nixos/ceph: use ConditionPathExists to delay ceph daemon start

This prevents services to be started before they're initialized, and
renders the `systemd.targets.ceph.wantedBy = lib.mkForce [];` hack in
the vm tests obsolete - The config now starts up ceph after a reboot,
too.

Let's take advantage of that, crash all VMs, and boot them up again.
This commit is contained in:
Florian Klink 2019-11-09 16:05:58 +01:00
parent 67e0777f62
commit ffd0060869
3 changed files with 27 additions and 32 deletions

View file

@ -14,7 +14,9 @@ let
{ "ceph-${daemonType}-${daemonId}" = makeService daemonType daemonId cfg.global.clusterName pkgs.ceph; })
daemonIds));
makeService = (daemonType: daemonId: clusterName: ceph: {
makeService = (daemonType: daemonId: clusterName: ceph:
let
stateDirectory = "ceph/${if daemonType == "rgw" then "radosgw" else daemonType}/${clusterName}-${daemonId}"; in {
enable = true;
description = "Ceph ${builtins.replaceStrings lowerChars upperChars daemonType} daemon ${daemonId}";
after = [ "network-online.target" "time-sync.target" ] ++ optional (daemonType == "osd") "ceph-mon.target";
@ -24,6 +26,9 @@ let
path = [ pkgs.getopt ];
# Don't start services that are not yet initialized
unitConfig.ConditionPathExists = "/var/lib/${stateDirectory}/keyring";
serviceConfig = {
LimitNOFILE = 1048576;
LimitNPROC = 1048576;
@ -36,7 +41,7 @@ let
Restart = "on-failure";
StartLimitBurst = "5";
StartLimitInterval = "30min";
StateDirectory = "ceph/${if daemonType == "rgw" then "radosgw" else daemonType}/${clusterName}-${daemonId}";
StateDirectory = stateDirectory;
User = "ceph";
Group = if daemonType == "osd" then "disk" else "ceph";
ExecStart = ''${ceph.out}/bin/${if daemonType == "rgw" then "radosgw" else "ceph-${daemonType}"} \
@ -60,6 +65,7 @@ let
partOf = [ "ceph.target" ];
wantedBy = [ "ceph.target" ];
before = [ "ceph.target" ];
unitConfig.StopWhenUnneeded = true;
};
}
);
@ -389,9 +395,12 @@ in
systemd.targets = let
targets = [
{ ceph = { description = "Ceph target allowing to start/stop all ceph service instances at once";
wantedBy = [ "multi-user.target" ]; }; }
] ++ optional cfg.mon.enable (makeTarget "mon")
{ ceph = {
description = "Ceph target allowing to start/stop all ceph service instances at once";
wantedBy = [ "multi-user.target" ];
unitConfig.StopWhenUnneeded = true;
}; } ]
++ optional cfg.mon.enable (makeTarget "mon")
++ optional cfg.mds.enable (makeTarget "mds")
++ optional cfg.osd.enable (makeTarget "osd")
++ optional cfg.rgw.enable (makeTarget "rgw")

View file

@ -49,9 +49,6 @@ let
boot.kernelModules = [ "xfs" ];
services.ceph = cephConfig;
# So that we don't have to battle systemd when bootstraping
systemd.targets.ceph.wantedBy = lib.mkForce [];
};
networkMonA = {
@ -191,22 +188,17 @@ let
"ceph osd pool delete multi-node-other-test multi-node-other-test --yes-i-really-really-mean-it"
);
# As we disable the target in the config, we still want to test that it works as intended
$osd0->mustSucceed("systemctl stop ceph-osd-${cfg.osd0.name}");
$osd1->mustSucceed("systemctl stop ceph-osd-${cfg.osd1.name}");
$monA->mustSucceed(
"systemctl stop ceph-mgr-${cfg.monA.name}",
"systemctl stop ceph-mon-${cfg.monA.name}"
);
# Shut down ceph on all machines in a very unpolite way
$monA->crash;
$osd0->crash;
$osd1->crash;
$monA->succeed("systemctl start ceph.target");
$monA->waitForUnit("ceph-mon-${cfg.monA.name}");
$monA->waitForUnit("ceph-mgr-${cfg.monA.name}");
$osd0->succeed("systemctl start ceph.target");
$osd0->waitForUnit("ceph-osd-${cfg.osd0.name}");
$osd1->succeed("systemctl start ceph.target");
$osd1->waitForUnit("ceph-osd-${cfg.osd1.name}");
# Start it up
$osd0->start;
$osd1->start;
$monA->start;
# Ensure the cluster comes back up again
$monA->succeed("ceph -s | grep 'mon: 1 daemons'");
$monA->waitUntilSucceeds("ceph -s | grep 'quorum ${cfg.monA.name}'");
$monA->waitUntilSucceeds("ceph osd stat | grep -e '2 osds: 2 up[^,]*, 2 in'");

View file

@ -46,9 +46,6 @@ let
boot.kernelModules = [ "xfs" ];
services.ceph = cephConfig;
# So that we don't have to battle systemd when bootstraping
systemd.targets.ceph.wantedBy = lib.mkForce [];
};
networkMonA = {
@ -151,20 +148,17 @@ let
"ceph osd pool delete single-node-other-test single-node-other-test --yes-i-really-really-mean-it"
);
# As we disable the target in the config, we still want to test that it works as intended
$monA->mustSucceed(
"systemctl stop ceph-osd-${cfg.osd0.name}",
"systemctl stop ceph-osd-${cfg.osd1.name}",
"systemctl stop ceph-mgr-${cfg.monA.name}",
"systemctl stop ceph-mon-${cfg.monA.name}"
);
# Shut down ceph by stopping ceph.target.
$monA->mustSucceed("systemctl stop ceph.target");
# Start it up
$monA->succeed("systemctl start ceph.target");
$monA->waitForUnit("ceph-mon-${cfg.monA.name}");
$monA->waitForUnit("ceph-mgr-${cfg.monA.name}");
$monA->waitForUnit("ceph-osd-${cfg.osd0.name}");
$monA->waitForUnit("ceph-osd-${cfg.osd1.name}");
# Ensure the cluster comes back up again
$monA->succeed("ceph -s | grep 'mon: 1 daemons'");
$monA->waitUntilSucceeds("ceph -s | grep 'quorum ${cfg.monA.name}'");
$monA->waitUntilSucceeds("ceph osd stat | grep -e '2 osds: 2 up[^,]*, 2 in'");