant_id.tenant_id, t.timeline_id)),
903 0 : },
904 0 : )
905 0 : })
906 0 : .collect::<HashMap<_, _>>();
907 :
908 : // Memorize all direct children of each timeline.
909 0 : for timeline in timelines.iter() {
910 0 : if let Some(ancestor_timeline_id) = timeline.ancestor_timeline_id {
911 0 : timelines_hash
912 0 : .get_mut(&ancestor_timeline_id)
913 0 : .context("missing timeline info in the HashMap")?
914 : .children
915 0 : .insert(timeline.timeline_id);
916 0 : }
917 : }
918 :
919 0 : for timeline in timelines_hash.values() {
920 : // Start with root local timelines (no ancestors) first.
921 0 : if timeline.info.ancestor_timeline_id.is_none() {
922 0 : print_timeline(0, &Vec::from([true]), timeline, &timelines_hash)?;
923 0 : }
924 : }
925 :
926 0 : Ok(())
927 0 : }
928 :
929 : ///
930 : /// Recursively prints timeline info with all its children.
931 : ///
932 0 : fn print_timeline(
933 0 : nesting_level: usize,
934 0 : is_last: &[bool],
935 0 : timeline: &TimelineTreeEl,
936 0 : timelines: &HashMap<TimelineId, TimelineTreeEl>,
937 0 : ) -> Result<()> {
938 0 : if nesting_level > 0 {
939 0 : let ancestor_lsn = match timeline.info.ancestor_lsn {
940 0 : Some(lsn) => lsn.to_string(),
941 0 : None => "Unknown Lsn".to_string(),
942 : };
943 :
944 0 : let mut br_sym = "┣━";
945 :
946 : // Draw each nesting padding with proper style
947 : // depending on whether its timeline ended or not.
948 0 : if nesting_level > 1 {
949 0 : for l in &is_last[1..is_last.len() - 1] {
950 0 : if *l {
951 0 : print!(" ");
952 0 : } else {
953 0 : print!("┃ ");
954 0 : }
955 : }
956 0 : }
957 :
958 : // We are the last in this sub-timeline
959 0 : if *is_last.last().unwrap() {
960 0 : br_sym = "┗━";
961 0 : }
962 :
963 0 : print!("{br_sym} @{ancestor_lsn}: ");
964 0 : }
965 :
966 : // Finally print a timeline id and name with new line
967 0 : println!(
968 0 : "{} [{}]",
969 0 : timeline.name.as_deref().unwrap_or("_no_name_"),
970 : timeline.info.timeline_id
971 : );
972 :
973 0 : let len = timeline.children.len();
974 0 : let mut i: usize = 0;
975 0 : let mut is_last_new = Vec::from(is_last);
976 0 : is_last_new.push(false);
977 :
978 0 : for child in &timeline.children {
979 0 : i += 1;
980 :
981 : // Mark that the last padding is the end of the timeline
982 0 : if i == len {
983 0 : if let Some(last) = is_last_new.last_mut() {
984 0 : *last = true;
985 0 : }
986 0 : }
987 :
988 0 : print_timeline(
989 0 : nesting_level + 1,
990 0 : &is_last_new,
991 0 : timelines
992 0 : .get(child)
993 0 : .context("missing timeline info in the HashMap")?,
994 0 : timelines,
995 0 : )?;
996 : }
997 :
998 0 : Ok(())
999 0 : }
1000 :
1001 : /// Helper function to get tenant id from an optional --tenant_id option or from the config file
1002 0 : fn get_tenant_id(
1003 0 : tenant_id_arg: Option<TenantId>,
1004 0 : env: &local_env::LocalEnv,
1005 0 : ) -> anyhow::Result<TenantId> {
1006 0 : if let Some(tenant_id_from_arguments) = tenant_id_arg {
1007 0 : Ok(tenant_id_from_arguments)
1008 0 : } else if let Some(default_id) = env.default_tenant_id {
1009 0 : Ok(default_id)
1010 : } else {
1011 0 : anyhow::bail!("No tenant id. Use --tenant-id, or set a default tenant");
1012 : }
1013 0 : }
1014 :
1015 : /// Helper function to get tenant-shard ID from an optional --tenant_id option or from the config file,
1016 : /// for commands that accept a shard suffix
1017 0 : fn get_tenant_shard_id(
1018 0 : tenant_shard_id_arg: Option<TenantShardId>,
1019 0 : env: &local_env::LocalEnv,
1020 0 : ) -> anyhow::Result<TenantShardId> {
1021 0 : if let Some(tenant_id_from_arguments) = tenant_shard_id_arg {
1022 0 : Ok(tenant_id_from_arguments)
1023 0 : } else if let Some(default_id) = env.default_tenant_id {
1024 0 : Ok(TenantShardId::unsharded(default_id))
1025 : } else {
1026 0 : anyhow::bail!("No tenant shard id. Use --tenant-id, or set a default tenant");
1027 : }
1028 0 : }
1029 :
1030 0 : fn handle_init(args: &InitCmdArgs) -> anyhow::Result<LocalEnv> {
1031 : // Create the in-memory `LocalEnv` that we'd normally load from disk in `load_config`.
1032 0 : let init_conf: NeonLocalInitConf = if let Some(config_path) = &args.config {
1033 : // User (likely the Python test suite) provided a description of the environment.
1034 0 : if args.num_pageservers.is_some() {
1035 0 : bail!(
1036 0 : "Cannot specify both --num-pageservers and --config, use key `pageservers` in the --config file instead"
1037 : );
1038 0 : }
1039 : // load and parse the file
1040 0 : let contents = std::fs::read_to_string(config_path).with_context(|| {
1041 0 : format!(
1042 0 : "Could not read configuration file '{}'",
1043 0 : config_path.display()
1044 : )
1045 0 : })?;
1046 0 : toml_edit::de::from_str(&contents)?
1047 : } else {
1048 : // User (likely interactive) did not provide a description of the environment, give them the default
1049 : NeonLocalInitConf {
1050 0 : control_plane_api: Some(DEFAULT_PAGESERVER_CONTROL_PLANE_API.parse().unwrap()),
1051 0 : broker: NeonBroker {
1052 0 : listen_addr: Some(DEFAULT_BROKER_ADDR.parse().unwrap()),
1053 0 : listen_https_addr: None,
1054 0 : },
1055 0 : safekeepers: vec![SafekeeperConf {
1056 0 : id: DEFAULT_SAFEKEEPER_ID,
1057 0 : pg_port: DEFAULT_SAFEKEEPER_PG_PORT,
1058 0 : http_port: DEFAULT_SAFEKEEPER_HTTP_PORT,
1059 0 : ..Default::default()
1060 0 : }],
1061 0 : pageservers: (0..args.num_pageservers.unwrap_or(1))
1062 0 : .map(|i| {
1063 0 : let pageserver_id = NodeId(DEFAULT_PAGESERVER_ID.0 + i as u64);
1064 0 : let pg_port = DEFAULT_PAGESERVER_PG_PORT + i;
1065 0 : let http_port = DEFAULT_PAGESERVER_HTTP_PORT + i;
1066 0 : let grpc_port = DEFAULT_PAGESERVER_GRPC_PORT + i;
1067 0 : NeonLocalInitPageserverConf {
1068 0 : id: pageserver_id,
1069 0 : listen_pg_addr: format!("127.0.0.1:{pg_port}"),
1070 0 : listen_http_addr: format!("127.0.0.1:{http_port}"),
1071 0 : listen_https_addr: None,
1072 0 : listen_grpc_addr: Some(format!("127.0.0.1:{grpc_port}")),
1073 0 : pg_auth_type: AuthType::Trust,
1074 0 : http_auth_type: AuthType::Trust,
1075 0 : grpc_auth_type: AuthType::Trust,
1076 0 : other: Default::default(),
1077 0 : // Typical developer machines use disks with slow fsync, and we don't care
1078 0 : // about data integrity: disable disk syncs.
1079 0 : no_sync: true,
1080 0 : }
1081 0 : })
1082 0 : .collect(),
1083 0 : endpoint_storage: EndpointStorageConf {
1084 0 : listen_addr: ENDPOINT_STORAGE_DEFAULT_ADDR,
1085 0 : },
1086 0 : pg_distrib_dir: None,
1087 0 : neon_distrib_dir: None,
1088 0 : default_tenant_id: TenantId::from_array(std::array::from_fn(|_| 0)),
1089 0 : storage_controller: None,
1090 0 : control_plane_hooks_api: None,
1091 : generate_local_ssl_certs: false,
1092 : }
1093 : };
1094 :
1095 0 : LocalEnv::init(init_conf, &args.force)
1096 0 : .context("materialize initial neon_local environment on disk")?;
1097 0 : Ok(LocalEnv::load_config(&local_env::base_path())
1098 0 : .expect("freshly written config should be loadable"))
1099 0 : }
1100 :
1101 : /// The default pageserver is the one where CLI tenant/timeline operations are sent by default.
1102 : /// For typical interactive use, one would just run with a single pageserver. Scenarios with
1103 : /// tenant/timeline placement across multiple pageservers are managed by python test code rather
1104 : /// than this CLI.
1105 0 : fn get_default_pageserver(env: &local_env::LocalEnv) -> PageServerNode {
1106 0 : let ps_conf = env
1107 0 : .pageservers
1108 0 : .first()
1109 0 : .expect("Config is validated to contain at least one pageserver");
1110 0 : PageServerNode::from_env(env, ps_conf)
1111 0 : }
1112 :
1113 0 : async fn handle_tenant(subcmd: &TenantCmd, env: &mut local_env::LocalEnv) -> anyhow::Result<()> {
1114 0 : let pageserver = get_default_pageserver(env);
1115 0 : match subcmd {
1116 : TenantCmd::List => {
1117 0 : for t in pageserver.tenant_list().await? {
1118 0 : println!("{} {:?}", t.id, t.state);
1119 0 : }
1120 : }
1121 0 : TenantCmd::Import(args) => {
1122 0 : let tenant_id = args.tenant_id;
1123 :
1124 0 : let storage_controller = StorageController::from_env(env);
1125 0 : let create_response = storage_controller.tenant_import(tenant_id).await?;
1126 :
1127 0 : let shard_zero = create_response
1128 0 : .shards
1129 0 : .first()
1130 0 : .expect("Import response omitted shards");
1131 :
1132 0 : let attached_pageserver_id = shard_zero.node_id;
1133 0 : let pageserver =
1134 0 : PageServerNode::from_env(env, env.get_pageserver_conf(attached_pageserver_id)?);
1135 :
1136 0 : println!(
1137 0 : "Imported tenant {tenant_id}, attached to pageserver {attached_pageserver_id}"
1138 : );
1139 :
1140 0 : let timelines = pageserver
1141 0 : .http_client
1142 0 : .list_timelines(shard_zero.shard_id)
1143 0 : .await?;
1144 :
1145 : // Pick a 'main' timeline that has no ancestors, the rest will get arbitrary names
1146 0 : let main_timeline = timelines
1147 0 : .iter()
1148 0 : .find(|t| t.ancestor_timeline_id.is_none())
1149 0 : .expect("No timelines found")
1150 : .timeline_id;
1151 :
1152 0 : let mut branch_i = 0;
1153 0 : for timeline in timelines.iter() {
1154 0 : let branch_name = if timeline.timeline_id == main_timeline {
1155 0 : "main".to_string()
1156 : } else {
1157 0 : branch_i += 1;
1158 0 : format!("branch_{branch_i}")
1159 : };
1160 :
1161 0 : println!(
1162 0 : "Importing timeline {tenant_id}/{} as branch {branch_name}",
1163 : timeline.timeline_id
1164 : );
1165 :
1166 0 : env.register_branch_mapping(branch_name, tenant_id, timeline.timeline_id)?;
1167 : }
1168 : }
1169 0 : TenantCmd::Create(args) => {
1170 0 : let tenant_conf: HashMap<_, _> =
1171 0 : args.config.iter().flat_map(|c| c.split_once(':')).collect();
1172 :
1173 0 : let tenant_conf = PageServerNode::parse_config(tenant_conf)?;
1174 :
1175 : // If tenant ID was not specified, generate one
1176 0 : let tenant_id = args.tenant_id.unwrap_or_else(TenantId::generate);
1177 :
1178 : // We must register the tenant with the storage controller, so
1179 : // that when the pageserver restarts, it will be re-attached.
1180 0 : let storage_controller = StorageController::from_env(env);
1181 0 : storage_controller
1182 0 : .tenant_create(TenantCreateRequest {
1183 0 : // Note that ::unsharded here isn't actually because the tenant is unsharded, its because the
1184 0 : // storage controller expects a shard-naive tenant_id in this attribute, and the TenantCreateRequest
1185 0 : // type is used both in the storage controller (for creating tenants) and in the pageserver (for
1186 0 : // creating shards)
1187 0 : new_tenant_id: TenantShardId::unsharded(tenant_id),
1188 0 : generation: None,
1189 0 : shard_parameters: ShardParameters {
1190 0 : count: ShardCount::new(args.shard_count),
1191 0 : stripe_size: args
1192 0 : .shard_stripe_size
1193 0 : .map(ShardStripeSize)
1194 0 : .unwrap_or(DEFAULT_STRIPE_SIZE),
1195 0 : },
1196 0 : placement_policy: args.placement_policy.clone(),
1197 0 : config: tenant_conf,
1198 0 : })
1199 0 : .await?;
1200 0 : println!("tenant {tenant_id} successfully created on the pageserver");
1201 :
1202 : // Create an initial timeline for the new tenant
1203 0 : let new_timeline_id = args.timeline_id.unwrap_or(TimelineId::generate());
1204 :
1205 : // FIXME: passing None for ancestor_start_lsn is not kosher in a sharded world: we can't have
1206 : // different shards picking different start lsns. Maybe we have to teach storage controller
1207 : // to let shard 0 branch first and then propagate the chosen LSN to other shards.
1208 0 : storage_controller
1209 0 : .tenant_timeline_create(
1210 0 : tenant_id,
1211 0 : TimelineCreateRequest {
1212 0 : new_timeline_id,
1213 0 : mode: pageserver_api::models::TimelineCreateRequestMode::Bootstrap {
1214 0 : existing_initdb_timeline_id: None,
1215 0 : pg_version: Some(args.pg_version),
1216 0 : },
1217 0 : },
1218 0 : )
1219 0 : .await?;
1220 :
1221 0 : env.register_branch_mapping(
1222 0 : DEFAULT_BRANCH_NAME.to_string(),
1223 0 : tenant_id,
1224 0 : new_timeline_id,
1225 0 : )?;
1226 :
1227 0 : println!("Created an initial timeline '{new_timeline_id}' for tenant: {tenant_id}",);
1228 :
1229 0 : if args.set_default {
1230 0 : println!("Setting tenant {tenant_id} as a default one");
1231 0 : env.default_tenant_id = Some(tenant_id);
1232 0 : }
1233 : }
1234 0 : TenantCmd::SetDefault(args) => {
1235 0 : println!("Setting tenant {} as a default one", args.tenant_id);
1236 0 : env.default_tenant_id = Some(args.tenant_id);
1237 0 : }
1238 0 : TenantCmd::Config(args) => {
1239 0 : let tenant_id = get_tenant_id(args.tenant_id, env)?;
1240 0 : let tenant_conf: HashMap<_, _> =
1241 0 : args.config.iter().flat_map(|c| c.split_once(':')).collect();
1242 0 : let config = PageServerNode::parse_config(tenant_conf)?;
1243 :
1244 0 : let req = TenantConfigRequest { tenant_id, config };
1245 :
1246 0 : let storage_controller = StorageController::from_env(env);
1247 0 : storage_controller
1248 0 : .set_tenant_config(&req)
1249 0 : .await
1250 0 : .with_context(|| format!("Tenant config failed for tenant with id {tenant_id}"))?;
1251 0 : println!("tenant {tenant_id} successfully configured via storcon");
1252 : }
1253 : }
1254 0 : Ok(())
1255 0 : }
1256 :
1257 0 : async fn handle_timeline(cmd: &TimelineCmd, env: &mut local_env::LocalEnv) -> Result<()> {
1258 0 : let pageserver = get_default_pageserver(env);
1259 :
1260 0 : match cmd {
1261 0 : TimelineCmd::List(args) => {
1262 : // TODO(sharding): this command shouldn't have to specify a shard ID: we should ask the storage controller
1263 : // where shard 0 is attached, and query there.
1264 0 : let tenant_shard_id = get_tenant_shard_id(args.tenant_shard_id, env)?;
1265 0 : let timelines = pageserver.timeline_list(&tenant_shard_id).await?;
1266 0 : print_timelines_tree(timelines, env.timeline_name_mappings())?;
1267 : }
1268 0 : TimelineCmd::Create(args) => {
1269 0 : let tenant_id = get_tenant_id(args.tenant_id, env)?;
1270 0 : let new_branch_name = &args.branch_name;
1271 0 : let new_timeline_id_opt = args.timeline_id;
1272 0 : let new_timeline_id = new_timeline_id_opt.unwrap_or(TimelineId::generate());
1273 :
1274 0 : let storage_controller = StorageController::from_env(env);
1275 0 : let create_req = TimelineCreateRequest {
1276 0 : new_timeline_id,
1277 0 : mode: pageserver_api::models::TimelineCreateRequestMode::Bootstrap {
1278 0 : existing_initdb_timeline_id: None,
1279 0 : pg_version: Some(args.pg_version),
1280 0 : },
1281 0 : };
1282 0 : let timeline_info = storage_controller
1283 0 : .tenant_timeline_create(tenant_id, create_req)
1284 0 : .await?;
1285 :
1286 0 : let last_record_lsn = timeline_info.last_record_lsn;
1287 0 : env.register_branch_mapping(new_branch_name.to_string(), tenant_id, new_timeline_id)?;
1288 :
1289 0 : println!(
1290 0 : "Created timeline '{}' at Lsn {last_record_lsn} for tenant: {tenant_id}",
1291 : timeline_info.timeline_id
1292 : );
1293 : }
1294 : // TODO: rename to import-basebackup-plus-wal
1295 0 : TimelineCmd::Import(args) => {
1296 0 : let tenant_id = get_tenant_id(args.tenant_id, env)?;
1297 0 : let timeline_id = args.timeline_id;
1298 0 : let branch_name = &args.branch_name;
1299 :
1300 : // Parse base inputs
1301 0 : let base = (args.base_lsn, args.base_tarfile.clone());
1302 :
1303 : // Parse pg_wal inputs
1304 0 : let wal_tarfile = args.wal_tarfile.clone();
1305 0 : let end_lsn = args.end_lsn;
1306 : // TODO validate both or none are provided
1307 0 : let pg_wal = end_lsn.zip(wal_tarfile);
1308 :
1309 0 : println!("Importing timeline into pageserver ...");
1310 0 : pageserver
1311 0 : .timeline_import(tenant_id, timeline_id, base, pg_wal, args.pg_version)
1312 0 : .await?;
1313 0 : if env.storage_controller.timelines_onto_safekeepers {
1314 0 : println!("Creating timeline on safekeeper ...");
1315 0 : let timeline_info = pageserver
1316 0 : .timeline_info(
1317 0 : TenantShardId::unsharded(tenant_id),
1318 0 : timeline_id,
1319 0 : pageserver_client::mgmt_api::ForceAwaitLogicalSize::No,
1320 0 : )
1321 0 : .await?;
1322 0 : let default_sk = SafekeeperNode::from_env(env, env.safekeepers.first().unwrap());
1323 0 : let default_host = default_sk
1324 0 : .conf
1325 0 : .listen_addr
1326 0 : .clone()
1327 0 : .unwrap_or_else(|| "localhost".to_string());
1328 0 : let mconf = safekeeper_api::membership::Configuration {
1329 0 : generation: SafekeeperGeneration::new(1),
1330 0 : members: safekeeper_api::membership::MemberSet {
1331 0 : m: vec![SafekeeperId {
1332 0 : host: default_host,
1333 0 : id: default_sk.conf.id,
1334 0 : pg_port: default_sk.conf.pg_port,
1335 0 : }],
1336 0 : },
1337 0 : new_members: None,
1338 0 : };
1339 0 : let pg_version = PgVersionId::from(args.pg_version);
1340 0 : let req = safekeeper_api::models::TimelineCreateRequest {
1341 0 : tenant_id,
1342 0 : timeline_id,
1343 0 : mconf,
1344 0 : pg_version,
1345 0 : system_id: None,
1346 0 : wal_seg_size: None,
1347 0 : start_lsn: timeline_info.last_record_lsn,
1348 0 : commit_lsn: None,
1349 0 : };
1350 0 : default_sk.create_timeline(&req).await?;
1351 0 : }
1352 0 : env.register_branch_mapping(branch_name.to_string(), tenant_id, timeline_id)?;
1353 0 : println!("Done");
1354 : }
1355 0 : TimelineCmd::Branch(args) => {
1356 0 : let tenant_id = get_tenant_id(args.tenant_id, env)?;
1357 0 : let new_timeline_id = args.timeline_id.unwrap_or(TimelineId::generate());
1358 0 : let new_branch_name = &args.branch_name;
1359 0 : let ancestor_branch_name = args
1360 0 : .ancestor_branch_name
1361 0 : .clone()
1362 0 : .unwrap_or(DEFAULT_BRANCH_NAME.to_owned());
1363 0 : let ancestor_timeline_id = env
1364 0 : .get_branch_timeline_id(&ancestor_branch_name, tenant_id)
1365 0 : .ok_or_else(|| {
1366 0 : anyhow!("Found no timeline id for branch name '{ancestor_branch_name}'")
1367 0 : })?;
1368 :
1369 0 : let start_lsn = args.ancestor_start_lsn;
1370 0 : let storage_controller = StorageController::from_env(env);
1371 0 : let create_req = TimelineCreateRequest {
1372 0 : new_timeline_id,
1373 0 : mode: pageserver_api::models::TimelineCreateRequestMode::Branch {
1374 0 : ancestor_timeline_id,
1375 0 : ancestor_start_lsn: start_lsn,
1376 0 : read_only: false,
1377 0 : pg_version: None,
1378 0 : },
1379 0 : };
1380 0 : let timeline_info = storage_controller
1381 0 : .tenant_timeline_create(tenant_id, create_req)
1382 0 : .await?;
1383 :
1384 0 : let last_record_lsn = timeline_info.last_record_lsn;
1385 :
1386 0 : env.register_branch_mapping(new_branch_name.to_string(), tenant_id, new_timeline_id)?;
1387 :
1388 0 : println!(
1389 0 : "Created timeline '{}' at Lsn {last_record_lsn} for tenant: {tenant_id}. Ancestor timeline: '{ancestor_branch_name}'",
1390 : timeline_info.timeline_id
1391 : );
1392 : }
1393 : }
1394 :
1395 0 : Ok(())
1396 0 : }
1397 :
1398 0 : async fn handle_endpoint(subcmd: &EndpointCmd, env: &local_env::LocalEnv) -> Result<()> {
1399 0 : let mut cplane = ComputeControlPlane::load(env.clone())?;
1400 :
1401 0 : match subcmd {
1402 0 : EndpointCmd::List(args) => {
1403 : // TODO(sharding): this command shouldn't have to specify a shard ID: we should ask the storage controller
1404 : // where shard 0 is attached, and query there.
1405 0 : let tenant_shard_id = get_tenant_shard_id(args.tenant_shard_id, env)?;
1406 :
1407 0 : let timeline_name_mappings = env.timeline_name_mappings();
1408 :
1409 0 : let mut table = comfy_table::Table::new();
1410 :
1411 0 : table.load_preset(comfy_table::presets::NOTHING);
1412 :
1413 0 : table.set_header([
1414 0 : "ENDPOINT",
1415 0 : "ADDRESS",
1416 0 : "TIMELINE",
1417 0 : "BRANCH NAME",
1418 0 : "LSN",
1419 0 : "STATUS",
1420 0 : ]);
1421 :
1422 0 : for (endpoint_id, endpoint) in cplane
1423 0 : .endpoints
1424 0 : .iter()
1425 0 : .filter(|(_, endpoint)| endpoint.tenant_id == tenant_shard_id.tenant_id)
1426 : {
1427 0 : let lsn_str = match endpoint.mode {
1428 0 : ComputeMode::Static(lsn) => {
1429 : // -> read-only endpoint
1430 : // Use the node's LSN.
1431 0 : lsn.to_string()
1432 : }
1433 : _ => {
1434 : // As the LSN here refers to the one that the compute is started with,
1435 : // we display nothing as it is a primary/hot standby compute.
1436 0 : "---".to_string()
1437 : }
1438 : };
1439 :
1440 0 : let branch_name = timeline_name_mappings
1441 0 : .get(&TenantTimelineId::new(
1442 0 : tenant_shard_id.tenant_id,
1443 0 : endpoint.timeline_id,
1444 0 : ))
1445 0 : .map(|name| name.as_str())
1446 0 : .unwrap_or("?");
1447 :
1448 0 : table.add_row([
1449 0 : endpoint_id.as_str(),
1450 0 : &endpoint.pg_address.to_string(),
1451 0 : &endpoint.timeline_id.to_string(),
1452 0 : branch_name,
1453 0 : lsn_str.as_str(),
1454 0 : &format!("{}", endpoint.status()),
1455 0 : ]);
1456 : }
1457 :
1458 0 : println!("{table}");
1459 : }
1460 0 : EndpointCmd::Create(args) => {
1461 0 : let tenant_id = get_tenant_id(args.tenant_id, env)?;
1462 0 : let branch_name = args
1463 0 : .branch_name
1464 0 : .clone()
1465 0 : .unwrap_or(DEFAULT_BRANCH_NAME.to_owned());
1466 0 : let endpoint_id = args
1467 0 : .endpoint_id
1468 0 : .clone()
1469 0 : .unwrap_or_else(|| format!("ep-{branch_name}"));
1470 :
1471 0 : let timeline_id = env
1472 0 : .get_branch_timeline_id(&branch_name, tenant_id)
1473 0 : .ok_or_else(|| anyhow!("Found no timeline id for branch name '{branch_name}'"))?;
1474 :
1475 0 : let mode = match (args.lsn, args.hot_standby) {
1476 0 : (Some(lsn), false) => ComputeMode::Static(lsn),
1477 0 : (None, true) => ComputeMode::Replica,
1478 0 : (None, false) => ComputeMode::Primary,
1479 0 : (Some(_), true) => anyhow::bail!("cannot specify both lsn and hot-standby"),
1480 : };
1481 :
1482 0 : match (mode, args.hot_standby) {
1483 : (ComputeMode::Static(_), true) => {
1484 0 : bail!(
1485 0 : "Cannot start a node in hot standby mode when it is already configured as a static replica"
1486 : )
1487 : }
1488 : (ComputeMode::Primary, true) => {
1489 0 : bail!(
1490 0 : "Cannot start a node as a hot standby replica, it is already configured as primary node"
1491 : )
1492 : }
1493 0 : _ => {}
1494 : }
1495 :
1496 0 : if !args.allow_multiple {
1497 0 : cplane.check_conflicting_endpoints(mode, tenant_id, timeline_id)?;
1498 0 : }
1499 :
1500 0 : cplane.new_endpoint(
1501 0 : &endpoint_id,
1502 0 : tenant_id,
1503 0 : timeline_id,
1504 0 : args.pg_port,
1505 0 : args.external_http_port,
1506 0 : args.internal_http_port,
1507 0 : args.pg_version,
1508 0 : mode,
1509 0 : args.grpc,
1510 0 : !args.update_catalog,
1511 : false,
1512 0 : args.privileged_role_name.clone(),
1513 0 : )?;
1514 : }
1515 0 : EndpointCmd::Start(args) => {
1516 0 : let endpoint_id = &args.endpoint_id;
1517 0 : let pageserver_id = args.endpoint_pageserver_id;
1518 0 : let remote_ext_base_url = &args.remote_ext_base_url;
1519 :
1520 0 : let default_generation = env
1521 0 : .storage_controller
1522 0 : .timelines_onto_safekeepers
1523 0 : .then_some(1);
1524 0 : let safekeepers_generation = args
1525 0 : .safekeepers_generation
1526 0 : .or(default_generation)
1527 0 : .map(SafekeeperGeneration::new);
1528 : // If --safekeepers argument is given, use only the listed
1529 : // safekeeper nodes; otherwise all from the env.
1530 0 : let safekeepers = if let Some(safekeepers) = parse_safekeepers(&args.safekeepers)? {
1531 0 : safekeepers
1532 : } else {
1533 0 : env.safekeepers.iter().map(|sk| sk.id).collect()
1534 : };
1535 :
1536 0 : let endpoint = cplane
1537 0 : .endpoints
1538 0 : .get(endpoint_id.as_str())
1539 0 : .ok_or_else(|| anyhow!("endpoint {endpoint_id} not found"))?;
1540 :
1541 0 : if !args.allow_multiple {
1542 0 : cplane.check_conflicting_endpoints(
1543 0 : endpoint.mode,
1544 0 : endpoint.tenant_id,
1545 0 : endpoint.timeline_id,
1546 0 : )?;
1547 0 : }
1548 :
1549 0 : let (pageservers, stripe_size) = if let Some(pageserver_id) = pageserver_id {
1550 0 : let conf = env.get_pageserver_conf(pageserver_id).unwrap();
1551 : // Use gRPC if requested.
1552 0 : let pageserver = if endpoint.grpc {
1553 0 : let grpc_addr = conf.listen_grpc_addr.as_ref().expect("bad config");
1554 0 : let (host, port) = parse_host_port(grpc_addr)?;
1555 0 : let port = port.unwrap_or(DEFAULT_PAGESERVER_GRPC_PORT);
1556 0 : (PageserverProtocol::Grpc, host, port)
1557 : } else {
1558 0 : let (host, port) = parse_host_port(&conf.listen_pg_addr)?;
1559 0 : let port = port.unwrap_or(5432);
1560 0 : (PageserverProtocol::Libpq, host, port)
1561 : };
1562 : // If caller is telling us what pageserver to use, this is not a tenant which is
1563 : // fully managed by storage controller, therefore not sharded.
1564 0 : (vec![pageserver], DEFAULT_STRIPE_SIZE)
1565 : } else {
1566 : // Look up the currently attached location of the tenant, and its striping metadata,
1567 : // to pass these on to postgres.
1568 0 : let storage_controller = StorageController::from_env(env);
1569 0 : let locate_result = storage_controller.tenant_locate(endpoint.tenant_id).await?;
1570 0 : let pageservers = futures::future::try_join_all(
1571 0 : locate_result.shards.into_iter().map(|shard| async move {
1572 0 : if let ComputeMode::Static(lsn) = endpoint.mode {
1573 : // Initialize LSN leases for static computes.
1574 0 : let conf = env.get_pageserver_conf(shard.node_id).unwrap();
1575 0 : let pageserver = PageServerNode::from_env(env, conf);
1576 :
1577 0 : pageserver
1578 0 : .http_client
1579 0 : .timeline_init_lsn_lease(shard.shard_id, endpoint.timeline_id, lsn)
1580 0 : .await?;
1581 0 : }
1582 :
1583 0 : let pageserver = if endpoint.grpc {
1584 : (
1585 0 : PageserverProtocol::Grpc,
1586 0 : Host::parse(&shard.listen_grpc_addr.expect("no gRPC address"))?,
1587 0 : shard.listen_grpc_port.expect("no gRPC port"),
1588 : )
1589 : } else {
1590 : (
1591 0 : PageserverProtocol::Libpq,
1592 0 : Host::parse(&shard.listen_pg_addr)?,
1593 0 : shard.listen_pg_port,
1594 : )
1595 : };
1596 0 : anyhow::Ok(pageserver)
1597 0 : }),
1598 : )
1599 0 : .await?;
1600 0 : let stripe_size = locate_result.shard_params.stripe_size;
1601 :
1602 0 : (pageservers, stripe_size)
1603 : };
1604 0 : assert!(!pageservers.is_empty());
1605 :
1606 0 : let ps_conf = env.get_pageserver_conf(DEFAULT_PAGESERVER_ID)?;
1607 0 : let auth_token = if matches!(ps_conf.pg_auth_type, AuthType::NeonJWT) {
1608 0 : let claims = Claims::new(Some(endpoint.tenant_id), Scope::Tenant);
1609 :
1610 0 : Some(env.generate_auth_token(&claims)?)
1611 : } else {
1612 0 : None
1613 : };
1614 :
1615 0 : let exp = (std::time::SystemTime::now().duration_since(std::time::UNIX_EPOCH)?
1616 0 : + Duration::from_secs(86400))
1617 0 : .as_secs();
1618 0 : let claims = endpoint_storage::claims::EndpointStorageClaims {
1619 0 : tenant_id: endpoint.tenant_id,
1620 0 : timeline_id: endpoint.timeline_id,
1621 0 : endpoint_id: endpoint_id.to_string(),
1622 0 : exp,
1623 0 : };
1624 :
1625 0 : let endpoint_storage_token = env.generate_auth_token(&claims)?;
1626 0 : let endpoint_storage_addr = env.endpoint_storage.listen_addr.to_string();
1627 :
1628 0 : let args = control_plane::endpoint::EndpointStartArgs {
1629 0 : auth_token,
1630 0 : endpoint_storage_token,
1631 0 : endpoint_storage_addr,
1632 0 : safekeepers_generation,
1633 0 : safekeepers,
1634 0 : pageservers,
1635 0 : remote_ext_base_url: remote_ext_base_url.clone(),
1636 0 : shard_stripe_size: stripe_size.0 as usize,
1637 0 : create_test_user: args.create_test_user,
1638 0 : start_timeout: args.start_timeout,
1639 0 : autoprewarm: args.autoprewarm,
1640 0 : offload_lfc_interval_seconds: args.offload_lfc_interval_seconds,
1641 0 : dev: args.dev,
1642 0 : };
1643 :
1644 0 : println!("Starting existing endpoint {endpoint_id}...");
1645 0 : endpoint.start(args).await?;
1646 : }
1647 0 : EndpointCmd::UpdatePageservers(args) => {
1648 0 : let endpoint_id = &args.endpoint_id;
1649 0 : let endpoint = cplane
1650 0 : .endpoints
1651 0 : .get(endpoint_id.as_str())
1652 0 : .with_context(|| format!("postgres endpoint {endpoint_id} is not found"))?;
1653 0 : let pageservers = match args.pageserver_id {
1654 0 : Some(pageserver_id) => {
1655 0 : let pageserver =
1656 0 : PageServerNode::from_env(env, env.get_pageserver_conf(pageserver_id)?);
1657 :
1658 0 : vec![(
1659 0 : PageserverProtocol::Libpq,
1660 0 : pageserver.pg_connection_config.host().clone(),
1661 0 : pageserver.pg_connection_config.port(),
1662 0 : )]
1663 : }
1664 : None => {
1665 0 : let storage_controller = StorageController::from_env(env);
1666 0 : storage_controller
1667 0 : .tenant_locate(endpoint.tenant_id)
1668 0 : .await?
1669 : .shards
1670 0 : .into_iter()
1671 0 : .map(|shard| {
1672 0 : (
1673 0 : PageserverProtocol::Libpq,
1674 0 : Host::parse(&shard.listen_pg_addr)
1675 0 : .expect("Storage controller reported malformed host"),
1676 0 : shard.listen_pg_port,
1677 0 : )
1678 0 : })
1679 0 : .collect::<Vec<_>>()
1680 : }
1681 : };
1682 :
1683 0 : endpoint.update_pageservers_in_config(pageservers).await?;
1684 : }
1685 0 : EndpointCmd::Reconfigure(args) => {
1686 0 : let endpoint_id = &args.endpoint_id;
1687 0 : let endpoint = cplane
1688 0 : .endpoints
1689 0 : .get(endpoint_id.as_str())
1690 0 : .with_context(|| format!("postgres endpoint {endpoint_id} is not found"))?;
1691 0 : let pageservers = if let Some(ps_id) = args.endpoint_pageserver_id {
1692 0 : let conf = env.get_pageserver_conf(ps_id)?;
1693 : // Use gRPC if requested.
1694 0 : let pageserver = if endpoint.grpc {
1695 0 : let grpc_addr = conf.listen_grpc_addr.as_ref().expect("bad config");
1696 0 : let (host, port) = parse_host_port(grpc_addr)?;
1697 0 : let port = port.unwrap_or(DEFAULT_PAGESERVER_GRPC_PORT);
1698 0 : (PageserverProtocol::Grpc, host, port)
1699 : } else {
1700 0 : let (host, port) = parse_host_port(&conf.listen_pg_addr)?;
1701 0 : let port = port.unwrap_or(5432);
1702 0 : (PageserverProtocol::Libpq, host, port)
1703 : };
1704 0 : vec![pageserver]
1705 : } else {
1706 0 : let storage_controller = StorageController::from_env(env);
1707 0 : storage_controller
1708 0 : .tenant_locate(endpoint.tenant_id)
1709 0 : .await?
1710 : .shards
1711 0 : .into_iter()
1712 0 : .map(|shard| {
1713 : // Use gRPC if requested.
1714 0 : if endpoint.grpc {
1715 0 : (
1716 0 : PageserverProtocol::Grpc,
1717 0 : Host::parse(&shard.listen_grpc_addr.expect("no gRPC address"))
1718 0 : .expect("bad hostname"),
1719 0 : shard.listen_grpc_port.expect("no gRPC port"),
1720 0 : )
1721 : } else {
1722 0 : (
1723 0 : PageserverProtocol::Libpq,
1724 0 : Host::parse(&shard.listen_pg_addr).expect("bad hostname"),
1725 0 : shard.listen_pg_port,
1726 0 : )
1727 : }
1728 0 : })
1729 0 : .collect::<Vec<_>>()
1730 : };
1731 : // If --safekeepers argument is given, use only the listed
1732 : // safekeeper nodes; otherwise all from the env.
1733 0 : let safekeepers = parse_safekeepers(&args.safekeepers)?;
1734 0 : endpoint
1735 0 : .reconfigure(Some(pageservers), None, safekeepers, None)
1736 0 : .await?;
1737 : }
1738 0 : EndpointCmd::RefreshConfiguration(args) => {
1739 0 : let endpoint_id = &args.endpoint_id;
1740 0 : let endpoint = cplane
1741 0 : .endpoints
1742 0 : .get(endpoint_id.as_str())
1743 0 : .with_context(|| format!("postgres endpoint {endpoint_id} is not found"))?;
1744 0 : endpoint.refresh_configuration().await?;
1745 : }
1746 0 : EndpointCmd::Stop(args) => {
1747 0 : let endpoint_id = &args.endpoint_id;
1748 0 : let endpoint = cplane
1749 0 : .endpoints
1750 0 : .get(endpoint_id)
1751 0 : .with_context(|| format!("postgres endpoint {endpoint_id} is not found"))?;
1752 0 : match endpoint.stop(args.mode, args.destroy).await?.lsn {
1753 0 : Some(lsn) => println!("{lsn}"),
1754 0 : None => println!("null"),
1755 : }
1756 : }
1757 0 : EndpointCmd::GenerateJwt(args) => {
1758 0 : let endpoint = {
1759 0 : let endpoint_id = &args.endpoint_id;
1760 :
1761 0 : cplane
1762 0 : .endpoints
1763 0 : .get(endpoint_id)
1764 0 : .with_context(|| format!("postgres endpoint {endpoint_id} is not found"))?
1765 : };
1766 :
1767 0 : let jwt = endpoint.generate_jwt(args.scope)?;
1768 :
1769 0 : print!("{jwt}");
1770 : }
1771 : }
1772 :
1773 0 : Ok(())
1774 0 : }
1775 :
1776 : /// Parse --safekeepers as list of safekeeper ids.
1777 0 : fn parse_safekeepers(safekeepers_str: &Option<String>) -> Result<Option<Vec<NodeId>>> {
1778 0 : if let Some(safekeepers_str) = safekeepers_str {
1779 0 : let mut safekeepers: Vec<NodeId> = Vec::new();
1780 0 : for sk_id in safekeepers_str.split(',').map(str::trim) {
1781 0 : let sk_id = NodeId(
1782 0 : u64::from_str(sk_id)
1783 0 : .map_err(|_| anyhow!("invalid node ID \"{sk_id}\" in --safekeepers list"))?,
1784 : );
1785 0 : safekeepers.push(sk_id);
1786 : }
1787 0 : Ok(Some(safekeepers))
1788 : } else {
1789 0 : Ok(None)
1790 : }
1791 0 : }
1792 :
1793 0 : fn handle_mappings(subcmd: &MappingsCmd, env: &mut local_env::LocalEnv) -> Result<()> {
1794 0 : match subcmd {
1795 0 : MappingsCmd::Map(args) => {
1796 0 : env.register_branch_mapping(
1797 0 : args.branch_name.to_owned(),
1798 0 : args.tenant_id,
1799 0 : args.timeline_id,
1800 0 : )?;
1801 :
1802 0 : Ok(())
1803 : }
1804 : }
1805 0 : }
1806 :
1807 0 : fn get_pageserver(
1808 0 : env: &local_env::LocalEnv,
1809 0 : pageserver_id_arg: Option<NodeId>,
1810 0 : ) -> Result<PageServerNode> {
1811 0 : let node_id = pageserver_id_arg.unwrap_or(DEFAULT_PAGESERVER_ID);
1812 :
1813 0 : Ok(PageServerNode::from_env(
1814 0 : env,
1815 0 : env.get_pageserver_conf(node_id)?,
1816 : ))
1817 0 : }
1818 :
1819 0 : async fn handle_pageserver(subcmd: &PageserverCmd, env: &local_env::LocalEnv) -> Result<()> {
1820 0 : match subcmd {
1821 0 : PageserverCmd::Start(args) => {
1822 0 : if let Err(e) = get_pageserver(env, args.pageserver_id)?
1823 0 : .start(&args.start_timeout)
1824 0 : .await
1825 : {
1826 0 : eprintln!("pageserver start failed: {e}");
1827 0 : exit(1);
1828 0 : }
1829 : }
1830 :
1831 0 : PageserverCmd::Stop(args) => {
1832 0 : let immediate = match args.stop_mode {
1833 0 : StopMode::Fast => false,
1834 0 : StopMode::Immediate => true,
1835 : };
1836 0 : if let Err(e) = get_pageserver(env, args.pageserver_id)?.stop(immediate) {
1837 0 : eprintln!("pageserver stop failed: {e}");
1838 0 : exit(1);
1839 0 : }
1840 : }
1841 :
1842 0 : PageserverCmd::Restart(args) => {
1843 0 : let pageserver = get_pageserver(env, args.pageserver_id)?;
1844 : //TODO what shutdown strategy should we use here?
1845 0 : if let Err(e) = pageserver.stop(false) {
1846 0 : eprintln!("pageserver stop failed: {e}");
1847 0 : exit(1);
1848 0 : }
1849 :
1850 0 : if let Err(e) = pageserver.start(&args.start_timeout).await {
1851 0 : eprintln!("pageserver start failed: {e}");
1852 0 : exit(1);
1853 0 : }
1854 : }
1855 :
1856 0 : PageserverCmd::Status(args) => {
1857 0 : match get_pageserver(env, args.pageserver_id)?
1858 0 : .check_status()
1859 0 : .await
1860 : {
1861 0 : Ok(_) => println!("Page server is up and running"),
1862 0 : Err(err) => {
1863 0 : eprintln!("Page server is not available: {err}");
1864 0 : exit(1);
1865 : }
1866 : }
1867 : }
1868 : }
1869 0 : Ok(())
1870 0 : }
1871 :
1872 0 : async fn handle_storage_controller(
1873 0 : subcmd: &StorageControllerCmd,
1874 0 : env: &local_env::LocalEnv,
1875 0 : ) -> Result<()> {
1876 0 : let svc = StorageController::from_env(env);
1877 0 : match subcmd {
1878 0 : StorageControllerCmd::Start(args) => {
1879 0 : let start_args = NeonStorageControllerStartArgs {
1880 0 : instance_id: args.instance_id,
1881 0 : base_port: args.base_port,
1882 0 : start_timeout: args.start_timeout,
1883 0 : handle_ps_local_disk_loss: args.handle_ps_local_disk_loss,
1884 0 : };
1885 :
1886 0 : if let Err(e) = svc.start(start_args).await {
1887 0 : eprintln!("start failed: {e}");
1888 0 : exit(1);
1889 0 : }
1890 : }
1891 :
1892 0 : StorageControllerCmd::Stop(args) => {
1893 0 : let stop_args = NeonStorageControllerStopArgs {
1894 0 : instance_id: args.instance_id,
1895 0 : immediate: match args.stop_mode {
1896 0 : StopMode::Fast => false,
1897 0 : StopMode::Immediate => true,
1898 : },
1899 : };
1900 0 : if let Err(e) = svc.stop(stop_args).await {
1901 0 : eprintln!("stop failed: {e}");
1902 0 : exit(1);
1903 0 : }
1904 : }
1905 : }
1906 0 : Ok(())
1907 0 : }
1908 :
1909 0 : fn get_safekeeper(env: &local_env::LocalEnv, id: NodeId) -> Result<SafekeeperNode> {
1910 0 : if let Some(node) = env.safekeepers.iter().find(|node| node.id == id) {
1911 0 : Ok(SafekeeperNode::from_env(env, node))
1912 : } else {
1913 0 : bail!("could not find safekeeper {id}")
1914 : }
1915 0 : }
1916 :
1917 0 : async fn handle_safekeeper(subcmd: &SafekeeperCmd, env: &local_env::LocalEnv) -> Result<()> {
1918 0 : match subcmd {
1919 0 : SafekeeperCmd::Start(args) => {
1920 0 : let safekeeper = get_safekeeper(env, args.id)?;
1921 :
1922 0 : if let Err(e) = safekeeper.start(&args.extra_opt, &args.start_timeout).await {
1923 0 : eprintln!("safekeeper start failed: {e}");
1924 0 : exit(1);
1925 0 : }
1926 : }
1927 :
1928 0 : SafekeeperCmd::Stop(args) => {
1929 0 : let safekeeper = get_safekeeper(env, args.id)?;
1930 0 : let immediate = match args.stop_mode {
1931 0 : StopMode::Fast => false,
1932 0 : StopMode::Immediate => true,
1933 : };
1934 0 : if let Err(e) = safekeeper.stop(immediate) {
1935 0 : eprintln!("safekeeper stop failed: {e}");
1936 0 : exit(1);
1937 0 : }
1938 : }
1939 :
1940 0 : SafekeeperCmd::Restart(args) => {
1941 0 : let safekeeper = get_safekeeper(env, args.id)?;
1942 0 : let immediate = match args.stop_mode {
1943 0 : StopMode::Fast => false,
1944 0 : StopMode::Immediate => true,
1945 : };
1946 :
1947 0 : if let Err(e) = safekeeper.stop(immediate) {
1948 0 : eprintln!("safekeeper stop failed: {e}");
1949 0 : exit(1);
1950 0 : }
1951 :
1952 0 : if let Err(e) = safekeeper.start(&args.extra_opt, &args.start_timeout).await {
1953 0 : eprintln!("safekeeper start failed: {e}");
1954 0 : exit(1);
1955 0 : }
1956 : }
1957 : }
1958 0 : Ok(())
1959 0 : }
1960 :
1961 0 : async fn handle_endpoint_storage(
1962 0 : subcmd: &EndpointStorageCmd,
1963 0 : env: &local_env::LocalEnv,
1964 0 : ) -> Result<()> {
1965 : use EndpointStorageCmd::*;
1966 0 : let storage = EndpointStorage::from_env(env);
1967 :
1968 : // In tests like test_forward_compatibility or test_graceful_cluster_restart
1969 : // old neon binaries (without endpoint_storage) are present
1970 0 : if !storage.bin.exists() {
1971 0 : eprintln!(
1972 0 : "{} binary not found. Ignore if this is a compatibility test",
1973 : storage.bin
1974 : );
1975 0 : return Ok(());
1976 0 : }
1977 :
1978 0 : match subcmd {
1979 0 : Start(EndpointStorageStartCmd { start_timeout }) => {
1980 0 : if let Err(e) = storage.start(start_timeout).await {
1981 0 : eprintln!("endpoint_storage start failed: {e}");
1982 0 : exit(1);
1983 0 : }
1984 : }
1985 0 : Stop(EndpointStorageStopCmd { stop_mode }) => {
1986 0 : let immediate = match stop_mode {
1987 0 : StopMode::Fast => false,
1988 0 : StopMode::Immediate => true,
1989 : };
1990 0 : if let Err(e) = storage.stop(immediate) {
1991 0 : eprintln!("proxy stop failed: {e}");
1992 0 : exit(1);
1993 0 : }
1994 : }
1995 : };
1996 0 : Ok(())
1997 0 : }
1998 :
1999 0 : async fn handle_storage_broker(subcmd: &StorageBrokerCmd, env: &local_env::LocalEnv) -> Result<()> {
2000 0 : match subcmd {
2001 0 : StorageBrokerCmd::Start(args) => {
2002 0 : let storage_broker = StorageBroker::from_env(env);
2003 0 : if let Err(e) = storage_broker.start(&args.start_timeout).await {
2004 0 : eprintln!("broker start failed: {e}");
2005 0 : exit(1);
2006 0 : }
2007 : }
2008 :
2009 0 : StorageBrokerCmd::Stop(_args) => {
2010 : // FIXME: stop_mode unused
2011 0 : let storage_broker = StorageBroker::from_env(env);
2012 0 : if let Err(e) = storage_broker.stop() {
2013 0 : eprintln!("broker stop failed: {e}");
2014 0 : exit(1);
2015 0 : }
2016 : }
2017 : }
2018 0 : Ok(())
2019 0 : }
2020 :
2021 0 : async fn handle_start_all(
2022 0 : args: &StartCmdArgs,
2023 0 : env: &'static local_env::LocalEnv,
2024 0 : ) -> anyhow::Result<()> {
2025 : // FIXME: this was called "retry_timeout", is it right?
2026 0 : let Err(errors) = handle_start_all_impl(env, args.timeout).await else {
2027 0 : neon_start_status_check(env, args.timeout.as_ref())
2028 0 : .await
2029 0 : .context("status check after successful startup of all services")?;
2030 0 : return Ok(());
2031 : };
2032 :
2033 0 : eprintln!("startup failed because one or more services could not be started");
2034 :
2035 0 : for e in errors {
2036 0 : eprintln!("{e}");
2037 0 : let debug_repr = format!("{e:?}");
2038 0 : for line in debug_repr.lines() {
2039 0 : eprintln!(" {line}");
2040 0 : }
2041 : }
2042 :
2043 0 : try_stop_all(env, true).await;
2044 :
2045 0 : exit(2);
2046 0 : }
2047 :
2048 : /// Returns Ok() if and only if all services could be started successfully.
2049 : /// Otherwise, returns the list of errors that occurred during startup.
2050 0 : async fn handle_start_all_impl(
2051 0 : env: &'static local_env::LocalEnv,
2052 0 : retry_timeout: humantime::Duration,
2053 0 : ) -> Result<(), Vec<anyhow::Error>> {
2054 : // Endpoints are not started automatically
2055 :
2056 0 : let mut js = JoinSet::new();
2057 :
2058 : // force infalliblity through closure
2059 : #[allow(clippy::redundant_closure_call)]
2060 0 : (|| {
2061 0 : js.spawn(async move {
2062 0 : let storage_broker = StorageBroker::from_env(env);
2063 0 : storage_broker
2064 0 : .start(&retry_timeout)
2065 0 : .await
2066 0 : .map_err(|e| e.context("start storage_broker"))
2067 0 : });
2068 :
2069 0 : js.spawn(async move {
2070 0 : let storage_controller = StorageController::from_env(env);
2071 0 : storage_controller
2072 0 : .start(NeonStorageControllerStartArgs::with_default_instance_id(
2073 0 : retry_timeout,
2074 0 : ))
2075 0 : .await
2076 0 : .map_err(|e| e.context("start storage_controller"))
2077 0 : });
2078 :
2079 0 : for ps_conf in &env.pageservers {
2080 0 : js.spawn(async move {
2081 0 : let pageserver = PageServerNode::from_env(env, ps_conf);
2082 0 : pageserver
2083 0 : .start(&retry_timeout)
2084 0 : .await
2085 0 : .map_err(|e| e.context(format!("start pageserver {}", ps_conf.id)))
2086 0 : });
2087 : }
2088 :
2089 0 : for node in env.safekeepers.iter() {
2090 0 : js.spawn(async move {
2091 0 : let safekeeper = SafekeeperNode::from_env(env, node);
2092 0 : safekeeper
2093 0 : .start(&[], &retry_timeout)
2094 0 : .await
2095 0 : .map_err(|e| e.context(format!("start safekeeper {}", safekeeper.id)))
2096 0 : });
2097 : }
2098 :
2099 0 : js.spawn(async move {
2100 0 : EndpointStorage::from_env(env)
2101 0 : .start(&retry_timeout)
2102 0 : .await
2103 0 : .map_err(|e| e.context("start endpoint_storage"))
2104 0 : });
2105 : })();
2106 :
2107 0 : let mut errors = Vec::new();
2108 0 : while let Some(result) = js.join_next().await {
2109 0 : let result = result.expect("we don't panic or cancel the tasks");
2110 0 : if let Err(e) = result {
2111 0 : errors.push(e);
2112 0 : }
2113 : }
2114 :
2115 0 : if !errors.is_empty() {
2116 0 : return Err(errors);
2117 0 : }
2118 :
2119 0 : Ok(())
2120 0 : }
2121 :
2122 0 : async fn neon_start_status_check(
2123 0 : env: &local_env::LocalEnv,
2124 0 : retry_timeout: &Duration,
2125 0 : ) -> anyhow::Result<()> {
2126 : const RETRY_INTERVAL: Duration = Duration::from_millis(100);
2127 : const NOTICE_AFTER_RETRIES: Duration = Duration::from_secs(5);
2128 :
2129 0 : let storcon = StorageController::from_env(env);
2130 :
2131 0 : let retries = retry_timeout.as_millis() / RETRY_INTERVAL.as_millis();
2132 0 : let notice_after_retries = retry_timeout.as_millis() / NOTICE_AFTER_RETRIES.as_millis();
2133 :
2134 0 : println!("\nRunning neon status check");
2135 :
2136 0 : for retry in 0..retries {
2137 0 : if retry == notice_after_retries {
2138 0 : println!("\nNeon status check has not passed yet, continuing to wait")
2139 0 : }
2140 :
2141 0 : let mut passed = true;
2142 0 : let mut nodes = storcon.node_list().await?;
2143 0 : let mut pageservers = env.pageservers.clone();
2144 :
2145 0 : if nodes.len() != pageservers.len() {
2146 0 : continue;
2147 0 : }
2148 :
2149 0 : nodes.sort_by_key(|ps| ps.id);
2150 0 : pageservers.sort_by_key(|ps| ps.id);
2151 :
2152 0 : for (idx, pageserver) in pageservers.iter().enumerate() {
2153 0 : let node = &nodes[idx];
2154 0 : if node.id != pageserver.id {
2155 0 : passed = false;
2156 0 : break;
2157 0 : }
2158 :
2159 0 : if !matches!(node.availability, NodeAvailabilityWrapper::Active) {
2160 0 : passed = false;
2161 0 : break;
2162 0 : }
2163 : }
2164 :
2165 0 : if passed {
2166 0 : println!("\nNeon started and passed status check");
2167 0 : return Ok(());
2168 0 : }
2169 :
2170 0 : tokio::time::sleep(RETRY_INTERVAL).await;
2171 : }
2172 :
2173 0 : anyhow::bail!("\nNeon passed status check")
2174 0 : }
2175 :
2176 0 : async fn handle_stop_all(args: &StopCmdArgs, env: &local_env::LocalEnv) -> Result<()> {
2177 0 : let immediate = match args.mode {
2178 0 : StopMode::Fast => false,
2179 0 : StopMode::Immediate => true,
2180 : };
2181 :
2182 0 : try_stop_all(env, immediate).await;
2183 :
2184 0 : Ok(())
2185 0 : }
2186 :
2187 0 : async fn try_stop_all(env: &local_env::LocalEnv, immediate: bool) {
2188 0 : let mode = if immediate {
2189 0 : EndpointTerminateMode::Immediate
2190 : } else {
2191 0 : EndpointTerminateMode::Fast
2192 : };
2193 : // Stop all endpoints
2194 0 : match ComputeControlPlane::load(env.clone()) {
2195 0 : Ok(cplane) => {
2196 0 : for (_k, node) in cplane.endpoints {
2197 0 : if let Err(e) = node.stop(mode, false).await {
2198 0 : eprintln!("postgres stop failed: {e:#}");
2199 0 : }
2200 : }
2201 : }
2202 0 : Err(e) => {
2203 0 : eprintln!("postgres stop failed, could not restore control plane data from env: {e:#}")
2204 : }
2205 : }
2206 :
2207 0 : let storage = EndpointStorage::from_env(env);
2208 0 : if let Err(e) = storage.stop(immediate) {
2209 0 : eprintln!("endpoint_storage stop failed: {e:#}");
2210 0 : }
2211 :
2212 0 : for ps_conf in &env.pageservers {
2213 0 : let pageserver = PageServerNode::from_env(env, ps_conf);
2214 0 : if let Err(e) = pageserver.stop(immediate) {
2215 0 : eprintln!("pageserver {} stop failed: {:#}", ps_conf.id, e);
2216 0 : }
2217 : }
2218 :
2219 0 : for node in env.safekeepers.iter() {
2220 0 : let safekeeper = SafekeeperNode::from_env(env, node);
2221 0 : if let Err(e) = safekeeper.stop(immediate) {
2222 0 : eprintln!("safekeeper {} stop failed: {:#}", safekeeper.id, e);
2223 0 : }
2224 : }
2225 :
2226 0 : let storage_broker = StorageBroker::from_env(env);
2227 0 : if let Err(e) = storage_broker.stop() {
2228 0 : eprintln!("neon broker stop failed: {e:#}");
2229 0 : }
2230 :
2231 : // Stop all storage controller instances. In the most common case there's only one,
2232 : // but iterate though the base data directory in order to discover the instances.
2233 0 : let storcon_instances = env
2234 0 : .storage_controller_instances()
2235 0 : .await
2236 0 : .expect("Must inspect data dir");
2237 0 : for (instance_id, _instance_dir_path) in storcon_instances {
2238 0 : let storage_controller = StorageController::from_env(env);
2239 0 : let stop_args = NeonStorageControllerStopArgs {
2240 0 : instance_id,
2241 0 : immediate,
2242 0 : };
2243 :
2244 0 : if let Err(e) = storage_controller.stop(stop_args).await {
2245 0 : eprintln!("Storage controller instance {instance_id} stop failed: {e:#}");
2246 0 : }
2247 : }
2248 0 : }
|
