Make the composite store router an atomic pointer

[masih]

CompositeCommitStore keeps its migration router in a plain field. SetWriteMode replaces it when the write mode changes, while child-store views handed out by GetChildStoreByName read it on every call through a provider, and those accesses weren't synchronized.

The router now lives in an atomic.Pointer and every access goes through loadRouter and storeRouter. A new test reads through a cached child-store view across a SetWriteMode transition under -race. Behaviour and the AppHash are unchanged.

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access that start with 'Devin'.

• Disable automatic comment, CI, and merge conflict monitoring

[github-actions]

The latest Buf updates on your PR. Results from workflow Buf / buf (pull_request).

Build	Format	Lint	Breaking	Updated (UTC)
✅ passed	✅ passed	✅ passed	✅ passed	Oct 6, 2026, 12:38 PM

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 56.85%. Comparing base (bf6c14f) to head (1682d7d).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4470      +/-   ##
==========================================
- Coverage   56.90%   56.85%   -0.05%     
==========================================
  Files        2130     2130              
  Lines      167505   167515      +10     
==========================================
- Hits        95315    95240      -75     
- Misses      72185    72270      +85     
  Partials        5        5              

Flag	Coverage Δ
sei-chain	55.04% <ø> (-0.06%)	⬇️
sei-db	75.17% <ø> (ø)
sei-db-state-db	78.86% <100.00%> (+0.02%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
sei-db/state_db/sc/composite/store.go	79.47% <100.00%> (+0.31%)	⬆️

... and 26 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[cursor]

PR Summary

Medium Risk
Touches hot paths for state reads and write-mode transitions during migration; behavior is unchanged but incorrect synchronization could cause rare query corruption under concurrent mode switches.

Overview
Synchronizes access to the composite store’s migration router so SetWriteMode can swap the router while cached child-store views from GetChildStoreByName still resolve routing on every read.

The router field is now an atomic.Pointer[migration.Router] with loadRouter / storeRouter used everywhere it is installed, cleared, or invoked (commits, reads, proofs, batch-size updates, close). GetChildStoreByName passes cs.loadRouter into NewRouterCommitKVStore instead of closing over a stale router value, so long-lived views pick up post-transition routing.

Adds TestComposite_Auto_ChildStoreReadsDuringWriteModeSwitch (intended for -race) to read through a cached child view while SetWriteMode(MigrateEVM) runs. Existing tests assert the router via loadRouter(). AppHash and routing semantics are unchanged aside from safe concurrent visibility.

^{Reviewed by Cursor Bugbot for commit 1682d7d. Bugbot is set up for automated code reviews on this repo. Configure here.}

[seidroid]

Moves CompositeCommitStore's migration router into an atomic.Pointer reached only through loadRouter/storeRouter, so child-store views stop racing with SetWriteMode. The router accesses are now consistently synchronized and nothing blocks; the open points are that the new test does not reliably overlap reads with the switch, and that views can still race on cs.flatKV. The tree at the merge ref was read in full, and no Go toolchain was available, so the race test was not run. Kept codex's test-synchronization finding, which checks out.

Non-blocking

1 finding on the changed lines, as inline comments.

Pre-existing

Already true on the base branch, not introduced here.

• suggestion — Views from GetChildStoreByName still read cs.flatKV without synchronization through their Iterator closure (cs.iterate, store.go ~1722). On the MemiavlOnly -> MigrateEVM edge, SetWriteMode assigns that field via materializeFlatKV/adoptFlatKV. That is the same concurrent-query scenario this PR addresses, so iteration during the kickoff block is still a data race, and the new test exercises only Get, so it would not catch it.

_{seidroid review · decision approve · session f47bf29309674702b9a3a0b73c556f88 · turn resp_claude_cf756772eda9d3ca8a3629f14b30d69b · item 81e309703d2551c989068341474d7445}

Findings: 0 blocking | 1 non-blocking | 1 posted inline | 1 pre-existing

[masih]

@seidroid review

[seidroid]

Since the last review only the new test changed. It now waits for the reader's first successful Get before calling SetWriteMode, and a deferred, once-guarded stop always shuts the reader down before Close, which addresses my earlier thread (raised with codex). The store.go change is unchanged and correct, so nothing blocks; the pre-existing race on cs.flatKV during iteration still stands. Codex found nothing this round, which agrees with this reading. The tree at the merge ref was read in full.

Pre-existing

Already true on the base branch, not introduced here.

• suggestion — Views from GetChildStoreByName still read cs.flatKV without synchronization through their Iterator closure (cs.iterate, store.go ~1722). On the MemiavlOnly -> MigrateEVM edge, SetWriteMode assigns that field via materializeFlatKV/adoptFlatKV, so iteration during the kickoff block is still a data race. The new test exercises only Get, so it would not catch it.

_{seidroid review · decision approve · session f47bf29309674702b9a3a0b73c556f88 · turn resp_claude_fdf838def8fe9a3c7489f30c9b57f0fd · item 1821a50a3093553c9ca054db2026b13e}

Findings: 0 blocking | 0 non-blocking | 0 posted inline | 1 pre-existing

[seidroid]

Successfully created backport PR for release/v6.7:

• Backport release/v6.7: Make the composite store router an atomic pointer #4473