Line data Source code
1 : use futures::StreamExt;
2 : use std::{
3 : collections::{HashMap, HashSet},
4 : str::FromStr,
5 : time::Duration,
6 : };
7 :
8 : use clap::{Parser, Subcommand};
9 : use pageserver_api::{
10 : controller_api::{
11 : AvailabilityZone, NodeAvailabilityWrapper, NodeDescribeResponse, NodeShardResponse,
12 : SafekeeperDescribeResponse, SafekeeperSchedulingPolicyRequest, ShardSchedulingPolicy,
13 : ShardsPreferredAzsRequest, ShardsPreferredAzsResponse, SkSchedulingPolicy,
14 : TenantCreateRequest, TenantDescribeResponse, TenantPolicyRequest,
15 : },
16 : models::{
17 : EvictionPolicy, EvictionPolicyLayerAccessThreshold, LocationConfigSecondary,
18 : ShardParameters, TenantConfig, TenantConfigPatchRequest, TenantConfigRequest,
19 : TenantShardSplitRequest, TenantShardSplitResponse,
20 : },
21 : shard::{ShardStripeSize, TenantShardId},
22 : };
23 : use pageserver_client::mgmt_api::{self};
24 : use reqwest::{Method, StatusCode, Url};
25 : use utils::id::{NodeId, TenantId, TimelineId};
26 :
27 : use pageserver_api::controller_api::{
28 : NodeConfigureRequest, NodeRegisterRequest, NodeSchedulingPolicy, PlacementPolicy,
29 : TenantShardMigrateRequest, TenantShardMigrateResponse,
30 : };
31 : use storage_controller_client::control_api::Client;
32 :
33 : #[derive(Subcommand, Debug)]
34 : enum Command {
35 : /// Register a pageserver with the storage controller. This shouldn't usually be necessary,
36 : /// since pageservers auto-register when they start up
37 : NodeRegister {
38 : #[arg(long)]
39 0 : node_id: NodeId,
40 :
41 : #[arg(long)]
42 0 : listen_pg_addr: String,
43 : #[arg(long)]
44 0 : listen_pg_port: u16,
45 :
46 : #[arg(long)]
47 0 : listen_http_addr: String,
48 : #[arg(long)]
49 0 : listen_http_port: u16,
50 : #[arg(long)]
51 : listen_https_port: Option<u16>,
52 :
53 : #[arg(long)]
54 0 : availability_zone_id: String,
55 : },
56 :
57 : /// Modify a node's configuration in the storage controller
58 : NodeConfigure {
59 : #[arg(long)]
60 0 : node_id: NodeId,
61 :
62 : /// Availability is usually auto-detected based on heartbeats. Set 'offline' here to
63 : /// manually mark a node offline
64 : #[arg(long)]
65 : availability: Option<NodeAvailabilityArg>,
66 : /// Scheduling policy controls whether tenant shards may be scheduled onto this node.
67 : #[arg(long)]
68 : scheduling: Option<NodeSchedulingPolicy>,
69 : },
70 : NodeDelete {
71 : #[arg(long)]
72 0 : node_id: NodeId,
73 : },
74 : /// Modify a tenant's policies in the storage controller
75 : TenantPolicy {
76 : #[arg(long)]
77 0 : tenant_id: TenantId,
78 : /// Placement policy controls whether a tenant is `detached`, has only a secondary location (`secondary`),
79 : /// or is in the normal attached state with N secondary locations (`attached:N`)
80 : #[arg(long)]
81 : placement: Option<PlacementPolicyArg>,
82 : /// Scheduling policy enables pausing the controller's scheduling activity involving this tenant. `active` is normal,
83 : /// `essential` disables optimization scheduling changes, `pause` disables all scheduling changes, and `stop` prevents
84 : /// all reconciliation activity including for scheduling changes already made. `pause` and `stop` can make a tenant
85 : /// unavailable, and are only for use in emergencies.
86 : #[arg(long)]
87 : scheduling: Option<ShardSchedulingPolicyArg>,
88 : },
89 : /// List nodes known to the storage controller
90 : Nodes {},
91 : /// List tenants known to the storage controller
92 : Tenants {
93 : /// If this field is set, it will list the tenants on a specific node
94 : node_id: Option<NodeId>,
95 : },
96 : /// Create a new tenant in the storage controller, and by extension on pageservers.
97 : TenantCreate {
98 : #[arg(long)]
99 0 : tenant_id: TenantId,
100 : },
101 : /// Delete a tenant in the storage controller, and by extension on pageservers.
102 : TenantDelete {
103 : #[arg(long)]
104 0 : tenant_id: TenantId,
105 : },
106 : /// Split an existing tenant into a higher number of shards than its current shard count.
107 : TenantShardSplit {
108 : #[arg(long)]
109 0 : tenant_id: TenantId,
110 : #[arg(long)]
111 0 : shard_count: u8,
112 : /// Optional, in 8kiB pages. e.g. set 2048 for 16MB stripes.
113 : #[arg(long)]
114 : stripe_size: Option<u32>,
115 : },
116 : /// Migrate the attached location for a tenant shard to a specific pageserver.
117 : TenantShardMigrate {
118 : #[arg(long)]
119 0 : tenant_shard_id: TenantShardId,
120 : #[arg(long)]
121 0 : node: NodeId,
122 : },
123 : /// Migrate the secondary location for a tenant shard to a specific pageserver.
124 : TenantShardMigrateSecondary {
125 : #[arg(long)]
126 0 : tenant_shard_id: TenantShardId,
127 : #[arg(long)]
128 0 : node: NodeId,
129 : },
130 : /// Cancel any ongoing reconciliation for this shard
131 : TenantShardCancelReconcile {
132 : #[arg(long)]
133 0 : tenant_shard_id: TenantShardId,
134 : },
135 : /// Set the pageserver tenant configuration of a tenant: this is the configuration structure
136 : /// that is passed through to pageservers, and does not affect storage controller behavior.
137 : /// Any previous tenant configs are overwritten.
138 : SetTenantConfig {
139 : #[arg(long)]
140 0 : tenant_id: TenantId,
141 : #[arg(long)]
142 0 : config: String,
143 : },
144 : /// Patch the pageserver tenant configuration of a tenant. Any fields with null values in the
145 : /// provided JSON are unset from the tenant config and all fields with non-null values are set.
146 : /// Unspecified fields are not changed.
147 : PatchTenantConfig {
148 : #[arg(long)]
149 0 : tenant_id: TenantId,
150 : #[arg(long)]
151 0 : config: String,
152 : },
153 : /// Print details about a particular tenant, including all its shards' states.
154 : TenantDescribe {
155 : #[arg(long)]
156 0 : tenant_id: TenantId,
157 : },
158 : /// For a tenant which hasn't been onboarded to the storage controller yet, add it in secondary
159 : /// mode so that it can warm up content on a pageserver.
160 : TenantWarmup {
161 : #[arg(long)]
162 0 : tenant_id: TenantId,
163 : },
164 : TenantSetPreferredAz {
165 : #[arg(long)]
166 0 : tenant_id: TenantId,
167 : #[arg(long)]
168 : preferred_az: Option<String>,
169 : },
170 : /// Uncleanly drop a tenant from the storage controller: this doesn't delete anything from pageservers. Appropriate
171 : /// if you e.g. used `tenant-warmup` by mistake on a tenant ID that doesn't really exist, or is in some other region.
172 : TenantDrop {
173 : #[arg(long)]
174 0 : tenant_id: TenantId,
175 : #[arg(long)]
176 0 : unclean: bool,
177 : },
178 : NodeDrop {
179 : #[arg(long)]
180 0 : node_id: NodeId,
181 : #[arg(long)]
182 0 : unclean: bool,
183 : },
184 : TenantSetTimeBasedEviction {
185 : #[arg(long)]
186 0 : tenant_id: TenantId,
187 : #[arg(long)]
188 0 : period: humantime::Duration,
189 : #[arg(long)]
190 0 : threshold: humantime::Duration,
191 : },
192 : // Migrate away from a set of specified pageservers by moving the primary attachments to pageservers
193 : // outside of the specified set.
194 : BulkMigrate {
195 : // Set of pageserver node ids to drain.
196 : #[arg(long)]
197 0 : nodes: Vec<NodeId>,
198 : // Optional: migration concurrency (default is 8)
199 : #[arg(long)]
200 : concurrency: Option<usize>,
201 : // Optional: maximum number of shards to migrate
202 : #[arg(long)]
203 : max_shards: Option<usize>,
204 : // Optional: when set to true, nothing is migrated, but the plan is printed to stdout
205 : #[arg(long)]
206 : dry_run: Option<bool>,
207 : },
208 : /// Start draining the specified pageserver.
209 : /// The drain is complete when the schedulling policy returns to active.
210 : StartDrain {
211 : #[arg(long)]
212 0 : node_id: NodeId,
213 : },
214 : /// Cancel draining the specified pageserver and wait for `timeout`
215 : /// for the operation to be canceled. May be retried.
216 : CancelDrain {
217 : #[arg(long)]
218 0 : node_id: NodeId,
219 : #[arg(long)]
220 0 : timeout: humantime::Duration,
221 : },
222 : /// Start filling the specified pageserver.
223 : /// The drain is complete when the schedulling policy returns to active.
224 : StartFill {
225 : #[arg(long)]
226 0 : node_id: NodeId,
227 : },
228 : /// Cancel filling the specified pageserver and wait for `timeout`
229 : /// for the operation to be canceled. May be retried.
230 : CancelFill {
231 : #[arg(long)]
232 0 : node_id: NodeId,
233 : #[arg(long)]
234 0 : timeout: humantime::Duration,
235 : },
236 : /// List safekeepers known to the storage controller
237 : Safekeepers {},
238 : /// Set the scheduling policy of the specified safekeeper
239 : SafekeeperScheduling {
240 : #[arg(long)]
241 0 : node_id: NodeId,
242 : #[arg(long)]
243 0 : scheduling_policy: SkSchedulingPolicyArg,
244 : },
245 : /// Downloads any missing heatmap layers for all shard for a given timeline
246 : DownloadHeatmapLayers {
247 : /// Tenant ID or tenant shard ID. When an unsharded tenant ID is specified,
248 : /// the operation is performed on all shards. When a sharded tenant ID is
249 : /// specified, the operation is only performed on the specified shard.
250 : #[arg(long)]
251 0 : tenant_shard_id: TenantShardId,
252 : #[arg(long)]
253 0 : timeline_id: TimelineId,
254 : /// Optional: Maximum download concurrency (default is 16)
255 : #[arg(long)]
256 : concurrency: Option<usize>,
257 : },
258 : }
259 :
260 : #[derive(Parser)]
261 : #[command(
262 : author,
263 : version,
264 : about,
265 : long_about = "CLI for Storage Controller Support/Debug"
266 : )]
267 : #[command(arg_required_else_help(true))]
268 : struct Cli {
269 : #[arg(long)]
270 : /// URL to storage controller. e.g. http://127.0.0.1:1234 when using `neon_local`
271 0 : api: Url,
272 :
273 : #[arg(long)]
274 : /// JWT token for authenticating with storage controller. Depending on the API used, this
275 : /// should have either `pageserverapi` or `admin` scopes: for convenience, you should mint
276 : /// a token with both scopes to use with this tool.
277 : jwt: Option<String>,
278 :
279 : #[command(subcommand)]
280 : command: Command,
281 : }
282 :
283 : #[derive(Debug, Clone)]
284 : struct PlacementPolicyArg(PlacementPolicy);
285 :
286 : impl FromStr for PlacementPolicyArg {
287 : type Err = anyhow::Error;
288 :
289 0 : fn from_str(s: &str) -> Result<Self, Self::Err> {
290 0 : match s {
291 0 : "detached" => Ok(Self(PlacementPolicy::Detached)),
292 0 : "secondary" => Ok(Self(PlacementPolicy::Secondary)),
293 0 : _ if s.starts_with("attached:") => {
294 0 : let mut splitter = s.split(':');
295 0 : let _prefix = splitter.next().unwrap();
296 0 : match splitter.next().and_then(|s| s.parse::<usize>().ok()) {
297 0 : Some(n) => Ok(Self(PlacementPolicy::Attached(n))),
298 0 : None => Err(anyhow::anyhow!(
299 0 : "Invalid format '{s}', a valid example is 'attached:1'"
300 0 : )),
301 : }
302 : }
303 0 : _ => Err(anyhow::anyhow!(
304 0 : "Unknown placement policy '{s}', try detached,secondary,attached:<n>"
305 0 : )),
306 : }
307 0 : }
308 : }
309 :
310 : #[derive(Debug, Clone)]
311 : struct SkSchedulingPolicyArg(SkSchedulingPolicy);
312 :
313 : impl FromStr for SkSchedulingPolicyArg {
314 : type Err = anyhow::Error;
315 :
316 0 : fn from_str(s: &str) -> Result<Self, Self::Err> {
317 0 : SkSchedulingPolicy::from_str(s).map(Self)
318 0 : }
319 : }
320 :
321 : #[derive(Debug, Clone)]
322 : struct ShardSchedulingPolicyArg(ShardSchedulingPolicy);
323 :
324 : impl FromStr for ShardSchedulingPolicyArg {
325 : type Err = anyhow::Error;
326 :
327 0 : fn from_str(s: &str) -> Result<Self, Self::Err> {
328 0 : match s {
329 0 : "active" => Ok(Self(ShardSchedulingPolicy::Active)),
330 0 : "essential" => Ok(Self(ShardSchedulingPolicy::Essential)),
331 0 : "pause" => Ok(Self(ShardSchedulingPolicy::Pause)),
332 0 : "stop" => Ok(Self(ShardSchedulingPolicy::Stop)),
333 0 : _ => Err(anyhow::anyhow!(
334 0 : "Unknown scheduling policy '{s}', try active,essential,pause,stop"
335 0 : )),
336 : }
337 0 : }
338 : }
339 :
340 : #[derive(Debug, Clone)]
341 : struct NodeAvailabilityArg(NodeAvailabilityWrapper);
342 :
343 : impl FromStr for NodeAvailabilityArg {
344 : type Err = anyhow::Error;
345 :
346 0 : fn from_str(s: &str) -> Result<Self, Self::Err> {
347 0 : match s {
348 0 : "active" => Ok(Self(NodeAvailabilityWrapper::Active)),
349 0 : "offline" => Ok(Self(NodeAvailabilityWrapper::Offline)),
350 0 : _ => Err(anyhow::anyhow!("Unknown availability state '{s}'")),
351 : }
352 0 : }
353 : }
354 :
355 0 : async fn wait_for_scheduling_policy<F>(
356 0 : client: Client,
357 0 : node_id: NodeId,
358 0 : timeout: Duration,
359 0 : f: F,
360 0 : ) -> anyhow::Result<NodeSchedulingPolicy>
361 0 : where
362 0 : F: Fn(NodeSchedulingPolicy) -> bool,
363 0 : {
364 0 : let waiter = tokio::time::timeout(timeout, async move {
365 : loop {
366 0 : let node = client
367 0 : .dispatch::<(), NodeDescribeResponse>(
368 0 : Method::GET,
369 0 : format!("control/v1/node/{node_id}"),
370 0 : None,
371 0 : )
372 0 : .await?;
373 :
374 0 : if f(node.scheduling) {
375 0 : return Ok::<NodeSchedulingPolicy, mgmt_api::Error>(node.scheduling);
376 0 : }
377 : }
378 0 : });
379 0 :
380 0 : Ok(waiter.await??)
381 0 : }
382 :
383 : #[tokio::main]
384 0 : async fn main() -> anyhow::Result<()> {
385 0 : let cli = Cli::parse();
386 0 :
387 0 : let storcon_client = Client::new(cli.api.clone(), cli.jwt.clone());
388 0 :
389 0 : let mut trimmed = cli.api.to_string();
390 0 : trimmed.pop();
391 0 : let vps_client = mgmt_api::Client::new(trimmed, cli.jwt.as_deref());
392 0 :
393 0 : match cli.command {
394 0 : Command::NodeRegister {
395 0 : node_id,
396 0 : listen_pg_addr,
397 0 : listen_pg_port,
398 0 : listen_http_addr,
399 0 : listen_http_port,
400 0 : listen_https_port,
401 0 : availability_zone_id,
402 0 : } => {
403 0 : storcon_client
404 0 : .dispatch::<_, ()>(
405 0 : Method::POST,
406 0 : "control/v1/node".to_string(),
407 0 : Some(NodeRegisterRequest {
408 0 : node_id,
409 0 : listen_pg_addr,
410 0 : listen_pg_port,
411 0 : listen_http_addr,
412 0 : listen_http_port,
413 0 : listen_https_port,
414 0 : availability_zone_id: AvailabilityZone(availability_zone_id),
415 0 : }),
416 0 : )
417 0 : .await?;
418 0 : }
419 0 : Command::TenantCreate { tenant_id } => {
420 0 : storcon_client
421 0 : .dispatch::<_, ()>(
422 0 : Method::POST,
423 0 : "v1/tenant".to_string(),
424 0 : Some(TenantCreateRequest {
425 0 : new_tenant_id: TenantShardId::unsharded(tenant_id),
426 0 : generation: None,
427 0 : shard_parameters: ShardParameters::default(),
428 0 : placement_policy: Some(PlacementPolicy::Attached(1)),
429 0 : config: TenantConfig::default(),
430 0 : }),
431 0 : )
432 0 : .await?;
433 0 : }
434 0 : Command::TenantDelete { tenant_id } => {
435 0 : let status = vps_client
436 0 : .tenant_delete(TenantShardId::unsharded(tenant_id))
437 0 : .await?;
438 0 : tracing::info!("Delete status: {}", status);
439 0 : }
440 0 : Command::Nodes {} => {
441 0 : let mut resp = storcon_client
442 0 : .dispatch::<(), Vec<NodeDescribeResponse>>(
443 0 : Method::GET,
444 0 : "control/v1/node".to_string(),
445 0 : None,
446 0 : )
447 0 : .await?;
448 0 :
449 0 : resp.sort_by(|a, b| a.listen_http_addr.cmp(&b.listen_http_addr));
450 0 :
451 0 : let mut table = comfy_table::Table::new();
452 0 : table.set_header(["Id", "Hostname", "AZ", "Scheduling", "Availability"]);
453 0 : for node in resp {
454 0 : table.add_row([
455 0 : format!("{}", node.id),
456 0 : node.listen_http_addr,
457 0 : node.availability_zone_id,
458 0 : format!("{:?}", node.scheduling),
459 0 : format!("{:?}", node.availability),
460 0 : ]);
461 0 : }
462 0 : println!("{table}");
463 0 : }
464 0 : Command::NodeConfigure {
465 0 : node_id,
466 0 : availability,
467 0 : scheduling,
468 0 : } => {
469 0 : let req = NodeConfigureRequest {
470 0 : node_id,
471 0 : availability: availability.map(|a| a.0),
472 0 : scheduling,
473 0 : };
474 0 : storcon_client
475 0 : .dispatch::<_, ()>(
476 0 : Method::PUT,
477 0 : format!("control/v1/node/{node_id}/config"),
478 0 : Some(req),
479 0 : )
480 0 : .await?;
481 0 : }
482 0 : Command::Tenants {
483 0 : node_id: Some(node_id),
484 0 : } => {
485 0 : let describe_response = storcon_client
486 0 : .dispatch::<(), NodeShardResponse>(
487 0 : Method::GET,
488 0 : format!("control/v1/node/{node_id}/shards"),
489 0 : None,
490 0 : )
491 0 : .await?;
492 0 : let shards = describe_response.shards;
493 0 : let mut table = comfy_table::Table::new();
494 0 : table.set_header([
495 0 : "Shard",
496 0 : "Intended Primary/Secondary",
497 0 : "Observed Primary/Secondary",
498 0 : ]);
499 0 : for shard in shards {
500 0 : table.add_row([
501 0 : format!("{}", shard.tenant_shard_id),
502 0 : match shard.is_intended_secondary {
503 0 : None => "".to_string(),
504 0 : Some(true) => "Secondary".to_string(),
505 0 : Some(false) => "Primary".to_string(),
506 0 : },
507 0 : match shard.is_observed_secondary {
508 0 : None => "".to_string(),
509 0 : Some(true) => "Secondary".to_string(),
510 0 : Some(false) => "Primary".to_string(),
511 0 : },
512 0 : ]);
513 0 : }
514 0 : println!("{table}");
515 0 : }
516 0 : Command::Tenants { node_id: None } => {
517 0 : // Set up output formatting
518 0 : let mut table = comfy_table::Table::new();
519 0 : table.set_header([
520 0 : "TenantId",
521 0 : "Preferred AZ",
522 0 : "ShardCount",
523 0 : "StripeSize",
524 0 : "Placement",
525 0 : "Scheduling",
526 0 : ]);
527 0 :
528 0 : // Pagination loop over listing API
529 0 : let mut start_after = None;
530 0 : const LIMIT: usize = 1000;
531 0 : loop {
532 0 : let path = match start_after {
533 0 : None => format!("control/v1/tenant?limit={LIMIT}"),
534 0 : Some(start_after) => {
535 0 : format!("control/v1/tenant?limit={LIMIT}&start_after={start_after}")
536 0 : }
537 0 : };
538 0 :
539 0 : let resp = storcon_client
540 0 : .dispatch::<(), Vec<TenantDescribeResponse>>(Method::GET, path, None)
541 0 : .await?;
542 0 :
543 0 : if resp.is_empty() {
544 0 : // End of data reached
545 0 : break;
546 0 : }
547 0 :
548 0 : // Give some visual feedback while we're building up the table (comfy_table doesn't have
549 0 : // streaming output)
550 0 : if resp.len() >= LIMIT {
551 0 : eprint!(".");
552 0 : }
553 0 :
554 0 : start_after = Some(resp.last().unwrap().tenant_id);
555 0 :
556 0 : for tenant in resp {
557 0 : let shard_zero = tenant.shards.into_iter().next().unwrap();
558 0 : table.add_row([
559 0 : format!("{}", tenant.tenant_id),
560 0 : shard_zero
561 0 : .preferred_az_id
562 0 : .as_ref()
563 0 : .cloned()
564 0 : .unwrap_or("".to_string()),
565 0 : format!("{}", shard_zero.tenant_shard_id.shard_count.literal()),
566 0 : format!("{:?}", tenant.stripe_size),
567 0 : format!("{:?}", tenant.policy),
568 0 : format!("{:?}", shard_zero.scheduling_policy),
569 0 : ]);
570 0 : }
571 0 : }
572 0 :
573 0 : // Terminate progress dots
574 0 : if table.row_count() > LIMIT {
575 0 : eprint!("");
576 0 : }
577 0 :
578 0 : println!("{table}");
579 0 : }
580 0 : Command::TenantPolicy {
581 0 : tenant_id,
582 0 : placement,
583 0 : scheduling,
584 0 : } => {
585 0 : let req = TenantPolicyRequest {
586 0 : scheduling: scheduling.map(|s| s.0),
587 0 : placement: placement.map(|p| p.0),
588 0 : };
589 0 : storcon_client
590 0 : .dispatch::<_, ()>(
591 0 : Method::PUT,
592 0 : format!("control/v1/tenant/{tenant_id}/policy"),
593 0 : Some(req),
594 0 : )
595 0 : .await?;
596 0 : }
597 0 : Command::TenantShardSplit {
598 0 : tenant_id,
599 0 : shard_count,
600 0 : stripe_size,
601 0 : } => {
602 0 : let req = TenantShardSplitRequest {
603 0 : new_shard_count: shard_count,
604 0 : new_stripe_size: stripe_size.map(ShardStripeSize),
605 0 : };
606 0 :
607 0 : let response = storcon_client
608 0 : .dispatch::<TenantShardSplitRequest, TenantShardSplitResponse>(
609 0 : Method::PUT,
610 0 : format!("control/v1/tenant/{tenant_id}/shard_split"),
611 0 : Some(req),
612 0 : )
613 0 : .await?;
614 0 : println!(
615 0 : "Split tenant {} into {} shards: {}",
616 0 : tenant_id,
617 0 : shard_count,
618 0 : response
619 0 : .new_shards
620 0 : .iter()
621 0 : .map(|s| format!("{:?}", s))
622 0 : .collect::<Vec<_>>()
623 0 : .join(",")
624 0 : );
625 0 : }
626 0 : Command::TenantShardMigrate {
627 0 : tenant_shard_id,
628 0 : node,
629 0 : } => {
630 0 : let req = TenantShardMigrateRequest {
631 0 : node_id: node,
632 0 : migration_config: None,
633 0 : };
634 0 :
635 0 : storcon_client
636 0 : .dispatch::<TenantShardMigrateRequest, TenantShardMigrateResponse>(
637 0 : Method::PUT,
638 0 : format!("control/v1/tenant/{tenant_shard_id}/migrate"),
639 0 : Some(req),
640 0 : )
641 0 : .await?;
642 0 : }
643 0 : Command::TenantShardMigrateSecondary {
644 0 : tenant_shard_id,
645 0 : node,
646 0 : } => {
647 0 : let req = TenantShardMigrateRequest {
648 0 : node_id: node,
649 0 : migration_config: None,
650 0 : };
651 0 :
652 0 : storcon_client
653 0 : .dispatch::<TenantShardMigrateRequest, TenantShardMigrateResponse>(
654 0 : Method::PUT,
655 0 : format!("control/v1/tenant/{tenant_shard_id}/migrate_secondary"),
656 0 : Some(req),
657 0 : )
658 0 : .await?;
659 0 : }
660 0 : Command::TenantShardCancelReconcile { tenant_shard_id } => {
661 0 : storcon_client
662 0 : .dispatch::<(), ()>(
663 0 : Method::PUT,
664 0 : format!("control/v1/tenant/{tenant_shard_id}/cancel_reconcile"),
665 0 : None,
666 0 : )
667 0 : .await?;
668 0 : }
669 0 : Command::SetTenantConfig { tenant_id, config } => {
670 0 : let tenant_conf = serde_json::from_str(&config)?;
671 0 :
672 0 : vps_client
673 0 : .set_tenant_config(&TenantConfigRequest {
674 0 : tenant_id,
675 0 : config: tenant_conf,
676 0 : })
677 0 : .await?;
678 0 : }
679 0 : Command::PatchTenantConfig { tenant_id, config } => {
680 0 : let tenant_conf = serde_json::from_str(&config)?;
681 0 :
682 0 : vps_client
683 0 : .patch_tenant_config(&TenantConfigPatchRequest {
684 0 : tenant_id,
685 0 : config: tenant_conf,
686 0 : })
687 0 : .await?;
688 0 : }
689 0 : Command::TenantDescribe { tenant_id } => {
690 0 : let TenantDescribeResponse {
691 0 : tenant_id,
692 0 : shards,
693 0 : stripe_size,
694 0 : policy,
695 0 : config,
696 0 : } = storcon_client
697 0 : .dispatch::<(), TenantDescribeResponse>(
698 0 : Method::GET,
699 0 : format!("control/v1/tenant/{tenant_id}"),
700 0 : None,
701 0 : )
702 0 : .await?;
703 0 :
704 0 : let nodes = storcon_client
705 0 : .dispatch::<(), Vec<NodeDescribeResponse>>(
706 0 : Method::GET,
707 0 : "control/v1/node".to_string(),
708 0 : None,
709 0 : )
710 0 : .await?;
711 0 : let nodes = nodes
712 0 : .into_iter()
713 0 : .map(|n| (n.id, n))
714 0 : .collect::<HashMap<_, _>>();
715 0 :
716 0 : println!("Tenant {tenant_id}");
717 0 : let mut table = comfy_table::Table::new();
718 0 : table.add_row(["Policy", &format!("{:?}", policy)]);
719 0 : table.add_row(["Stripe size", &format!("{:?}", stripe_size)]);
720 0 : table.add_row(["Config", &serde_json::to_string_pretty(&config).unwrap()]);
721 0 : println!("{table}");
722 0 : println!("Shards:");
723 0 : let mut table = comfy_table::Table::new();
724 0 : table.set_header([
725 0 : "Shard",
726 0 : "Attached",
727 0 : "Attached AZ",
728 0 : "Secondary",
729 0 : "Last error",
730 0 : "status",
731 0 : ]);
732 0 : for shard in shards {
733 0 : let secondary = shard
734 0 : .node_secondary
735 0 : .iter()
736 0 : .map(|n| format!("{}", n))
737 0 : .collect::<Vec<_>>()
738 0 : .join(",");
739 0 :
740 0 : let mut status_parts = Vec::new();
741 0 : if shard.is_reconciling {
742 0 : status_parts.push("reconciling");
743 0 : }
744 0 :
745 0 : if shard.is_pending_compute_notification {
746 0 : status_parts.push("pending_compute");
747 0 : }
748 0 :
749 0 : if shard.is_splitting {
750 0 : status_parts.push("splitting");
751 0 : }
752 0 : let status = status_parts.join(",");
753 0 :
754 0 : let attached_node = shard
755 0 : .node_attached
756 0 : .as_ref()
757 0 : .map(|id| nodes.get(id).expect("Shard references nonexistent node"));
758 0 :
759 0 : table.add_row([
760 0 : format!("{}", shard.tenant_shard_id),
761 0 : attached_node
762 0 : .map(|n| format!("{} ({})", n.listen_http_addr, n.id))
763 0 : .unwrap_or(String::new()),
764 0 : attached_node
765 0 : .map(|n| n.availability_zone_id.clone())
766 0 : .unwrap_or(String::new()),
767 0 : secondary,
768 0 : shard.last_error,
769 0 : status,
770 0 : ]);
771 0 : }
772 0 : println!("{table}");
773 0 : }
774 0 : Command::TenantSetPreferredAz {
775 0 : tenant_id,
776 0 : preferred_az,
777 0 : } => {
778 0 : // First learn about the tenant's shards
779 0 : let describe_response = storcon_client
780 0 : .dispatch::<(), TenantDescribeResponse>(
781 0 : Method::GET,
782 0 : format!("control/v1/tenant/{tenant_id}"),
783 0 : None,
784 0 : )
785 0 : .await?;
786 0 :
787 0 : // Learn about nodes to validate the AZ ID
788 0 : let nodes = storcon_client
789 0 : .dispatch::<(), Vec<NodeDescribeResponse>>(
790 0 : Method::GET,
791 0 : "control/v1/node".to_string(),
792 0 : None,
793 0 : )
794 0 : .await?;
795 0 :
796 0 : if let Some(preferred_az) = &preferred_az {
797 0 : let azs = nodes
798 0 : .into_iter()
799 0 : .map(|n| (n.availability_zone_id))
800 0 : .collect::<HashSet<_>>();
801 0 : if !azs.contains(preferred_az) {
802 0 : anyhow::bail!(
803 0 : "AZ {} not found on any node: known AZs are: {:?}",
804 0 : preferred_az,
805 0 : azs
806 0 : );
807 0 : }
808 0 : } else {
809 0 : // Make it obvious to the user that since they've omitted an AZ, we're clearing it
810 0 : eprintln!("Clearing preferred AZ for tenant {}", tenant_id);
811 0 : }
812 0 :
813 0 : // Construct a request that modifies all the tenant's shards
814 0 : let req = ShardsPreferredAzsRequest {
815 0 : preferred_az_ids: describe_response
816 0 : .shards
817 0 : .into_iter()
818 0 : .map(|s| {
819 0 : (
820 0 : s.tenant_shard_id,
821 0 : preferred_az.clone().map(AvailabilityZone),
822 0 : )
823 0 : })
824 0 : .collect(),
825 0 : };
826 0 : storcon_client
827 0 : .dispatch::<ShardsPreferredAzsRequest, ShardsPreferredAzsResponse>(
828 0 : Method::PUT,
829 0 : "control/v1/preferred_azs".to_string(),
830 0 : Some(req),
831 0 : )
832 0 : .await?;
833 0 : }
834 0 : Command::TenantWarmup { tenant_id } => {
835 0 : let describe_response = storcon_client
836 0 : .dispatch::<(), TenantDescribeResponse>(
837 0 : Method::GET,
838 0 : format!("control/v1/tenant/{tenant_id}"),
839 0 : None,
840 0 : )
841 0 : .await;
842 0 : match describe_response {
843 0 : Ok(describe) => {
844 0 : if matches!(describe.policy, PlacementPolicy::Secondary) {
845 0 : // Fine: it's already known to controller in secondary mode: calling
846 0 : // again to put it into secondary mode won't cause problems.
847 0 : } else {
848 0 : anyhow::bail!("Tenant already present with policy {:?}", describe.policy);
849 0 : }
850 0 : }
851 0 : Err(mgmt_api::Error::ApiError(StatusCode::NOT_FOUND, _)) => {
852 0 : // Fine: this tenant isn't know to the storage controller yet.
853 0 : }
854 0 : Err(e) => {
855 0 : // Unexpected API error
856 0 : return Err(e.into());
857 0 : }
858 0 : }
859 0 :
860 0 : vps_client
861 0 : .location_config(
862 0 : TenantShardId::unsharded(tenant_id),
863 0 : pageserver_api::models::LocationConfig {
864 0 : mode: pageserver_api::models::LocationConfigMode::Secondary,
865 0 : generation: None,
866 0 : secondary_conf: Some(LocationConfigSecondary { warm: true }),
867 0 : shard_number: 0,
868 0 : shard_count: 0,
869 0 : shard_stripe_size: ShardParameters::DEFAULT_STRIPE_SIZE.0,
870 0 : tenant_conf: TenantConfig::default(),
871 0 : },
872 0 : None,
873 0 : true,
874 0 : )
875 0 : .await?;
876 0 :
877 0 : let describe_response = storcon_client
878 0 : .dispatch::<(), TenantDescribeResponse>(
879 0 : Method::GET,
880 0 : format!("control/v1/tenant/{tenant_id}"),
881 0 : None,
882 0 : )
883 0 : .await?;
884 0 :
885 0 : let secondary_ps_id = describe_response
886 0 : .shards
887 0 : .first()
888 0 : .unwrap()
889 0 : .node_secondary
890 0 : .first()
891 0 : .unwrap();
892 0 :
893 0 : println!("Tenant {tenant_id} warming up on pageserver {secondary_ps_id}");
894 0 : loop {
895 0 : let (status, progress) = vps_client
896 0 : .tenant_secondary_download(
897 0 : TenantShardId::unsharded(tenant_id),
898 0 : Some(Duration::from_secs(10)),
899 0 : )
900 0 : .await?;
901 0 : println!(
902 0 : "Progress: {}/{} layers, {}/{} bytes",
903 0 : progress.layers_downloaded,
904 0 : progress.layers_total,
905 0 : progress.bytes_downloaded,
906 0 : progress.bytes_total
907 0 : );
908 0 : match status {
909 0 : StatusCode::OK => {
910 0 : println!("Download complete");
911 0 : break;
912 0 : }
913 0 : StatusCode::ACCEPTED => {
914 0 : // Loop
915 0 : }
916 0 : _ => {
917 0 : anyhow::bail!("Unexpected download status: {status}");
918 0 : }
919 0 : }
920 0 : }
921 0 : }
922 0 : Command::TenantDrop { tenant_id, unclean } => {
923 0 : if !unclean {
924 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.")
925 0 : }
926 0 : storcon_client
927 0 : .dispatch::<(), ()>(
928 0 : Method::POST,
929 0 : format!("debug/v1/tenant/{tenant_id}/drop"),
930 0 : None,
931 0 : )
932 0 : .await?;
933 0 : }
934 0 : Command::NodeDrop { node_id, unclean } => {
935 0 : if !unclean {
936 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.")
937 0 : }
938 0 : storcon_client
939 0 : .dispatch::<(), ()>(Method::POST, format!("debug/v1/node/{node_id}/drop"), None)
940 0 : .await?;
941 0 : }
942 0 : Command::NodeDelete { node_id } => {
943 0 : storcon_client
944 0 : .dispatch::<(), ()>(Method::DELETE, format!("control/v1/node/{node_id}"), None)
945 0 : .await?;
946 0 : }
947 0 : Command::TenantSetTimeBasedEviction {
948 0 : tenant_id,
949 0 : period,
950 0 : threshold,
951 0 : } => {
952 0 : vps_client
953 0 : .set_tenant_config(&TenantConfigRequest {
954 0 : tenant_id,
955 0 : config: TenantConfig {
956 0 : eviction_policy: Some(EvictionPolicy::LayerAccessThreshold(
957 0 : EvictionPolicyLayerAccessThreshold {
958 0 : period: period.into(),
959 0 : threshold: threshold.into(),
960 0 : },
961 0 : )),
962 0 : heatmap_period: Some("300s".to_string()),
963 0 : ..Default::default()
964 0 : },
965 0 : })
966 0 : .await?;
967 0 : }
968 0 : Command::BulkMigrate {
969 0 : nodes,
970 0 : concurrency,
971 0 : max_shards,
972 0 : dry_run,
973 0 : } => {
974 0 : // Load the list of nodes, split them up into the drained and filled sets,
975 0 : // and validate that draining is possible.
976 0 : let node_descs = storcon_client
977 0 : .dispatch::<(), Vec<NodeDescribeResponse>>(
978 0 : Method::GET,
979 0 : "control/v1/node".to_string(),
980 0 : None,
981 0 : )
982 0 : .await?;
983 0 :
984 0 : let mut node_to_drain_descs = Vec::new();
985 0 : let mut node_to_fill_descs = Vec::new();
986 0 :
987 0 : for desc in node_descs {
988 0 : let to_drain = nodes.iter().any(|id| *id == desc.id);
989 0 : if to_drain {
990 0 : node_to_drain_descs.push(desc);
991 0 : } else {
992 0 : node_to_fill_descs.push(desc);
993 0 : }
994 0 : }
995 0 :
996 0 : if nodes.len() != node_to_drain_descs.len() {
997 0 : anyhow::bail!("Bulk migration requested away from node which doesn't exist.")
998 0 : }
999 0 :
1000 0 : node_to_fill_descs.retain(|desc| {
1001 0 : matches!(desc.availability, NodeAvailabilityWrapper::Active)
1002 0 : && matches!(
1003 0 : desc.scheduling,
1004 0 : NodeSchedulingPolicy::Active | NodeSchedulingPolicy::Filling
1005 0 : )
1006 0 : });
1007 0 :
1008 0 : if node_to_fill_descs.is_empty() {
1009 0 : anyhow::bail!("There are no nodes to migrate to")
1010 0 : }
1011 0 :
1012 0 : // Set the node scheduling policy to draining for the nodes which
1013 0 : // we plan to drain.
1014 0 : for node_desc in node_to_drain_descs.iter() {
1015 0 : let req = NodeConfigureRequest {
1016 0 : node_id: node_desc.id,
1017 0 : availability: None,
1018 0 : scheduling: Some(NodeSchedulingPolicy::Draining),
1019 0 : };
1020 0 :
1021 0 : storcon_client
1022 0 : .dispatch::<_, ()>(
1023 0 : Method::PUT,
1024 0 : format!("control/v1/node/{}/config", node_desc.id),
1025 0 : Some(req),
1026 0 : )
1027 0 : .await?;
1028 0 : }
1029 0 :
1030 0 : // Perform the migration: move each tenant shard scheduled on a node to
1031 0 : // be drained to a node which is being filled. A simple round robin
1032 0 : // strategy is used to pick the new node.
1033 0 : let tenants = storcon_client
1034 0 : .dispatch::<(), Vec<TenantDescribeResponse>>(
1035 0 : Method::GET,
1036 0 : "control/v1/tenant".to_string(),
1037 0 : None,
1038 0 : )
1039 0 : .await?;
1040 0 :
1041 0 : let mut selected_node_idx = 0;
1042 0 :
1043 0 : struct MigrationMove {
1044 0 : tenant_shard_id: TenantShardId,
1045 0 : from: NodeId,
1046 0 : to: NodeId,
1047 0 : }
1048 0 :
1049 0 : let mut moves: Vec<MigrationMove> = Vec::new();
1050 0 :
1051 0 : let shards = tenants
1052 0 : .into_iter()
1053 0 : .flat_map(|tenant| tenant.shards.into_iter());
1054 0 : for shard in shards {
1055 0 : if let Some(max_shards) = max_shards {
1056 0 : if moves.len() >= max_shards {
1057 0 : println!(
1058 0 : "Stop planning shard moves since the requested maximum was reached"
1059 0 : );
1060 0 : break;
1061 0 : }
1062 0 : }
1063 0 :
1064 0 : let should_migrate = {
1065 0 : if let Some(attached_to) = shard.node_attached {
1066 0 : node_to_drain_descs
1067 0 : .iter()
1068 0 : .map(|desc| desc.id)
1069 0 : .any(|id| id == attached_to)
1070 0 : } else {
1071 0 : false
1072 0 : }
1073 0 : };
1074 0 :
1075 0 : if !should_migrate {
1076 0 : continue;
1077 0 : }
1078 0 :
1079 0 : moves.push(MigrationMove {
1080 0 : tenant_shard_id: shard.tenant_shard_id,
1081 0 : from: shard
1082 0 : .node_attached
1083 0 : .expect("We only migrate attached tenant shards"),
1084 0 : to: node_to_fill_descs[selected_node_idx].id,
1085 0 : });
1086 0 : selected_node_idx = (selected_node_idx + 1) % node_to_fill_descs.len();
1087 0 : }
1088 0 :
1089 0 : let total_moves = moves.len();
1090 0 :
1091 0 : if dry_run == Some(true) {
1092 0 : println!("Dryrun requested. Planned {total_moves} moves:");
1093 0 : for mv in &moves {
1094 0 : println!("{}: {} -> {}", mv.tenant_shard_id, mv.from, mv.to)
1095 0 : }
1096 0 :
1097 0 : return Ok(());
1098 0 : }
1099 0 :
1100 0 : const DEFAULT_MIGRATE_CONCURRENCY: usize = 8;
1101 0 : let mut stream = futures::stream::iter(moves)
1102 0 : .map(|mv| {
1103 0 : let client = Client::new(cli.api.clone(), cli.jwt.clone());
1104 0 : async move {
1105 0 : client
1106 0 : .dispatch::<TenantShardMigrateRequest, TenantShardMigrateResponse>(
1107 0 : Method::PUT,
1108 0 : format!("control/v1/tenant/{}/migrate", mv.tenant_shard_id),
1109 0 : Some(TenantShardMigrateRequest {
1110 0 : node_id: mv.to,
1111 0 : migration_config: None,
1112 0 : }),
1113 0 : )
1114 0 : .await
1115 0 : .map_err(|e| (mv.tenant_shard_id, mv.from, mv.to, e))
1116 0 : }
1117 0 : })
1118 0 : .buffered(concurrency.unwrap_or(DEFAULT_MIGRATE_CONCURRENCY));
1119 0 :
1120 0 : let mut success = 0;
1121 0 : let mut failure = 0;
1122 0 :
1123 0 : while let Some(res) = stream.next().await {
1124 0 : match res {
1125 0 : Ok(_) => {
1126 0 : success += 1;
1127 0 : }
1128 0 : Err((tenant_shard_id, from, to, error)) => {
1129 0 : failure += 1;
1130 0 : println!(
1131 0 : "Failed to migrate {} from node {} to node {}: {}",
1132 0 : tenant_shard_id, from, to, error
1133 0 : );
1134 0 : }
1135 0 : }
1136 0 :
1137 0 : if (success + failure) % 20 == 0 {
1138 0 : println!(
1139 0 : "Processed {}/{} shards: {} succeeded, {} failed",
1140 0 : success + failure,
1141 0 : total_moves,
1142 0 : success,
1143 0 : failure
1144 0 : );
1145 0 : }
1146 0 : }
1147 0 :
1148 0 : println!(
1149 0 : "Processed {}/{} shards: {} succeeded, {} failed",
1150 0 : success + failure,
1151 0 : total_moves,
1152 0 : success,
1153 0 : failure
1154 0 : );
1155 0 : }
1156 0 : Command::StartDrain { node_id } => {
1157 0 : storcon_client
1158 0 : .dispatch::<(), ()>(
1159 0 : Method::PUT,
1160 0 : format!("control/v1/node/{node_id}/drain"),
1161 0 : None,
1162 0 : )
1163 0 : .await?;
1164 0 : println!("Drain started for {node_id}");
1165 0 : }
1166 0 : Command::CancelDrain { node_id, timeout } => {
1167 0 : storcon_client
1168 0 : .dispatch::<(), ()>(
1169 0 : Method::DELETE,
1170 0 : format!("control/v1/node/{node_id}/drain"),
1171 0 : None,
1172 0 : )
1173 0 : .await?;
1174 0 :
1175 0 : println!("Waiting for node {node_id} to quiesce on scheduling policy ...");
1176 0 :
1177 0 : let final_policy =
1178 0 : wait_for_scheduling_policy(storcon_client, node_id, *timeout, |sched| {
1179 0 : use NodeSchedulingPolicy::*;
1180 0 : matches!(sched, Active | PauseForRestart)
1181 0 : })
1182 0 : .await?;
1183 0 :
1184 0 : println!(
1185 0 : "Drain was cancelled for node {node_id}. Schedulling policy is now {final_policy:?}"
1186 0 : );
1187 0 : }
1188 0 : Command::StartFill { node_id } => {
1189 0 : storcon_client
1190 0 : .dispatch::<(), ()>(Method::PUT, format!("control/v1/node/{node_id}/fill"), None)
1191 0 : .await?;
1192 0 :
1193 0 : println!("Fill started for {node_id}");
1194 0 : }
1195 0 : Command::CancelFill { node_id, timeout } => {
1196 0 : storcon_client
1197 0 : .dispatch::<(), ()>(
1198 0 : Method::DELETE,
1199 0 : format!("control/v1/node/{node_id}/fill"),
1200 0 : None,
1201 0 : )
1202 0 : .await?;
1203 0 :
1204 0 : println!("Waiting for node {node_id} to quiesce on scheduling policy ...");
1205 0 :
1206 0 : let final_policy =
1207 0 : wait_for_scheduling_policy(storcon_client, node_id, *timeout, |sched| {
1208 0 : use NodeSchedulingPolicy::*;
1209 0 : matches!(sched, Active)
1210 0 : })
1211 0 : .await?;
1212 0 :
1213 0 : println!(
1214 0 : "Fill was cancelled for node {node_id}. Schedulling policy is now {final_policy:?}"
1215 0 : );
1216 0 : }
1217 0 : Command::Safekeepers {} => {
1218 0 : let mut resp = storcon_client
1219 0 : .dispatch::<(), Vec<SafekeeperDescribeResponse>>(
1220 0 : Method::GET,
1221 0 : "control/v1/safekeeper".to_string(),
1222 0 : None,
1223 0 : )
1224 0 : .await?;
1225 0 :
1226 0 : resp.sort_by(|a, b| a.id.cmp(&b.id));
1227 0 :
1228 0 : let mut table = comfy_table::Table::new();
1229 0 : table.set_header([
1230 0 : "Id",
1231 0 : "Version",
1232 0 : "Host",
1233 0 : "Port",
1234 0 : "Http Port",
1235 0 : "AZ Id",
1236 0 : "Scheduling",
1237 0 : ]);
1238 0 : for sk in resp {
1239 0 : table.add_row([
1240 0 : format!("{}", sk.id),
1241 0 : format!("{}", sk.version),
1242 0 : sk.host,
1243 0 : format!("{}", sk.port),
1244 0 : format!("{}", sk.http_port),
1245 0 : sk.availability_zone_id.clone(),
1246 0 : String::from(sk.scheduling_policy),
1247 0 : ]);
1248 0 : }
1249 0 : println!("{table}");
1250 0 : }
1251 0 : Command::SafekeeperScheduling {
1252 0 : node_id,
1253 0 : scheduling_policy,
1254 0 : } => {
1255 0 : let scheduling_policy = scheduling_policy.0;
1256 0 : storcon_client
1257 0 : .dispatch::<SafekeeperSchedulingPolicyRequest, ()>(
1258 0 : Method::POST,
1259 0 : format!("control/v1/safekeeper/{node_id}/scheduling_policy"),
1260 0 : Some(SafekeeperSchedulingPolicyRequest { scheduling_policy }),
1261 0 : )
1262 0 : .await?;
1263 0 : println!(
1264 0 : "Scheduling policy of {node_id} set to {}",
1265 0 : String::from(scheduling_policy)
1266 0 : );
1267 0 : }
1268 0 : Command::DownloadHeatmapLayers {
1269 0 : tenant_shard_id,
1270 0 : timeline_id,
1271 0 : concurrency,
1272 0 : } => {
1273 0 : let mut path = format!(
1274 0 : "/v1/tenant/{}/timeline/{}/download_heatmap_layers",
1275 0 : tenant_shard_id, timeline_id,
1276 0 : );
1277 0 :
1278 0 : if let Some(c) = concurrency {
1279 0 : path = format!("{path}?concurrency={c}");
1280 0 : }
1281 0 :
1282 0 : storcon_client
1283 0 : .dispatch::<(), ()>(Method::POST, path, None)
1284 0 : .await?;
1285 0 : }
1286 0 : }
1287 0 :
1288 0 : Ok(())
1289 0 : }
|
