[RyoKusnadi]

Linked issue (required)

Fixes #2245

Summary

Problem: repman raises a blocking ERR00007 ("User must have REPLICATION SLAVE privilege on ...") for a replica even though the replication user has the grant and replication works.

How it happens: db-servers-credential (monitor user) and replication-credential (repl user) are different accounts, and the monitor user has no account matching a replica's address, for example monitor@'10.1.1.%' but a replica in 10.2.2.0/24. With monitoring-check-grants on (the default path), the check asks the privileges of the monitor user for that replica, finds no match, and reports "N". repl@'10.2.2.%' is never looked at.

Cause: in ServerMonitor.CheckPrivileges (cluster/srv_chk.go), the loop under "Check replication user has correct privs" passed cluster.GetDbUser() to dbhelper.GetPrivileges. The comment and the ERR00015 message both name the replication user.

Fix: pass cluster.GetRplUser(). One word. No other change, since the first check in the function (monitor user on the repman host) is meant to use the monitor user.

Scope: I checked the other GetPrivileges call in the same function; it checks the monitor user for ERR00006/8/9 and is correct. If both credentials are the same user, nothing changes.

How it was tested

New test TestCheckPrivilegesLooksUpReplicationUser in cluster/srv_chk_privileges_test.go, using sqlmock with the issue's layout (monitor account only for the repman host, repl account for 10.2.2.%).

• Before the fix: fails, the second mysql.user query is asked for the monitor user so the expected repl lookup is never made.
• After the fix: passes.

Commands: GOTOOLCHAIN=local go test ./cluster -run TestCheckPrivilegesLooksUpReplicationUser, gofmt -l.
Not run: the full cluster package (it crashes on develop with a concurrent map write, unrelated) and a real Docker regtest with MariaDB/MySQL.

Checklist

• Linked an issue above with Fixes #NNNN (or Refs #NNNN for partial) — T9/T10
• Targets develop (or a topology branch for failover/rejoin/split-brain changes) — T10/T20
• Synced from latest origin/develop and checked no in-flight branch/PR already does this — T20
• New config is a TOML key with an off-switch where applicable — T14/T16 (n/a, no new config)
• New API endpoint ships with its GUI surface (no API-only features) — T6 (n/a)
• User doc (docs.signal18.io) + technical doc (doc/implementation/) updated — T8 (n/a, bug fix)
• Real regtest / Docker coverage for the behavior — T13 (not run; unit test with sqlmock only)
• No unbounded buffer/queue; producer→sink buffers are bounded + drops tracked as state — T18 (n/a)