Line data Source code
1 : use futures::StreamExt;
2 : use std::{str::FromStr, time::Duration};
3 :
4 : use clap::{Parser, Subcommand};
5 : use pageserver_api::{
6 : controller_api::{
7 : NodeAvailabilityWrapper, NodeDescribeResponse, NodeShardResponse, ShardSchedulingPolicy,
8 : TenantCreateRequest, TenantDescribeResponse, TenantPolicyRequest,
9 : },
10 : models::{
11 : EvictionPolicy, EvictionPolicyLayerAccessThreshold, LocationConfigSecondary,
12 : ShardParameters, TenantConfig, TenantConfigRequest, TenantShardSplitRequest,
13 : TenantShardSplitResponse,
14 : },
15 : shard::{ShardStripeSize, TenantShardId},
16 : };
17 : use pageserver_client::mgmt_api::{self};
18 : use reqwest::{Method, StatusCode, Url};
19 : use utils::id::{NodeId, TenantId};
20 :
21 : use pageserver_api::controller_api::{
22 : NodeConfigureRequest, NodeRegisterRequest, NodeSchedulingPolicy, PlacementPolicy,
23 : TenantShardMigrateRequest, TenantShardMigrateResponse,
24 : };
25 : use storage_controller_client::control_api::Client;
26 :
27 0 : #[derive(Subcommand, Debug)]
28 : enum Command {
29 : /// Register a pageserver with the storage controller. This shouldn't usually be necessary,
30 : /// since pageservers auto-register when they start up
31 : NodeRegister {
32 : #[arg(long)]
33 0 : node_id: NodeId,
34 :
35 : #[arg(long)]
36 0 : listen_pg_addr: String,
37 : #[arg(long)]
38 0 : listen_pg_port: u16,
39 :
40 : #[arg(long)]
41 0 : listen_http_addr: String,
42 : #[arg(long)]
43 0 : listen_http_port: u16,
44 : #[arg(long)]
45 0 : availability_zone_id: String,
46 : },
47 :
48 : /// Modify a node's configuration in the storage controller
49 : NodeConfigure {
50 : #[arg(long)]
51 0 : node_id: NodeId,
52 :
53 : /// Availability is usually auto-detected based on heartbeats. Set 'offline' here to
54 : /// manually mark a node offline
55 : #[arg(long)]
56 : availability: Option<NodeAvailabilityArg>,
57 : /// Scheduling policy controls whether tenant shards may be scheduled onto this node.
58 : #[arg(long)]
59 : scheduling: Option<NodeSchedulingPolicy>,
60 : },
61 : NodeDelete {
62 : #[arg(long)]
63 0 : node_id: NodeId,
64 : },
65 : /// Modify a tenant's policies in the storage controller
66 : TenantPolicy {
67 : #[arg(long)]
68 0 : tenant_id: TenantId,
69 : /// Placement policy controls whether a tenant is `detached`, has only a secondary location (`secondary`),
70 : /// or is in the normal attached state with N secondary locations (`attached:N`)
71 : #[arg(long)]
72 : placement: Option<PlacementPolicyArg>,
73 : /// Scheduling policy enables pausing the controller's scheduling activity involving this tenant. `active` is normal,
74 : /// `essential` disables optimization scheduling changes, `pause` disables all scheduling changes, and `stop` prevents
75 : /// all reconciliation activity including for scheduling changes already made. `pause` and `stop` can make a tenant
76 : /// unavailable, and are only for use in emergencies.
77 : #[arg(long)]
78 : scheduling: Option<ShardSchedulingPolicyArg>,
79 : },
80 : /// List nodes known to the storage controller
81 : Nodes {},
82 : /// List tenants known to the storage controller
83 : Tenants {
84 : /// If this field is set, it will list the tenants on a specific node
85 : node_id: Option<NodeId>,
86 : },
87 : /// Create a new tenant in the storage controller, and by extension on pageservers.
88 : TenantCreate {
89 : #[arg(long)]
90 0 : tenant_id: TenantId,
91 : },
92 : /// Delete a tenant in the storage controller, and by extension on pageservers.
93 : TenantDelete {
94 : #[arg(long)]
95 0 : tenant_id: TenantId,
96 : },
97 : /// Split an existing tenant into a higher number of shards than its current shard count.
98 : TenantShardSplit {
99 : #[arg(long)]
100 0 : tenant_id: TenantId,
101 : #[arg(long)]
102 0 : shard_count: u8,
103 : /// Optional, in 8kiB pages. e.g. set 2048 for 16MB stripes.
104 : #[arg(long)]
105 : stripe_size: Option<u32>,
106 : },
107 : /// Migrate the attached location for a tenant shard to a specific pageserver.
108 : TenantShardMigrate {
109 : #[arg(long)]
110 0 : tenant_shard_id: TenantShardId,
111 : #[arg(long)]
112 0 : node: NodeId,
113 : },
114 : /// Modify the pageserver tenant configuration of a tenant: this is the configuration structure
115 : /// that is passed through to pageservers, and does not affect storage controller behavior.
116 : TenantConfig {
117 : #[arg(long)]
118 0 : tenant_id: TenantId,
119 : #[arg(long)]
120 0 : config: String,
121 : },
122 : /// Print details about a particular tenant, including all its shards' states.
123 : TenantDescribe {
124 : #[arg(long)]
125 0 : tenant_id: TenantId,
126 : },
127 : /// For a tenant which hasn't been onboarded to the storage controller yet, add it in secondary
128 : /// mode so that it can warm up content on a pageserver.
129 : TenantWarmup {
130 : #[arg(long)]
131 0 : tenant_id: TenantId,
132 : },
133 : /// Uncleanly drop a tenant from the storage controller: this doesn't delete anything from pageservers. Appropriate
134 : /// if you e.g. used `tenant-warmup` by mistake on a tenant ID that doesn't really exist, or is in some other region.
135 : TenantDrop {
136 : #[arg(long)]
137 0 : tenant_id: TenantId,
138 : #[arg(long)]
139 0 : unclean: bool,
140 : },
141 : NodeDrop {
142 : #[arg(long)]
143 0 : node_id: NodeId,
144 : #[arg(long)]
145 0 : unclean: bool,
146 : },
147 : TenantSetTimeBasedEviction {
148 : #[arg(long)]
149 0 : tenant_id: TenantId,
150 : #[arg(long)]
151 0 : period: humantime::Duration,
152 : #[arg(long)]
153 0 : threshold: humantime::Duration,
154 : },
155 : // Migrate away from a set of specified pageservers by moving the primary attachments to pageservers
156 : // outside of the specified set.
157 : BulkMigrate {
158 : // Set of pageserver node ids to drain.
159 : #[arg(long)]
160 0 : nodes: Vec<NodeId>,
161 : // Optional: migration concurrency (default is 8)
162 : #[arg(long)]
163 : concurrency: Option<usize>,
164 : // Optional: maximum number of shards to migrate
165 : #[arg(long)]
166 : max_shards: Option<usize>,
167 : // Optional: when set to true, nothing is migrated, but the plan is printed to stdout
168 : #[arg(long)]
169 : dry_run: Option<bool>,
170 : },
171 : /// Start draining the specified pageserver.
172 : /// The drain is complete when the schedulling policy returns to active.
173 : StartDrain {
174 : #[arg(long)]
175 0 : node_id: NodeId,
176 : },
177 : /// Cancel draining the specified pageserver and wait for `timeout`
178 : /// for the operation to be canceled. May be retried.
179 : CancelDrain {
180 : #[arg(long)]
181 0 : node_id: NodeId,
182 : #[arg(long)]
183 0 : timeout: humantime::Duration,
184 : },
185 : /// Start filling the specified pageserver.
186 : /// The drain is complete when the schedulling policy returns to active.
187 : StartFill {
188 : #[arg(long)]
189 0 : node_id: NodeId,
190 : },
191 : /// Cancel filling the specified pageserver and wait for `timeout`
192 : /// for the operation to be canceled. May be retried.
193 : CancelFill {
194 : #[arg(long)]
195 0 : node_id: NodeId,
196 : #[arg(long)]
197 0 : timeout: humantime::Duration,
198 : },
199 : }
200 :
201 0 : #[derive(Parser)]
202 : #[command(
203 : author,
204 : version,
205 : about,
206 : long_about = "CLI for Storage Controller Support/Debug"
207 : )]
208 : #[command(arg_required_else_help(true))]
209 : struct Cli {
210 : #[arg(long)]
211 : /// URL to storage controller. e.g. http://127.0.0.1:1234 when using `neon_local`
212 0 : api: Url,
213 :
214 : #[arg(long)]
215 : /// JWT token for authenticating with storage controller. Depending on the API used, this
216 : /// should have either `pageserverapi` or `admin` scopes: for convenience, you should mint
217 : /// a token with both scopes to use with this tool.
218 : jwt: Option<String>,
219 :
220 : #[command(subcommand)]
221 : command: Command,
222 : }
223 :
224 : #[derive(Debug, Clone)]
225 : struct PlacementPolicyArg(PlacementPolicy);
226 :
227 : impl FromStr for PlacementPolicyArg {
228 : type Err = anyhow::Error;
229 :
230 0 : fn from_str(s: &str) -> Result<Self, Self::Err> {
231 0 : match s {
232 0 : "detached" => Ok(Self(PlacementPolicy::Detached)),
233 0 : "secondary" => Ok(Self(PlacementPolicy::Secondary)),
234 0 : _ if s.starts_with("attached:") => {
235 0 : let mut splitter = s.split(':');
236 0 : let _prefix = splitter.next().unwrap();
237 0 : match splitter.next().and_then(|s| s.parse::<usize>().ok()) {
238 0 : Some(n) => Ok(Self(PlacementPolicy::Attached(n))),
239 0 : None => Err(anyhow::anyhow!(
240 0 : "Invalid format '{s}', a valid example is 'attached:1'"
241 0 : )),
242 : }
243 : }
244 0 : _ => Err(anyhow::anyhow!(
245 0 : "Unknown placement policy '{s}', try detached,secondary,attached:<n>"
246 0 : )),
247 : }
248 0 : }
249 : }
250 :
251 : #[derive(Debug, Clone)]
252 : struct ShardSchedulingPolicyArg(ShardSchedulingPolicy);
253 :
254 : impl FromStr for ShardSchedulingPolicyArg {
255 : type Err = anyhow::Error;
256 :
257 0 : fn from_str(s: &str) -> Result<Self, Self::Err> {
258 0 : match s {
259 0 : "active" => Ok(Self(ShardSchedulingPolicy::Active)),
260 0 : "essential" => Ok(Self(ShardSchedulingPolicy::Essential)),
261 0 : "pause" => Ok(Self(ShardSchedulingPolicy::Pause)),
262 0 : "stop" => Ok(Self(ShardSchedulingPolicy::Stop)),
263 0 : _ => Err(anyhow::anyhow!(
264 0 : "Unknown scheduling policy '{s}', try active,essential,pause,stop"
265 0 : )),
266 : }
267 0 : }
268 : }
269 :
270 : #[derive(Debug, Clone)]
271 : struct NodeAvailabilityArg(NodeAvailabilityWrapper);
272 :
273 : impl FromStr for NodeAvailabilityArg {
274 : type Err = anyhow::Error;
275 :
276 0 : fn from_str(s: &str) -> Result<Self, Self::Err> {
277 0 : match s {
278 0 : "active" => Ok(Self(NodeAvailabilityWrapper::Active)),
279 0 : "offline" => Ok(Self(NodeAvailabilityWrapper::Offline)),
280 0 : _ => Err(anyhow::anyhow!("Unknown availability state '{s}'")),
281 : }
282 0 : }
283 : }
284 :
285 0 : async fn wait_for_scheduling_policy<F>(
286 0 : client: Client,
287 0 : node_id: NodeId,
288 0 : timeout: Duration,
289 0 : f: F,
290 0 : ) -> anyhow::Result<NodeSchedulingPolicy>
291 0 : where
292 0 : F: Fn(NodeSchedulingPolicy) -> bool,
293 0 : {
294 0 : let waiter = tokio::time::timeout(timeout, async move {
295 : loop {
296 0 : let node = client
297 0 : .dispatch::<(), NodeDescribeResponse>(
298 0 : Method::GET,
299 0 : format!("control/v1/node/{node_id}"),
300 0 : None,
301 0 : )
302 0 : .await?;
303 :
304 0 : if f(node.scheduling) {
305 0 : return Ok::<NodeSchedulingPolicy, mgmt_api::Error>(node.scheduling);
306 0 : }
307 : }
308 0 : });
309 0 :
310 0 : Ok(waiter.await??)
311 0 : }
312 :
313 : #[tokio::main]
314 0 : async fn main() -> anyhow::Result<()> {
315 0 : let cli = Cli::parse();
316 0 :
317 0 : let storcon_client = Client::new(cli.api.clone(), cli.jwt.clone());
318 0 :
319 0 : let mut trimmed = cli.api.to_string();
320 0 : trimmed.pop();
321 0 : let vps_client = mgmt_api::Client::new(trimmed, cli.jwt.as_deref());
322 0 :
323 0 : match cli.command {
324 0 : Command::NodeRegister {
325 0 : node_id,
326 0 : listen_pg_addr,
327 0 : listen_pg_port,
328 0 : listen_http_addr,
329 0 : listen_http_port,
330 0 : availability_zone_id,
331 0 : } => {
332 0 : storcon_client
333 0 : .dispatch::<_, ()>(
334 0 : Method::POST,
335 0 : "control/v1/node".to_string(),
336 0 : Some(NodeRegisterRequest {
337 0 : node_id,
338 0 : listen_pg_addr,
339 0 : listen_pg_port,
340 0 : listen_http_addr,
341 0 : listen_http_port,
342 0 : availability_zone_id,
343 0 : }),
344 0 : )
345 0 : .await?;
346 0 : }
347 0 : Command::TenantCreate { tenant_id } => {
348 0 : storcon_client
349 0 : .dispatch::<_, ()>(
350 0 : Method::POST,
351 0 : "v1/tenant".to_string(),
352 0 : Some(TenantCreateRequest {
353 0 : new_tenant_id: TenantShardId::unsharded(tenant_id),
354 0 : generation: None,
355 0 : shard_parameters: ShardParameters::default(),
356 0 : placement_policy: Some(PlacementPolicy::Attached(1)),
357 0 : config: TenantConfig::default(),
358 0 : }),
359 0 : )
360 0 : .await?;
361 0 : }
362 0 : Command::TenantDelete { tenant_id } => {
363 0 : let status = vps_client
364 0 : .tenant_delete(TenantShardId::unsharded(tenant_id))
365 0 : .await?;
366 0 : tracing::info!("Delete status: {}", status);
367 0 : }
368 0 : Command::Nodes {} => {
369 0 : let mut resp = storcon_client
370 0 : .dispatch::<(), Vec<NodeDescribeResponse>>(
371 0 : Method::GET,
372 0 : "control/v1/node".to_string(),
373 0 : None,
374 0 : )
375 0 : .await?;
376 0 :
377 0 : resp.sort_by(|a, b| a.listen_http_addr.cmp(&b.listen_http_addr));
378 0 :
379 0 : let mut table = comfy_table::Table::new();
380 0 : table.set_header(["Id", "Hostname", "Scheduling", "Availability"]);
381 0 : for node in resp {
382 0 : table.add_row([
383 0 : format!("{}", node.id),
384 0 : node.listen_http_addr,
385 0 : format!("{:?}", node.scheduling),
386 0 : format!("{:?}", node.availability),
387 0 : ]);
388 0 : }
389 0 : println!("{table}");
390 0 : }
391 0 : Command::NodeConfigure {
392 0 : node_id,
393 0 : availability,
394 0 : scheduling,
395 0 : } => {
396 0 : let req = NodeConfigureRequest {
397 0 : node_id,
398 0 : availability: availability.map(|a| a.0),
399 0 : scheduling,
400 0 : };
401 0 : storcon_client
402 0 : .dispatch::<_, ()>(
403 0 : Method::PUT,
404 0 : format!("control/v1/node/{node_id}/config"),
405 0 : Some(req),
406 0 : )
407 0 : .await?;
408 0 : }
409 0 : Command::Tenants {
410 0 : node_id: Some(node_id),
411 0 : } => {
412 0 : let describe_response = storcon_client
413 0 : .dispatch::<(), NodeShardResponse>(
414 0 : Method::GET,
415 0 : format!("control/v1/node/{node_id}/shards"),
416 0 : None,
417 0 : )
418 0 : .await?;
419 0 : let shards = describe_response.shards;
420 0 : let mut table = comfy_table::Table::new();
421 0 : table.set_header([
422 0 : "Shard",
423 0 : "Intended Primary/Secondary",
424 0 : "Observed Primary/Secondary",
425 0 : ]);
426 0 : for shard in shards {
427 0 : table.add_row([
428 0 : format!("{}", shard.tenant_shard_id),
429 0 : match shard.is_intended_secondary {
430 0 : None => "".to_string(),
431 0 : Some(true) => "Secondary".to_string(),
432 0 : Some(false) => "Primary".to_string(),
433 0 : },
434 0 : match shard.is_observed_secondary {
435 0 : None => "".to_string(),
436 0 : Some(true) => "Secondary".to_string(),
437 0 : Some(false) => "Primary".to_string(),
438 0 : },
439 0 : ]);
440 0 : }
441 0 : println!("{table}");
442 0 : }
443 0 : Command::Tenants { node_id: None } => {
444 0 : let mut resp = storcon_client
445 0 : .dispatch::<(), Vec<TenantDescribeResponse>>(
446 0 : Method::GET,
447 0 : "control/v1/tenant".to_string(),
448 0 : None,
449 0 : )
450 0 : .await?;
451 0 :
452 0 : resp.sort_by(|a, b| a.tenant_id.cmp(&b.tenant_id));
453 0 :
454 0 : let mut table = comfy_table::Table::new();
455 0 : table.set_header([
456 0 : "TenantId",
457 0 : "ShardCount",
458 0 : "StripeSize",
459 0 : "Placement",
460 0 : "Scheduling",
461 0 : ]);
462 0 : for tenant in resp {
463 0 : let shard_zero = tenant.shards.into_iter().next().unwrap();
464 0 : table.add_row([
465 0 : format!("{}", tenant.tenant_id),
466 0 : format!("{}", shard_zero.tenant_shard_id.shard_count.literal()),
467 0 : format!("{:?}", tenant.stripe_size),
468 0 : format!("{:?}", tenant.policy),
469 0 : format!("{:?}", shard_zero.scheduling_policy),
470 0 : ]);
471 0 : }
472 0 :
473 0 : println!("{table}");
474 0 : }
475 0 : Command::TenantPolicy {
476 0 : tenant_id,
477 0 : placement,
478 0 : scheduling,
479 0 : } => {
480 0 : let req = TenantPolicyRequest {
481 0 : scheduling: scheduling.map(|s| s.0),
482 0 : placement: placement.map(|p| p.0),
483 0 : };
484 0 : storcon_client
485 0 : .dispatch::<_, ()>(
486 0 : Method::PUT,
487 0 : format!("control/v1/tenant/{tenant_id}/policy"),
488 0 : Some(req),
489 0 : )
490 0 : .await?;
491 0 : }
492 0 : Command::TenantShardSplit {
493 0 : tenant_id,
494 0 : shard_count,
495 0 : stripe_size,
496 0 : } => {
497 0 : let req = TenantShardSplitRequest {
498 0 : new_shard_count: shard_count,
499 0 : new_stripe_size: stripe_size.map(ShardStripeSize),
500 0 : };
501 0 :
502 0 : let response = storcon_client
503 0 : .dispatch::<TenantShardSplitRequest, TenantShardSplitResponse>(
504 0 : Method::PUT,
505 0 : format!("control/v1/tenant/{tenant_id}/shard_split"),
506 0 : Some(req),
507 0 : )
508 0 : .await?;
509 0 : println!(
510 0 : "Split tenant {} into {} shards: {}",
511 0 : tenant_id,
512 0 : shard_count,
513 0 : response
514 0 : .new_shards
515 0 : .iter()
516 0 : .map(|s| format!("{:?}", s))
517 0 : .collect::<Vec<_>>()
518 0 : .join(",")
519 0 : );
520 0 : }
521 0 : Command::TenantShardMigrate {
522 0 : tenant_shard_id,
523 0 : node,
524 0 : } => {
525 0 : let req = TenantShardMigrateRequest {
526 0 : tenant_shard_id,
527 0 : node_id: node,
528 0 : };
529 0 :
530 0 : storcon_client
531 0 : .dispatch::<TenantShardMigrateRequest, TenantShardMigrateResponse>(
532 0 : Method::PUT,
533 0 : format!("control/v1/tenant/{tenant_shard_id}/migrate"),
534 0 : Some(req),
535 0 : )
536 0 : .await?;
537 0 : }
538 0 : Command::TenantConfig { tenant_id, config } => {
539 0 : let tenant_conf = serde_json::from_str(&config)?;
540 0 :
541 0 : vps_client
542 0 : .tenant_config(&TenantConfigRequest {
543 0 : tenant_id,
544 0 : config: tenant_conf,
545 0 : })
546 0 : .await?;
547 0 : }
548 0 : Command::TenantDescribe { tenant_id } => {
549 0 : let describe_response = storcon_client
550 0 : .dispatch::<(), TenantDescribeResponse>(
551 0 : Method::GET,
552 0 : format!("control/v1/tenant/{tenant_id}"),
553 0 : None,
554 0 : )
555 0 : .await?;
556 0 : let shards = describe_response.shards;
557 0 : let mut table = comfy_table::Table::new();
558 0 : table.set_header(["Shard", "Attached", "Secondary", "Last error", "status"]);
559 0 : for shard in shards {
560 0 : let secondary = shard
561 0 : .node_secondary
562 0 : .iter()
563 0 : .map(|n| format!("{}", n))
564 0 : .collect::<Vec<_>>()
565 0 : .join(",");
566 0 :
567 0 : let mut status_parts = Vec::new();
568 0 : if shard.is_reconciling {
569 0 : status_parts.push("reconciling");
570 0 : }
571 0 :
572 0 : if shard.is_pending_compute_notification {
573 0 : status_parts.push("pending_compute");
574 0 : }
575 0 :
576 0 : if shard.is_splitting {
577 0 : status_parts.push("splitting");
578 0 : }
579 0 : let status = status_parts.join(",");
580 0 :
581 0 : table.add_row([
582 0 : format!("{}", shard.tenant_shard_id),
583 0 : shard
584 0 : .node_attached
585 0 : .map(|n| format!("{}", n))
586 0 : .unwrap_or(String::new()),
587 0 : secondary,
588 0 : shard.last_error,
589 0 : status,
590 0 : ]);
591 0 : }
592 0 : println!("{table}");
593 0 : }
594 0 : Command::TenantWarmup { tenant_id } => {
595 0 : let describe_response = storcon_client
596 0 : .dispatch::<(), TenantDescribeResponse>(
597 0 : Method::GET,
598 0 : format!("control/v1/tenant/{tenant_id}"),
599 0 : None,
600 0 : )
601 0 : .await;
602 0 : match describe_response {
603 0 : Ok(describe) => {
604 0 : if matches!(describe.policy, PlacementPolicy::Secondary) {
605 0 : // Fine: it's already known to controller in secondary mode: calling
606 0 : // again to put it into secondary mode won't cause problems.
607 0 : } else {
608 0 : anyhow::bail!("Tenant already present with policy {:?}", describe.policy);
609 0 : }
610 0 : }
611 0 : Err(mgmt_api::Error::ApiError(StatusCode::NOT_FOUND, _)) => {
612 0 : // Fine: this tenant isn't know to the storage controller yet.
613 0 : }
614 0 : Err(e) => {
615 0 : // Unexpected API error
616 0 : return Err(e.into());
617 0 : }
618 0 : }
619 0 :
620 0 : vps_client
621 0 : .location_config(
622 0 : TenantShardId::unsharded(tenant_id),
623 0 : pageserver_api::models::LocationConfig {
624 0 : mode: pageserver_api::models::LocationConfigMode::Secondary,
625 0 : generation: None,
626 0 : secondary_conf: Some(LocationConfigSecondary { warm: true }),
627 0 : shard_number: 0,
628 0 : shard_count: 0,
629 0 : shard_stripe_size: ShardParameters::DEFAULT_STRIPE_SIZE.0,
630 0 : tenant_conf: TenantConfig::default(),
631 0 : },
632 0 : None,
633 0 : true,
634 0 : )
635 0 : .await?;
636 0 :
637 0 : let describe_response = storcon_client
638 0 : .dispatch::<(), TenantDescribeResponse>(
639 0 : Method::GET,
640 0 : format!("control/v1/tenant/{tenant_id}"),
641 0 : None,
642 0 : )
643 0 : .await?;
644 0 :
645 0 : let secondary_ps_id = describe_response
646 0 : .shards
647 0 : .first()
648 0 : .unwrap()
649 0 : .node_secondary
650 0 : .first()
651 0 : .unwrap();
652 0 :
653 0 : println!("Tenant {tenant_id} warming up on pageserver {secondary_ps_id}");
654 0 : loop {
655 0 : let (status, progress) = vps_client
656 0 : .tenant_secondary_download(
657 0 : TenantShardId::unsharded(tenant_id),
658 0 : Some(Duration::from_secs(10)),
659 0 : )
660 0 : .await?;
661 0 : println!(
662 0 : "Progress: {}/{} layers, {}/{} bytes",
663 0 : progress.layers_downloaded,
664 0 : progress.layers_total,
665 0 : progress.bytes_downloaded,
666 0 : progress.bytes_total
667 0 : );
668 0 : match status {
669 0 : StatusCode::OK => {
670 0 : println!("Download complete");
671 0 : break;
672 0 : }
673 0 : StatusCode::ACCEPTED => {
674 0 : // Loop
675 0 : }
676 0 : _ => {
677 0 : anyhow::bail!("Unexpected download status: {status}");
678 0 : }
679 0 : }
680 0 : }
681 0 : }
682 0 : Command::TenantDrop { tenant_id, unclean } => {
683 0 : if !unclean {
684 0 : anyhow::bail!("This command is not a tenant deletion, and uncleanly drops all controller state for the tenant. If you know what you're doing, add `--unclean` to proceed.")
685 0 : }
686 0 : storcon_client
687 0 : .dispatch::<(), ()>(
688 0 : Method::POST,
689 0 : format!("debug/v1/tenant/{tenant_id}/drop"),
690 0 : None,
691 0 : )
692 0 : .await?;
693 0 : }
694 0 : Command::NodeDrop { node_id, unclean } => {
695 0 : if !unclean {
696 0 : anyhow::bail!("This command is not a clean node decommission, and uncleanly drops all controller state for the node, without checking if any tenants still refer to it. If you know what you're doing, add `--unclean` to proceed.")
697 0 : }
698 0 : storcon_client
699 0 : .dispatch::<(), ()>(Method::POST, format!("debug/v1/node/{node_id}/drop"), None)
700 0 : .await?;
701 0 : }
702 0 : Command::NodeDelete { node_id } => {
703 0 : storcon_client
704 0 : .dispatch::<(), ()>(Method::DELETE, format!("control/v1/node/{node_id}"), None)
705 0 : .await?;
706 0 : }
707 0 : Command::TenantSetTimeBasedEviction {
708 0 : tenant_id,
709 0 : period,
710 0 : threshold,
711 0 : } => {
712 0 : vps_client
713 0 : .tenant_config(&TenantConfigRequest {
714 0 : tenant_id,
715 0 : config: TenantConfig {
716 0 : eviction_policy: Some(EvictionPolicy::LayerAccessThreshold(
717 0 : EvictionPolicyLayerAccessThreshold {
718 0 : period: period.into(),
719 0 : threshold: threshold.into(),
720 0 : },
721 0 : )),
722 0 : heatmap_period: Some("300s".to_string()),
723 0 : ..Default::default()
724 0 : },
725 0 : })
726 0 : .await?;
727 0 : }
728 0 : Command::BulkMigrate {
729 0 : nodes,
730 0 : concurrency,
731 0 : max_shards,
732 0 : dry_run,
733 0 : } => {
734 0 : // Load the list of nodes, split them up into the drained and filled sets,
735 0 : // and validate that draining is possible.
736 0 : let node_descs = storcon_client
737 0 : .dispatch::<(), Vec<NodeDescribeResponse>>(
738 0 : Method::GET,
739 0 : "control/v1/node".to_string(),
740 0 : None,
741 0 : )
742 0 : .await?;
743 0 :
744 0 : let mut node_to_drain_descs = Vec::new();
745 0 : let mut node_to_fill_descs = Vec::new();
746 0 :
747 0 : for desc in node_descs {
748 0 : let to_drain = nodes.iter().any(|id| *id == desc.id);
749 0 : if to_drain {
750 0 : node_to_drain_descs.push(desc);
751 0 : } else {
752 0 : node_to_fill_descs.push(desc);
753 0 : }
754 0 : }
755 0 :
756 0 : if nodes.len() != node_to_drain_descs.len() {
757 0 : anyhow::bail!("Bulk migration requested away from node which doesn't exist.")
758 0 : }
759 0 :
760 0 : node_to_fill_descs.retain(|desc| {
761 0 : matches!(desc.availability, NodeAvailabilityWrapper::Active)
762 0 : && matches!(
763 0 : desc.scheduling,
764 0 : NodeSchedulingPolicy::Active | NodeSchedulingPolicy::Filling
765 0 : )
766 0 : });
767 0 :
768 0 : if node_to_fill_descs.is_empty() {
769 0 : anyhow::bail!("There are no nodes to migrate to")
770 0 : }
771 0 :
772 0 : // Set the node scheduling policy to draining for the nodes which
773 0 : // we plan to drain.
774 0 : for node_desc in node_to_drain_descs.iter() {
775 0 : let req = NodeConfigureRequest {
776 0 : node_id: node_desc.id,
777 0 : availability: None,
778 0 : scheduling: Some(NodeSchedulingPolicy::Draining),
779 0 : };
780 0 :
781 0 : storcon_client
782 0 : .dispatch::<_, ()>(
783 0 : Method::PUT,
784 0 : format!("control/v1/node/{}/config", node_desc.id),
785 0 : Some(req),
786 0 : )
787 0 : .await?;
788 0 : }
789 0 :
790 0 : // Perform the migration: move each tenant shard scheduled on a node to
791 0 : // be drained to a node which is being filled. A simple round robin
792 0 : // strategy is used to pick the new node.
793 0 : let tenants = storcon_client
794 0 : .dispatch::<(), Vec<TenantDescribeResponse>>(
795 0 : Method::GET,
796 0 : "control/v1/tenant".to_string(),
797 0 : None,
798 0 : )
799 0 : .await?;
800 0 :
801 0 : let mut selected_node_idx = 0;
802 0 :
803 0 : struct MigrationMove {
804 0 : tenant_shard_id: TenantShardId,
805 0 : from: NodeId,
806 0 : to: NodeId,
807 0 : }
808 0 :
809 0 : let mut moves: Vec<MigrationMove> = Vec::new();
810 0 :
811 0 : let shards = tenants
812 0 : .into_iter()
813 0 : .flat_map(|tenant| tenant.shards.into_iter());
814 0 : for shard in shards {
815 0 : if let Some(max_shards) = max_shards {
816 0 : if moves.len() >= max_shards {
817 0 : println!(
818 0 : "Stop planning shard moves since the requested maximum was reached"
819 0 : );
820 0 : break;
821 0 : }
822 0 : }
823 0 :
824 0 : let should_migrate = {
825 0 : if let Some(attached_to) = shard.node_attached {
826 0 : node_to_drain_descs
827 0 : .iter()
828 0 : .map(|desc| desc.id)
829 0 : .any(|id| id == attached_to)
830 0 : } else {
831 0 : false
832 0 : }
833 0 : };
834 0 :
835 0 : if !should_migrate {
836 0 : continue;
837 0 : }
838 0 :
839 0 : moves.push(MigrationMove {
840 0 : tenant_shard_id: shard.tenant_shard_id,
841 0 : from: shard
842 0 : .node_attached
843 0 : .expect("We only migrate attached tenant shards"),
844 0 : to: node_to_fill_descs[selected_node_idx].id,
845 0 : });
846 0 : selected_node_idx = (selected_node_idx + 1) % node_to_fill_descs.len();
847 0 : }
848 0 :
849 0 : let total_moves = moves.len();
850 0 :
851 0 : if dry_run == Some(true) {
852 0 : println!("Dryrun requested. Planned {total_moves} moves:");
853 0 : for mv in &moves {
854 0 : println!("{}: {} -> {}", mv.tenant_shard_id, mv.from, mv.to)
855 0 : }
856 0 :
857 0 : return Ok(());
858 0 : }
859 0 :
860 0 : const DEFAULT_MIGRATE_CONCURRENCY: usize = 8;
861 0 : let mut stream = futures::stream::iter(moves)
862 0 : .map(|mv| {
863 0 : let client = Client::new(cli.api.clone(), cli.jwt.clone());
864 0 : async move {
865 0 : client
866 0 : .dispatch::<TenantShardMigrateRequest, TenantShardMigrateResponse>(
867 0 : Method::PUT,
868 0 : format!("control/v1/tenant/{}/migrate", mv.tenant_shard_id),
869 0 : Some(TenantShardMigrateRequest {
870 0 : tenant_shard_id: mv.tenant_shard_id,
871 0 : node_id: mv.to,
872 0 : }),
873 0 : )
874 0 : .await
875 0 : .map_err(|e| (mv.tenant_shard_id, mv.from, mv.to, e))
876 0 : }
877 0 : })
878 0 : .buffered(concurrency.unwrap_or(DEFAULT_MIGRATE_CONCURRENCY));
879 0 :
880 0 : let mut success = 0;
881 0 : let mut failure = 0;
882 0 :
883 0 : while let Some(res) = stream.next().await {
884 0 : match res {
885 0 : Ok(_) => {
886 0 : success += 1;
887 0 : }
888 0 : Err((tenant_shard_id, from, to, error)) => {
889 0 : failure += 1;
890 0 : println!(
891 0 : "Failed to migrate {} from node {} to node {}: {}",
892 0 : tenant_shard_id, from, to, error
893 0 : );
894 0 : }
895 0 : }
896 0 :
897 0 : if (success + failure) % 20 == 0 {
898 0 : println!(
899 0 : "Processed {}/{} shards: {} succeeded, {} failed",
900 0 : success + failure,
901 0 : total_moves,
902 0 : success,
903 0 : failure
904 0 : );
905 0 : }
906 0 : }
907 0 :
908 0 : println!(
909 0 : "Processed {}/{} shards: {} succeeded, {} failed",
910 0 : success + failure,
911 0 : total_moves,
912 0 : success,
913 0 : failure
914 0 : );
915 0 : }
916 0 : Command::StartDrain { node_id } => {
917 0 : storcon_client
918 0 : .dispatch::<(), ()>(
919 0 : Method::PUT,
920 0 : format!("control/v1/node/{node_id}/drain"),
921 0 : None,
922 0 : )
923 0 : .await?;
924 0 : println!("Drain started for {node_id}");
925 0 : }
926 0 : Command::CancelDrain { node_id, timeout } => {
927 0 : storcon_client
928 0 : .dispatch::<(), ()>(
929 0 : Method::DELETE,
930 0 : format!("control/v1/node/{node_id}/drain"),
931 0 : None,
932 0 : )
933 0 : .await?;
934 0 :
935 0 : println!("Waiting for node {node_id} to quiesce on scheduling policy ...");
936 0 :
937 0 : let final_policy =
938 0 : wait_for_scheduling_policy(storcon_client, node_id, *timeout, |sched| {
939 0 : use NodeSchedulingPolicy::*;
940 0 : matches!(sched, Active | PauseForRestart)
941 0 : })
942 0 : .await?;
943 0 :
944 0 : println!(
945 0 : "Drain was cancelled for node {node_id}. Schedulling policy is now {final_policy:?}"
946 0 : );
947 0 : }
948 0 : Command::StartFill { node_id } => {
949 0 : storcon_client
950 0 : .dispatch::<(), ()>(Method::PUT, format!("control/v1/node/{node_id}/fill"), None)
951 0 : .await?;
952 0 :
953 0 : println!("Fill started for {node_id}");
954 0 : }
955 0 : Command::CancelFill { node_id, timeout } => {
956 0 : storcon_client
957 0 : .dispatch::<(), ()>(
958 0 : Method::DELETE,
959 0 : format!("control/v1/node/{node_id}/fill"),
960 0 : None,
961 0 : )
962 0 : .await?;
963 0 :
964 0 : println!("Waiting for node {node_id} to quiesce on scheduling policy ...");
965 0 :
966 0 : let final_policy =
967 0 : wait_for_scheduling_policy(storcon_client, node_id, *timeout, |sched| {
968 0 : use NodeSchedulingPolicy::*;
969 0 : matches!(sched, Active)
970 0 : })
971 0 : .await?;
972 0 :
973 0 : println!(
974 0 : "Fill was cancelled for node {node_id}. Schedulling policy is now {final_policy:?}"
975 0 : );
976 0 : }
977 0 : }
978 0 :
979 0 : Ok(())
980 0 : }
|
