nixos/tests/pacemaker: init

nixos/doc/manual/from_md/release-notes/rl-2205.section.xml

@@ -280,6 +280,12 @@
           with many features.
         </para>
       </listitem>
+      <listitem>
+        <para>
+          <link xlink:href="https://clusterlabs.org/pacemaker/">pacemaker</link>
+          cluster resource manager
+        </para>
+      </listitem>
     </itemizedlist>
   </section>
   <section xml:id="sec-release-22.05-incompatibilities">

nixos/doc/manual/release-notes/rl-2205.section.md

@@ -81,6 +81,8 @@ In addition to numerous new and upgraded packages, this release has the followin
 
 - [blocky](https://0xerr0r.github.io/blocky/), fast and lightweight DNS proxy as ad-blocker for local network with many features.
 
+- [pacemaker](https://clusterlabs.org/pacemaker/) cluster resource manager
+
 <!-- To avoid merge conflicts, consider adding your item at an arbitrary place in the list instead. -->
 
 ## Backward Incompatibilities {#sec-release-22.05-incompatibilities}

nixos/tests/all-tests.nix

@@ -383,6 +383,7 @@ in
   os-prober = handleTestOn ["x86_64-linux"] ./os-prober.nix {};
   osrm-backend = handleTest ./osrm-backend.nix {};
   overlayfs = handleTest ./overlayfs.nix {};
+  pacemaker = handleTest ./pacemaker.nix {};
   packagekit = handleTest ./packagekit.nix {};
   pam-file-contents = handleTest ./pam/pam-file-contents.nix {};
   pam-oath-login = handleTest ./pam/pam-oath-login.nix {};

nixos/tests/pacemaker.nix

@@ -0,0 +1,110 @@
+import ./make-test-python.nix  ({ pkgs, lib, ... }: rec {
+  name = "pacemaker";
+  meta = with pkgs.lib.maintainers; {
+    maintainers = [ astro ];
+  };
+
+  nodes =
+    let
+      node = i: {
+        networking.interfaces.eth1.ipv4.addresses = [ {
+          address = "192.168.0.${toString i}";
+          prefixLength = 24;
+        } ];
+
+        services.corosync = {
+          enable = true;
+          clusterName = "zentralwerk-network";
+          nodelist = lib.imap (i: name: {
+            nodeid = i;
+            inherit name;
+            ring_addrs = [
+              (builtins.head nodes.${name}.networking.interfaces.eth1.ipv4.addresses).address
+            ];
+          }) (builtins.attrNames nodes);
+        };
+        environment.etc."corosync/authkey" = {
+          source = builtins.toFile "authkey"
+            # minimum length: 128 bytes
+            "testtesttesttesttesttesttesttesttesttesttesttesttesttesttesttesttesttesttesttesttesttesttesttesttesttesttesttesttesttesttesttest";
+          mode = "0400";
+        };
+
+        services.pacemaker.enable = true;
+
+        # used for pacemaker resource
+        systemd.services.ha-cat = {
+          description = "Highly available netcat";
+          serviceConfig.ExecStart = "${pkgs.netcat}/bin/nc -l discard";
+        };
+      };
+    in {
+      node1 = node 1;
+      node2 = node 2;
+      node3 = node 3;
+    };
+
+  # sets up pacemaker with resources configuration, then crashes a
+  # node and waits for service restart on another node
+  testScript =
+    let
+      resources = builtins.toFile "cib-resources.xml" ''
+        <resources>
+          <primitive id="cat" class="systemd" type="ha-cat">
+            <operations>
+              <op id="stop-cat" name="start" interval="0" timeout="1s"/>
+              <op id="start-cat" name="start" interval="0" timeout="1s"/>
+              <op id="monitor-cat" name="monitor" interval="1s" timeout="1s"/>
+            </operations>
+          </primitive>
+        </resources>
+      '';
+    in ''
+      import re
+      import time
+
+      start_all()
+
+      ${lib.concatMapStrings (node: ''
+        ${node}.wait_until_succeeds("corosync-quorumtool")
+        ${node}.wait_for_unit("pacemaker.service")
+      '') (builtins.attrNames nodes)}
+
+      # No STONITH device
+      node1.succeed("crm_attribute -t crm_config -n stonith-enabled -v false")
+      # Configure the cat resource
+      node1.succeed("cibadmin --replace --scope resources --xml-file ${resources}")
+
+      # wait until the service is started
+      while True:
+        output = node1.succeed("crm_resource -r cat --locate")
+        match = re.search("is running on: (.+)", output)
+        if match:
+          for machine in machines:
+            if machine.name == match.group(1):
+              current_node = machine
+          break
+        time.sleep(1)
+
+      current_node.log("Service running here!")
+      current_node.crash()
+
+      # pick another node that's still up
+      for machine in machines:
+        if machine.booted:
+          check_node = machine
+      # find where the service has been started next
+      while True:
+        output = check_node.succeed("crm_resource -r cat --locate")
+        match = re.search("is running on: (.+)", output)
+        # output will remain the old current_node until the crash is detected by pacemaker
+        if match and match.group(1) != current_node.name:
+          for machine in machines:
+            if machine.name == match.group(1):
+              next_node = machine
+          break
+        time.sleep(1)
+
+      next_node.log("Service migrated here!")
+  '';
+})

pkgs/misc/logging/pacemaker/default.nix

@@ -17,6 +17,7 @@
 , pam
 , pkg-config
 , python3
+, nixosTests
 
 # Pacemaker is compiled twice, once with forOCF = true to extract its
 # OCF definitions for use in the ocf-resource-agents derivation, then
@@ -87,11 +88,15 @@ stdenv.mkDerivation rec {
     rm -r $out/nix
   '';
 
+  passthru.tests = {
+    inherit (nixosTests) pacemaker;
+  };
+
   meta = with lib; {
     homepage = "https://clusterlabs.org/pacemaker/";
     description = "Pacemaker is an open source, high availability resource manager suitable for both small and large clusters.";
     license = licenses.gpl2Plus;
     platforms = platforms.linux;
-    maintainers = with maintainers; [ ryantm ];
+    maintainers = with maintainers; [ ryantm astro ];
   };
 }

pkgs/servers/corosync/default.nix

@@ -65,6 +65,10 @@ stdenv.mkDerivation rec {
       --prefix PATH ":" "$out/sbin:${libqb}/sbin"
   '';
 
+  passthru.tests = {
+    inherit (nixosTests) pacemaker;
+  };
+
   meta = {
     homepage = "http://corosync.org/";
     description = "A Group Communication System with features for implementing high availability within applications";