Files
neon/pageserver/src/tenant/storage_layer.rs
Heikki Linnakangas 7ff591ffbf On-Demand Download
The code in this change was extracted from #2595 (Heikki’s on-demand
download draft PR).

High-Level Changes

- New RemoteLayer Type
- On-Demand Download As An Effect Of Page Reconstruction
- Breaking Semantics For Physical Size Metrics

There are several follow-up work items planned.
Refer to the Epic issue on GitHub: https://github.com/neondatabase/neon/issues/2029

closes https://github.com/neondatabase/neon/pull/3013

Co-authored-by: Kirill Bulatov <kirill@neon.tech>
Co-authored-by: Christian Schwarz <christian@neon.tech>

New RemoteLayer Type
====================

Instead of downloading all layers during tenant attach, we create
RemoteLayer instances for each of them and add them to the layer map.

On-Demand Download As An Effect Of Page Reconstruction
======================================================

At the heart of pageserver is Timeline::get_reconstruct_data(). It
traverses the layer map until it has collected all the data it needs to
produce the page image. Most code in the code base uses it, though many
layers of indirection.

Before this patch, the function would use synchronous filesystem IO to
load data from disk-resident layer files if the data was not cached.

That is not possible with RemoteLayer, because the layer file has not
been downloaded yet. So, we do the download when get_reconstruct_data
gets there, i.e., “on demand”.

The mechanics of how the download is done are rather involved, because
of the infamous async-sync-async sandwich problem that plagues the async
Rust world. We use the new PageReconstructResult type to work around
this. Its introduction is the cause for a good amount of code churn in
this patch. Refer to the block comment on `with_ondemand_download()`
for details.

Breaking Semantics For Physical Size Metrics
============================================

We rename prometheus metric pageserver_{current,resident}_physical_size to
reflect what this metric actually represents with on-demand download.
This intentionally BREAKS existing grafana dashboard and the cost model data
pipeline. Breaking is desirable because the meaning of this metrics has changed
with on-demand download. See
 https://docs.google.com/document/d/12AFpvKY-7FZdR5a4CaD6Ir_rI3QokdCLSPJ6upHxJBo/edit#
for how we will handle this breakage.

Likewise, we rename the new billing_metrics’s PhysicalSize => ResidentSize.
This is not yet used anywhere, so, this is not a breaking change.

There is still a field called TimelineInfo::current_physical_size. It
is now the sum of the layer sizes in layer map, regardless of whether
local or remote. To compute that sum, we added a new trait method
PersistentLayer::file_size().

When updating the Python tests, we got rid of
current_physical_size_non_incremental. An earlier commit removed it from
the OpenAPI spec already, so this is not a breaking change.

test_timeline_size.py has grown additional assertions on the
resident_physical_size metric.
2022-12-21 19:16:39 +01:00

192 lines
6.7 KiB
Rust

//!
//! Common traits and structs for layers
//!
use crate::repository::{Key, Value};
use crate::walrecord::NeonWalRecord;
use anyhow::Result;
use bytes::Bytes;
use std::ops::Range;
use std::path::PathBuf;
use std::sync::Arc;
use utils::{
id::{TenantId, TimelineId},
lsn::Lsn,
};
use super::filename::LayerFileName;
use super::remote_layer::RemoteLayer;
pub fn range_overlaps<T>(a: &Range<T>, b: &Range<T>) -> bool
where
T: PartialOrd<T>,
{
if a.start < b.start {
a.end > b.start
} else {
b.end > a.start
}
}
pub fn range_eq<T>(a: &Range<T>, b: &Range<T>) -> bool
where
T: PartialEq<T>,
{
a.start == b.start && a.end == b.end
}
/// Struct used to communicate across calls to 'get_value_reconstruct_data'.
///
/// Before first call, you can fill in 'page_img' if you have an older cached
/// version of the page available. That can save work in
/// 'get_value_reconstruct_data', as it can stop searching for page versions
/// when all the WAL records going back to the cached image have been collected.
///
/// When get_value_reconstruct_data returns Complete, 'img' is set to an image
/// of the page, or the oldest WAL record in 'records' is a will_init-type
/// record that initializes the page without requiring a previous image.
///
/// If 'get_page_reconstruct_data' returns Continue, some 'records' may have
/// been collected, but there are more records outside the current layer. Pass
/// the same ValueReconstructState struct in the next 'get_value_reconstruct_data'
/// call, to collect more records.
///
#[derive(Debug)]
pub struct ValueReconstructState {
pub records: Vec<(Lsn, NeonWalRecord)>,
pub img: Option<(Lsn, Bytes)>,
}
/// Return value from Layer::get_page_reconstruct_data
#[derive(Clone, Copy, Debug)]
pub enum ValueReconstructResult {
/// Got all the data needed to reconstruct the requested page
Complete,
/// This layer didn't contain all the required data, the caller should look up
/// the predecessor layer at the returned LSN and collect more data from there.
Continue,
/// This layer didn't contain data needed to reconstruct the page version at
/// the returned LSN. This is usually considered an error, but might be OK
/// in some circumstances.
Missing,
}
/// Supertrait of the [`Layer`] trait that captures the bare minimum interface
/// required by [`LayerMap`].
pub trait Layer: Send + Sync {
/// Range of keys that this layer covers
fn get_key_range(&self) -> Range<Key>;
/// Inclusive start bound of the LSN range that this layer holds
/// Exclusive end bound of the LSN range that this layer holds.
///
/// - For an open in-memory layer, this is MAX_LSN.
/// - For a frozen in-memory layer or a delta layer, this is a valid end bound.
/// - An image layer represents snapshot at one LSN, so end_lsn is always the snapshot LSN + 1
fn get_lsn_range(&self) -> Range<Lsn>;
/// Does this layer only contain some data for the key-range (incremental),
/// or does it contain a version of every page? This is important to know
/// for garbage collecting old layers: an incremental layer depends on
/// the previous non-incremental layer.
fn is_incremental(&self) -> bool;
///
/// Return data needed to reconstruct given page at LSN.
///
/// It is up to the caller to collect more data from previous layer and
/// perform WAL redo, if necessary.
///
/// See PageReconstructResult for possible return values. The collected data
/// is appended to reconstruct_data; the caller should pass an empty struct
/// on first call, or a struct with a cached older image of the page if one
/// is available. If this returns PageReconstructResult::Continue, look up
/// the predecessor layer and call again with the same 'reconstruct_data' to
/// collect more data.
fn get_value_reconstruct_data(
&self,
key: Key,
lsn_range: Range<Lsn>,
reconstruct_data: &mut ValueReconstructState,
) -> Result<ValueReconstructResult>;
/// A short ID string that uniquely identifies the given layer within a [`LayerMap`].
fn short_id(&self) -> String;
/// Dump summary of the contents of the layer to stdout
fn dump(&self, verbose: bool) -> Result<()>;
}
/// Returned by [`Layer::iter`]
pub type LayerIter<'i> = Box<dyn Iterator<Item = Result<(Key, Lsn, Value)>> + 'i>;
/// Returned by [`Layer::key_iter`]
pub type LayerKeyIter<'i> = Box<dyn Iterator<Item = (Key, Lsn, u64)> + 'i>;
/// A Layer contains all data in a "rectangle" consisting of a range of keys and
/// range of LSNs.
///
/// There are two kinds of layers, in-memory and on-disk layers. In-memory
/// layers are used to ingest incoming WAL, and provide fast access to the
/// recent page versions. On-disk layers are stored as files on disk, and are
/// immutable. This trait presents the common functionality of in-memory and
/// on-disk layers.
///
/// Furthermore, there are two kinds of on-disk layers: delta and image layers.
/// A delta layer contains all modifications within a range of LSNs and keys.
/// An image layer is a snapshot of all the data in a key-range, at a single
/// LSN
///
pub trait PersistentLayer: Layer {
fn get_tenant_id(&self) -> TenantId;
/// Identify the timeline this layer belongs to
fn get_timeline_id(&self) -> TimelineId;
/// File name used for this layer, both in the pageserver's local filesystem
/// state as well as in the remote storage.
fn filename(&self) -> LayerFileName;
// Path to the layer file in the local filesystem.
// `None` for `RemoteLayer`.
fn local_path(&self) -> Option<PathBuf>;
/// Iterate through all keys and values stored in the layer
fn iter(&self) -> Result<LayerIter<'_>>;
/// Iterate through all keys stored in the layer. Returns key, lsn and value size
/// It is used only for compaction and so is currently implemented only for DeltaLayer
fn key_iter(&self) -> Result<LayerKeyIter<'_>> {
panic!("Not implemented")
}
/// Permanently remove this layer from disk.
fn delete(&self) -> Result<()>;
fn downcast_remote_layer(self: Arc<Self>) -> Option<std::sync::Arc<RemoteLayer>> {
None
}
fn is_remote_layer(&self) -> bool {
false
}
/// Returns None if the layer file size is not known.
///
/// Should not change over the lifetime of the layer object because
/// current_physical_size is computed as the som of this value.
fn file_size(&self) -> Option<u64>;
}
pub fn downcast_remote_layer(
layer: &Arc<dyn PersistentLayer>,
) -> Option<std::sync::Arc<RemoteLayer>> {
if layer.is_remote_layer() {
Arc::clone(layer).downcast_remote_layer()
} else {
None
}
}