[fabwa]

Replaces #4283, which tried to solve this inside Meta.Run by unsubscribing. That deadlocks, and as @carpawell put it there, the ordering "is hard to say inside this Meta service" — so this does it in cmd/neofs-node, where both lifetimes are owned.

Problem

SideChain.Stop is what actually takes the chain down (core.Close), and it is registered as a node closer. Closers run in shutdown(), after wait() has already cancelled the node context — and the metadata service, like every other worker, stops the moment that context is done. So on every shutdown the consumer tears down first and the still-running chain keeps writing into the channels it was handed, which is where send on closed channel comes from.

Nothing drains those subscriptions in that window either. If a buffer fills, the chain's dispatcher blocks, the next AddBlock holds addLock waiting for it, and core.Close — i.e. the closer itself — hangs with it, as @End-rey pointed out on #4283.

Fix

The service now runs until the closer stops it, after the chain:

c.closers = append(c.closers, func() {
    c.sidechain.Stop()
    stopMeta()
})

so the order is: chain stops delivering → service stops reading → c.wg.Wait() collects it. The service keeps draining while Stop runs, so the dispatcher never blocks and core.Close cannot wedge on addLock. Its context is deliberately not the node's; that is the one deviation from the other workers and the comment says why.

The two chain-fed channels are no longer closed. Blockchain.Close closes stopCh and waits on runToExitCh, which is the Run loop only — notificationDispatcher observes the same stopCh but nothing waits for it, and its sends sit outside its select. So there is no point at which a consumer can know the sends have stopped, whatever the ordering. m.taskQueue is still closed: its only writer is notifyReceived, called from notificationHandler, which Run does wait on.

Test plan

• TestMeta_Run_DoesNotCloseChainFedChannels feeds both subscriptions from goroutines Run does not wait on, as the real dispatcher does, and fails with send on closed channel if either close is restored — confirmed over -count=3 -race.
• -race green on pkg/services/meta and cmd/neofs-node, including -shuffle=1794309779113128012 (the seed from meta: unsubscribe from the chain before closing its delivery channels #4283's metrics-registry report).
• golangci-lint with the shared config, go build ./... and go vet ./... clean.
• The shutdown ordering itself is not unit-tested: it needs a real cfg and a real chain, and cmd/neofs-node has no harness for either.

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 0% with 6 lines in your changes missing coverage. Please review.
✅ Project coverage is 32.83%. Comparing base (34d007f) to head (0b95959).
⚠️ Report is 1 commits behind head on master.

Files with missing lines	Patch %	Lines
cmd/neofs-node/meta.go	0.00%	6 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #4299      +/-   ##
==========================================
+ Coverage   32.70%   32.83%   +0.12%     
==========================================
  Files         673      673              
  Lines       41698    41697       -1     
==========================================
+ Hits        13638    13691      +53     
+ Misses      28060    28006      -54     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.