mbutrovich commented on code in PR #3224:
URL: https://github.com/apache/iceberg-rust/pull/3224#discussion_r4235390475
##########
crates/iceberg/src/spec/manifest_list/mod.rs:
##########
@@ -58,22 +57,39 @@ pub struct ManifestList {
impl ManifestList {
/// Parse manifest list from bytes.
+ /// Convert each Avro value immediately, retaining only the final typed
entries.
+ /// A decoding or conversion error still discards the entire result.
pub fn parse_with_version(bs: &[u8], version: FormatVersion) ->
Result<ManifestList> {
match version {
FormatVersion::V1 => {
let reader =
Reader::with_schema(&MANIFEST_LIST_AVRO_SCHEMA_V1, bs)?;
- let values =
Value::Array(reader.collect::<std::result::Result<Vec<Value>, _>>()?);
- from_value::<_serde::ManifestListV1>(&values)?.try_into()
+ let entries = reader
+ .map(|value| {
+ let value = value?;
+
from_value::<_serde::ManifestFileV1>(&value)?.try_into()
+ })
+ .collect::<Result<Vec<_>>>()?;
+ Ok(ManifestList { entries })
}
FormatVersion::V2 => {
let reader = Reader::new(bs)?;
- let values =
Value::Array(reader.collect::<std::result::Result<Vec<Value>, _>>()?);
- from_value::<_serde::ManifestListV2>(&values)?.try_into()
+ let entries = reader
+ .map(|value| {
+ let value = value?;
+
from_value::<_serde::ManifestFileV2>(&value)?.try_into()
+ })
+ .collect::<Result<Vec<_>>>()?;
+ Ok(ManifestList { entries })
}
FormatVersion::V3 => {
let reader = Reader::new(bs)?;
- let values =
Value::Array(reader.collect::<std::result::Result<Vec<Value>, _>>()?);
- from_value::<_serde::ManifestListV3>(&values)?.try_into()
+ let entries = reader
+ .map(|value| {
+ let value = value?;
+
from_value::<_serde::ManifestFileV3>(&value)?.try_into()
+ })
+ .collect::<Result<Vec<_>>>()?;
+ Ok(ManifestList { entries })
}
}
}
Review Comment:
#3354 merged today. It upgraded to `apache-avro` 0.22 and reads manifest
entries with `Reader::into_deser_iter`, which decodes straight into the serde
types without building an `apache_avro::types::Value`. The
[`Resolved`](https://github.com/apache/iceberg-rust/blob/e8ec7f8403554496db976700eb993459e0ab7f9b/crates/iceberg/src/avro/deserializer.rs#L56-L64)
adapter it added handles writer record names, unions, and numeric widening
against the writer schema.
What do you think about rebasing and reading manifest lists the same way as
the [manifest entry
reader](https://github.com/apache/iceberg-rust/blob/e8ec7f8403554496db976700eb993459e0ab7f9b/crates/iceberg/src/spec/manifest/mod.rs#L114-L136)?
Something like:
```rust
/// Parse manifest list from bytes.
pub fn parse_with_version(bs: &[u8], version: FormatVersion) ->
Result<ManifestList> {
let reader = Reader::new(bs)?;
let entries = match version {
FormatVersion::V1 => reader
.into_deser_iter::<Resolved<_serde::ManifestFileV1>>()
.map(|entry| entry?.0.try_into())
.collect::<Result<Vec<_>>>()?,
FormatVersion::V2 => reader
.into_deser_iter::<Resolved<_serde::ManifestFileV2>>()
.map(|entry| entry?.0.try_into())
.collect::<Result<Vec<_>>>()?,
FormatVersion::V3 => reader
.into_deser_iter::<Resolved<_serde::ManifestFileV3>>()
.map(|entry| entry?.0.try_into())
.collect::<Result<Vec<_>>>()?,
};
Ok(ManifestList { entries })
}
```
This removes the per-record `Value` as well as the whole-list one. With it,
your `manifest_list_memory` test reports a peak of 4,355,177 bytes for the
uncompressed list on my machine, against 6,040,920 at the head commit. The
field aliases and V1 defaults on `ManifestFileV2` and `ManifestFileV3` are
serde attributes, so they carry over. V1 would read against the writer schema
instead of `MANIFEST_LIST_AVRO_SCHEMA_V1`, as the manifest entry reader already
does for V1 manifests. The two doc lines about converting each Avro value would
go too.
This is also the "Read manifest lists without `Value`" task in #3262.
##########
crates/iceberg/tests/manifest_list_memory.rs:
##########
@@ -0,0 +1,200 @@
+// Licensed to the Apache Software Foundation (ASF) under one
+// or more contributor license agreements. See the NOTICE file
+// distributed with this work for additional information
+// regarding copyright ownership. The ASF licenses this file
+// to you under the Apache License, Version 2.0 (the
+// "License"); you may not use this file except in compliance
+// with the License. You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing,
+// software distributed under the License is distributed on an
+// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+// KIND, either express or implied. See the License for the
+// specific language governing permissions and limitations
+// under the License.
+
+//! Manifest-list compatibility and peak live-allocation regression.
+
+use std::alloc::{GlobalAlloc, Layout, System};
+use std::sync::atomic::AtomicUsize;
+use std::sync::atomic::Ordering::Relaxed;
+use std::time::Instant;
+
+use apache_avro::types::Value;
+use apache_avro::{Codec, Reader, Writer};
+use iceberg::io::FileIO;
+use iceberg::spec::{
+ FieldSummary, FormatVersion, ManifestContentType, ManifestFile,
ManifestList,
+ ManifestListWriter,
+};
+
+struct CountingAllocator;
+static LIVE: AtomicUsize = AtomicUsize::new(0);
+static PEAK: AtomicUsize = AtomicUsize::new(0);
+
+fn allocated(bytes: usize) {
+ PEAK.fetch_max(LIVE.fetch_add(bytes, Relaxed) + bytes, Relaxed);
+}
+
+// The allocator is confined to this test executable. Every allocation is
tracked,
+// including those made before measurement, so deallocation cannot underflow
LIVE.
+unsafe impl GlobalAlloc for CountingAllocator {
+ unsafe fn alloc(&self, layout: Layout) -> *mut u8 {
+ let ptr = unsafe { System.alloc(layout) };
+ if !ptr.is_null() {
+ allocated(layout.size());
+ }
+ ptr
+ }
+
+ unsafe fn alloc_zeroed(&self, layout: Layout) -> *mut u8 {
+ let ptr = unsafe { System.alloc_zeroed(layout) };
+ if !ptr.is_null() {
+ allocated(layout.size());
+ }
+ ptr
+ }
+
+ unsafe fn dealloc(&self, ptr: *mut u8, layout: Layout) {
+ unsafe { System.dealloc(ptr, layout) };
+ LIVE.fetch_sub(layout.size(), Relaxed);
+ }
+
+ unsafe fn realloc(&self, ptr: *mut u8, layout: Layout, size: usize) ->
*mut u8 {
+ let new_ptr = unsafe { System.realloc(ptr, layout, size) };
+ if !new_ptr.is_null() {
+ LIVE.fetch_sub(layout.size(), Relaxed);
+ allocated(size);
+ }
+ new_ptr
+ }
+}
+
+#[global_allocator]
+static ALLOCATOR: CountingAllocator = CountingAllocator;
Review Comment:
This would be the first integration test in `crates/iceberg`. It brings a
new test binary and an `unsafe` global allocator, and it folds the format
checks into one test so they can share the allocator. Could the round trips,
the empty lists, the truncated input, and the V1 file read as V2 move into the
unit tests in `spec/manifest_list/mod.rs`, next to
`test_parse_manifest_list_v1`? Those cases aren't covered there today, and as
separate tests a failure in one wouldn't hide the others.
For the allocation budget, what do you think about leaving it out of the
suite and keeping the before and after numbers in the PR description? The test
also times the parse with `Instant` and prints it without checking it.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
