Line data Source code
1 : //! Helper functions to upload files to remote storage with a RemoteStorage
2 :
3 : use anyhow::{bail, Context};
4 : use bytes::Bytes;
5 : use camino::Utf8Path;
6 : use fail::fail_point;
7 : use pageserver_api::shard::TenantShardId;
8 : use std::io::{ErrorKind, SeekFrom};
9 : use std::time::SystemTime;
10 : use tokio::fs::{self, File};
11 : use tokio::io::AsyncSeekExt;
12 : use tokio_util::sync::CancellationToken;
13 : use utils::{backoff, pausable_failpoint};
14 :
15 : use super::index::IndexPart;
16 : use super::manifest::TenantManifest;
17 : use super::Generation;
18 : use crate::tenant::remote_timeline_client::{
19 : remote_index_path, remote_initdb_archive_path, remote_initdb_preserved_archive_path,
20 : remote_tenant_manifest_path,
21 : };
22 : use remote_storage::{GenericRemoteStorage, RemotePath, TimeTravelError};
23 : use utils::id::{TenantId, TimelineId};
24 :
25 : use tracing::info;
26 :
27 : /// Serializes and uploads the given index part data to the remote storage.
28 1414 : pub(crate) async fn upload_index_part<'a>(
29 1414 : storage: &'a GenericRemoteStorage,
30 1414 : tenant_shard_id: &TenantShardId,
31 1414 : timeline_id: &TimelineId,
32 1414 : generation: Generation,
33 1414 : index_part: &IndexPart,
34 1414 : cancel: &CancellationToken,
35 1414 : ) -> anyhow::Result<()> {
36 1414 : tracing::trace!("uploading new index part");
37 :
38 1414 : fail_point!("before-upload-index", |_| {
39 0 : bail!("failpoint before-upload-index")
40 1414 : });
41 1414 : pausable_failpoint!("before-upload-index-pausable");
42 :
43 : // FIXME: this error comes too late
44 1414 : let serialized = index_part.to_json_bytes()?;
45 1414 : let serialized = Bytes::from(serialized);
46 1414 :
47 1414 : let index_part_size = serialized.len();
48 1414 :
49 1414 : let remote_path = remote_index_path(tenant_shard_id, timeline_id, generation);
50 1414 : storage
51 1414 : .upload_storage_object(
52 1414 : futures::stream::once(futures::future::ready(Ok(serialized))),
53 1414 : index_part_size,
54 1414 : &remote_path,
55 1414 : cancel,
56 1414 : )
57 3917 : .await
58 1410 : .with_context(|| format!("upload index part for '{tenant_shard_id} / {timeline_id}'"))
59 1410 : }
60 : /// Serializes and uploads the given tenant manifest data to the remote storage.
61 2 : pub(crate) async fn upload_tenant_manifest(
62 2 : storage: &GenericRemoteStorage,
63 2 : tenant_shard_id: &TenantShardId,
64 2 : generation: Generation,
65 2 : tenant_manifest: &TenantManifest,
66 2 : cancel: &CancellationToken,
67 2 : ) -> anyhow::Result<()> {
68 2 : tracing::trace!("uploading new tenant manifest");
69 :
70 2 : fail_point!("before-upload-manifest", |_| {
71 0 : bail!("failpoint before-upload-manifest")
72 2 : });
73 2 : pausable_failpoint!("before-upload-manifest-pausable");
74 :
75 2 : let serialized = tenant_manifest.to_json_bytes()?;
76 2 : let serialized = Bytes::from(serialized);
77 2 :
78 2 : let tenant_manifest_site = serialized.len();
79 2 :
80 2 : let remote_path = remote_tenant_manifest_path(tenant_shard_id, generation);
81 2 : storage
82 2 : .upload_storage_object(
83 2 : futures::stream::once(futures::future::ready(Ok(serialized))),
84 2 : tenant_manifest_site,
85 2 : &remote_path,
86 2 : cancel,
87 2 : )
88 3 : .await
89 2 : .with_context(|| format!("upload tenant manifest for '{tenant_shard_id}'"))
90 2 : }
91 :
92 : /// Attempts to upload given layer files.
93 : /// No extra checks for overlapping files is made and any files that are already present remotely will be overwritten, if submitted during the upload.
94 : ///
95 : /// On an error, bumps the retries count and reschedules the entire task.
96 1436 : pub(super) async fn upload_timeline_layer<'a>(
97 1436 : storage: &'a GenericRemoteStorage,
98 1436 : local_path: &'a Utf8Path,
99 1436 : remote_path: &'a RemotePath,
100 1436 : metadata_size: u64,
101 1436 : cancel: &CancellationToken,
102 1436 : ) -> anyhow::Result<()> {
103 1436 : fail_point!("before-upload-layer", |_| {
104 0 : bail!("failpoint before-upload-layer")
105 1436 : });
106 :
107 1436 : pausable_failpoint!("before-upload-layer-pausable");
108 :
109 1378 : let source_file_res = fs::File::open(&local_path).await;
110 1372 : let source_file = match source_file_res {
111 1372 : Ok(source_file) => source_file,
112 0 : Err(e) if e.kind() == ErrorKind::NotFound => {
113 0 : // If we encounter this arm, it wasn't intended, but it's also not
114 0 : // a big problem, if it's because the file was deleted before an
115 0 : // upload. However, a nonexistent file can also be indicative of
116 0 : // something worse, like when a file is scheduled for upload before
117 0 : // it has been written to disk yet.
118 0 : //
119 0 : // This is tested against `test_compaction_delete_before_upload`
120 0 : info!(path = %local_path, "File to upload doesn't exist. Likely the file has been deleted and an upload is not required any more.");
121 0 : return Ok(());
122 : }
123 0 : Err(e) => Err(e).with_context(|| format!("open a source file for layer {local_path:?}"))?,
124 : };
125 :
126 1372 : let fs_size = source_file
127 1372 : .metadata()
128 1258 : .await
129 1372 : .with_context(|| format!("get the source file metadata for layer {local_path:?}"))?
130 1372 : .len();
131 1372 :
132 1372 : if metadata_size != fs_size {
133 0 : bail!("File {local_path:?} has its current FS size {fs_size} diferent from initially determined {metadata_size}");
134 1372 : }
135 :
136 1372 : let fs_size = usize::try_from(fs_size)
137 1372 : .with_context(|| format!("convert {local_path:?} size {fs_size} usize"))?;
138 :
139 1372 : let reader = tokio_util::io::ReaderStream::with_capacity(source_file, super::BUFFER_SIZE);
140 1372 :
141 1372 : storage
142 1372 : .upload(reader, fs_size, remote_path, None, cancel)
143 35172 : .await
144 1342 : .with_context(|| format!("upload layer from local path '{local_path}'"))
145 1342 : }
146 :
147 0 : pub(super) async fn copy_timeline_layer(
148 0 : storage: &GenericRemoteStorage,
149 0 : source_path: &RemotePath,
150 0 : target_path: &RemotePath,
151 0 : cancel: &CancellationToken,
152 0 : ) -> anyhow::Result<()> {
153 0 : fail_point!("before-copy-layer", |_| {
154 0 : bail!("failpoint before-copy-layer")
155 0 : });
156 :
157 0 : pausable_failpoint!("before-copy-layer-pausable");
158 :
159 0 : storage
160 0 : .copy_object(source_path, target_path, cancel)
161 0 : .await
162 0 : .with_context(|| format!("copy layer {source_path} to {target_path}"))
163 0 : }
164 :
165 : /// Uploads the given `initdb` data to the remote storage.
166 0 : pub(crate) async fn upload_initdb_dir(
167 0 : storage: &GenericRemoteStorage,
168 0 : tenant_id: &TenantId,
169 0 : timeline_id: &TimelineId,
170 0 : mut initdb_tar_zst: File,
171 0 : size: u64,
172 0 : cancel: &CancellationToken,
173 0 : ) -> anyhow::Result<()> {
174 0 : tracing::trace!("uploading initdb dir");
175 :
176 : // We might have read somewhat into the file already in the prior retry attempt
177 0 : initdb_tar_zst.seek(SeekFrom::Start(0)).await?;
178 :
179 0 : let file = tokio_util::io::ReaderStream::with_capacity(initdb_tar_zst, super::BUFFER_SIZE);
180 0 :
181 0 : let remote_path = remote_initdb_archive_path(tenant_id, timeline_id);
182 0 : storage
183 0 : .upload_storage_object(file, size as usize, &remote_path, cancel)
184 0 : .await
185 0 : .with_context(|| format!("upload initdb dir for '{tenant_id} / {timeline_id}'"))
186 0 : }
187 :
188 0 : pub(crate) async fn preserve_initdb_archive(
189 0 : storage: &GenericRemoteStorage,
190 0 : tenant_id: &TenantId,
191 0 : timeline_id: &TimelineId,
192 0 : cancel: &CancellationToken,
193 0 : ) -> anyhow::Result<()> {
194 0 : let source_path = remote_initdb_archive_path(tenant_id, timeline_id);
195 0 : let dest_path = remote_initdb_preserved_archive_path(tenant_id, timeline_id);
196 0 : storage
197 0 : .copy_object(&source_path, &dest_path, cancel)
198 0 : .await
199 0 : .with_context(|| format!("backing up initdb archive for '{tenant_id} / {timeline_id}'"))
200 0 : }
201 :
202 0 : pub(crate) async fn time_travel_recover_tenant(
203 0 : storage: &GenericRemoteStorage,
204 0 : tenant_shard_id: &TenantShardId,
205 0 : timestamp: SystemTime,
206 0 : done_if_after: SystemTime,
207 0 : cancel: &CancellationToken,
208 0 : ) -> Result<(), TimeTravelError> {
209 0 : let warn_after = 3;
210 0 : let max_attempts = 10;
211 0 : let mut prefixes = Vec::with_capacity(2);
212 0 : if tenant_shard_id.is_shard_zero() {
213 0 : // Also recover the unsharded prefix for a shard of zero:
214 0 : // - if the tenant is totally unsharded, the unsharded prefix contains all the data
215 0 : // - if the tenant is sharded, we still want to recover the initdb data, but we only
216 0 : // want to do it once, so let's do it on the 0 shard
217 0 : let timelines_path_unsharded =
218 0 : super::remote_timelines_path_unsharded(&tenant_shard_id.tenant_id);
219 0 : prefixes.push(timelines_path_unsharded);
220 0 : }
221 0 : if !tenant_shard_id.is_unsharded() {
222 0 : // If the tenant is sharded, we need to recover the sharded prefix
223 0 : let timelines_path = super::remote_timelines_path(tenant_shard_id);
224 0 : prefixes.push(timelines_path);
225 0 : }
226 0 : for prefix in &prefixes {
227 0 : backoff::retry(
228 0 : || async {
229 0 : storage
230 0 : .time_travel_recover(Some(prefix), timestamp, done_if_after, cancel)
231 0 : .await
232 0 : },
233 0 : |e| !matches!(e, TimeTravelError::Other(_)),
234 0 : warn_after,
235 0 : max_attempts,
236 0 : "time travel recovery of tenant prefix",
237 0 : cancel,
238 0 : )
239 0 : .await
240 0 : .ok_or_else(|| TimeTravelError::Cancelled)
241 0 : .and_then(|x| x)?;
242 : }
243 0 : Ok(())
244 0 : }
|
