gsadmin/rustfs
1
0
Fork 0
mirror of https://github.com/rustfs/rustfs.git synced 2026-08-31 17:28:12 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix(get): remove GET chunk fast path (#2507)

Browse Source
This commit is contained in:
安正超
2026-04-12 23:12:23 +08:00
committed by GitHub
parent 0e697fa7e1
commit 458086dc80
22 changed files with 80 additions and 4746 deletions
Download Patch File Download Diff File Expand all files Collapse all files
crates/ecstore/README.md
+1 -34
View File
@@ -34,40 +34,7 @@ For comprehensive documentation, examples, and usage guides, please visit the ma
## 📈 Benchmarks
ECStore ships several Criterion benchmarks under [`crates/ecstore/benches/`](./benches/).
### Direct Chunk Path
Use the direct chunk benchmark to compare the current slice-forwarding path against the previous assembled-copy path:
```bash
cargo bench -p rustfs-ecstore --bench direct_chunk_benchmark
```
To run only the end-to-end ECStore range-read benchmark:
```bash
cargo bench -p rustfs-ecstore --bench direct_chunk_benchmark ecstore_get_object_chunks
```
To run the reconstructed multi-disk range-read benchmark:
```bash
cargo bench -p rustfs-ecstore --bench reconstructed_chunk_benchmark
```
### Saved Comparison Points
Latest local measurements on this branch:
- `direct_chunk_path/slice_forwarding/single_block_aligned`: about `477 ns`
- `direct_chunk_path/assembled_copy/single_block_aligned`: about `3.25 us`
- `direct_chunk_path/slice_forwarding/multi_block_unaligned`: about `963 ns`
- `direct_chunk_path/assembled_copy/multi_block_unaligned`: about `7.25 us`
- `ecstore_get_object_chunks/drain/multi_disk_range`: about `644-654 us`, throughput about `2.86-2.90 GiB/s`
- `reconstructed_chunk_path/drain/multi_disk_missing_shard`: about `1.292-1.304 ms`, throughput about `1.43-1.45 GiB/s`
These numbers are intended as branch-local reference points. Re-run the benchmark on your target machine before treating them as a regression baseline.
ECStore ships Criterion benchmarks under [`crates/ecstore/benches/`](./benches/) for the remaining erasure-code and comparison workloads. Re-run benchmarks on your target machine before treating any local measurement as a regression baseline.
## 📄 License