[pve-devel] [PATCH v2 manager 5/7] api: ceph: add cmd-safety endpoint

[Aaron Lauterer]

Ceph provides us with several safety checks to verify that an action is
safe to perform. This endpoint provides means to acces them.
The actual mon commands are not exposed directly. Instead the two
actions "stop" and "destroy" are offered.

In case it is not okay to perform an action, Ceph provides a status
message explaining why. This message is part of the returned values.

For now there are the following checks for these services:

MON:
  - ok-to-stop
  - ok-to-rm
OSD:
  - ok-to-stop
  - safe-to-destroy
MDS:
  - ok-to-stop

Even though OSDs have a check if it is okay to destroy them, it is for
now not really usable in our workflow because it needs the OSD to be up
and running to return useful information. Our workflow in the GUI
currently is that the OSD needs to be stopped in order to destroy it.

There are no checks if the service actually exists. Ceph will report
back that it is safe to stop/destroy if the service does not exist.

Signed-off-by: Aaron Lauterer <a.lauterer at proxmox.com>
---
changes:
* remove repetitive endpoints for each service type in favor for a
  central one

 PVE/API2/Ceph.pm | 96 ++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 96 insertions(+)

diff --git a/PVE/API2/Ceph.pm b/PVE/API2/Ceph.pm
index 1e1b1edd..69ae746a 100644
--- a/PVE/API2/Ceph.pm
+++ b/PVE/API2/Ceph.pm
@@ -641,4 +641,100 @@ __PACKAGE__->register_method ({
 	return $res;
     }});
 
+__PACKAGE__->register_method ({
+    name => 'cmd_safety',
+    path => 'cmd-safety',
+    method => 'GET',
+    description => "Heuristical check if it is safe to perform an action.",
+    proxyto => 'node',
+    protected => 1,
+    permissions => {
+	check => ['perm', '/', [ 'Sys.audit' ]],
+    },
+    parameters => {
+	additionalProperties => 0,
+	properties => {
+	    node => get_standard_option('pve-node'),
+	    service => {
+		description => 'Service type',
+		type => 'string',
+		enum => ['osd', 'mon', 'mds'],
+	    },
+	    id => {
+		description => 'ID of the service',
+		type => 'string',
+	    },
+	    action => {
+		description => 'Action to check',
+		type => 'string',
+		enum => ['stop', 'destroy'],
+	    },
+	},
+    },
+    returns => {
+	type => 'object',
+	properties => {
+	   safe  => {
+		type => 'boolean',
+		description => 'If it is safe to run the command.',
+	    },
+	    status => {
+		type => 'string',
+		optional => 1,
+		description => 'Status message given by Ceph.'
+	    },
+	},
+    },
+    code => sub {
+	my ($param) = @_;
+
+	PVE::Ceph::Tools::check_ceph_inited();
+
+	my $id = $param->{id};
+	my $service = $param->{service};
+	my $action = $param->{action};
+
+	my $rados = PVE::RADOS->new();
+
+	my $supported_actions = {
+	    osd => {
+		stop => 'ok-to-stop',
+		destroy => 'safe-to-destroy',
+	    },
+	    mon => {
+		stop => 'ok-to-stop',
+		destroy => 'ok-to-rm',
+	    },
+	    mds => {
+		stop => 'ok-to-stop',
+	    },
+	};
+
+	die "Service does not support this action: ${service}: ${action}\n"
+	    if !$supported_actions->{$service}->{$action};
+
+	my $result = {
+	    safe => 0,
+	    status => '',
+	};
+
+	my $params = {
+	    prefix => "${service} $supported_actions->{$service}->{$action}",
+	    format => 'plain',
+	};
+	if ($service eq 'mon' && $action eq 'destroy') {
+	    $params->{id} = $id;
+	} else {
+	    $params->{ids} = [ $id ];
+	}
+
+	$result = $rados->mon_command($params, 1);
+	die $@ if $@;
+
+	$result->{safe} = $result->{return_code} == 0 ? 1 : 0;
+	$result->{status} = $result->{status_message};
+
+	return $result;
+    }});
+
 1;
-- 
2.30.2