fix: start monit after the units of Core's processes (keel#61)

packages/keel-core/debian/changelog

@@ -1,3 +1,15 @@
+keel-core (0.1.1) trixie; urgency=medium
+
+  * Monit starts after the units of Core's processes, ssh.service,
+    webmin.service and postfix.service: a drop-in,
+    /usr/lib/systemd/system/monit.service.d/keel-core.conf, ordering only.
+    Started beside them, monit's first cycle after a boot found webmin and
+    postfix inactive (Keel-Linux/keel#61, item 2). keel writes no
+    `set daemon ... with start delay` instead: that line is global, and
+    keel leaves the operator's cycle alone (decision 0021).
+
+ -- Marcos Mendez <mendez.foto@gmail.com>  Fri, 02 Oct 2026 06:00:00 +0000
+
 keel-core (0.1.0) trixie; urgency=medium
 
   * First release (handbook decision 0041, first implementation, step 4).

packages/keel-core/debian/control

@@ -23,7 +23,9 @@ Description: Keel Core, the appliance every Keel appliance is built on
  The appliance manifest of Keel Core (handbook decisions 0036 and 0041),
  installed as /usr/share/keel/appliances/core.yaml: the four overlays Core
  carries with their state in each installation mode, and Core's own
- processes (sshd, Webmin and postfix) with their ports and checks.
+ processes (sshd, Webmin and postfix) with their ports and checks. A
+ systemd drop-in starts monit after those three units, so its first cycle
+ does not find them still starting.
  .
  It depends on the four overlay packages, installer, wireguard, etcd and
  crowdsec, and on keel 0.13.0, the first keel that converges the overlays'

packages/keel-core/debian/rules

@@ -9,3 +9,5 @@
 execute_after_dh_auto_install:
 	install -D -m 0644 manifest.yaml \
 		debian/keel-core/usr/share/keel/appliances/core.yaml
+	install -D -m 0644 monit-after-core.conf \
+		debian/keel-core/usr/lib/systemd/system/monit.service.d/keel-core.conf

packages/keel-core/monit-after-core.conf

@@ -0,0 +1,8 @@
+# Installed by keel-core as /usr/lib/systemd/system/monit.service.d/.
+# Monit's checks of Keel Core are the units of the manifest's processes
+# (manifest.yaml, beside debian/). Started beside them, monit's first
+# cycle after a boot found webmin and postfix inactive (keel#61), so
+# monit starts after them. Ordering only: a unit that fails does not
+# keep monit from starting, and monit then reports it.
+[Unit]
+After=ssh.service webmin.service postfix.service

tests/README.md

@@ -70,9 +70,10 @@ container:
     keel inspect --output /root/emitted.yaml
     keel spec apply --system --spec /root/emitted.yaml   # 0 change(s)
 
-`inspect` cannot read the monitor's channels back (they are in
-`/etc/keel/monitor.json`, which it does not repeat), so the channel is
-added to the emitted file before it is applied, as its report says. And
+Before keel 0.15.1, `inspect` could not read the monitor's channels back
+(they are in `/etc/keel/monitor.json`), so the channel was added to the
+emitted file before it was applied; keel 0.15.1 reads them back, tokens
+by file (keel#60), and the emitted file applies as it is. And
 Monit's `/etc/keel/monit/keel-manifest.conf` checks sshd, webmin and
 postfix and nothing else. Core has no web shell: TurnKey removed
 shellinabox in 18.0, so the manifest does not declare the `webshell` of

tests/package.bats

@@ -53,10 +53,43 @@ print(eval(sys.argv[2]))' "$PACKAGE_DIR/manifest.yaml" "$1"
     [[ "$output" == "-rw-r--r-- root/root "* ]]
 }
 
-@test "it installs nothing but the manifest and its documentation" {
+@test "it installs the manifest, monit's ordering and its documentation" {
     run bash -c "dpkg-deb -c '$DEB' | awk '{print \$6}' | grep -v '/\$' | sort"
     [ "$status" -eq 0 ]
-    [ "$output" = $'./usr/share/doc/keel-core/changelog.gz\n./usr/share/doc/keel-core/copyright\n./usr/share/keel/appliances/core.yaml' ]
+    [ "$output" = $'./usr/lib/systemd/system/monit.service.d/keel-core.conf\n./usr/share/doc/keel-core/changelog.gz\n./usr/share/doc/keel-core/copyright\n./usr/share/keel/appliances/core.yaml' ]
+}
+
+# keel#61: monit's first cycle after boot found webmin and postfix
+# inactive, because monit started beside them; its checks are the units
+# of the manifest's processes, so monit starts after those units
+@test "monit starts after the units of Core's processes, and only that" {
+    dpkg-deb -x "$DEB" "$BUILD/root"
+    dropin="$BUILD/root/usr/lib/systemd/system/monit.service.d/keel-core.conf"
+    run grep -v '^#' "$dropin"
+    [ "$status" -eq 0 ]
+    [ "$output" = $'[Unit]\nAfter=ssh.service webmin.service postfix.service' ]
+    units=$(manifest '" ".join(p["unit"] for p in m["processes"])')
+    [ "$(sed -n 's/^After=//p' "$dropin")" = "$units" ]
+}
+
+@test "monit's ordering is a plain file of mode 0644, owned by root" {
+    run bash -c "dpkg-deb -c '$DEB' | grep ' ./usr/lib/systemd/system/monit.service.d/keel-core.conf$'"
+    [ "$status" -eq 0 ]
+    [[ "$output" == "-rw-r--r-- root/root "* ]]
+}
+
+@test "systemd reads the ordering without a complaint" {
+    if ! command -v systemd-analyze >/dev/null; then
+        skip "systemd-analyze is not installed"
+    fi
+    dpkg-deb -x "$DEB" "$BUILD/root"
+    mkdir -p "$BUILD/units"
+    printf '[Service]\nExecStart=/bin/true\n' > "$BUILD/units/monit.service"
+    mkdir -p "$BUILD/units/monit.service.d"
+    cp "$BUILD/root/usr/lib/systemd/system/monit.service.d/keel-core.conf" \
+        "$BUILD/units/monit.service.d/"
+    run systemd-analyze verify --man=no "$BUILD/units/monit.service"
+    [[ "$output" != *"keel-core.conf"* ]]
 }
 
 # the manifest: the Core table of decision 0041