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