chore(defaults): refresh RPC/LCD endpoints (2026-05-02)#28
Merged
Sentinel-Bluebuilder merged 1 commit intomasterfrom May 2, 2026
Merged
Conversation
The bundled `RPC_ENDPOINTS` was 5 entries verified 2026-03-08, with
`rpc.sentinel.co` first. As of 2026-05-02 that node has been ~22k blocks
behind tip while still reporting `catching_up: false` -- so consumers
calling `createRpcQueryClient()` without an explicit URL got a node that
appeared healthy via /status but served stale ABCI state on every read.
Concrete failure: a Plan Manager wallet endpoint returning 0 P2P for an
address that holds 10,000 P2P, traced through the SDK fallback to
rpc.sentinel.co.
Audited 22 candidate endpoints (existing list + cosmos chain-registry +
suchnode) end-to-end: connect, /status sync flag, AND ABCI bank balance
query against a known funded address. 12 are healthy and serving correct
state. New ordering is by measured latency.
Changes:
- `defaults.js`: replace RPC_ENDPOINTS (5 -> 13) and LCD_ENDPOINTS (4 -> 6)
with the audit results. Stamped verified=2026-05-02. `rpc.sentinel.co`
and `lcd.sentinel.co` kept LAST as fallback (some integrators hardcode
them); flagged in name as "stale fallback".
- `tools/audit-rpc-endpoints.mjs`: new script that any contributor or CI
job can run before a release. Tests connect + sync + balance correctness
(not just /status -- that's exactly the check that lied). Outputs a
paste-ready RPC_ENDPOINTS block.
- `ai-path/FAILURES.md`: new entry C17 + Quick Rule 43 documenting the
"/status lies" failure mode so future builders verify ABCI correctness,
not just sync flag.
No API changes. Existing `addRpcEndpoint`/`removeRpcEndpoint`/`setEndpoints`/
`optimizeEndpoints` continue to work as before. Apps that override
endpoints at runtime are unaffected.
Tested:
- `node tools/audit-rpc-endpoints.mjs` -> 12/22 healthy, list matches
what's now in defaults.js
- `import('./defaults.js')` smoke test: DEFAULT_RPC resolves to
rpc-sentinel.busurnode.com (fastest verified, 125ms)
- Patched Plan Manager (the consumer that hit the bug) confirmed wallet
endpoint now returns correct 10,000 P2P balance after the SDK list is
reordered via addRpcEndpoint() at startup.
Sentinel-Bluebuilder
added a commit
to Sentinel-Bluebuilder/sentinel-node-tester
that referenced
this pull request
May 2, 2026
…026-05-02)
rpc.sentinel.co was reporting `catching_up: false` while serving ABCI state
~22k blocks behind tip. Audit-script run against a known funded address
(sent1uav3z70yynp4jnt39c6pg3d6ujw78m52v2h7gs) returned `0 udvpn` from
sentinel.co while every other endpoint returned the correct balance — the
exact failure mode that breaks balance/feegrant flows on the tester.
- core/constants.js: replace 4-entry RPC_ENDPOINTS with 13 audited entries
(busurnode first, sentinel.co kept last as stale-fallback). Default RPC
flips to busurnode (~125ms). LCD list refreshed similarly.
- core/tkd-bridge.js: defer to RPC_ENDPOINTS in constants.js so future
refreshes only land in one place.
- server.js: replace two hardcoded `rpc.sentinel.co:443` SigningStargate
connects (initial balance + periodic refresh) with a connectWithRpcFailover
helper that walks RPC_ENDPOINTS in order.
- .env.example, README.md: flip RPC default to busurnode and document why.
- scripts/audit-rpc-endpoints.mjs: end-to-end audit (connect + /status +
ABCI bank balance + latency). Run before each release.
Smoke-tested: 12/13 endpoints healthy, deltas within 1 block. Mirrors the
parallel SDK PR Sentinel-Bluebuilder/blue-js-sdk#28.
4 tasks
Sentinel-Bluebuilder
added a commit
that referenced
this pull request
May 2, 2026
…-02) (#29) The 2026-05-02 follow-up to PR #28 surfaced two issues with the audit: 1. The previous script hardcoded EXPECTED_UDVPN as a constant. The audited wallet had a small outflow between PR authoring and merge, so re-running the script declared 0/22 healthy when in fact 12 endpoints all agreed on the new (correct) balance and only rpc.sentinel.co was actually broken. 2. defaults.js still listed rpc.sentinel.co / lcd.sentinel.co as a "stale fallback" entry. Today's audit confirms both are still ~22k blocks behind tip and return 0 for funded addresses while reporting catching_up=false. Keeping them in the array meant a tryWithFallback() call could still hit them after every other endpoint failed and silently return wrong data. Changes: - tools/audit-rpc-endpoints.mjs — replace hardcoded EXPECTED_UDVPN with a consensus check (modal balance across responding candidates, within 50 blocks of tip). Survives any future outflow from the audit address. Also audits LCD endpoints in the same run; the previous script only covered RPC. - defaults.js — drop rpc.sentinel.co and lcd.sentinel.co. Re-sort RPC list by today's measured latency. Expand LCD list 6 -> 9 entries (Roomit, ChainTools, ChainVibes, Validatus all newly verified). Audit results (2026-05-02, consensus mode): - RPC: 12/22 healthy, 12/13 responding agree on balance - LCD: 9/12 healthy, 9/10 responding agree on balance Run: node tools/audit-rpc-endpoints.mjs Test plan - node -e "import('./defaults.js').then(d => console.log(d.DEFAULT_RPC, d.DEFAULT_LCD))" -> Busurnode for both - Audit script exits 0 with both RPC and LCD tier-1 lists populated Co-authored-by: Sentinel-Autonomybuilder <[email protected]>
Merged
2 tasks
Sentinel-Bluebuilder
added a commit
that referenced
this pull request
May 2, 2026
Patch release shipping the consensus-audited RPC + LCD endpoint defaults from #28 and #29 to npm consumers. - package.json: 2.7.0 -> 2.7.1 - defaults.js: SDK_VERSION 2.4.0 -> 2.7.1 (was lagging package.json) No API surface changes. Defaults-only update -- consumers calling createRpcQueryClientWithFallback() / SentinelClient with no overrides will now skip the stale rpc.sentinel.co / lcd.sentinel.co endpoints that were returning balance=0 for funded addresses. Co-authored-by: Sentinel-Autonomybuilder <[email protected]>
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
1 participant
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.
Why
rpc.sentinel.co:443was reportingcatching_up: falsewhile serving ABCI state ~22k blocks behind tip. Consumers querying balances against a known funded address (sent1uav3z70yynp4jnt39c6pg3d6ujw78m52v2h7gs, expected10000000000 udvpn) saw0. This shipped through to the Plan Manager wallet UI as "0 P2P balance" for funded operators, and any other SDK consumer relying on the default RPC list was equally broken.The root cause is a class of failure:
/statuslies — a Tendermint node can reportcatching_up: falseand still serve stale ABCI state. The only reliable health check is to query a known funded address and verify the balance matches an expected value.What
defaults.js— replaceRPC_ENDPOINTS(5 → 13) andLCD_ENDPOINTS(4 → 6) with audited, latency-sorted lists.rpc.sentinel.co:443andlcd.sentinel.coare kept last as "stale fallback" so consumers don't lose them entirely, but every other endpoint precedes them.LAST_VERIFIEDbumped to2026-05-02T00:00:00Z.tools/audit-rpc-endpoints.mjs(new) — end-to-end audit script: connect →/status→ ABCI bank balance → latency. ImportsRPC_ENDPOINTSfromdefaults.jsso the SDK list is the canonical candidate source; outputs paste-ready entries. Run before each release.ai-path/FAILURES.md— Quick Rule 43 + CHAIN section C17 entry documenting the failure mode and the audit-script prevention rule (per the SDK PR-workflow rule: every code fix ships its doc update in the same diff).Audit results (2026-05-02)
12 / 22 candidates healthy, sorted by latency:
Failed (excluded):
rpc.sentinel.co(stale ABCI), Notional, BadgerBite, ValidatorNode, Stakewolle, Decloud Nodes Lab, dvpn.me, ro.mathnodes, Noncompliant, Quasar.Consumer-app trace (Plan Manager)
Desktop/plans(Plan Manager) wallet UI returned0 P2Pfor the configured operator wallet. Trace landed inlib/chain.js'screateRpcQueryClientWithFallback()→ first endpoint =rpc.sentinel.co:443→ balance0returned. Plan Manager has applied a temporary local mutation (removeRpcEndpoint+addRpcEndpoint) which will be reverted once this PR ships, so the SDK's defaults are the only list of record.Test plan
node tools/audit-rpc-endpoints.mjs— 12/22 healthy, deduped against SDK list, paste-ready output stamped today's datenode -e "import('./defaults.js').then(d => console.log(d.RPC_ENDPOINTS[0]))"— first entry is busurnode (125ms)10000000000(10000.00 P2P)node tools/audit-rpc-endpoints.mjson a fresh checkout to confirm list still passes