docs(changelog): correct the #768 entry's account of the planted fault (#770) - #771
Merged
Merged
Conversation
#770) The entry claimed the first planted fault -- dropping a val_decref in shared_incr -- leaked nothing because make_num serves request-path Values from the arena. That is wrong. A probe in the HTTP worker reports g_arena.active=0 and new_v->arena=0: nothing in the HTTP path opens an arena window (only the user-callable arena_mark builtin does), so the fault leaks 72 heap bytes per request and is unbounded -- ~145 B/req sustained across five consecutive 20,000-request batches. The gate stayed green for a different and worse reason: its entire decision happens inside the first 7,000 requests, which for that rate is still inside the window where the leak is absorbed by resident free heap the process already owns. Filed as #770. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
Pull request overview
This PR corrects a causal claim in the CHANGELOG record for the RSS leak gate validation in #768/#769, clarifying that a planted missing-val_decref fault does leak and that the gate stayed green due to a measurement-window blind spot (tracked in #770), not because the fault was arena-allocated.
Changes:
- Rewrites the #768 CHANGELOG entry to state the planted missing-
val_decreffault leaks on the heap and can be missed within the gate’s first-7,000-requests decision window. - Adds the proposed direction for fixing the gate (#770), referencing allocator accounting (
mallinfo2().uordblks) as a replacement for the RSS proxy.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+200
to
+203
| green. That is not because the fault is harmless. It leaks 72 heap bytes per | ||
| request (probed: `g_arena.active=0`, `new_v->arena=0`, so it is not an | ||
| arena-served Value), and at 20,000-request batches it shows as ~145 B/req | ||
| sustained and unbounded. The gate misses it because its whole decision happens |
Comment on lines
+207
to
+209
| instrument for — tracked as #770, with exact allocator accounting | ||
| (`mallinfo2().uordblks`) as the proposed replacement for the RSS proxy. A gate | ||
| this class depends on has to be shown turning red by something that actually |
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.
Refs #770.
The #768 CHANGELOG entry I landed in #769 contains a wrong causal claim, and it is worth correcting in place rather than leaving a plausible-but-false explanation in the record.
Claimed: the first planted fault (dropping a
val_decrefinshared_incr) leaked nothing measurable becausemake_numserves request-path Values from the arena, where they die at arena reset regardless of refcount.Measured: a probe inside the HTTP worker reports
g_arena.active=0,new_v->arena=0,sizeof(Value)=72. Nothing in the HTTP path opens an arena window — only the user-callablearena_markbuiltin does (builtins.c:3828). The fault leaks 72 heap bytes per request, unbounded: ~145 B/req sustained across five consecutive 20,000-request batches, +15 MB by 150,000 requests on one connection.The gate stayed green for a worse reason, now tracked as #770: its entire decision happens within the first 7,000 requests, which for that leak rate is inside the window where the leak is absorbed by resident free heap the process already owns.
Docs-only; no code or test change.
🤖 Generated with Claude Code