Line data Source code
1 : use anyhow::bail;
2 : use camino::Utf8PathBuf;
3 : use pageserver_api::shard::TenantShardId;
4 : use storage_scrubber::garbage::{find_garbage, purge_garbage, PurgeMode};
5 : use storage_scrubber::pageserver_physical_gc::GcMode;
6 : use storage_scrubber::scan_pageserver_metadata::scan_metadata;
7 : use storage_scrubber::tenant_snapshot::SnapshotDownloader;
8 : use storage_scrubber::{
9 : init_logging, pageserver_physical_gc::pageserver_physical_gc,
10 : scan_safekeeper_metadata::scan_safekeeper_metadata, BucketConfig, ConsoleConfig, NodeKind,
11 : TraversingDepth,
12 : };
13 :
14 : use clap::{Parser, Subcommand};
15 : use utils::id::TenantId;
16 :
17 0 : #[derive(Parser)]
18 : #[command(author, version, about, long_about = None)]
19 : #[command(arg_required_else_help(true))]
20 : struct Cli {
21 : #[command(subcommand)]
22 : command: Command,
23 :
24 0 : #[arg(short, long, default_value_t = false)]
25 0 : delete: bool,
26 : }
27 :
28 0 : #[derive(Subcommand, Debug)]
29 : enum Command {
30 : FindGarbage {
31 : #[arg(short, long)]
32 0 : node_kind: NodeKind,
33 0 : #[arg(short, long, default_value_t=TraversingDepth::Tenant)]
34 0 : depth: TraversingDepth,
35 0 : #[arg(short, long, default_value_t = String::from("garbage.json"))]
36 0 : output_path: String,
37 : },
38 : PurgeGarbage {
39 : #[arg(short, long)]
40 0 : input_path: String,
41 0 : #[arg(short, long, default_value_t = PurgeMode::DeletedOnly)]
42 0 : mode: PurgeMode,
43 : },
44 : #[command(verbatim_doc_comment)]
45 : ScanMetadata {
46 : #[arg(short, long)]
47 0 : node_kind: NodeKind,
48 0 : #[arg(short, long, default_value_t = false)]
49 0 : json: bool,
50 : #[arg(long = "tenant-id", num_args = 0..)]
51 0 : tenant_ids: Vec<TenantShardId>,
52 : #[arg(long, default_value = None)]
53 : /// For safekeeper node_kind only, points to db with debug dump
54 : dump_db_connstr: Option<String>,
55 : /// For safekeeper node_kind only, table in the db with debug dump
56 : #[arg(long, default_value = None)]
57 : dump_db_table: Option<String>,
58 : },
59 : TenantSnapshot {
60 : #[arg(long = "tenant-id")]
61 0 : tenant_id: TenantId,
62 0 : #[arg(long = "concurrency", short = 'j', default_value_t = 8)]
63 0 : concurrency: usize,
64 : #[arg(short, long)]
65 0 : output_path: Utf8PathBuf,
66 : },
67 : PageserverPhysicalGc {
68 : #[arg(long = "tenant-id", num_args = 0..)]
69 0 : tenant_ids: Vec<TenantShardId>,
70 : #[arg(long = "min-age")]
71 0 : min_age: humantime::Duration,
72 0 : #[arg(short, long, default_value_t = GcMode::IndicesOnly)]
73 0 : mode: GcMode,
74 : },
75 : }
76 :
77 : #[tokio::main]
78 0 : async fn main() -> anyhow::Result<()> {
79 0 : let cli = Cli::parse();
80 0 :
81 0 : let bucket_config = BucketConfig::from_env()?;
82 0 :
83 0 : let command_log_name = match &cli.command {
84 0 : Command::ScanMetadata { .. } => "scan",
85 0 : Command::FindGarbage { .. } => "find-garbage",
86 0 : Command::PurgeGarbage { .. } => "purge-garbage",
87 0 : Command::TenantSnapshot { .. } => "tenant-snapshot",
88 0 : Command::PageserverPhysicalGc { .. } => "pageserver-physical-gc",
89 0 : };
90 0 : let _guard = init_logging(&format!(
91 0 : "{}_{}_{}_{}.log",
92 0 : std::env::args().next().unwrap(),
93 0 : command_log_name,
94 0 : bucket_config.bucket,
95 0 : chrono::Utc::now().format("%Y_%m_%d__%H_%M_%S")
96 0 : ));
97 0 :
98 0 : match cli.command {
99 0 : Command::ScanMetadata {
100 0 : json,
101 0 : tenant_ids,
102 0 : node_kind,
103 0 : dump_db_connstr,
104 0 : dump_db_table,
105 0 : } => {
106 0 : if let NodeKind::Safekeeper = node_kind {
107 0 : let dump_db_connstr =
108 0 : dump_db_connstr.ok_or(anyhow::anyhow!("dump_db_connstr not specified"))?;
109 0 : let dump_db_table =
110 0 : dump_db_table.ok_or(anyhow::anyhow!("dump_db_table not specified"))?;
111 0 :
112 0 : let summary = scan_safekeeper_metadata(
113 0 : bucket_config.clone(),
114 0 : tenant_ids.iter().map(|tshid| tshid.tenant_id).collect(),
115 0 : dump_db_connstr,
116 0 : dump_db_table,
117 0 : )
118 0 : .await?;
119 0 : if json {
120 0 : println!("{}", serde_json::to_string(&summary).unwrap())
121 0 : } else {
122 0 : println!("{}", summary.summary_string());
123 0 : }
124 0 : if summary.is_fatal() {
125 0 : bail!("Fatal scrub errors detected");
126 0 : }
127 0 : if summary.is_empty() {
128 0 : // Strictly speaking an empty bucket is a valid bucket, but if someone ran the
129 0 : // scrubber they were likely expecting to scan something, and if we see no timelines
130 0 : // at all then it's likely due to some configuration issues like a bad prefix
131 0 : bail!(
132 0 : "No timelines found in bucket {} prefix {}",
133 0 : bucket_config.bucket,
134 0 : bucket_config
135 0 : .prefix_in_bucket
136 0 : .unwrap_or("<none>".to_string())
137 0 : );
138 0 : }
139 0 : Ok(())
140 0 : } else {
141 0 : match scan_metadata(bucket_config.clone(), tenant_ids).await {
142 0 : Err(e) => {
143 0 : tracing::error!("Failed: {e}");
144 0 : Err(e)
145 0 : }
146 0 : Ok(summary) => {
147 0 : if json {
148 0 : println!("{}", serde_json::to_string(&summary).unwrap())
149 0 : } else {
150 0 : println!("{}", summary.summary_string());
151 0 : }
152 0 : if summary.is_fatal() {
153 0 : Err(anyhow::anyhow!("Fatal scrub errors detected"))
154 0 : } else if summary.is_empty() {
155 0 : // Strictly speaking an empty bucket is a valid bucket, but if someone ran the
156 0 : // scrubber they were likely expecting to scan something, and if we see no timelines
157 0 : // at all then it's likely due to some configuration issues like a bad prefix
158 0 : Err(anyhow::anyhow!(
159 0 : "No timelines found in bucket {} prefix {}",
160 0 : bucket_config.bucket,
161 0 : bucket_config
162 0 : .prefix_in_bucket
163 0 : .unwrap_or("<none>".to_string())
164 0 : ))
165 0 : } else {
166 0 : Ok(())
167 0 : }
168 0 : }
169 0 : }
170 0 : }
171 0 : }
172 0 : Command::FindGarbage {
173 0 : node_kind,
174 0 : depth,
175 0 : output_path,
176 0 : } => {
177 0 : let console_config = ConsoleConfig::from_env()?;
178 0 : find_garbage(bucket_config, console_config, depth, node_kind, output_path).await
179 0 : }
180 0 : Command::PurgeGarbage { input_path, mode } => {
181 0 : purge_garbage(input_path, mode, !cli.delete).await
182 0 : }
183 0 : Command::TenantSnapshot {
184 0 : tenant_id,
185 0 : output_path,
186 0 : concurrency,
187 0 : } => {
188 0 : let downloader =
189 0 : SnapshotDownloader::new(bucket_config, tenant_id, output_path, concurrency)?;
190 0 : downloader.download().await
191 0 : }
192 0 : Command::PageserverPhysicalGc {
193 0 : tenant_ids,
194 0 : min_age,
195 0 : mode,
196 0 : } => {
197 0 : let summary =
198 0 : pageserver_physical_gc(bucket_config, tenant_ids, min_age.into(), mode).await?;
199 0 : println!("{}", serde_json::to_string(&summary).unwrap());
200 0 : Ok(())
201 0 : }
202 0 : }
203 0 : }
|
