Repository navigation
Revert sparse snapshot blob writes (#1788) - #1898
Merged
Merged
Conversation
This reverts commit 3ef1a36. On Windows, a snapshot blob marked with FSCTL_SET_SPARSE is slow to map under WHP. The first guest function call after restoring from it takes about 35 ms longer than from a dense blob. Having the flag set is enough: a blob with no holes and a single extent is just as slow, and clearing the flag afterwards does not help. Only a fresh dense write does. Signed-off-by: danbugs <danilochiarlone@gmail.com>
danbugs
requested review from
andreiltd,
dblnz,
devigned,
jprendes,
jsturtevant,
ludfjig,
simongdavies,
squillace and
syntactically
as code owners
October 7, 2026 18:54
danbugs
enabled auto-merge (squash)
October 7, 2026 18:54
jprendes
approved these changes
Oct 7, 2026
Contributor
There was a problem hiding this comment.
🟡 Changes recommended
Existing valid sparse blobs are reused without the dense rewrite needed to remove the Windows penalty.
1 open finding
What changed in this PR
Restores dense snapshot blob writes to avoid WHP restore latency.
Changes:
- Removes sparse file writing and related tests.
- Reduces Windows API features.
- Removes the sparse writing changelog entry.
| File | Description |
|---|---|
fsutil.rs |
Restores dense blob writes. |
Cargo.toml |
Removes sparse file API features. |
CHANGELOG.md |
Removes the reverted feature entry. |
🧠 Review effort: Balanced
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
Benchmark ResultsMeasured commit: kvm / amd (Linux) (➖ stable)No benchmark improved or regressed. Benchmark Resultsfunction_call_codec
payload_allocation
sandboxes
slot_pool
snapshot_files
virtq_readonly
virtq_readwrite
kvm / intel (Linux) (➖ stable)No benchmark improved or regressed. Benchmark Resultsfunction_call_codec
payload_allocation
sandboxes
slot_pool
snapshot_files
virtq_readonly
virtq_readwrite
mshv3 / amd (Linux) (
|
encode_control |
decode_vec_bytes_copy |
|
|---|---|---|
byte_chunks |
922.04 ns (➖ 1.02x faster) |
|
vec_bytes |
703.20 ns ( |
|
373.28 µs (➖ 1.30x slower) |
payload_allocation
slot_pool_segmented |
|
|---|---|
262144 |
730.38 ns (➖ 1.01x faster) |
65536 |
254.31 ns (➖ 1.30x slower) |
sandboxes
create_initialized_and_drop |
|
|---|---|
medium |
59.28 ms (➖ 1.03x slower) |
slot_pool
alloc_dealloc_1500 |
alloc_dealloc_4096 |
alloc_dealloc_128 |
|
|---|---|---|---|
9.73 ns (➖ 1.00x slower) |
9.71 ns (➖ 1.00x faster) |
9.67 ns (➖ 1.01x faster) |
snapshot_files
load_snapshot_unverified |
|
|---|---|
small |
84.69 µs (➖ 1.00x slower) |
virtq_readonly
slot_pool_segmented_fragmented |
slot_pool_segmented |
|
|---|---|---|
262144 |
9.84 µs (➖ 1.09x faster) |
8.78 µs (➖ 1.04x faster) |
65536 |
2.28 µs (➖ 1.02x slower) |
2.22 µs (➖ 1.10x faster) |
virtq_readwrite
slot_pool_segmented_fragmented |
slot_pool_segmented |
|
|---|---|---|
65536 |
8.14 µs (➖ 1.00x faster) |
8.10 µs (➖ 1.03x faster) |
8192 |
1.29 µs (➖ 1.00x faster) |
1.39 µs (➖ 1.04x slower) |
262144 |
36.96 µs (➖ 1.01x slower) |
mshv3 / intel (Linux) (➖ stable)
No benchmark improved or regressed.
Benchmark Results
function_call_codec
encode_control |
decode_vec_bytes_copy |
|
|---|---|---|
byte_chunks |
929.64 ns (➖ 1.01x faster) |
|
vec_bytes |
663.85 ns (➖ 1.01x faster) |
|
701.44 µs (➖ 1.05x slower) |
payload_allocation
slot_pool_segmented |
|
|---|---|
262144 |
672.68 ns (➖ 1.13x slower) |
65536 |
170.80 ns (➖ 1.02x slower) |
sandboxes
create_initialized_and_drop |
|
|---|---|
medium |
83.40 ms (➖ 1.31x slower) |
slot_pool
alloc_dealloc_1500 |
alloc_dealloc_4096 |
alloc_dealloc_128 |
|
|---|---|---|---|
9.61 ns (➖ 1.01x slower) |
8.82 ns (➖ 1.13x faster) |
8.81 ns (➖ 1.00x slower) |
snapshot_files
load_snapshot_unverified |
|
|---|---|
small |
48.48 µs (➖ 1.06x slower) |
virtq_readonly
slot_pool_segmented_fragmented |
slot_pool_segmented |
|
|---|---|---|
262144 |
7.78 µs (➖ 1.00x faster) |
7.78 µs (➖ 1.03x faster) |
65536 |
2.27 µs (➖ 1.01x slower) |
2.27 µs (➖ 1.00x faster) |
virtq_readwrite
slot_pool_segmented_fragmented |
slot_pool_segmented |
|
|---|---|---|
65536 |
7.57 µs (➖ 1.01x slower) |
7.54 µs (➖ 1.01x faster) |
8192 |
881.98 ns (➖ 1.01x slower) |
880.92 ns (➖ 1.01x slower) |
262144 |
35.81 µs (➖ 1.12x faster) |
hyperv-ws2025 / amd (Windows) (➖ stable)
No benchmark improved or regressed.
Benchmark Results
function_call_codec
encode_control |
decode_vec_bytes_copy |
|
|---|---|---|
byte_chunks |
1.15 µs (➖ 1.01x faster) |
|
vec_bytes |
760.35 ns (➖ 1.04x faster) |
|
2.18 ms (➖ 1.03x slower) |
payload_allocation
slot_pool_segmented |
|
|---|---|
262144 |
793.69 ns (➖ 1.05x faster) |
65536 |
227.86 ns (➖ 1.04x faster) |
sandboxes
create_initialized_and_drop |
|
|---|---|
medium |
102.45 ms (➖ 1.10x slower) |
slot_pool
alloc_dealloc_1500 |
alloc_dealloc_4096 |
alloc_dealloc_128 |
|
|---|---|---|---|
11.22 ns (➖ 1.11x slower) |
10.57 ns (➖ 1.01x slower) |
10.75 ns (➖ 1.06x slower) |
snapshot_files
load_snapshot_unverified |
|
|---|---|
small |
732.70 µs (➖ 1.10x faster) |
virtq_readonly
slot_pool_segmented_fragmented |
slot_pool_segmented |
|
|---|---|---|
262144 |
8.68 µs (➖ 1.11x faster) |
8.60 µs (➖ 1.24x faster) |
65536 |
2.54 µs (➖ 1.06x slower) |
2.55 µs (➖ 1.08x slower) |
virtq_readwrite
slot_pool_segmented_fragmented |
slot_pool_segmented |
|
|---|---|---|
65536 |
9.07 µs (➖ 1.02x slower) |
8.96 µs (➖ 1.06x faster) |
8192 |
1.41 µs (➖ 1.05x slower) |
1.37 µs (➖ 1.03x slower) |
262144 |
48.40 µs (➖ 1.25x slower) |
hyperv-ws2025 / intel (Windows) (➖ stable)
No benchmark improved or regressed.
Benchmark Results
function_call_codec
encode_control |
decode_vec_bytes_copy |
|
|---|---|---|
byte_chunks |
1.24 µs (➖ 1.24x slower) |
|
vec_bytes |
816.77 ns (➖ 1.26x slower) |
|
3.07 ms (➖ 1.17x faster) |
payload_allocation
slot_pool_segmented |
|
|---|---|
262144 |
742.03 ns (➖ 1.10x slower) |
65536 |
218.37 ns (➖ 1.14x slower) |
sandboxes
create_initialized_and_drop |
|
|---|---|
medium |
112.59 ms (➖ 1.10x faster) |
slot_pool
alloc_dealloc_1500 |
alloc_dealloc_4096 |
alloc_dealloc_128 |
|
|---|---|---|---|
10.29 ns (➖ 1.20x slower) |
10.11 ns (➖ 1.20x slower) |
10.63 ns (➖ 1.31x slower) |
snapshot_files
load_snapshot_unverified |
|
|---|---|
small |
476.54 µs (➖ 1.11x faster) |
virtq_readonly
slot_pool_segmented_fragmented |
slot_pool_segmented |
|
|---|---|---|
262144 |
7.62 µs (➖ 1.14x faster) |
7.72 µs (➖ 1.13x faster) |
65536 |
2.33 µs (➖ 1.09x faster) |
2.31 µs (➖ 1.09x faster) |
virtq_readwrite
slot_pool_segmented_fragmented |
slot_pool_segmented |
|
|---|---|---|
65536 |
8.54 µs (➖ 1.08x faster) |
8.60 µs (➖ 1.15x faster) |
8192 |
1.11 µs (➖ 1.01x faster) |
1.18 µs (➖ 1.02x slower) |
262144 |
42.27 µs (➖ 1.18x slower) |
Reported by cargo ci bench-report --candidate run:37670356285 --baseline run:37551535479 --config-file bench_report.toml.
Contributor
|
Why doesn't bot comment show an improvement on windows? |
ludfjig
approved these changes
Oct 7, 2026
Contributor
Author
We don't have a benchmark in hyperlight-core that catches the regression. It was observed in hyperlight-unikraft. |
Contributor
also, it currently only shows the results for a subset of the benchmarks, due to the high variance of some benches (#1894 is trying to fix that) |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This reverts #1788 because sparse snapshot blobs slow down the first guest function call after a restore on Windows (WHP). Any blob written with
FSCTL_SET_SPARSEcosts about 35 ms extra on that first guest function call, even when the blob has no holes. Clearing the flag later doesn't fix it; only a dense rewrite does.Replaces #1895, which was opened from a fork and hit the shorter CI timeouts.