Skip to content

LLP 0381 request: an overflowing watermark second re-appends its overflow every tick, and the exit needs a decision - #1368

Open
philcunliffe wants to merge 1 commit into
masterfrom
llp/boundary-overflow-request
Open

LLP 0381 request: an overflowing watermark second re-appends its overflow every tick, and the exit needs a decision#1368
philcunliffe wants to merge 1 commit into
masterfrom
llp/boundary-overflow-request

Conversation

@philcunliffe

Copy link
Copy Markdown
Contributor

Feature or issue

Issue #1332 (deferred from PR #1330 review round 2): when more than MAX_BOUNDARY_IDS (1000) items share one watermark second, the GitHub boundary gate's overflow sits outside the guard and is re-appended, with its sub-resource requests re-spent, on every tick until unrelated activity moves the watermark. Reviewer probe at the PR head: N=1500 gives tick1 1500 rows / 1500 listCommitFiles, then 500 rows and 500 requests every tick with no convergence; N=900 converges after tick1. Under the production 400-request budget the repository never finishes its commits phase and starves the round-robin rotation. Every exit forks against a settled document (LLP 0361#page-work's equal-timestamp anti-loss rule, or LLP 0360#resource-bounds' identity cap), so this is not fixable as a bug: issue #1332's own acceptance condition opens with "An LLP records the chosen exit".

Solution

  • Mints request LLP 0381 (llp/0381-boundary-overflow-past-the-cap.rfc.md, Draft RFC), which states the loop, the measured probe, the settled constraints, and the candidate exits: A, a bounded declared loss (advance the watermark past a second that will not fit); B, a population-bounded carry (lift the cap to the second's whole population); C, an absorbing scheme (rotation plus a completion marker, or overflow-scoped downstream dedup). It chooses none of them; choosing is a later design tick's job and ultimately a human's.
  • Kept distinct from Draft LLP 0373 (the identity-key question), whose #constraints sets this capacity question aside by name; the relationship is stated in both documents.
  • Appends Extended-by: forward-refs to LLP 0360 and LLP 0361 and a Related: entry to LLP 0373. node scripts/llp-numbers.js check reports no cross-branch collision; npm test (6067 pass) and npm run typecheck are green.

Code: +0 / -0 lines

…boundary cap

Issue #1332 carries neutral:fix but its own acceptance condition opens with
an LLP recording the chosen exit, and every exit forks against a settled
document (LLP 0361#page-work's anti-loss rule or LLP 0360#resource-bounds'
cap), so this is a design request, not a review-rung repair. The RFC states
the loop, the measured probe from PR #1330 review round 2, the settled
constraints, and three candidate exits (bounded declared loss, a
population-bounded carry, an absorbing scheme). It chooses none of them.

Forward-refs: Extended-by on LLP 0360 and 0361, a Related entry on the
Draft LLP 0373, whose #constraints sets this question aside by name.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant