[fabwa]

Summary

Two related bugs in SIGHUP config reload handling for gRPC servers.

H9: reload can permanently misattribute a running server to the wrong endpoint after a prior partial failure

reloadGRPC matched currently-running servers to a freshly recomputed config snapshot by pairing c.cfgGRPC.servers[i] with oldCfg[i] - plain positional index. c.appCfg (and so oldCfg, recomputed from it on every reload) always reflects the full declared config file regardless of which servers actually started; c.cfgGRPC.servers only ever contains the ones that did. The moment any single endpoint ever fails to build - a temporarily unbindable address, a bad cert path - that slice has one fewer entry than oldCfg, and every positional pairing from that point on is off by one, silently reusing the wrong server under the wrong endpoint's label on every subsequent reload until restart.

H10: NetmapService missing from gRPC servers (re)created by reload

initNetmapService registered NetmapServiceServer by looping c.cfgGRPC.servers directly, instead of using cfgGRPC.registerService like every other service (Accounting, Container, Session, Object, Reputation all use it). registerService both applies the registrator to current servers and stores it for future ones; looping the current servers directly does only the former. Any server (re)created by a later reload - endpoint, conn_limit, or TLS cert change, not just a brand new endpoint - therefore serves every other registered service but not Netmap, until the whole node is restarted.

Fix

• Track each server together with its own endpoint/config snapshot in a single cfgGRPCServer struct, instead of in parallel slices indexed separately from the config. reloadGRPC now builds its "old" lookup directly from c.cfgGRPC.servers's own stored snapshots - each entry is self-describing, so there's no positional pairing left to desync, regardless of how many prior reloads partially failed. The matching/reconciliation logic itself is extracted into reconcileGRPCServers for direct testing without needing a real nCli/TLS-dependent build path.
• Register NetmapServiceServer the same way every other service does, via cfgGRPC.registerService.

Test plan

• Added a regression test reproducing H9's exact two-reload scenario: build servers for endpoints A/B/C, fail B's rebuild on reload Add Issue template #1 (A/C reused, B dropped), then reload [#1] Add GitHub issue templates #2 with no further config changes - asserting C's original server is correctly reused (not misattributed to B) and only B is rebuilt.
• go vet ./... clean; go test ./cmd/neofs-node/... passes; full repo build (go build ./...) clean.
• Patch verified to apply cleanly and pass tests/build in a separate clean worktree against the current master tip.

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 25.00000% with 33 lines in your changes missing coverage. Please review.
✅ Project coverage is 32.83%. Comparing base (fd6e171) to head (a193853).
⚠️ Report is 1 commits behind head on master.

Files with missing lines	Patch %	Lines
cmd/neofs-node/grpc.go	26.82%	30 Missing ⚠️
cmd/neofs-node/config.go	0.00%	2 Missing ⚠️
cmd/neofs-node/netmap.go	0.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #4286      +/-   ##
==========================================
+ Coverage   32.78%   32.83%   +0.05%     
==========================================
  Files         673      673              
  Lines       41707    41706       -1     
==========================================
+ Hits        13673    13696      +23     
+ Misses      28034    28010      -24     

☔ 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.

[roman-khimov]

Conflicts.

[fabwa]

Rebased onto fd6e171. Only CHANGELOG conflicted; the code patch is unchanged.