Repository navigation
chore(release): bump bridge-core to 0.1.22 - #92
Merged
Merged
Conversation
Ships #91 — `RPC_RETRY_DELAY_MS` 250ms -> 1000ms, so viem's first retry lands in a FRESH rate period instead of re-spending a per-second budget that has not refilled. This one is worth releasing promptly, because 0.1.21 has a field regression that 0.1.22 is the fix for. 0.1.20 and earlier ran `retryCount: 6, retryDelay: 250` — seven attempts across 15.75s of doubling backoff. That backoff was an accidental rate limiter: a cold chain scan breached the provider's compute-units-per-SECOND ceiling constantly, but every 429 was absorbed and re-sent slowly enough that nothing ever surfaced to the user. 0.1.21 cut the budget to `retryCount: 2, retryDelay: 250` — 0.75s of total backoff, 1/21st of the cushion — and at the same time made each scan wave cheaper AND faster. Lower cost per wave, but waves issued far more often: the SUSTAINED rate went up, and with the cushion gone the breach became a visible "exceeded its compute units per second capacity" error. 0.1.22 keeps the small attempt count (six retries could outlast a caller's whole scan timeout, which is why it was cut) while restoring real pacing: 3.0s of backoff, spread ACROSS rate periods rather than inside one. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
PR SummaryLow Risk Overview Reviewed by Cursor Bugbot for commit 58a52d5. Bugbot is set up for automated code reviews on this repo. Configure here. |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Ships #91. One line:
packages/bridge-core/package.json0.1.21→0.1.22.Why this release is urgent
0.1.21 has a field regression, and 0.1.22 is the fix.
0.1.20and earlier ranretryCount: 6, retryDelay: 250— seven attempts across 15.75s of doubling backoff. That backoff was functioning as an accidental rate limiter. A cold chain scan was breaching the provider's compute-units-per-second ceiling routinely, but every429was absorbed and re-sent slowly enough that nothing ever surfaced to the user. The scan was slow; it was not broken.0.1.21did two things at once:retryCount: 2, retryDelay: 250— 0.75s of total backoff, 1/21st of the cushion; andaggregate3) and faster.Cost per wave went down; waves per second went up. Because the provider's limit is a rate, not a total, the sustained CU/s actually rose — and with the retry cushion gone, the breach stopped being invisible and started surfacing as
exceeded its compute units per second capacity.What 0.1.22 changes
0.1.22keeps the small attempt count — six retries could outlast a caller's entire scan timeout, which is why it was cut and that reasoning still holds — while restoring genuine pacing:0.1.200.1.210.1.22The
retryDelaymatters more than the count here: viem waits exactlyretryDelaybefore attempt two, so at 250ms all three attempts land inside one second and none of them can pass a per-second ceiling. At 1000ms each attempt gets a fresh window.Note
This is a real mitigation, not a complete fix. A cold scan is still structurally over budget on a free-tier key — 0.1.22 makes it degrade gracefully (slow, self-paced, absorbed) instead of erroring. Fitting inside the ceiling needs explicit CU pacing on the consumer side, tracked separately.
🤖 Generated with Claude Code
This change is