[SharovBot] node/eth: fix data race between MergeLoop and Aggregator.Close

[erigon-copilot]

[SharovBot]

Problem

TestImportClosesChaindataOnInitError was triggering a DATA RACE detected by -race:

WARNING: DATA RACE
Write at 0x00c0653acb90 by goroutine 321:
  github.com/erigontech/erigon/db/state.(*Aggregator).Close()
      db/state/aggregator.go:630

Previous read at 0x00c0653acb90 by goroutine 373:
  github.com/erigontech/erigon/db/state.(*Aggregator).MergeLoop()
      db/state/aggregator.go:1228

  github.com/erigontech/erigon/node/eth.New.func16()
      node/eth/backend.go:1133

Root Cause

eth.New creates its own sentryCtx/sentryCancel pair (derived from context.Background(), not from the ctx parameter). Stop() cancels via sentryCancel(), then waits on bgComponentsEg.

However, MergeLoop was launched with the outer ctx parameter — not backend.sentryCtx. So sentryCancel() did not terminate the MergeLoop goroutine. bgComponentsEg.Wait() returned immediately (the goroutine was still running), and then chainDB.Close() → Aggregator.Close() would write Aggregator fields while MergeLoop concurrently read them.

This race surfaces in the test because importChain calls defer ethereum.Stop() and runErigonCommand calls defer cancel() — the outer context is only cancelled after Stop() has already returned.

Fix

Capture backend.sentryCtx into a local variable and pass it to MergeLoop, so Stop() → sentryCancel() reliably cancels the goroutine before bgComponentsEg.Wait() returns.

Verification

go test -race -count=3 -run TestImportClosesChaindataOnInitError ./cmd/utils/app/...
ok  github.com/erigontech/erigon/cmd/utils/app  5.669s

3 consecutive passing runs with -race, no DATA RACE warnings.

CI job: https://github.com/erigontech/erigon/actions/runs/28674562917/job/85045122311

[erigon-copilot]

[SharovBot]

Closing this PR as it's superseded by the already-merged PR #22244 which fixes the same DATA RACE via a complementary approach (tracking MergeLoop in bgComponentsEg). The root cause analysis was correct — MergeLoop was using the outer ctx instead of sentryCtx — but #22244 + #22203 together fully resolve the race with a cleaner approach.