feat(ha): add NixOS modules for DRBD+XFS+LIO+Corosync+Pacemaker HA stack

All 7 acceptance tests pass on live NixOS 25.11 VMs (VMIDs 200/201 on
pve1).  Failover completes in ~5 s with data integrity verified.

modules/ha/pacemaker-stack.nix — fixes four NixOS-specific breakages:
  - systemd StateDirectory resets /var/lib/pacemaker to root:root; removed
    and replaced with ExecStartPre to create/chown dirs as hacluster
  - HA_SBIN_DIR points to a non-existent Nix store path; overridden to
    /run/current-system/sw/bin so crm_master resolves correctly
  - OCF agents need an explicit broad PATH (iproute2, util-linux, xfsprogs,
    drbd, bash, etc.) — NixOS services have no implicit PATH
  - FUSER=true bypasses the psmisc fuser check_binary call in the
    Filesystem OCF agent (psmisc not installed on minimal hosts)

modules/ha/iscsi-target.nix — LIO iSCSI target via targetctl with a
Python/rtslib_fb ExecStop that explicitly clears the kernel LIO state
(not just saves JSON), so the XFS backing store's file descriptor is
released before umount — preventing EBUSY stop timeouts on failover.
Includes an empty-config guard so the secondary node never overwrites
the primary's saveconfig.json with an empty one.

test-lab/ha/common.nix — updated to import both modules, use fencing
dont-care (no STONITH in test lab), omit LVM handlers (non-existent on
NixOS paths), and merge repeated services/networking attr sets to satisfy
statix W20.  test-lab/ha/acceptance-tests.sh — final v4 with crm_standby
fix (pacemaker 3.x API).

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HaH1cSGvhogRP5ExoF6nD8
This commit is contained in:
2026-07-27 09:13:55 +10:00
co-authored by Claude Sonnet 4.6
parent 23634134f0
commit 724d9a45af
4 changed files with 454 additions and 188 deletions
+99
View File
@@ -0,0 +1,99 @@
# LIO iSCSI target service (targetctl) for NixOS HA clusters.
#
# Provides the targetctl.service that saves/restores LIO configuration from
# /etc/target/saveconfig.json. Pacemaker manages this service via its
# systemd resource agent (class="systemd" type="targetctl").
#
# Why ExecStop is not simply "targetctl save":
# targetctl save writes the LIO config to JSON but does NOT remove the LIO
# target from the kernel's configfs. As a result, any fileio backing store
# that LIO has open (e.g. iscsi-lun.img on an XFS-over-DRBD filesystem)
# stays referenced in the kernel. The subsequent XFS umount from the
# Filesystem OCF resource then returns EBUSY and either hangs for the full
# op-stop timeout or fails outright, blocking the entire failover.
#
# The ExecStop script here additionally tears down the kernel LIO state
# via rtslib_fb after saving, so the backing-store file descriptor is
# released and umount succeeds immediately.
#
# Empty-config guard:
# The save step is skipped when no iSCSI targets are currently active.
# This prevents the secondary node (where LIO was never started) from
# overwriting a valid saveconfig.json with an empty one when Pacemaker
# stops the iscsi-target resource as part of a failover or cleanup.
{ pkgs, ... }:
let
python3 = pkgs.python3.withPackages (ps: [ ps.rtslib-fb ]);
targetctl = "${pkgs.targetcli-fb}/bin/targetctl";
targetctlStop = pkgs.writeScript "targetctl-stop" ''
#!${python3}/bin/python3
import subprocess, sys
import rtslib_fb
root = rtslib_fb.RTSRoot()
targets = list(root.targets)
if targets:
subprocess.run(
["${targetctl}", "save", "/etc/target/saveconfig.json"],
capture_output=True,
)
print(f"saved {len(targets)} iSCSI target(s)")
else:
print("no active LIO targets saveconfig.json unchanged")
for target in targets:
try:
for tpg in list(target.tpgs):
tpg.enable = False
target.delete()
except Exception as e:
print(f"warn (target): {e}", file=sys.stderr)
for so in list(root.storage_objects):
try:
so.delete()
except Exception as e:
print(f"warn (backstore): {e}", file=sys.stderr)
print("LIO kernel target cleared")
'';
in
{
boot.kernelModules = [
"target_core_mod"
"iscsi_target_mod"
"target_core_file"
"target_core_pscsi"
"target_core_user"
"configfs"
];
systemd = {
mounts = [{
where = "/sys/kernel/config";
what = "configfs";
type = "configfs";
wantedBy = [ "multi-user.target" ];
before = [ "targetctl.service" ];
}];
services.targetctl = {
description = "LIO iSCSI target config save/restore";
wantedBy = [ "multi-user.target" ];
after = [ "sys-kernel-config.mount" "network.target" ];
requires = [ "sys-kernel-config.mount" ];
serviceConfig = {
Type = "oneshot";
RemainAfterExit = true;
ExecStart = "${targetctl} restore /etc/target/saveconfig.json";
ExecStop = "${targetctlStop}";
};
unitConfig.ConditionFileNotEmpty = "/etc/target/saveconfig.json";
};
tmpfiles.rules = [
"d /etc/target 0750 root root -"
"f /etc/target/saveconfig.json 0640 root root -"
];
};
environment.systemPackages = [ pkgs.targetcli-fb ];
}
+94
View File
@@ -0,0 +1,94 @@
# Pacemaker + Corosync HA stack for NixOS with known-good workarounds.
#
# Issues fixed here (confirmed through live testing on NixOS 25.11):
#
# 1. StateDirectory ownership reset: systemd's StateDirectory=pacemaker
# creates /var/lib/pacemaker owned root:root. pacemaker-based (the CIB
# daemon) runs as the hacluster user and calls pcmk__daemon_can_write,
# which requires the CIB directory to be owned by hacluster or be
# group-writable by haclient. Workaround: remove StateDirectory and let
# ExecStartPre create every required subdirectory with correct ownership.
#
# 2. HA_SBIN_DIR wrong path: ocf-shellfuncs sets HA_SBIN_DIR to the Nix
# store path of the resource-agents derivation's /sbin, which doesn't
# exist. The DRBD OCF agent uses ${HA_SBIN_DIR}/crm_master, so it exits
# 127 without this override. Fix: export HA_SBIN_DIR=/run/current-system/sw/bin.
#
# 3. Broad PATH for OCF agents: the resource executor (pacemaker-execd) runs
# OCF agent scripts as children. NixOS provides no implicit PATH for
# system services; without an explicit PATH the agents can't find ip, ss,
# mount, umount, drbdadm, etc.
#
# 4. FUSER=true: the Filesystem OCF agent calls check_binary $FUSER (default:
# fuser from psmisc), which is not installed. Setting FUSER=true makes
# check_binary succeed (true is always in PATH) and the subsequent
# "$FUSER -km $mountpoint" becomes a no-op. Pair with force_unmount=false
# on each Filesystem resource unless you want lazy unmount behaviour.
{ lib, pkgs, ... }:
let
ocfBinPath = lib.concatStringsSep ":" [
"${pkgs.iproute2}/bin"
"${pkgs.iproute2}/sbin"
"${pkgs.iputils}/bin"
"${pkgs.util-linux}/bin"
"${pkgs.util-linux}/sbin"
"${pkgs.gawk}/bin"
"${pkgs.gnugrep}/bin"
"${pkgs.gnused}/bin"
"${pkgs.coreutils}/bin"
"${pkgs.bash}/bin"
"${pkgs.procps}/bin"
"${pkgs.xfsprogs}/bin"
"${pkgs.drbd}/bin"
"${pkgs.python3}/bin"
"/run/current-system/sw/bin"
"/run/current-system/sw/sbin"
"/usr/local/sbin"
"/usr/local/bin"
"/usr/sbin"
"/usr/bin"
"/sbin"
"/bin"
];
# Single pre-start script: schemas symlink + directory ownership.
# Runs before pacemakerd so pacemaker-based finds hacluster-owned dirs.
preStartCmd = "${pkgs.bash}/bin/bash -c '"
+ "ln -sfn ${pkgs.pacemaker}/share/pacemaker /var/lib/pacemaker/schemas; "
+ "for d in /var/lib/pacemaker /var/lib/pacemaker/cib /var/lib/pacemaker/cores "
+ "/var/lib/pacemaker/pengine /var/lib/pacemaker/blackbox "
+ "/var/lib/pacemaker/hostcache; do "
+ "mkdir -p \"\\$d\" && chown hacluster:pacemaker \"\\$d\" && chmod 2770 \"\\$d\"; "
+ "done'";
ocfEnv = {
PATH = lib.mkForce ocfBinPath;
OCF_ROOT = "${pkgs.ocf-resource-agents}/usr/lib/ocf";
HA_SBIN_DIR = "/run/current-system/sw/bin";
FUSER = "true";
};
in
{
users.groups.haclient = { };
services.corosync.enable = true;
services.pacemaker.enable = true;
systemd.services = {
pacemaker = {
serviceConfig = {
StateDirectory = lib.mkForce "";
ExecStartPre = lib.mkBefore [ preStartCmd ];
};
environment = ocfEnv;
};
pacemaker-execd.environment = ocfEnv;
};
environment.systemPackages = with pkgs; [
corosync
pacemaker
ocf-resource-agents
];
}