perf(delivery): stop re-reading hard links to verify them
Deploy / deploy (push) Successful in 6m25s

A 25-shot episode took about an hour to package when the linking itself
is instant. All of it was verification: every delivered file was read
back and hashed, and at ~100 MB/s over SMB one 2.8 GB shot costs 30 s.

Re-hashing a hard link cannot disagree with validation. The entry is a
second name for the same data, so it re-reads the exact bytes that were
already hashed — half an hour per episode to prove something that cannot
be false. Verification is now proportional to what was actually done:
copies are read back in full, links carry over the validation-time digest.

Linking has one gap hashing covered incidentally: a source changed after
it passed QC, which a link would carry faithfully. Source timestamps are
now checked against the export's validation time, which costs a stat per
file rather than a full read.

"verifyMode": "always-hash" restores unconditional read-back for a
production that requires it.

Worker 1.3.0.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
twotalesanimation
2026-08-14 09:29:04 +02:00
parent 991abe072c
commit af810258e6
6 changed files with 165 additions and 24 deletions
+19 -6
View File
@@ -114,16 +114,28 @@ never claims to have linked when it copied.
## Verification
Every file is read back after packaging and hashed, and the sequence digest is
compared against the one recorded at validation time. `fsutil hardlink list` does
not work over SMB, so link identity is never queried — reading the bytes covers
links and copies through one path.
Verification is proportional to what the builder actually did, which is what
keeps a package fast:
| Mode | How it verifies | Cost for a 25-shot episode |
|---|---|---|
| **Hard-linked** (default) | The entry *is* the source data, so the validation-time digest already describes it. Checks each file exists, and that no source was modified after it passed QC. | seconds |
| **Copied** | Read back in full and hashed; the sequence digest must match the one recorded at validation. | ~30 s per 2.8 GB shot |
Re-reading a hard link cannot disagree with validation — it is the same bytes
under a second name — and at ~100 MB/s over SMB it costs roughly half an hour
per episode to prove something that cannot be false. A copy is genuinely
different: it can be truncated or corrupted, so it is always read back.
Set `"verifyMode": "always-hash"` in the delivery config to read everything back
regardless, for a production that requires it contractually.
An item fails, and the whole package fails with it, when:
| Failure | Meaning |
|---|---|
| Checksum mismatch | The delivered bytes differ from what passed QC |
| Checksum mismatch (copied items) | The delivered bytes differ from what passed QC |
| A source file modified after validation | The render changed after it passed QC — a link would carry the new bytes faithfully |
| Frame count differs from the export record | The render folder changed after QC — someone re-rendered in place |
| Source folder missing | The render is not where the export says it is |
| No matching files | The folder exists but holds nothing matching the output pattern |
@@ -186,7 +198,8 @@ key by key, so a project can override just `deliveryRoot`.
"includeMovs": true,
"checksumAlgo": "xxh64",
"reportFormats": ["csv", "json"],
"copyMode": "auto"
"copyMode": "auto",
"verifyMode": "auto"
}
```