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