## Problem
The vectored read path proposed in
https://github.com/neondatabase/neon/pull/6576 seems
to be functionally correct, but in my testing (see below) it is about 10-20% slower than the naive
sequential vectored implementation.
## Summary of changes
There's three parts to this PR:
1. Supporting vectored blob reads. This is actually trickier than it
sounds because on disk blobs are prefixed with a variable length size header.
Since the blobs are not necessarily fixed size, we need to juggle the offsets
such that the callers can retrieve the blobs from the resulting buffer.
2. Merge disk read requests issued by the vectored read path up to a
maximum size. Again, the merging is complicated by the fact that blobs
are not fixed size. We keep track of the begin and end offset of each blob
and pass them into the vectored blob reader. In turn, the reader will return
a buffer and the offsets at which the blobs begin and end.
3. A benchmark for basebackup requests against tenant with large SLRU
block counts is added. This required a small change to pagebench and a new config
variable for the pageserver which toggles the vectored get validation.
We can probably optimise things further by adding a little bit of
concurrency for our IO. In principle, it's as simple as spawning a task which deals with issuing
IO and doing the serialisation and handling on the parent task which receives input via a
channel.
Problem
=======
Prior to this PR, when we had a cache miss, we'd get back a write guard,
fill it, the drop it and retry the read from cache.
If there's severe contention for the cache, it could happen that the
just-filled data gets evicted before our retry, resulting in lost work
and no forward progress.
Solution
========
This PR leverages the now-available `tokio::sync::RwLockWriteGuard`'s
`downgrade()` functionality to turn the filled slot write guard into a
read guard.
We don't drop the guard at any point, so, forward progress is ensured.
Refs
====
Stacked atop https://github.com/neondatabase/neon/pull/5480
part of https://github.com/neondatabase/neon/issues/4743
specifically part of https://github.com/neondatabase/neon/issues/5479
This PR adds a `task_kind` label to page cache access metrics.
These are to validate our hypothesis that the high hit page cache rate
we observe in prod is due to internal tasks, not getpage requests from
compute.
We believe the latter should near-always be a pageserver-page-cache
_miss_ because compute has it's own page cache, and hence there is no
locality of reference for its accesses to pageserver page cache.
Before this PR, we didn't have `RequestContext` propagation to any code
below the on-demand downloader.
The vast majority of changes in this PR is concerned with adding that
propagation.
## Problem
We want to convert the `VirtualFile` APIs to async fn so that we can
adopt one of the async I/O solutions.
## Summary of changes
This PR is a follow-up of #5189, #5190, and #5195, and does the
following:
* Move the used `Write` trait functions of `VirtualFile` into inherent
functions
* Add optional buffering to `WriteBlobWriter`. The buffer is discarded
on drop, similarly to how tokio's `BufWriter` does it: drop is neither
async nor does it support errors.
* Remove the generics by `Write` impl of `WriteBlobWriter`, alwaays
using `VirtualFile`
* Rename `WriteBlobWriter` to `BlobWriter`
* Make various functions in the write path async, like
`VirtualFile::{write,write_all}`.
Part of #4743.
## Problem
`VirtualFile` does both reading and writing, and it would be nice if
both could be converted to async, so that it doesn't have to support an
async read path and a blocking write path (especially for the locks this
is annoying as none of the lock implementations in std, tokio or
parking_lot have support for both async and blocking access).
## Summary of changes
This PR is some initial work on making the `VirtualFile` APIs async. It
can be reviewed commit-by-commit.
* Introduce the `MaybeVirtualFile` enum to be generic in a test that
compares real files with virtual files.
* Make various APIs of `VirtualFile` async, including `write_all_at`,
`read_at`, `read_exact_at`.
Part of #4743 , successor of #5180.
Co-authored-by: Christian Schwarz <me@cschwarz.com>
## Problem
`read_blk` does I/O and thus we would like to make it async. We can't
make the function async as long as the `PageReadGuard` returned by
`read_blk` isn't `Send`. The page cache is called by `read_blk`, and
thus it can't be async without `read_blk` being async. Thus, we have a
circular dependency.
## Summary of changes
Due to the circular dependency, we convert both the page cache and
`read_blk` to async at the same time:
We make the page cache use `tokio::sync` synchronization primitives as
those are `Send`. This makes all the places that acquire a lock require
async though, which we then also do. This includes also asyncification
of the `read_blk` function.
Builds upon #4994, #5015, #5056, and #5129.
Part of #4743.
## Problem
We want to make `read_blk` an async function, but outside of
`async_trait`, which allocates, and nightly features, we can't use async
fn's in traits.
## Summary of changes
* Remove all uses of `BlockReader::read_blk` in favour of using block
cursors, at least where the type of the `BlockReader` is behind a
generic
* Introduce a `BlockReaderRef` enum that lists all implementors of
`BlockReader::read_blk`.
* Remove `BlockReader::read_blk` and move its implementations into
inherent functions on the types instead.
We don't turn `read_blk` into an async fn yet, for that we also need to
modify the page cache. So this is a preparatory PR, albeit an important
one.
Part of #4743.
(This PR is the successor of https://github.com/neondatabase/neon/pull/4984 )
## Summary
The current way in which `EphemeralFile` uses `PageCache` complicates
the Pageserver code base to a degree that isn't worth it.
This PR refactors how we cache `EphemeralFile` contents, by exploiting
the append-only nature of `EphemeralFile`.
The result is that `PageCache` only holds `ImmutableFilePage` and
`MaterializedPage`.
These types of pages are read-only and evictable without write-back.
This allows us to remove the writeback code from `PageCache`, also
eliminating an entire failure mode.
Futher, many great open-source libraries exist to solve the problem of a
read-only cache,
much better than our `page_cache.rs` (e.g., better replacement policy,
less global locking).
With this PR, we can now explore using them.
## Problem & Analysis
Before this PR, `PageCache` had three types of pages:
* `ImmutableFilePage`: caches Delta / Image layer file contents
* `MaterializedPage`: caches results of Timeline::get (page
materialization)
* `EphemeralPage`: caches `EphemeralFile` contents
`EphemeralPage` is quite different from `ImmutableFilePage` and
`MaterializedPage`:
* Immutable and materialized pages are for the acceleration of (future)
reads of the same data using `PAGE_CACHE_SIZE * PAGE_SIZE` bytes of
DRAM.
* Ephemeral pages are a write-back cache of `EphemeralFile` contents,
i.e., if there is pressure in the page cache, we spill `EphemeralFile`
contents to disk.
`EphemeralFile` is only used by `InMemoryLayer`, for the following
purposes:
* **write**: when filling up the `InMemoryLayer`, via `impl BlobWriter
for EphemeralFile`
* **read**: when doing **page reconstruction** for a page@lsn that isn't
written to disk
* **read**: when writing L0 layer files, we re-read the `InMemoryLayer`
and put the contents into the L0 delta writer
(**`create_delta_layer`**). This happens every 10min or when
InMemoryLayer reaches 256MB in size.
The access patterns of the `InMemoryLayer` use case are as follows:
* **write**: via `BlobWriter`, strictly append-only
* **read for page reconstruction**: via `BlobReader`, random
* **read for `create_delta_layer`**: via `BlobReader`, dependent on
data, but generally random. Why?
* in classical LSM terms, this function is what writes the
memory-resident `C0` tree into the disk-resident `C1` tree
* in our system, though, the values of InMemoryLayer are stored in an
EphemeralFile, and hence they are not guaranteed to be memory-resident
* the function reads `Value`s in `Key, LSN` order, which is `!=` insert
order
What do these `EphemeralFile`-level access patterns mean for the page
cache?
* **write**:
* the common case is that `Value` is a WAL record, and if it isn't a
full-page-image WAL record, then it's smaller than `PAGE_SIZE`
* So, the `EphemeralPage` pages act as a buffer for these `< PAGE_CACHE`
sized writes.
* If there's no page cache eviction between subsequent
`InMemoryLayer::put_value` calls, the `EphemeralPage` is still resident,
so the page cache avoids doing a `write` system call.
* In practice, a busy page server will have page cache evictions because
we only configure 64MB of page cache size.
* **reads for page reconstruction**: read acceleration, just as for the
other page types.
* **reads for `create_delta_layer`**:
* The `Value` reads happen through a `BlockCursor`, which optimizes the
case of repeated reads from the same page.
* So, the best case is that subsequent values are located on the same
page; hence `BlockCursor`s buffer is maximally effective.
* The worst case is that each `Value` is on a different page; hence the
`BlockCursor`'s 1-page-sized buffer is ineffective.
* The best case translates into `256MB/PAGE_SIZE` page cache accesses,
one per page.
* the worst case translates into `#Values` page cache accesses
* again, the page cache accesses must be assumed to be random because
the `Value`s aren't accessed in insertion order but `Key, LSN` order.
## Summary of changes
Preliminaries for this PR were:
- #5003
- #5004
- #5005
- uncommitted microbenchmark in #5011
Based on the observations outlined above, this PR makes the following
changes:
* Rip out `EphemeralPage` from `page_cache.rs`
* Move the `block_io::FileId` to `page_cache::FileId`
* Add a `PAGE_SIZE`d buffer to the `EphemeralPage` struct.
It's called `mutable_tail`.
* Change `write_blob` to use `mutable_tail` for the write buffering
instead of a page cache page.
* if `mutable_tail` is full, it writes it out to disk, zeroes it out,
and re-uses it.
* There is explicitly no double-buffering, so that memory allocation per
`EphemeralFile` instance is fixed.
* Change `read_blob` to return different `BlockLease` variants depending
on `blknum`
* for the `blknum` that corresponds to the `mutable_tail`, return a ref
to it
* Rust borrowing rules prevent `write_blob` calls while refs are
outstanding.
* for all non-tail blocks, return a page-cached `ImmutablePage`
* It is safe to page-cache these as ImmutablePage because EphemeralFile
is append-only.
## Performance
How doe the changes above affect performance?
M claim is: not significantly.
* **write path**:
* before this PR, the `EphemeralFile::write_blob` didn't issue its own
`write` system calls.
* If there were enough free pages, it didn't issue *any* `write` system
calls.
* If it had to evict other `EphemeralPage`s to get pages a page for its
writes (`get_buf_for_write`), the page cache code would implicitly issue
the writeback of victim pages as needed.
* With this PR, `EphemeralFile::write_blob` *always* issues *all* of its
*own* `write` system calls.
* Also, the writes are explicit instead of implicit through page cache
write back, which will help #4743
* The perf impact of always doing the writes is the CPU overhead and
syscall latency.
* Before this PR, we might have never issued them if there were enough
free pages.
* We don't issue `fsync` and can expect the writes to only hit the
kernel page cache.
* There is also an advantage in issuing the writes directly: the perf
impact is paid by the tenant that caused the writes, instead of whatever
tenant evicts the `EphemeralPage`.
* **reads for page reconstruction**: no impact.
* The `write_blob` function pre-warms the page cache when it writes the
`mutable_tail` to disk.
* So, the behavior is the same as with the EphemeralPages before this
PR.
* **reads for `create_delta_layer`**: no impact.
* Same argument as for page reconstruction.
* Note for the future:
* going through the page cache likely causes read amplification here.
Why?
* Due to the `Key,Lsn`-ordered access pattern, we don't read all the
values in the page before moving to the next page. In the worst case, we
might read the same page multiple times to read different `Values` from
it.
* So, it might be better to bypass the page cache here.
* Idea drafts:
* bypass PS page cache + prefetch pipeline + iovec-based IO
* bypass PS page cache + use `copy_file_range` to copy from ephemeral
file into the L0 delta file, without going through user space
This makes it more explicit that these are different u64-sized
namespaces.
Re-using one in place of the other would be catastrophic.
Prep for https://github.com/neondatabase/neon/pull/4994
which will eliminate the ephemeral_file::FileId and move the
blob_io::FileId into page_cache.
It makes sense to have this preliminary commit though,
to minimize amount of new concept in #4994 and other
preliminaries that depend on that work.
## Problem
The `BlockReader` trait is not ready to be asyncified, as associated
types are not supported by asyncification strategies like via the
`async_trait` macro, or via adopting enums.
## Summary of changes
Remove the `BlockLease` associated type from the `BlockReader` trait and
turn it into an enum instead, bearing the same name. The enum has two
variants, one of which is gated by `#[cfg(test)]`. Therefore, outside of
test settings, the enum has zero overhead over just having the
`PageReadGuard`. Using the enum allows us to impl `BlockReader` without
needing the page cache.
Part of https://github.com/neondatabase/neon/issues/4743
## Problem
The k-merge in pageserver compaction currently relies on iterators over
the keys and also over the values. This approach does not support async
code because we are using iterators and those don't support async in
general. Also, the k-merge implementation we use doesn't support async
either. Instead, as we already load all the keys into memory, the plan
is to just do the sorting in-memory for now, switch to async, and then
once we want to support workloads that don't have all keys stored in
memory, we can look into switching to a k-merge implementation that
supports async instead.
## Summary of changes
The core of this PR is the move from functions on the `PersistentLayer`
trait to return custom iterator types to inherent functions on `DeltaLayer`
that return buffers with all keys or value references.
Value references are a type we created in this PR, containing a
`BlobRef` as well as an `Arc` pointer to the `DeltaLayerInner`, so that
we can lazily load the values during compaction. This preserves the
property of the current code.
This PR does not switch us to doing the k-merge via sort on slices, but
with this PR, doing such a switch is relatively easy and only requires
changes of the compaction code itself.
Part of https://github.com/neondatabase/neon/issues/4743
## Describe your changes
Do not pin current block in BlockCursor
## Issue ticket number and link
See #3712
There are places (see get_reconstruct_data) in our code when thread is
holding read layers lock and then try to read file and so lock page
cache slot. So we have edge in dependency graph layers->page cache slot.
At the same time (as Christian noticed) we can lock page cache slot in
BlockCursor and then try obtain shard lock on layers.
So there is backward edge in dependency graph page cache slot>layers
which forms loop and may cause deadlock.
There are three possible fixes of the problem:
1. Perform compaction under `layers` shared lock. See PR #3732. It fixes
the problem but make it not possible to append any data to pageserver
until compaction is completed.
2. Do not hold `layers` lock while accessing layers (not sure if it is
possible to do because it definitely introduce some new race
conditions).
3. Do not pin current pages in BockCursor (this PR).
My experiments shows that this cache in BlockCursor is not so useful:
the number of hits/misses for cursor cache on pgbench workload (-i -s
10/-c 10 -T 100/-c 10 -S -T 100):
```
hits: 163011
misses: 1023602
```
So number of cache misses is 10x times larger.
And results for read-only pgbench are mostly the same:
```
with cache: 14581
w/out cache: 14429
```
## Checklist before requesting a review
- [ ] I have performed a self-review of my code.
- [ ] If it is a core feature, I have added thorough tests.
- [ ] Do we need to implement analytics? if so did you add the relevant
metrics to the dashboard?
- [ ] If this PR requires public announcement, mark it with
/release-notes label and add several sentences in this section.
refactor: use new type LayerFileName when referring to layer file names in PathBuf/RemotePath
Before this patch, we would sometimes carry around plain file names in
`Path` types and/or awkwardly "rebase" paths to have a unified
representation of the layer file name between local and remote.
This patch introduces a new type `LayerFileName` which replaces the use
of `Path` / `PathBuf` / `RemotePath` in the `storage_sync2` APIs.
Instead of holding a string, it contains the parsed representation of
the image and delta file name.
When we need the file name, e.g., to construct a local path or
remote object key, we construct the name ad-hoc.
`LayerFileName` is also serde {Dese,Se}rializable, and in an initial
version of this patch, it was supposed to be used directly inside
`IndexPart`, replacing `RemotePath`.
However,
commit 3122f3282f
Ignore backup files (ones with .n.old suffix) in download_missing
fixed handling of `*.old` backup file names in IndexPart, and we need
to carry that behavior forward.
The solution is to remove `*.old` backup files names during
deserialization. When we re-serialize the IndexPart, the `*.old` file
will be gone.
This leaks the `.old` file in the remote storage, but makes it safe
to clean it up later.
There is additional churn by a preliminary refactoring that got squashed
into this change:
split off LayerMap's needs from trait Layer into super trait
That refactoring renames `Layer` to `PersistentLayer` and splits off a subset
of the functions into a super-trait called `Layer`.
The upser trait implements just the functions needed by `LayerMap`, whereas
`PersisentLayer` adds the context of the pageserver.
The naming is imperfect as some functions that reside in `PersistentLayer`
have nothing persistence-specific to it. But it's a step in the right direction.
Vlad Lazar
Christian Schwarz
Arpad Müller
Konstantin Knizhnik
Kirill Bulatov