Line data Source code
1 : use anyhow::{anyhow, bail, Context};
2 : use camino::Utf8PathBuf;
3 : use pageserver_api::controller_api::{MetadataHealthUpdateRequest, MetadataHealthUpdateResponse};
4 : use pageserver_api::shard::TenantShardId;
5 : use reqwest::{Method, Url};
6 : use storage_controller_client::control_api;
7 : use storage_scrubber::garbage::{find_garbage, purge_garbage, PurgeMode};
8 : use storage_scrubber::pageserver_physical_gc::GcMode;
9 : use storage_scrubber::scan_pageserver_metadata::scan_pageserver_metadata;
10 : use storage_scrubber::scan_safekeeper_metadata::DatabaseOrList;
11 : use storage_scrubber::tenant_snapshot::SnapshotDownloader;
12 : use storage_scrubber::{find_large_objects, ControllerClientConfig};
13 : use storage_scrubber::{
14 : init_logging, pageserver_physical_gc::pageserver_physical_gc,
15 : scan_safekeeper_metadata::scan_safekeeper_metadata, BucketConfig, ConsoleConfig, NodeKind,
16 : TraversingDepth,
17 : };
18 :
19 : use clap::{Parser, Subcommand};
20 : use utils::id::TenantId;
21 :
22 : use utils::{project_build_tag, project_git_version};
23 :
24 : project_git_version!(GIT_VERSION);
25 : project_build_tag!(BUILD_TAG);
26 :
27 0 : #[derive(Parser)]
28 : #[command(author, version, about, long_about = None)]
29 : #[command(arg_required_else_help(true))]
30 : struct Cli {
31 : #[command(subcommand)]
32 : command: Command,
33 :
34 0 : #[arg(short, long, default_value_t = false)]
35 0 : delete: bool,
36 :
37 : #[arg(long)]
38 : /// URL to storage controller. e.g. http://127.0.0.1:1234 when using `neon_local`
39 : controller_api: Option<Url>,
40 :
41 : #[arg(long)]
42 : /// JWT token for authenticating with storage controller. Requires scope 'scrubber' or 'admin'.
43 : controller_jwt: Option<String>,
44 : }
45 :
46 0 : #[derive(Subcommand, Debug)]
47 : enum Command {
48 : FindGarbage {
49 : #[arg(short, long)]
50 0 : node_kind: NodeKind,
51 0 : #[arg(short, long, default_value_t=TraversingDepth::Tenant)]
52 0 : depth: TraversingDepth,
53 0 : #[arg(short, long, default_value_t = String::from("garbage.json"))]
54 0 : output_path: String,
55 : },
56 : PurgeGarbage {
57 : #[arg(short, long)]
58 0 : input_path: String,
59 0 : #[arg(short, long, default_value_t = PurgeMode::DeletedOnly)]
60 0 : mode: PurgeMode,
61 : #[arg(long = "min-age")]
62 0 : min_age: humantime::Duration,
63 : },
64 : #[command(verbatim_doc_comment)]
65 : ScanMetadata {
66 : #[arg(short, long)]
67 0 : node_kind: NodeKind,
68 0 : #[arg(short, long, default_value_t = false)]
69 0 : json: bool,
70 : #[arg(long = "tenant-id", num_args = 0..)]
71 0 : tenant_ids: Vec<TenantShardId>,
72 0 : #[arg(long = "post", default_value_t = false)]
73 0 : post_to_storcon: bool,
74 : #[arg(long, default_value = None)]
75 : /// For safekeeper node_kind only, points to db with debug dump
76 : dump_db_connstr: Option<String>,
77 : /// For safekeeper node_kind only, table in the db with debug dump
78 : #[arg(long, default_value = None)]
79 : dump_db_table: Option<String>,
80 : /// For safekeeper node_kind only, json list of timelines and their lsn info
81 : #[arg(long, default_value = None)]
82 : timeline_lsns: Option<String>,
83 : },
84 : TenantSnapshot {
85 : #[arg(long = "tenant-id")]
86 0 : tenant_id: TenantId,
87 0 : #[arg(long = "concurrency", short = 'j', default_value_t = 8)]
88 0 : concurrency: usize,
89 : #[arg(short, long)]
90 0 : output_path: Utf8PathBuf,
91 : },
92 : PageserverPhysicalGc {
93 : #[arg(long = "tenant-id", num_args = 0..)]
94 0 : tenant_ids: Vec<TenantShardId>,
95 : #[arg(long = "min-age")]
96 0 : min_age: humantime::Duration,
97 0 : #[arg(short, long, default_value_t = GcMode::IndicesOnly)]
98 0 : mode: GcMode,
99 : },
100 : FindLargeObjects {
101 : #[arg(long = "min-size")]
102 0 : min_size: u64,
103 0 : #[arg(short, long, default_value_t = false)]
104 0 : ignore_deltas: bool,
105 0 : #[arg(long = "concurrency", short = 'j', default_value_t = 64)]
106 0 : concurrency: usize,
107 : },
108 : CronJob {
109 : // PageserverPhysicalGc
110 : #[arg(long = "min-age")]
111 0 : gc_min_age: humantime::Duration,
112 0 : #[arg(short, long, default_value_t = GcMode::IndicesOnly)]
113 0 : gc_mode: GcMode,
114 : // ScanMetadata
115 0 : #[arg(long = "post", default_value_t = false)]
116 0 : post_to_storcon: bool,
117 : },
118 : }
119 :
120 : #[tokio::main]
121 0 : async fn main() -> anyhow::Result<()> {
122 0 : let cli = Cli::parse();
123 0 :
124 0 : tracing::info!("version: {}, build_tag {}", GIT_VERSION, BUILD_TAG);
125 0 :
126 0 : let bucket_config = BucketConfig::from_env()?;
127 0 :
128 0 : let command_log_name = match &cli.command {
129 0 : Command::ScanMetadata { .. } => "scan",
130 0 : Command::FindGarbage { .. } => "find-garbage",
131 0 : Command::PurgeGarbage { .. } => "purge-garbage",
132 0 : Command::TenantSnapshot { .. } => "tenant-snapshot",
133 0 : Command::PageserverPhysicalGc { .. } => "pageserver-physical-gc",
134 0 : Command::FindLargeObjects { .. } => "find-large-objects",
135 0 : Command::CronJob { .. } => "cron-job",
136 0 : };
137 0 : let _guard = init_logging(&format!(
138 0 : "{}_{}_{}_{}.log",
139 0 : std::env::args().next().unwrap(),
140 0 : command_log_name,
141 0 : bucket_config.bucket,
142 0 : chrono::Utc::now().format("%Y_%m_%d__%H_%M_%S")
143 0 : ));
144 0 :
145 0 : let controller_client = cli.controller_api.map(|controller_api| {
146 0 : ControllerClientConfig {
147 0 : controller_api,
148 0 : // Default to no key: this is a convenience when working in a development environment
149 0 : controller_jwt: cli.controller_jwt.unwrap_or("".to_owned()),
150 0 : }
151 0 : .build_client()
152 0 : });
153 0 :
154 0 : match cli.command {
155 0 : Command::ScanMetadata {
156 0 : json,
157 0 : tenant_ids,
158 0 : node_kind,
159 0 : post_to_storcon,
160 0 : dump_db_connstr,
161 0 : dump_db_table,
162 0 : timeline_lsns,
163 0 : } => {
164 0 : if let NodeKind::Safekeeper = node_kind {
165 0 : let db_or_list = match (timeline_lsns, dump_db_connstr) {
166 0 : (Some(timeline_lsns), _) => {
167 0 : let timeline_lsns = serde_json::from_str(&timeline_lsns).context("parsing timeline_lsns")?;
168 0 : DatabaseOrList::List(timeline_lsns)
169 0 : }
170 0 : (None, Some(dump_db_connstr)) => {
171 0 : let dump_db_table = dump_db_table.ok_or_else(|| anyhow::anyhow!("dump_db_table not specified"))?;
172 0 : let tenant_ids = tenant_ids.iter().map(|tshid| tshid.tenant_id).collect();
173 0 : DatabaseOrList::Database { tenant_ids, connstr: dump_db_connstr, table: dump_db_table }
174 0 : }
175 0 : (None, None) => anyhow::bail!("neither `timeline_lsns` specified, nor `dump_db_connstr` and `dump_db_table`"),
176 0 : };
177 0 : let summary = scan_safekeeper_metadata(bucket_config.clone(), db_or_list).await?;
178 0 : if json {
179 0 : println!("{}", serde_json::to_string(&summary).unwrap())
180 0 : } else {
181 0 : println!("{}", summary.summary_string());
182 0 : }
183 0 : if summary.is_fatal() {
184 0 : bail!("Fatal scrub errors detected");
185 0 : }
186 0 : if summary.is_empty() {
187 0 : // Strictly speaking an empty bucket is a valid bucket, but if someone ran the
188 0 : // scrubber they were likely expecting to scan something, and if we see no timelines
189 0 : // at all then it's likely due to some configuration issues like a bad prefix
190 0 : bail!(
191 0 : "No timelines found in bucket {} prefix {}",
192 0 : bucket_config.bucket,
193 0 : bucket_config
194 0 : .prefix_in_bucket
195 0 : .unwrap_or("<none>".to_string())
196 0 : );
197 0 : }
198 0 : Ok(())
199 0 : } else {
200 0 : scan_pageserver_metadata_cmd(
201 0 : bucket_config,
202 0 : controller_client.as_ref(),
203 0 : tenant_ids,
204 0 : json,
205 0 : post_to_storcon,
206 0 : )
207 0 : .await
208 0 : }
209 0 : }
210 0 : Command::FindGarbage {
211 0 : node_kind,
212 0 : depth,
213 0 : output_path,
214 0 : } => {
215 0 : let console_config = ConsoleConfig::from_env()?;
216 0 : find_garbage(bucket_config, console_config, depth, node_kind, output_path).await
217 0 : }
218 0 : Command::PurgeGarbage {
219 0 : input_path,
220 0 : mode,
221 0 : min_age,
222 0 : } => purge_garbage(input_path, mode, min_age.into(), !cli.delete).await,
223 0 : Command::TenantSnapshot {
224 0 : tenant_id,
225 0 : output_path,
226 0 : concurrency,
227 0 : } => {
228 0 : let downloader =
229 0 : SnapshotDownloader::new(bucket_config, tenant_id, output_path, concurrency).await?;
230 0 : downloader.download().await
231 0 : }
232 0 : Command::PageserverPhysicalGc {
233 0 : tenant_ids,
234 0 : min_age,
235 0 : mode,
236 0 : } => {
237 0 : pageserver_physical_gc_cmd(
238 0 : &bucket_config,
239 0 : controller_client.as_ref(),
240 0 : tenant_ids,
241 0 : min_age,
242 0 : mode,
243 0 : )
244 0 : .await
245 0 : }
246 0 : Command::FindLargeObjects {
247 0 : min_size,
248 0 : ignore_deltas,
249 0 : concurrency,
250 0 : } => {
251 0 : let summary = find_large_objects::find_large_objects(
252 0 : bucket_config,
253 0 : min_size,
254 0 : ignore_deltas,
255 0 : concurrency,
256 0 : )
257 0 : .await?;
258 0 : println!("{}", serde_json::to_string(&summary).unwrap());
259 0 : Ok(())
260 0 : }
261 0 : Command::CronJob {
262 0 : gc_min_age,
263 0 : gc_mode,
264 0 : post_to_storcon,
265 0 : } => {
266 0 : run_cron_job(
267 0 : bucket_config,
268 0 : controller_client.as_ref(),
269 0 : gc_min_age,
270 0 : gc_mode,
271 0 : post_to_storcon,
272 0 : )
273 0 : .await
274 0 : }
275 0 : }
276 0 : }
277 :
278 : /// Runs the scrubber cron job.
279 : /// 1. Do pageserver physical gc
280 : /// 2. Scan pageserver metadata
281 0 : pub async fn run_cron_job(
282 0 : bucket_config: BucketConfig,
283 0 : controller_client: Option<&control_api::Client>,
284 0 : gc_min_age: humantime::Duration,
285 0 : gc_mode: GcMode,
286 0 : post_to_storcon: bool,
287 0 : ) -> anyhow::Result<()> {
288 0 : tracing::info!(%gc_min_age, %gc_mode, "Running pageserver-physical-gc");
289 0 : pageserver_physical_gc_cmd(
290 0 : &bucket_config,
291 0 : controller_client,
292 0 : Vec::new(),
293 0 : gc_min_age,
294 0 : gc_mode,
295 0 : )
296 0 : .await?;
297 0 : tracing::info!(%post_to_storcon, node_kind = %NodeKind::Pageserver, "Running scan-metadata");
298 0 : scan_pageserver_metadata_cmd(
299 0 : bucket_config,
300 0 : controller_client,
301 0 : Vec::new(),
302 0 : true,
303 0 : post_to_storcon,
304 0 : )
305 0 : .await?;
306 :
307 0 : Ok(())
308 0 : }
309 :
310 0 : pub async fn pageserver_physical_gc_cmd(
311 0 : bucket_config: &BucketConfig,
312 0 : controller_client: Option<&control_api::Client>,
313 0 : tenant_shard_ids: Vec<TenantShardId>,
314 0 : min_age: humantime::Duration,
315 0 : mode: GcMode,
316 0 : ) -> anyhow::Result<()> {
317 0 : match (controller_client, mode) {
318 0 : (Some(_), _) => {
319 0 : // Any mode may run when controller API is set
320 0 : }
321 : (None, GcMode::Full) => {
322 : // The part of physical GC where we erase ancestor layers cannot be done safely without
323 : // confirming the most recent complete shard split with the controller. Refuse to run, rather
324 : // than doing it unsafely.
325 0 : return Err(anyhow!(
326 0 : "Full physical GC requires `--controller-api` and `--controller-jwt` to run"
327 0 : ));
328 : }
329 0 : (None, GcMode::DryRun | GcMode::IndicesOnly) => {
330 0 : // These GcModes do not require the controller to run.
331 0 : }
332 : }
333 :
334 0 : let summary = pageserver_physical_gc(
335 0 : bucket_config,
336 0 : controller_client,
337 0 : tenant_shard_ids,
338 0 : min_age.into(),
339 0 : mode,
340 0 : )
341 0 : .await?;
342 0 : println!("{}", serde_json::to_string(&summary).unwrap());
343 0 : Ok(())
344 0 : }
345 :
346 0 : pub async fn scan_pageserver_metadata_cmd(
347 0 : bucket_config: BucketConfig,
348 0 : controller_client: Option<&control_api::Client>,
349 0 : tenant_shard_ids: Vec<TenantShardId>,
350 0 : json: bool,
351 0 : post_to_storcon: bool,
352 0 : ) -> anyhow::Result<()> {
353 0 : if controller_client.is_none() && post_to_storcon {
354 0 : return Err(anyhow!("Posting pageserver scan health status to storage controller requires `--controller-api` and `--controller-jwt` to run"));
355 0 : }
356 0 : match scan_pageserver_metadata(bucket_config.clone(), tenant_shard_ids).await {
357 0 : Err(e) => {
358 0 : tracing::error!("Failed: {e}");
359 0 : Err(e)
360 : }
361 0 : Ok(summary) => {
362 0 : if json {
363 0 : println!("{}", serde_json::to_string(&summary).unwrap())
364 0 : } else {
365 0 : println!("{}", summary.summary_string());
366 0 : }
367 :
368 0 : if post_to_storcon {
369 0 : if let Some(client) = controller_client {
370 0 : let body = summary.build_health_update_request();
371 0 : client
372 0 : .dispatch::<MetadataHealthUpdateRequest, MetadataHealthUpdateResponse>(
373 0 : Method::POST,
374 0 : "control/v1/metadata_health/update".to_string(),
375 0 : Some(body),
376 0 : )
377 0 : .await?;
378 0 : }
379 0 : }
380 :
381 0 : if summary.is_fatal() {
382 0 : tracing::error!("Fatal scrub errors detected");
383 0 : } else if summary.is_empty() {
384 : // Strictly speaking an empty bucket is a valid bucket, but if someone ran the
385 : // scrubber they were likely expecting to scan something, and if we see no timelines
386 : // at all then it's likely due to some configuration issues like a bad prefix
387 0 : tracing::error!(
388 0 : "No timelines found in bucket {} prefix {}",
389 0 : bucket_config.bucket,
390 0 : bucket_config
391 0 : .prefix_in_bucket
392 0 : .unwrap_or("<none>".to_string())
393 : );
394 0 : }
395 :
396 0 : Ok(())
397 : }
398 : }
399 0 : }
|
