docs(infra): record redis role tools as binary names, not units (#16060, #16071)

autobot-infrastructure/shared/scripts/vm-management/fix-architecture-issues.sh

@@ -49,10 +49,28 @@ log "Checking Redis architecture compliance..."
 if pgrep redis-server >/dev/null || docker ps | grep redis >/dev/null 2>&1; then
     warning "Redis is running locally on main instance - this violates architecture!"
     echo "Stopping local Redis instances..."
-    sudo systemctl stop redis-server 2>/dev/null || true
-    sudo systemctl stop redis-stack-server 2>/dev/null || true
+    # #16071 AC4: distinguish "unit not installed on this host" (nothing to do,
+    # e.g. redis-server on a node that only ever had redis-stack-server) from
+    # "unit exists but the stop failed" (a real problem) -- a blanket `|| true`
+    # discarded both alike. list-unit-files exits non-zero for an absent unit,
+    # the same existence check already used for autobot-redis/redis-stack-server
+    # elsewhere in this tree.
+    stop_failed=0
+    for unit in redis-server redis-stack-server; do
+        if ! systemctl list-unit-files "${unit}.service" &>/dev/null; then
+            continue
+        fi
+        if ! sudo systemctl stop "${unit}"; then
+            error "Failed to stop ${unit}"
+            stop_failed=1
+        fi
+    done
     docker stop $(docker ps -q -f name=redis) 2>/dev/null || true
-    success "Local Redis instances stopped"
+    if [ "${stop_failed}" -eq 0 ]; then
+        success "Local Redis instances stopped"
+    else
+        error "One or more local Redis units failed to stop -- see above"
+    fi
 else
     success "✅ No local Redis instances found (correct)"
 fi

autobot-slm-backend/api/deployments.py

@@ -80,6 +80,10 @@
     "redis": {
         "description": "Redis Stack server for data persistence",
         "category": "data",
+        # Binary names, not systemd units (#16060/#16071): Redis Stack ships
+        # redis-server/redis-cli at /opt/redis-stack/bin/. The systemd unit it
+        # runs under is redis-stack-server, tracked separately by
+        # services/role_units.py.
         "tools": ["redis-server", "redis-cli", "redis-sentinel"],
     },
     "ai-stack": {

autobot-slm-frontend/src/constants/node-roles.ts

@@ -92,6 +92,10 @@ export const NODE_ROLE_METADATA: Record<NodeRole, RoleMetadata> = {
     displayName: 'Redis',
     description: 'Redis Stack server for data persistence',
     category: 'data',
+    // Binary names, not systemd units (#16060/#16071): Redis Stack ships
+    // redis-server/redis-cli at /opt/redis-stack/bin/, same as api/deployments.py.
+    // The systemd unit it runs under is redis-stack-server, tracked separately
+    // by services/role_units.py.
     tools: ['redis-server', 'redis-cli', 'redis-sentinel'],
   },
   'postgres': {

changelog/unreleased/16060-16071-redis-naming-completion.md

@@ -0,0 +1,9 @@
+---
+type: fix
+scope: infra
+issue: 16060
+pr: 0
+---
+Recorded why `node-roles.ts` and `api/deployments.py` list `redis-server`/`redis-cli` for the redis role — binary names Redis Stack genuinely ships, not the systemd unit (`redis-stack-server`) the fleet manages via `services/role_units.py` — completing #16060's remaining "correct or record as display-only" criterion.
+
+`fix-architecture-issues.sh`'s Redis-stop step no longer discards its result with a blanket `|| true`: it now checks whether each unit (`redis-server`, `redis-stack-server`) is actually installed before attempting to stop it, and reports a real stop failure instead of swallowing it identically to an absent unit — completing #16071's AC4.

repo_tests/fix_architecture_issues_redis_stop_result_16071_test.py

@@ -0,0 +1,73 @@
+# Copyright 2025-2026 mrveiss
+# SPDX-License-Identifier: Apache-2.0
+# AutoBot - AI-Powered Automation Platform
+# Author: mrveiss
+"""fix-architecture-issues.sh reports a failed Redis stop, not just an absent
+unit (#16071 AC4).
+
+The script stops BOTH `redis-server` and `redis-stack-server` deliberately
+(a node provisioned before #16060/#16071 may still carry the legacy apt
+package) -- that dual-stop is a declared, tested exemption in
+`redis_unit_is_the_stack_unit_16071_test.py::_LEGACY_CLEANUP`, not the
+defect. The defect was `2>/dev/null || true` on each stop: it discarded
+"unit not installed on this host" (expected, nothing to do) and "unit
+exists but the stop failed" (a real problem) identically.
+
+Not executed here -- a real `systemctl stop` needs a live systemd host,
+and this repo never runs its own scripts locally (evidence comes from CI
+or a sanctioned hook). These are static/structural checks on the script
+text, the same evidence shape as the ansible YAML guards in this session.
+"""
+
+from __future__ import annotations
+
+import re
+
+from repo_tests._paths import repo_root
+
+_SCRIPT = "autobot-infrastructure/shared/scripts/vm-management/fix-architecture-issues.sh"
+
+
+def _text() -> str:
+    return (repo_root() / _SCRIPT).read_text(encoding="utf-8")
+
+
+def test_script_exists():
+    assert (repo_root() / _SCRIPT).is_file(), f"{_SCRIPT} missing or moved"
+
+
+def test_no_blanket_swallow_on_a_redis_unit_stop():
+    """The exact defect: a `stop <unit> ... || true` line can't tell an
+    absent unit from a real failure. None should remain for either name."""
+    offenders = [
+        line.strip()
+        for line in _text().splitlines()
+        if re.search(r"systemctl\s+stop\s+redis(-stack)?-server", line) and "|| true" in line
+    ]
+    assert not offenders, f"still discards the stop result: {offenders}"
+
+
+def test_checks_unit_existence_before_stopping():
+    """AC4's fix: skip a unit that was never installed, rather than
+    attempting (and then swallowing the failure of) stopping it. The script
+    loops a single variable-driven check over both unit names rather than
+    writing the check out twice, so this asserts the loop covers both names
+    AND that the check itself is present, rather than checking each name
+    against a check that is actually parameterised by shell variable."""
+    text = _text()
+    assert re.search(
+        r'list-unit-files\s+"\$\{unit\}\.service"', text
+    ), "no existence check gates the per-unit stop loop"
+    assert re.search(
+        r"for\s+unit\s+in\s+redis-server\s+redis-stack-server\b", text
+    ), "the existence-gated loop no longer covers both redis unit names"
+
+
+def test_a_real_stop_failure_is_reported_not_swallowed():
+    """The other half of AC4: once a unit is confirmed present, a failed
+    stop must surface (via error()), not disappear into `|| true`."""
+    text = _text()
+    assert re.search(
+        r"if\s+!\s+sudo\s+systemctl\s+stop\s+\"?\$\{?unit\}?\"?", text
+    ), 'expected an `if ! sudo systemctl stop "$unit"` guard so a real failure is caught'
+    assert "Failed to stop" in text, "a failed stop is no longer reported"