Line data Source code
1 : //! An ImageLayer represents an image or a snapshot of a key-range at
2 : //! one particular LSN.
3 : //!
4 : //! It contains an image of all key-value pairs in its key-range. Any key
5 : //! that falls into the image layer's range but does not exist in the layer,
6 : //! does not exist.
7 : //!
8 : //! An image layer is stored in a file on disk. The file is stored in
9 : //! timelines/<timeline_id> directory. Currently, there are no
10 : //! subdirectories, and each image layer file is named like this:
11 : //!
12 : //! ```text
13 : //! <key start>-<key end>__<LSN>
14 : //! ```
15 : //!
16 : //! For example:
17 : //!
18 : //! ```text
19 : //! 000000067F000032BE0000400000000070B6-000000067F000032BE0000400000000080B6__00000000346BC568
20 : //! ```
21 : //!
22 : //! Every image layer file consists of three parts: "summary",
23 : //! "index", and "values". The summary is a fixed size header at the
24 : //! beginning of the file, and it contains basic information about the
25 : //! layer, and offsets to the other parts. The "index" is a B-tree,
26 : //! mapping from Key to an offset in the "values" part. The
27 : //! actual page images are stored in the "values" part.
28 : use crate::config::PageServerConf;
29 : use crate::context::{PageContentKind, RequestContext, RequestContextBuilder};
30 : use crate::page_cache::{self, FileId, PAGE_SZ};
31 : use crate::tenant::blob_io::BlobWriter;
32 : use crate::tenant::block_io::{BlockBuf, FileBlockReader};
33 : use crate::tenant::disk_btree::{
34 : DiskBtreeBuilder, DiskBtreeIterator, DiskBtreeReader, VisitDirection,
35 : };
36 : use crate::tenant::timeline::GetVectoredError;
37 : use crate::tenant::vectored_blob_io::{
38 : BlobFlag, BufView, StreamingVectoredReadPlanner, VectoredBlobReader, VectoredRead,
39 : VectoredReadPlanner,
40 : };
41 : use crate::virtual_file::owned_buffers_io::io_buf_ext::IoBufExt;
42 : use crate::virtual_file::IoBufferMut;
43 : use crate::virtual_file::{self, MaybeFatalIo, VirtualFile};
44 : use crate::{IMAGE_FILE_MAGIC, STORAGE_FORMAT_VERSION, TEMP_FILE_SUFFIX};
45 : use anyhow::{bail, ensure, Context, Result};
46 : use bytes::Bytes;
47 : use camino::{Utf8Path, Utf8PathBuf};
48 : use hex;
49 : use itertools::Itertools;
50 : use pageserver_api::config::MaxVectoredReadBytes;
51 : use pageserver_api::key::DBDIR_KEY;
52 : use pageserver_api::key::{Key, KEY_SIZE};
53 : use pageserver_api::keyspace::KeySpace;
54 : use pageserver_api::shard::{ShardIdentity, TenantShardId};
55 : use pageserver_api::value::Value;
56 : use rand::{distributions::Alphanumeric, Rng};
57 : use serde::{Deserialize, Serialize};
58 : use std::collections::{HashMap, VecDeque};
59 : use std::fs::File;
60 : use std::io::SeekFrom;
61 : use std::ops::Range;
62 : use std::os::unix::prelude::FileExt;
63 : use std::str::FromStr;
64 : use std::sync::Arc;
65 : use tokio::sync::OnceCell;
66 : use tokio_stream::StreamExt;
67 : use tracing::*;
68 :
69 : use utils::{
70 : bin_ser::BeSer,
71 : id::{TenantId, TimelineId},
72 : lsn::Lsn,
73 : };
74 :
75 : use super::layer_name::ImageLayerName;
76 : use super::{
77 : AsLayerDesc, LayerName, OnDiskValue, OnDiskValueIo, PersistentLayerDesc, ResidentLayer,
78 : ValuesReconstructState,
79 : };
80 :
81 : ///
82 : /// Header stored in the beginning of the file
83 : ///
84 : /// After this comes the 'values' part, starting on block 1. After that,
85 : /// the 'index' starts at the block indicated by 'index_start_blk'
86 : ///
87 0 : #[derive(Debug, Serialize, Deserialize, PartialEq, Eq)]
88 : pub struct Summary {
89 : /// Magic value to identify this as a neon image file. Always IMAGE_FILE_MAGIC.
90 : pub magic: u16,
91 : pub format_version: u16,
92 :
93 : pub tenant_id: TenantId,
94 : pub timeline_id: TimelineId,
95 : pub key_range: Range<Key>,
96 : pub lsn: Lsn,
97 :
98 : /// Block number where the 'index' part of the file begins.
99 : pub index_start_blk: u32,
100 : /// Block within the 'index', where the B-tree root page is stored
101 : pub index_root_blk: u32,
102 : // the 'values' part starts after the summary header, on block 1.
103 : }
104 :
105 : impl From<&ImageLayer> for Summary {
106 0 : fn from(layer: &ImageLayer) -> Self {
107 0 : Self::expected(
108 0 : layer.desc.tenant_shard_id.tenant_id,
109 0 : layer.desc.timeline_id,
110 0 : layer.desc.key_range.clone(),
111 0 : layer.lsn,
112 0 : )
113 0 : }
114 : }
115 :
116 : impl Summary {
117 268 : pub(super) fn expected(
118 268 : tenant_id: TenantId,
119 268 : timeline_id: TimelineId,
120 268 : key_range: Range<Key>,
121 268 : lsn: Lsn,
122 268 : ) -> Self {
123 268 : Self {
124 268 : magic: IMAGE_FILE_MAGIC,
125 268 : format_version: STORAGE_FORMAT_VERSION,
126 268 : tenant_id,
127 268 : timeline_id,
128 268 : key_range,
129 268 : lsn,
130 268 :
131 268 : index_start_blk: 0,
132 268 : index_root_blk: 0,
133 268 : }
134 268 : }
135 : }
136 :
137 : /// This is used only from `pagectl`. Within pageserver, all layers are
138 : /// [`crate::tenant::storage_layer::Layer`], which can hold an [`ImageLayerInner`].
139 : pub struct ImageLayer {
140 : path: Utf8PathBuf,
141 : pub desc: PersistentLayerDesc,
142 : // This entry contains an image of all pages as of this LSN, should be the same as desc.lsn
143 : pub lsn: Lsn,
144 : inner: OnceCell<ImageLayerInner>,
145 : }
146 :
147 : impl std::fmt::Debug for ImageLayer {
148 0 : fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
149 : use super::RangeDisplayDebug;
150 :
151 0 : f.debug_struct("ImageLayer")
152 0 : .field("key_range", &RangeDisplayDebug(&self.desc.key_range))
153 0 : .field("file_size", &self.desc.file_size)
154 0 : .field("lsn", &self.lsn)
155 0 : .field("inner", &self.inner)
156 0 : .finish()
157 0 : }
158 : }
159 :
160 : /// ImageLayer is the in-memory data structure associated with an on-disk image
161 : /// file.
162 : pub struct ImageLayerInner {
163 : // values copied from summary
164 : index_start_blk: u32,
165 : index_root_blk: u32,
166 :
167 : key_range: Range<Key>,
168 : lsn: Lsn,
169 :
170 : file: Arc<VirtualFile>,
171 : file_id: FileId,
172 :
173 : max_vectored_read_bytes: Option<MaxVectoredReadBytes>,
174 : }
175 :
176 : impl ImageLayerInner {
177 0 : pub(crate) fn layer_dbg_info(&self) -> String {
178 0 : format!(
179 0 : "image {}..{} {}",
180 0 : self.key_range().start,
181 0 : self.key_range().end,
182 0 : self.lsn()
183 0 : )
184 0 : }
185 : }
186 :
187 : impl std::fmt::Debug for ImageLayerInner {
188 0 : fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
189 0 : f.debug_struct("ImageLayerInner")
190 0 : .field("index_start_blk", &self.index_start_blk)
191 0 : .field("index_root_blk", &self.index_root_blk)
192 0 : .finish()
193 0 : }
194 : }
195 :
196 : impl ImageLayerInner {
197 0 : pub(super) async fn dump(&self, ctx: &RequestContext) -> anyhow::Result<()> {
198 0 : let block_reader = FileBlockReader::new(&self.file, self.file_id);
199 0 : let tree_reader = DiskBtreeReader::<_, KEY_SIZE>::new(
200 0 : self.index_start_blk,
201 0 : self.index_root_blk,
202 0 : block_reader,
203 0 : );
204 0 :
205 0 : tree_reader.dump().await?;
206 :
207 0 : tree_reader
208 0 : .visit(
209 0 : &[0u8; KEY_SIZE],
210 0 : VisitDirection::Forwards,
211 0 : |key, value| {
212 0 : println!("key: {} offset {}", hex::encode(key), value);
213 0 : true
214 0 : },
215 0 : ctx,
216 0 : )
217 0 : .await?;
218 :
219 0 : Ok(())
220 0 : }
221 : }
222 :
223 : /// Boilerplate to implement the Layer trait, always use layer_desc for persistent layers.
224 : impl std::fmt::Display for ImageLayer {
225 0 : fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
226 0 : write!(f, "{}", self.layer_desc().short_id())
227 0 : }
228 : }
229 :
230 : impl AsLayerDesc for ImageLayer {
231 0 : fn layer_desc(&self) -> &PersistentLayerDesc {
232 0 : &self.desc
233 0 : }
234 : }
235 :
236 : impl ImageLayer {
237 0 : pub async fn dump(&self, verbose: bool, ctx: &RequestContext) -> Result<()> {
238 0 : self.desc.dump();
239 0 :
240 0 : if !verbose {
241 0 : return Ok(());
242 0 : }
243 :
244 0 : let inner = self.load(ctx).await?;
245 :
246 0 : inner.dump(ctx).await?;
247 :
248 0 : Ok(())
249 0 : }
250 :
251 1168 : fn temp_path_for(
252 1168 : conf: &PageServerConf,
253 1168 : timeline_id: TimelineId,
254 1168 : tenant_shard_id: TenantShardId,
255 1168 : fname: &ImageLayerName,
256 1168 : ) -> Utf8PathBuf {
257 1168 : let rand_string: String = rand::thread_rng()
258 1168 : .sample_iter(&Alphanumeric)
259 1168 : .take(8)
260 1168 : .map(char::from)
261 1168 : .collect();
262 1168 :
263 1168 : conf.timeline_path(&tenant_shard_id, &timeline_id)
264 1168 : .join(format!("{fname}.{rand_string}.{TEMP_FILE_SUFFIX}"))
265 1168 : }
266 :
267 : ///
268 : /// Open the underlying file and read the metadata into memory, if it's
269 : /// not loaded already.
270 : ///
271 0 : async fn load(&self, ctx: &RequestContext) -> Result<&ImageLayerInner> {
272 0 : self.inner
273 0 : .get_or_try_init(|| self.load_inner(ctx))
274 0 : .await
275 0 : .with_context(|| format!("Failed to load image layer {}", self.path()))
276 0 : }
277 :
278 0 : async fn load_inner(&self, ctx: &RequestContext) -> Result<ImageLayerInner> {
279 0 : let path = self.path();
280 :
281 0 : let loaded =
282 0 : ImageLayerInner::load(&path, self.desc.image_layer_lsn(), None, None, ctx).await?;
283 :
284 : // not production code
285 0 : let actual_layer_name = LayerName::from_str(path.file_name().unwrap()).unwrap();
286 0 : let expected_layer_name = self.layer_desc().layer_name();
287 0 :
288 0 : if actual_layer_name != expected_layer_name {
289 0 : println!("warning: filename does not match what is expected from in-file summary");
290 0 : println!("actual: {:?}", actual_layer_name.to_string());
291 0 : println!("expected: {:?}", expected_layer_name.to_string());
292 0 : }
293 :
294 0 : Ok(loaded)
295 0 : }
296 :
297 : /// Create an ImageLayer struct representing an existing file on disk.
298 : ///
299 : /// This variant is only used for debugging purposes, by the 'pagectl' binary.
300 0 : pub fn new_for_path(path: &Utf8Path, file: File) -> Result<ImageLayer> {
301 0 : let mut summary_buf = vec![0; PAGE_SZ];
302 0 : file.read_exact_at(&mut summary_buf, 0)?;
303 0 : let summary = Summary::des_prefix(&summary_buf)?;
304 0 : let metadata = file
305 0 : .metadata()
306 0 : .context("get file metadata to determine size")?;
307 :
308 : // This function is never used for constructing layers in a running pageserver,
309 : // so it does not need an accurate TenantShardId.
310 0 : let tenant_shard_id = TenantShardId::unsharded(summary.tenant_id);
311 0 :
312 0 : Ok(ImageLayer {
313 0 : path: path.to_path_buf(),
314 0 : desc: PersistentLayerDesc::new_img(
315 0 : tenant_shard_id,
316 0 : summary.timeline_id,
317 0 : summary.key_range,
318 0 : summary.lsn,
319 0 : metadata.len(),
320 0 : ), // Now we assume image layer ALWAYS covers the full range. This may change in the future.
321 0 : lsn: summary.lsn,
322 0 : inner: OnceCell::new(),
323 0 : })
324 0 : }
325 :
326 0 : fn path(&self) -> Utf8PathBuf {
327 0 : self.path.clone()
328 0 : }
329 : }
330 :
331 : #[derive(thiserror::Error, Debug)]
332 : pub enum RewriteSummaryError {
333 : #[error("magic mismatch")]
334 : MagicMismatch,
335 : #[error(transparent)]
336 : Other(#[from] anyhow::Error),
337 : }
338 :
339 : impl From<std::io::Error> for RewriteSummaryError {
340 0 : fn from(e: std::io::Error) -> Self {
341 0 : Self::Other(anyhow::anyhow!(e))
342 0 : }
343 : }
344 :
345 : impl ImageLayer {
346 0 : pub async fn rewrite_summary<F>(
347 0 : path: &Utf8Path,
348 0 : rewrite: F,
349 0 : ctx: &RequestContext,
350 0 : ) -> Result<(), RewriteSummaryError>
351 0 : where
352 0 : F: Fn(Summary) -> Summary,
353 0 : {
354 0 : let mut file = VirtualFile::open_with_options(
355 0 : path,
356 0 : virtual_file::OpenOptions::new().read(true).write(true),
357 0 : ctx,
358 0 : )
359 0 : .await
360 0 : .with_context(|| format!("Failed to open file '{}'", path))?;
361 0 : let file_id = page_cache::next_file_id();
362 0 : let block_reader = FileBlockReader::new(&file, file_id);
363 0 : let summary_blk = block_reader.read_blk(0, ctx).await?;
364 0 : let actual_summary = Summary::des_prefix(summary_blk.as_ref()).context("deserialize")?;
365 0 : if actual_summary.magic != IMAGE_FILE_MAGIC {
366 0 : return Err(RewriteSummaryError::MagicMismatch);
367 0 : }
368 0 :
369 0 : let new_summary = rewrite(actual_summary);
370 0 :
371 0 : let mut buf = Vec::with_capacity(PAGE_SZ);
372 0 : // TODO: could use smallvec here but it's a pain with Slice<T>
373 0 : Summary::ser_into(&new_summary, &mut buf).context("serialize")?;
374 0 : file.seek(SeekFrom::Start(0)).await?;
375 0 : let (_buf, res) = file.write_all(buf.slice_len(), ctx).await;
376 0 : res?;
377 0 : Ok(())
378 0 : }
379 : }
380 :
381 : impl ImageLayerInner {
382 264 : pub(crate) fn key_range(&self) -> &Range<Key> {
383 264 : &self.key_range
384 264 : }
385 :
386 264 : pub(crate) fn lsn(&self) -> Lsn {
387 264 : self.lsn
388 264 : }
389 :
390 268 : pub(super) async fn load(
391 268 : path: &Utf8Path,
392 268 : lsn: Lsn,
393 268 : summary: Option<Summary>,
394 268 : max_vectored_read_bytes: Option<MaxVectoredReadBytes>,
395 268 : ctx: &RequestContext,
396 268 : ) -> anyhow::Result<Self> {
397 268 : let file = Arc::new(
398 268 : VirtualFile::open_v2(path, ctx)
399 268 : .await
400 268 : .context("open layer file")?,
401 : );
402 268 : let file_id = page_cache::next_file_id();
403 268 : let block_reader = FileBlockReader::new(&file, file_id);
404 268 : let summary_blk = block_reader
405 268 : .read_blk(0, ctx)
406 268 : .await
407 268 : .context("read first block")?;
408 :
409 : // length is the only way how this could fail, so it's not actually likely at all unless
410 : // read_blk returns wrong sized block.
411 : //
412 : // TODO: confirm and make this into assertion
413 268 : let actual_summary =
414 268 : Summary::des_prefix(summary_blk.as_ref()).context("deserialize first block")?;
415 :
416 268 : if let Some(mut expected_summary) = summary {
417 : // production code path
418 268 : expected_summary.index_start_blk = actual_summary.index_start_blk;
419 268 : expected_summary.index_root_blk = actual_summary.index_root_blk;
420 268 : // mask out the timeline_id, but still require the layers to be from the same tenant
421 268 : expected_summary.timeline_id = actual_summary.timeline_id;
422 268 :
423 268 : if actual_summary != expected_summary {
424 0 : bail!(
425 0 : "in-file summary does not match expected summary. actual = {:?} expected = {:?}",
426 0 : actual_summary,
427 0 : expected_summary
428 0 : );
429 268 : }
430 0 : }
431 :
432 268 : Ok(ImageLayerInner {
433 268 : index_start_blk: actual_summary.index_start_blk,
434 268 : index_root_blk: actual_summary.index_root_blk,
435 268 : lsn,
436 268 : file,
437 268 : file_id,
438 268 : max_vectored_read_bytes,
439 268 : key_range: actual_summary.key_range,
440 268 : })
441 268 : }
442 :
443 : // Look up the keys in the provided keyspace and update
444 : // the reconstruct state with whatever is found.
445 45116 : pub(super) async fn get_values_reconstruct_data(
446 45116 : &self,
447 45116 : this: ResidentLayer,
448 45116 : keyspace: KeySpace,
449 45116 : reconstruct_state: &mut ValuesReconstructState,
450 45116 : ctx: &RequestContext,
451 45116 : ) -> Result<(), GetVectoredError> {
452 45116 : let reads = self
453 45116 : .plan_reads(keyspace, None, ctx)
454 45116 : .await
455 45116 : .map_err(GetVectoredError::Other)?;
456 :
457 45116 : self.do_reads_and_update_state(this, reads, reconstruct_state, ctx)
458 45116 : .await;
459 :
460 45116 : reconstruct_state.on_image_layer_visited(&self.key_range);
461 45116 :
462 45116 : Ok(())
463 45116 : }
464 :
465 : /// Traverse the layer's index to build read operations on the overlap of the input keyspace
466 : /// and the keys in this layer.
467 : ///
468 : /// If shard_identity is provided, it will be used to filter keys down to those stored on
469 : /// this shard.
470 45132 : async fn plan_reads(
471 45132 : &self,
472 45132 : keyspace: KeySpace,
473 45132 : shard_identity: Option<&ShardIdentity>,
474 45132 : ctx: &RequestContext,
475 45132 : ) -> anyhow::Result<Vec<VectoredRead>> {
476 45132 : let mut planner = VectoredReadPlanner::new(
477 45132 : self.max_vectored_read_bytes
478 45132 : .expect("Layer is loaded with max vectored bytes config")
479 45132 : .0
480 45132 : .into(),
481 45132 : );
482 45132 :
483 45132 : let block_reader = FileBlockReader::new(&self.file, self.file_id);
484 45132 : let tree_reader =
485 45132 : DiskBtreeReader::new(self.index_start_blk, self.index_root_blk, block_reader);
486 45132 :
487 45132 : let ctx = RequestContextBuilder::extend(ctx)
488 45132 : .page_content_kind(PageContentKind::ImageLayerBtreeNode)
489 45132 : .build();
490 :
491 45148 : for range in keyspace.ranges.iter() {
492 45148 : let mut range_end_handled = false;
493 45148 : let mut search_key: [u8; KEY_SIZE] = [0u8; KEY_SIZE];
494 45148 : range.start.write_to_byte_slice(&mut search_key);
495 45148 :
496 45148 : let index_stream = tree_reader.clone().into_stream(&search_key, &ctx);
497 45148 : let mut index_stream = std::pin::pin!(index_stream);
498 :
499 2251468 : while let Some(index_entry) = index_stream.next().await {
500 2250860 : let (raw_key, offset) = index_entry?;
501 :
502 2250860 : let key = Key::from_slice(&raw_key[..KEY_SIZE]);
503 2250860 : assert!(key >= range.start);
504 :
505 2250860 : let flag = if let Some(shard_identity) = shard_identity {
506 2097152 : if shard_identity.is_key_disposable(&key) {
507 1572864 : BlobFlag::Ignore
508 : } else {
509 524288 : BlobFlag::None
510 : }
511 : } else {
512 153708 : BlobFlag::None
513 : };
514 :
515 2250860 : if key >= range.end {
516 44540 : planner.handle_range_end(offset);
517 44540 : range_end_handled = true;
518 44540 : break;
519 2206320 : } else {
520 2206320 : planner.handle(key, self.lsn, offset, flag);
521 2206320 : }
522 : }
523 :
524 45148 : if !range_end_handled {
525 608 : let payload_end = self.index_start_blk as u64 * PAGE_SZ as u64;
526 608 : planner.handle_range_end(payload_end);
527 44540 : }
528 : }
529 :
530 45132 : Ok(planner.finish())
531 45132 : }
532 :
533 : /// Given a key range, select the parts of that range that should be retained by the ShardIdentity,
534 : /// then execute vectored GET operations, passing the results of all read keys into the writer.
535 16 : pub(super) async fn filter(
536 16 : &self,
537 16 : shard_identity: &ShardIdentity,
538 16 : writer: &mut ImageLayerWriter,
539 16 : ctx: &RequestContext,
540 16 : ) -> anyhow::Result<usize> {
541 : // Fragment the range into the regions owned by this ShardIdentity
542 16 : let plan = self
543 16 : .plan_reads(
544 16 : KeySpace {
545 16 : // If asked for the total key space, plan_reads will give us all the keys in the layer
546 16 : ranges: vec![Key::MIN..Key::MAX],
547 16 : },
548 16 : Some(shard_identity),
549 16 : ctx,
550 16 : )
551 16 : .await?;
552 :
553 16 : let vectored_blob_reader = VectoredBlobReader::new(&self.file);
554 16 : let mut key_count = 0;
555 32 : for read in plan.into_iter() {
556 32 : let buf_size = read.size();
557 32 :
558 32 : let buf = IoBufferMut::with_capacity(buf_size);
559 32 : let blobs_buf = vectored_blob_reader.read_blobs(&read, buf, ctx).await?;
560 :
561 32 : let view = BufView::new_slice(&blobs_buf.buf);
562 :
563 524288 : for meta in blobs_buf.blobs.iter() {
564 524288 : let img_buf = meta.read(&view).await?;
565 :
566 524288 : key_count += 1;
567 524288 : writer
568 524288 : .put_image(meta.meta.key, img_buf.into_bytes(), ctx)
569 524288 : .await
570 524288 : .context(format!("Storing key {}", meta.meta.key))?;
571 : }
572 : }
573 :
574 16 : Ok(key_count)
575 16 : }
576 :
577 45116 : async fn do_reads_and_update_state(
578 45116 : &self,
579 45116 : this: ResidentLayer,
580 45116 : reads: Vec<VectoredRead>,
581 45116 : reconstruct_state: &mut ValuesReconstructState,
582 45116 : ctx: &RequestContext,
583 45116 : ) {
584 45116 : let max_vectored_read_bytes = self
585 45116 : .max_vectored_read_bytes
586 45116 : .expect("Layer is loaded with max vectored bytes config")
587 45116 : .0
588 45116 : .into();
589 :
590 45116 : for read in reads.into_iter() {
591 45040 : let mut ios: HashMap<(Key, Lsn), OnDiskValueIo> = Default::default();
592 109168 : for (_, blob_meta) in read.blobs_at.as_slice() {
593 109168 : let io = reconstruct_state.update_key(&blob_meta.key, blob_meta.lsn, true);
594 109168 : ios.insert((blob_meta.key, blob_meta.lsn), io);
595 109168 : }
596 :
597 45040 : let buf_size = read.size();
598 45040 :
599 45040 : if buf_size > max_vectored_read_bytes {
600 : // If the read is oversized, it should only contain one key.
601 0 : let offenders = read
602 0 : .blobs_at
603 0 : .as_slice()
604 0 : .iter()
605 0 : .filter_map(|(_, blob_meta)| {
606 0 : if blob_meta.key.is_rel_dir_key() || blob_meta.key == DBDIR_KEY {
607 : // The size of values for these keys is unbounded and can
608 : // grow very large in pathological cases.
609 0 : None
610 : } else {
611 0 : Some(format!("{}@{}", blob_meta.key, blob_meta.lsn))
612 : }
613 0 : })
614 0 : .join(", ");
615 0 :
616 0 : if !offenders.is_empty() {
617 0 : tracing::warn!(
618 0 : "Oversized vectored read ({} > {}) for keys {}",
619 : buf_size,
620 : max_vectored_read_bytes,
621 : offenders
622 : );
623 0 : }
624 45040 : }
625 :
626 45040 : let read_extend_residency = this.clone();
627 45040 : let read_from = self.file.clone();
628 45040 : let read_ctx = ctx.attached_child();
629 45040 : reconstruct_state
630 45040 : .spawn_io(async move {
631 45040 : let buf = IoBufferMut::with_capacity(buf_size);
632 45040 : let vectored_blob_reader = VectoredBlobReader::new(&read_from);
633 45040 : let res = vectored_blob_reader.read_blobs(&read, buf, &read_ctx).await;
634 :
635 45040 : match res {
636 45040 : Ok(blobs_buf) => {
637 45040 : let view = BufView::new_slice(&blobs_buf.buf);
638 109168 : for meta in blobs_buf.blobs.iter() {
639 109168 : let io: OnDiskValueIo =
640 109168 : ios.remove(&(meta.meta.key, meta.meta.lsn)).unwrap();
641 109168 : let img_buf = meta.read(&view).await;
642 :
643 109168 : let img_buf = match img_buf {
644 109168 : Ok(img_buf) => img_buf,
645 0 : Err(e) => {
646 0 : io.complete(Err(e));
647 0 : continue;
648 : }
649 : };
650 :
651 109168 : io.complete(Ok(OnDiskValue::RawImage(img_buf.into_bytes())));
652 : }
653 :
654 45040 : assert!(ios.is_empty());
655 : }
656 0 : Err(err) => {
657 0 : for (_, io) in ios {
658 0 : io.complete(Err(std::io::Error::new(
659 0 : err.kind(),
660 0 : "vec read failed",
661 0 : )));
662 0 : }
663 : }
664 : }
665 :
666 : // keep layer resident until this IO is done; this spawned IO future generally outlives the
667 : // call to `self` / the `Arc<DownloadedLayer>` / the `ResidentLayer` that guarantees residency
668 45040 : drop(read_extend_residency);
669 45040 : })
670 45040 : .await;
671 : }
672 45116 : }
673 :
674 244 : pub(crate) fn iter<'a>(&'a self, ctx: &'a RequestContext) -> ImageLayerIterator<'a> {
675 244 : let block_reader = FileBlockReader::new(&self.file, self.file_id);
676 244 : let tree_reader =
677 244 : DiskBtreeReader::new(self.index_start_blk, self.index_root_blk, block_reader);
678 244 : ImageLayerIterator {
679 244 : image_layer: self,
680 244 : ctx,
681 244 : index_iter: tree_reader.iter(&[0; KEY_SIZE], ctx),
682 244 : key_values_batch: VecDeque::new(),
683 244 : is_end: false,
684 244 : planner: StreamingVectoredReadPlanner::new(
685 244 : 1024 * 8192, // The default value. Unit tests might use a different value. 1024 * 8K = 8MB buffer.
686 244 : 1024, // The default value. Unit tests might use a different value
687 244 : ),
688 244 : }
689 244 : }
690 :
691 : /// NB: not super efficient, but not terrible either. Should prob be an iterator.
692 : //
693 : // We're reusing the index traversal logical in plan_reads; would be nice to
694 : // factor that out.
695 0 : pub(crate) async fn load_keys(&self, ctx: &RequestContext) -> anyhow::Result<Vec<Key>> {
696 0 : let plan = self
697 0 : .plan_reads(KeySpace::single(self.key_range.clone()), None, ctx)
698 0 : .await?;
699 0 : Ok(plan
700 0 : .into_iter()
701 0 : .flat_map(|read| read.blobs_at)
702 0 : .map(|(_, blob_meta)| blob_meta.key)
703 0 : .collect())
704 0 : }
705 : }
706 :
707 : /// A builder object for constructing a new image layer.
708 : ///
709 : /// Usage:
710 : ///
711 : /// 1. Create the ImageLayerWriter by calling ImageLayerWriter::new(...)
712 : ///
713 : /// 2. Write the contents by calling `put_page_image` for every key-value
714 : /// pair in the key range.
715 : ///
716 : /// 3. Call `finish`.
717 : ///
718 : struct ImageLayerWriterInner {
719 : conf: &'static PageServerConf,
720 : path: Utf8PathBuf,
721 : timeline_id: TimelineId,
722 : tenant_shard_id: TenantShardId,
723 : key_range: Range<Key>,
724 : lsn: Lsn,
725 :
726 : // Total uncompressed bytes passed into put_image
727 : uncompressed_bytes: u64,
728 :
729 : // Like `uncompressed_bytes`,
730 : // but only of images we might consider for compression
731 : uncompressed_bytes_eligible: u64,
732 :
733 : // Like `uncompressed_bytes`, but only of images
734 : // where we have chosen their compressed form
735 : uncompressed_bytes_chosen: u64,
736 :
737 : // Number of keys in the layer.
738 : num_keys: usize,
739 :
740 : blob_writer: BlobWriter<false>,
741 : tree: DiskBtreeBuilder<BlockBuf, KEY_SIZE>,
742 :
743 : #[cfg(feature = "testing")]
744 : last_written_key: Key,
745 : }
746 :
747 : impl ImageLayerWriterInner {
748 : ///
749 : /// Start building a new image layer.
750 : ///
751 1168 : async fn new(
752 1168 : conf: &'static PageServerConf,
753 1168 : timeline_id: TimelineId,
754 1168 : tenant_shard_id: TenantShardId,
755 1168 : key_range: &Range<Key>,
756 1168 : lsn: Lsn,
757 1168 : ctx: &RequestContext,
758 1168 : ) -> anyhow::Result<Self> {
759 1168 : // Create the file initially with a temporary filename.
760 1168 : // We'll atomically rename it to the final name when we're done.
761 1168 : let path = ImageLayer::temp_path_for(
762 1168 : conf,
763 1168 : timeline_id,
764 1168 : tenant_shard_id,
765 1168 : &ImageLayerName {
766 1168 : key_range: key_range.clone(),
767 1168 : lsn,
768 1168 : },
769 1168 : );
770 1168 : trace!("creating image layer {}", path);
771 1168 : let mut file = {
772 1168 : VirtualFile::open_with_options(
773 1168 : &path,
774 1168 : virtual_file::OpenOptions::new()
775 1168 : .write(true)
776 1168 : .create_new(true),
777 1168 : ctx,
778 1168 : )
779 1168 : .await?
780 : };
781 : // make room for the header block
782 1168 : file.seek(SeekFrom::Start(PAGE_SZ as u64)).await?;
783 1168 : let blob_writer = BlobWriter::new(file, PAGE_SZ as u64);
784 1168 :
785 1168 : // Initialize the b-tree index builder
786 1168 : let block_buf = BlockBuf::new();
787 1168 : let tree_builder = DiskBtreeBuilder::new(block_buf);
788 1168 :
789 1168 : let writer = Self {
790 1168 : conf,
791 1168 : path,
792 1168 : timeline_id,
793 1168 : tenant_shard_id,
794 1168 : key_range: key_range.clone(),
795 1168 : lsn,
796 1168 : tree: tree_builder,
797 1168 : blob_writer,
798 1168 : uncompressed_bytes: 0,
799 1168 : uncompressed_bytes_eligible: 0,
800 1168 : uncompressed_bytes_chosen: 0,
801 1168 : num_keys: 0,
802 1168 : #[cfg(feature = "testing")]
803 1168 : last_written_key: Key::MIN,
804 1168 : };
805 1168 :
806 1168 : Ok(writer)
807 1168 : }
808 :
809 : ///
810 : /// Write next value to the file.
811 : ///
812 : /// The page versions must be appended in blknum order.
813 : ///
814 1093244 : async fn put_image(
815 1093244 : &mut self,
816 1093244 : key: Key,
817 1093244 : img: Bytes,
818 1093244 : ctx: &RequestContext,
819 1093244 : ) -> anyhow::Result<()> {
820 1093244 : ensure!(self.key_range.contains(&key));
821 1093244 : let compression = self.conf.image_compression;
822 1093244 : let uncompressed_len = img.len() as u64;
823 1093244 : self.uncompressed_bytes += uncompressed_len;
824 1093244 : self.num_keys += 1;
825 1093244 : let (_img, res) = self
826 1093244 : .blob_writer
827 1093244 : .write_blob_maybe_compressed(img.slice_len(), ctx, compression)
828 1093244 : .await;
829 : // TODO: re-use the buffer for `img` further upstack
830 1093244 : let (off, compression_info) = res?;
831 1093244 : if compression_info.compressed_size.is_some() {
832 16004 : // The image has been considered for compression at least
833 16004 : self.uncompressed_bytes_eligible += uncompressed_len;
834 1077240 : }
835 1093244 : if compression_info.written_compressed {
836 0 : // The image has been compressed
837 0 : self.uncompressed_bytes_chosen += uncompressed_len;
838 1093244 : }
839 :
840 1093244 : let mut keybuf: [u8; KEY_SIZE] = [0u8; KEY_SIZE];
841 1093244 : key.write_to_byte_slice(&mut keybuf);
842 1093244 : self.tree.append(&keybuf, off)?;
843 :
844 : #[cfg(feature = "testing")]
845 1093244 : {
846 1093244 : self.last_written_key = key;
847 1093244 : }
848 1093244 :
849 1093244 : Ok(())
850 1093244 : }
851 :
852 : ///
853 : /// Finish writing the image layer.
854 : ///
855 708 : async fn finish(
856 708 : self,
857 708 : ctx: &RequestContext,
858 708 : end_key: Option<Key>,
859 708 : ) -> anyhow::Result<(PersistentLayerDesc, Utf8PathBuf)> {
860 708 : let temp_path = self.path.clone();
861 708 : let result = self.finish0(ctx, end_key).await;
862 708 : if let Err(ref e) = result {
863 0 : tracing::info!(%temp_path, "cleaning up temporary file after error during writing: {e}");
864 0 : if let Err(e) = std::fs::remove_file(&temp_path) {
865 0 : tracing::warn!(error=%e, %temp_path, "error cleaning up temporary layer file after error during writing");
866 0 : }
867 708 : }
868 708 : result
869 708 : }
870 :
871 : ///
872 : /// Finish writing the image layer.
873 : ///
874 708 : async fn finish0(
875 708 : self,
876 708 : ctx: &RequestContext,
877 708 : end_key: Option<Key>,
878 708 : ) -> anyhow::Result<(PersistentLayerDesc, Utf8PathBuf)> {
879 708 : let index_start_blk = self.blob_writer.size().div_ceil(PAGE_SZ as u64) as u32;
880 708 :
881 708 : // Calculate compression ratio
882 708 : let compressed_size = self.blob_writer.size() - PAGE_SZ as u64; // Subtract PAGE_SZ for header
883 708 : crate::metrics::COMPRESSION_IMAGE_INPUT_BYTES.inc_by(self.uncompressed_bytes);
884 708 : crate::metrics::COMPRESSION_IMAGE_INPUT_BYTES_CONSIDERED
885 708 : .inc_by(self.uncompressed_bytes_eligible);
886 708 : crate::metrics::COMPRESSION_IMAGE_INPUT_BYTES_CHOSEN.inc_by(self.uncompressed_bytes_chosen);
887 708 : crate::metrics::COMPRESSION_IMAGE_OUTPUT_BYTES.inc_by(compressed_size);
888 708 :
889 708 : let mut file = self.blob_writer.into_inner();
890 708 :
891 708 : // Write out the index
892 708 : file.seek(SeekFrom::Start(index_start_blk as u64 * PAGE_SZ as u64))
893 708 : .await?;
894 708 : let (index_root_blk, block_buf) = self.tree.finish()?;
895 2328 : for buf in block_buf.blocks {
896 1620 : let (_buf, res) = file.write_all(buf.slice_len(), ctx).await;
897 1620 : res?;
898 : }
899 :
900 708 : let final_key_range = if let Some(end_key) = end_key {
901 572 : self.key_range.start..end_key
902 : } else {
903 136 : self.key_range.clone()
904 : };
905 :
906 : // Fill in the summary on blk 0
907 708 : let summary = Summary {
908 708 : magic: IMAGE_FILE_MAGIC,
909 708 : format_version: STORAGE_FORMAT_VERSION,
910 708 : tenant_id: self.tenant_shard_id.tenant_id,
911 708 : timeline_id: self.timeline_id,
912 708 : key_range: final_key_range.clone(),
913 708 : lsn: self.lsn,
914 708 : index_start_blk,
915 708 : index_root_blk,
916 708 : };
917 708 :
918 708 : let mut buf = Vec::with_capacity(PAGE_SZ);
919 708 : // TODO: could use smallvec here but it's a pain with Slice<T>
920 708 : Summary::ser_into(&summary, &mut buf)?;
921 708 : file.seek(SeekFrom::Start(0)).await?;
922 708 : let (_buf, res) = file.write_all(buf.slice_len(), ctx).await;
923 708 : res?;
924 :
925 708 : let metadata = file
926 708 : .metadata()
927 708 : .await
928 708 : .context("get metadata to determine file size")?;
929 :
930 708 : let desc = PersistentLayerDesc::new_img(
931 708 : self.tenant_shard_id,
932 708 : self.timeline_id,
933 708 : final_key_range,
934 708 : self.lsn,
935 708 : metadata.len(),
936 708 : );
937 :
938 : #[cfg(feature = "testing")]
939 708 : if let Some(end_key) = end_key {
940 572 : assert!(
941 572 : self.last_written_key < end_key,
942 0 : "written key violates end_key range"
943 : );
944 136 : }
945 :
946 : // Note: Because we open the file in write-only mode, we cannot
947 : // reuse the same VirtualFile for reading later. That's why we don't
948 : // set inner.file here. The first read will have to re-open it.
949 :
950 : // fsync the file
951 708 : file.sync_all()
952 708 : .await
953 708 : .maybe_fatal_err("image_layer sync_all")?;
954 :
955 708 : trace!("created image layer {}", self.path);
956 :
957 708 : Ok((desc, self.path))
958 708 : }
959 : }
960 :
961 : /// A builder object for constructing a new image layer.
962 : ///
963 : /// Usage:
964 : ///
965 : /// 1. Create the ImageLayerWriter by calling ImageLayerWriter::new(...)
966 : ///
967 : /// 2. Write the contents by calling `put_page_image` for every key-value
968 : /// pair in the key range.
969 : ///
970 : /// 3. Call `finish`.
971 : ///
972 : /// # Note
973 : ///
974 : /// As described in <https://github.com/neondatabase/neon/issues/2650>, it's
975 : /// possible for the writer to drop before `finish` is actually called. So this
976 : /// could lead to odd temporary files in the directory, exhausting file system.
977 : /// This structure wraps `ImageLayerWriterInner` and also contains `Drop`
978 : /// implementation that cleans up the temporary file in failure. It's not
979 : /// possible to do this directly in `ImageLayerWriterInner` since `finish` moves
980 : /// out some fields, making it impossible to implement `Drop`.
981 : ///
982 : #[must_use]
983 : pub struct ImageLayerWriter {
984 : inner: Option<ImageLayerWriterInner>,
985 : }
986 :
987 : impl ImageLayerWriter {
988 : ///
989 : /// Start building a new image layer.
990 : ///
991 1168 : pub async fn new(
992 1168 : conf: &'static PageServerConf,
993 1168 : timeline_id: TimelineId,
994 1168 : tenant_shard_id: TenantShardId,
995 1168 : key_range: &Range<Key>,
996 1168 : lsn: Lsn,
997 1168 : ctx: &RequestContext,
998 1168 : ) -> anyhow::Result<ImageLayerWriter> {
999 1168 : Ok(Self {
1000 1168 : inner: Some(
1001 1168 : ImageLayerWriterInner::new(conf, timeline_id, tenant_shard_id, key_range, lsn, ctx)
1002 1168 : .await?,
1003 : ),
1004 : })
1005 1168 : }
1006 :
1007 : ///
1008 : /// Write next value to the file.
1009 : ///
1010 : /// The page versions must be appended in blknum order.
1011 : ///
1012 1093244 : pub async fn put_image(
1013 1093244 : &mut self,
1014 1093244 : key: Key,
1015 1093244 : img: Bytes,
1016 1093244 : ctx: &RequestContext,
1017 1093244 : ) -> anyhow::Result<()> {
1018 1093244 : self.inner.as_mut().unwrap().put_image(key, img, ctx).await
1019 1093244 : }
1020 :
1021 : /// Estimated size of the image layer.
1022 17080 : pub(crate) fn estimated_size(&self) -> u64 {
1023 17080 : let inner = self.inner.as_ref().unwrap();
1024 17080 : inner.blob_writer.size() + inner.tree.borrow_writer().size() + PAGE_SZ as u64
1025 17080 : }
1026 :
1027 17272 : pub(crate) fn num_keys(&self) -> usize {
1028 17272 : self.inner.as_ref().unwrap().num_keys
1029 17272 : }
1030 :
1031 : ///
1032 : /// Finish writing the image layer.
1033 : ///
1034 136 : pub(crate) async fn finish(
1035 136 : mut self,
1036 136 : ctx: &RequestContext,
1037 136 : ) -> anyhow::Result<(PersistentLayerDesc, Utf8PathBuf)> {
1038 136 : self.inner.take().unwrap().finish(ctx, None).await
1039 136 : }
1040 :
1041 : /// Finish writing the image layer with an end key, used in [`super::batch_split_writer::SplitImageLayerWriter`]. The end key determines the end of the image layer's covered range and is exclusive.
1042 572 : pub(super) async fn finish_with_end_key(
1043 572 : mut self,
1044 572 : end_key: Key,
1045 572 : ctx: &RequestContext,
1046 572 : ) -> anyhow::Result<(PersistentLayerDesc, Utf8PathBuf)> {
1047 572 : self.inner.take().unwrap().finish(ctx, Some(end_key)).await
1048 572 : }
1049 : }
1050 :
1051 : impl Drop for ImageLayerWriter {
1052 1168 : fn drop(&mut self) {
1053 1168 : if let Some(inner) = self.inner.take() {
1054 460 : inner.blob_writer.into_inner().remove();
1055 708 : }
1056 1168 : }
1057 : }
1058 :
1059 : pub struct ImageLayerIterator<'a> {
1060 : image_layer: &'a ImageLayerInner,
1061 : ctx: &'a RequestContext,
1062 : planner: StreamingVectoredReadPlanner,
1063 : index_iter: DiskBtreeIterator<'a>,
1064 : key_values_batch: VecDeque<(Key, Lsn, Value)>,
1065 : is_end: bool,
1066 : }
1067 :
1068 : impl ImageLayerIterator<'_> {
1069 0 : pub(crate) fn layer_dbg_info(&self) -> String {
1070 0 : self.image_layer.layer_dbg_info()
1071 0 : }
1072 :
1073 : /// Retrieve a batch of key-value pairs into the iterator buffer.
1074 38252 : async fn next_batch(&mut self) -> anyhow::Result<()> {
1075 38252 : assert!(self.key_values_batch.is_empty());
1076 38252 : assert!(!self.is_end);
1077 :
1078 38252 : let plan = loop {
1079 58072 : if let Some(res) = self.index_iter.next().await {
1080 57884 : let (raw_key, offset) = res?;
1081 57884 : if let Some(batch_plan) = self.planner.handle(
1082 57884 : Key::from_slice(&raw_key[..KEY_SIZE]),
1083 57884 : self.image_layer.lsn,
1084 57884 : offset,
1085 57884 : true,
1086 57884 : ) {
1087 38064 : break batch_plan;
1088 19820 : }
1089 : } else {
1090 188 : self.is_end = true;
1091 188 : let payload_end = self.image_layer.index_start_blk as u64 * PAGE_SZ as u64;
1092 188 : if let Some(item) = self.planner.handle_range_end(payload_end) {
1093 188 : break item;
1094 : } else {
1095 0 : return Ok(()); // TODO: a test case on empty iterator
1096 : }
1097 : }
1098 : };
1099 38252 : let vectored_blob_reader = VectoredBlobReader::new(&self.image_layer.file);
1100 38252 : let mut next_batch = std::collections::VecDeque::new();
1101 38252 : let buf_size = plan.size();
1102 38252 : let buf = IoBufferMut::with_capacity(buf_size);
1103 38252 : let blobs_buf = vectored_blob_reader
1104 38252 : .read_blobs(&plan, buf, self.ctx)
1105 38252 : .await?;
1106 38252 : let view = BufView::new_slice(&blobs_buf.buf);
1107 57828 : for meta in blobs_buf.blobs.iter() {
1108 57828 : let img_buf = meta.read(&view).await?;
1109 57828 : next_batch.push_back((
1110 57828 : meta.meta.key,
1111 57828 : self.image_layer.lsn,
1112 57828 : Value::Image(img_buf.into_bytes()),
1113 57828 : ));
1114 : }
1115 38252 : self.key_values_batch = next_batch;
1116 38252 : Ok(())
1117 38252 : }
1118 :
1119 57608 : pub async fn next(&mut self) -> anyhow::Result<Option<(Key, Lsn, Value)>> {
1120 57608 : if self.key_values_batch.is_empty() {
1121 38404 : if self.is_end {
1122 320 : return Ok(None);
1123 38084 : }
1124 38084 : self.next_batch().await?;
1125 19204 : }
1126 57288 : Ok(Some(
1127 57288 : self.key_values_batch
1128 57288 : .pop_front()
1129 57288 : .expect("should not be empty"),
1130 57288 : ))
1131 57608 : }
1132 : }
1133 :
1134 : #[cfg(test)]
1135 : mod test {
1136 : use std::{sync::Arc, time::Duration};
1137 :
1138 : use bytes::Bytes;
1139 : use itertools::Itertools;
1140 : use pageserver_api::{
1141 : key::Key,
1142 : shard::{ShardCount, ShardIdentity, ShardNumber, ShardStripeSize},
1143 : value::Value,
1144 : };
1145 : use utils::{
1146 : generation::Generation,
1147 : id::{TenantId, TimelineId},
1148 : lsn::Lsn,
1149 : };
1150 :
1151 : use crate::{
1152 : context::RequestContext,
1153 : tenant::{
1154 : config::TenantConf,
1155 : harness::{TenantHarness, TIMELINE_ID},
1156 : storage_layer::{Layer, ResidentLayer},
1157 : vectored_blob_io::StreamingVectoredReadPlanner,
1158 : Tenant, Timeline,
1159 : },
1160 : DEFAULT_PG_VERSION,
1161 : };
1162 :
1163 : use super::{ImageLayerIterator, ImageLayerWriter};
1164 :
1165 : #[tokio::test]
1166 4 : async fn image_layer_rewrite() {
1167 4 : let tenant_conf = TenantConf {
1168 4 : gc_period: Duration::ZERO,
1169 4 : compaction_period: Duration::ZERO,
1170 4 : ..TenantConf::default()
1171 4 : };
1172 4 : let tenant_id = TenantId::generate();
1173 4 : let mut gen = Generation::new(0xdead0001);
1174 20 : let mut get_next_gen = || {
1175 20 : let ret = gen;
1176 20 : gen = gen.next();
1177 20 : ret
1178 20 : };
1179 4 : // The LSN at which we will create an image layer to filter
1180 4 : let lsn = Lsn(0xdeadbeef0000);
1181 4 : let timeline_id = TimelineId::generate();
1182 4 :
1183 4 : //
1184 4 : // Create an unsharded parent with a layer.
1185 4 : //
1186 4 :
1187 4 : let harness = TenantHarness::create_custom(
1188 4 : "test_image_layer_rewrite--parent",
1189 4 : tenant_conf.clone(),
1190 4 : tenant_id,
1191 4 : ShardIdentity::unsharded(),
1192 4 : get_next_gen(),
1193 4 : )
1194 4 : .await
1195 4 : .unwrap();
1196 4 : let (tenant, ctx) = harness.load().await;
1197 4 : let timeline = tenant
1198 4 : .create_test_timeline(timeline_id, lsn, DEFAULT_PG_VERSION, &ctx)
1199 4 : .await
1200 4 : .unwrap();
1201 4 :
1202 4 : // This key range contains several 0x8000 page stripes, only one of which belongs to shard zero
1203 4 : let input_start = Key::from_hex("000000067f00000001000000ae0000000000").unwrap();
1204 4 : let input_end = Key::from_hex("000000067f00000001000000ae0000020000").unwrap();
1205 4 : let range = input_start..input_end;
1206 4 :
1207 4 : // Build an image layer to filter
1208 4 : let resident = {
1209 4 : let mut writer = ImageLayerWriter::new(
1210 4 : harness.conf,
1211 4 : timeline_id,
1212 4 : harness.tenant_shard_id,
1213 4 : &range,
1214 4 : lsn,
1215 4 : &ctx,
1216 4 : )
1217 4 : .await
1218 4 : .unwrap();
1219 4 :
1220 4 : let foo_img = Bytes::from_static(&[1, 2, 3, 4]);
1221 4 : let mut key = range.start;
1222 524292 : while key < range.end {
1223 524288 : writer.put_image(key, foo_img.clone(), &ctx).await.unwrap();
1224 524288 :
1225 524288 : key = key.next();
1226 4 : }
1227 4 : let (desc, path) = writer.finish(&ctx).await.unwrap();
1228 4 : Layer::finish_creating(tenant.conf, &timeline, desc, &path).unwrap()
1229 4 : };
1230 4 : let original_size = resident.metadata().file_size;
1231 4 :
1232 4 : //
1233 4 : // Create child shards and do the rewrite, exercising filter().
1234 4 : // TODO: abstraction in TenantHarness for splits.
1235 4 : //
1236 4 :
1237 4 : // Filter for various shards: this exercises cases like values at start of key range, end of key
1238 4 : // range, middle of key range.
1239 4 : let shard_count = ShardCount::new(4);
1240 16 : for shard_number in 0..shard_count.count() {
1241 4 : //
1242 4 : // mimic the shard split
1243 4 : //
1244 16 : let shard_identity = ShardIdentity::new(
1245 16 : ShardNumber(shard_number),
1246 16 : shard_count,
1247 16 : ShardStripeSize(0x8000),
1248 16 : )
1249 16 : .unwrap();
1250 16 : let harness = TenantHarness::create_custom(
1251 16 : Box::leak(Box::new(format!(
1252 16 : "test_image_layer_rewrite--child{}",
1253 16 : shard_identity.shard_slug()
1254 16 : ))),
1255 16 : tenant_conf.clone(),
1256 16 : tenant_id,
1257 16 : shard_identity,
1258 16 : // NB: in reality, the shards would each fork off their own gen number sequence from the parent.
1259 16 : // But here, all we care about is that the gen number is unique.
1260 16 : get_next_gen(),
1261 16 : )
1262 16 : .await
1263 16 : .unwrap();
1264 16 : let (tenant, ctx) = harness.load().await;
1265 16 : let timeline = tenant
1266 16 : .create_test_timeline(timeline_id, lsn, DEFAULT_PG_VERSION, &ctx)
1267 16 : .await
1268 16 : .unwrap();
1269 4 :
1270 4 : //
1271 4 : // use filter() and make assertions
1272 4 : //
1273 4 :
1274 16 : let mut filtered_writer = ImageLayerWriter::new(
1275 16 : harness.conf,
1276 16 : timeline_id,
1277 16 : harness.tenant_shard_id,
1278 16 : &range,
1279 16 : lsn,
1280 16 : &ctx,
1281 16 : )
1282 16 : .await
1283 16 : .unwrap();
1284 4 :
1285 16 : let wrote_keys = resident
1286 16 : .filter(&shard_identity, &mut filtered_writer, &ctx)
1287 16 : .await
1288 16 : .unwrap();
1289 16 : let replacement = if wrote_keys > 0 {
1290 12 : let (desc, path) = filtered_writer.finish(&ctx).await.unwrap();
1291 12 : let resident = Layer::finish_creating(tenant.conf, &timeline, desc, &path).unwrap();
1292 12 : Some(resident)
1293 4 : } else {
1294 4 : None
1295 4 : };
1296 4 :
1297 4 : // This exact size and those below will need updating as/when the layer encoding changes, but
1298 4 : // should be deterministic for a given version of the format, as we used no randomness generating the input.
1299 16 : assert_eq!(original_size, 1597440);
1300 4 :
1301 16 : match shard_number {
1302 4 : 0 => {
1303 4 : // We should have written out just one stripe for our shard identity
1304 4 : assert_eq!(wrote_keys, 0x8000);
1305 4 : let replacement = replacement.unwrap();
1306 4 :
1307 4 : // We should have dropped some of the data
1308 4 : assert!(replacement.metadata().file_size < original_size);
1309 4 : assert!(replacement.metadata().file_size > 0);
1310 4 :
1311 4 : // Assert that we dropped ~3/4 of the data.
1312 4 : assert_eq!(replacement.metadata().file_size, 417792);
1313 4 : }
1314 4 : 1 => {
1315 4 : // Shard 1 has no keys in our input range
1316 4 : assert_eq!(wrote_keys, 0x0);
1317 4 : assert!(replacement.is_none());
1318 4 : }
1319 4 : 2 => {
1320 4 : // Shard 2 has one stripes in the input range
1321 4 : assert_eq!(wrote_keys, 0x8000);
1322 4 : let replacement = replacement.unwrap();
1323 4 : assert!(replacement.metadata().file_size < original_size);
1324 4 : assert!(replacement.metadata().file_size > 0);
1325 4 : assert_eq!(replacement.metadata().file_size, 417792);
1326 4 : }
1327 4 : 3 => {
1328 4 : // Shard 3 has two stripes in the input range
1329 4 : assert_eq!(wrote_keys, 0x10000);
1330 4 : let replacement = replacement.unwrap();
1331 4 : assert!(replacement.metadata().file_size < original_size);
1332 4 : assert!(replacement.metadata().file_size > 0);
1333 4 : assert_eq!(replacement.metadata().file_size, 811008);
1334 4 : }
1335 4 : _ => unreachable!(),
1336 4 : }
1337 4 : }
1338 4 : }
1339 :
1340 4 : async fn produce_image_layer(
1341 4 : tenant: &Tenant,
1342 4 : tline: &Arc<Timeline>,
1343 4 : mut images: Vec<(Key, Bytes)>,
1344 4 : lsn: Lsn,
1345 4 : ctx: &RequestContext,
1346 4 : ) -> anyhow::Result<ResidentLayer> {
1347 4 : images.sort();
1348 4 : let (key_start, _) = images.first().unwrap();
1349 4 : let (key_last, _) = images.last().unwrap();
1350 4 : let key_end = key_last.next();
1351 4 : let key_range = *key_start..key_end;
1352 4 : let mut writer = ImageLayerWriter::new(
1353 4 : tenant.conf,
1354 4 : tline.timeline_id,
1355 4 : tenant.tenant_shard_id,
1356 4 : &key_range,
1357 4 : lsn,
1358 4 : ctx,
1359 4 : )
1360 4 : .await?;
1361 :
1362 4004 : for (key, img) in images {
1363 4000 : writer.put_image(key, img, ctx).await?;
1364 : }
1365 4 : let (desc, path) = writer.finish(ctx).await?;
1366 4 : let img_layer = Layer::finish_creating(tenant.conf, tline, desc, &path)?;
1367 :
1368 4 : Ok::<_, anyhow::Error>(img_layer)
1369 4 : }
1370 :
1371 56 : async fn assert_img_iter_equal(
1372 56 : img_iter: &mut ImageLayerIterator<'_>,
1373 56 : expect: &[(Key, Bytes)],
1374 56 : expect_lsn: Lsn,
1375 56 : ) {
1376 56 : let mut expect_iter = expect.iter();
1377 : loop {
1378 56056 : let o1 = img_iter.next().await.unwrap();
1379 56056 : let o2 = expect_iter.next();
1380 56056 : match (o1, o2) {
1381 56 : (None, None) => break,
1382 56000 : (Some((k1, l1, v1)), Some((k2, i2))) => {
1383 56000 : let Value::Image(i1) = v1 else {
1384 0 : panic!("expect Value::Image")
1385 : };
1386 56000 : assert_eq!(&k1, k2);
1387 56000 : assert_eq!(l1, expect_lsn);
1388 56000 : assert_eq!(&i1, i2);
1389 : }
1390 0 : (o1, o2) => panic!("iterators length mismatch: {:?}, {:?}", o1, o2),
1391 : }
1392 : }
1393 56 : }
1394 :
1395 : #[tokio::test]
1396 4 : async fn image_layer_iterator() {
1397 4 : let harness = TenantHarness::create("image_layer_iterator").await.unwrap();
1398 4 : let (tenant, ctx) = harness.load().await;
1399 4 :
1400 4 : let tline = tenant
1401 4 : .create_test_timeline(TIMELINE_ID, Lsn(0x10), DEFAULT_PG_VERSION, &ctx)
1402 4 : .await
1403 4 : .unwrap();
1404 4 :
1405 4000 : fn get_key(id: u32) -> Key {
1406 4000 : let mut key = Key::from_hex("000000000033333333444444445500000000").unwrap();
1407 4000 : key.field6 = id;
1408 4000 : key
1409 4000 : }
1410 4 : const N: usize = 1000;
1411 4 : let test_imgs = (0..N)
1412 4000 : .map(|idx| (get_key(idx as u32), Bytes::from(format!("img{idx:05}"))))
1413 4 : .collect_vec();
1414 4 : let resident_layer =
1415 4 : produce_image_layer(&tenant, &tline, test_imgs.clone(), Lsn(0x10), &ctx)
1416 4 : .await
1417 4 : .unwrap();
1418 4 : let img_layer = resident_layer.get_as_image(&ctx).await.unwrap();
1419 12 : for max_read_size in [1, 1024] {
1420 64 : for batch_size in [1, 2, 4, 8, 3, 7, 13] {
1421 56 : println!("running with batch_size={batch_size} max_read_size={max_read_size}");
1422 56 : // Test if the batch size is correctly determined
1423 56 : let mut iter = img_layer.iter(&ctx);
1424 56 : iter.planner = StreamingVectoredReadPlanner::new(max_read_size, batch_size);
1425 56 : let mut num_items = 0;
1426 224 : for _ in 0..3 {
1427 168 : iter.next_batch().await.unwrap();
1428 168 : num_items += iter.key_values_batch.len();
1429 168 : if max_read_size == 1 {
1430 4 : // every key should be a batch b/c the value is larger than max_read_size
1431 84 : assert_eq!(iter.key_values_batch.len(), 1);
1432 4 : } else {
1433 84 : assert!(iter.key_values_batch.len() <= batch_size);
1434 4 : }
1435 168 : if num_items >= N {
1436 4 : break;
1437 168 : }
1438 168 : iter.key_values_batch.clear();
1439 4 : }
1440 4 : // Test if the result is correct
1441 56 : let mut iter = img_layer.iter(&ctx);
1442 56 : iter.planner = StreamingVectoredReadPlanner::new(max_read_size, batch_size);
1443 56 : assert_img_iter_equal(&mut iter, &test_imgs, Lsn(0x10)).await;
1444 4 : }
1445 4 : }
1446 4 : }
1447 : }
|
