[pve-devel] [PATCH manager v2] Fix #1210: ceph: extend pveceph purge

[Alwin Antreich]

to clean service directories as well as disable and stop Ceph services.
Addtionally provide the option to remove crash and log information.

This patch is also in addtion to #2607, as the current cleanup doesn't
allow to re-configure Ceph, without manual steps during purge.

Signed-off-by: Alwin Antreich <a.antreich at proxmox.com>
---
v1 -> v2:
    * incorporate Thomas suggestions. Thanks.
	- add warning for failed ceph connection
	- use grep instead of map
	- change $ceph variable name to $service in purge methods

 PVE/CLI/pveceph.pm | 48 ++++++++++++++++++++++++++-----
 PVE/Ceph/Tools.pm  | 71 ++++++++++++++++++++++++++++++++++++++--------
 2 files changed, 100 insertions(+), 19 deletions(-)

diff --git a/PVE/CLI/pveceph.pm b/PVE/CLI/pveceph.pm
index 064ae545..448d3ec1 100755
--- a/PVE/CLI/pveceph.pm
+++ b/PVE/CLI/pveceph.pm
@@ -18,6 +18,7 @@ use PVE::Storage;
 use PVE::Tools qw(run_command);
 use PVE::JSONSchema qw(get_standard_option);
 use PVE::Ceph::Tools;
+use PVE::Ceph::Services;
 use PVE::API2::Ceph;
 use PVE::API2::Ceph::FS;
 use PVE::API2::Ceph::MDS;
@@ -49,25 +50,58 @@ __PACKAGE__->register_method ({
     parameters => {
 	additionalProperties => 0,
 	properties => {
+	    logs => {
+		description => 'Additionally purge Ceph logs, /var/log/ceph.',
+		type => 'boolean',
+		optional => 1,
+	    },
+	    crash => {
+		description => 'Additionally purge Ceph crash logs, /var/lib/ceph/crash.',
+		type => 'boolean',
+		optional => 1,
+	    },
 	},
     },
     returns => { type => 'null' },
     code => sub {
 	my ($param) = @_;
 
-	my $monstat;
+	my $message;
+	my $pools = [];
+	my $monstat = {};
+	my $mdsstat = {};
+	my $osdstat = [];
 
 	eval {
 	    my $rados = PVE::RADOS->new();
-	    my $monstat = $rados->mon_command({ prefix => 'mon_status' });
+	    $pools = PVE::Ceph::Tools::ls_pools(undef, $rados);
+	    $monstat = PVE::Ceph::Services::get_services_info('mon', undef, $rados);
+	    $mdsstat = PVE::Ceph::Services::get_services_info('mds', undef, $rados);
+	    $osdstat = $rados->mon_command({ prefix => 'osd metadata' });
 	};
-	my $err = $@;
+	warn "Could not connect: $@" if $@;
+
+	my $osd = grep { $_->{hostname} eq $nodename } @$osdstat;
+	my $mds = grep { $mdsstat->{$_}->{host} eq $nodename } keys %$mdsstat;
+	my $mon = grep { $monstat->{$_}->{host} eq $nodename } keys %$monstat;
+
+	# no pools = no data
+	$message .= "- remove pools, this will !!DESTROY DATA!!\n" if @$pools;
+	$message .= "- remove active OSD on $nodename\n" if $osd;
+	$message .= "- remove active MDS on $nodename\n" if $mds;
+	$message .= "- remove other MONs, $nodename is not the last MON\n"
+	    if scalar(keys %$monstat) > 1 && $mon;
+
+	# display all steps at once
+	die "Unable to purge Ceph!\n\nTo continue:\n$message" if $message;
 
-	die "detected running ceph services- unable to purge data\n"
-	    if !$err;
+	my $services = PVE::Ceph::Services::get_local_services();
+	$services->{mon} = $monstat if $mon;
+	$services->{crash}->{$nodename} = { direxists => 1 } if $param->{crash};
+	$services->{logs}->{$nodename} = { direxists => 1 } if $param->{logs};
 
-	# fixme: this is dangerous - should we really support this function?
-	PVE::Ceph::Tools::purge_all_ceph_files();
+	PVE::Ceph::Tools::purge_all_ceph_services($services);
+	PVE::Ceph::Tools::purge_all_ceph_files($services);
 
 	return undef;
     }});
diff --git a/PVE/Ceph/Tools.pm b/PVE/Ceph/Tools.pm
index e6225b78..acec746a 100644
--- a/PVE/Ceph/Tools.pm
+++ b/PVE/Ceph/Tools.pm
@@ -11,6 +11,8 @@ use JSON;
 use PVE::Tools qw(run_command dir_glob_foreach);
 use PVE::Cluster qw(cfs_read_file);
 use PVE::RADOS;
+use PVE::Ceph::Services;
+use PVE::CephConfig;
 
 my $ccname = 'ceph'; # ceph cluster name
 my $ceph_cfgdir = "/etc/ceph";
@@ -42,6 +44,7 @@ my $config_hash = {
     ceph_bootstrap_mds_keyring => $ceph_bootstrap_mds_keyring,
     ceph_mds_data_dir => $ceph_mds_data_dir,
     long_rados_timeout => 60,
+    ceph_cfgpath => $ceph_cfgpath,
 };
 
 sub get_local_version {
@@ -89,20 +92,64 @@ sub get_config {
 }
 
 sub purge_all_ceph_files {
-    # fixme: this is very dangerous - should we really support this function?
-
-    unlink $ceph_cfgpath;
-
-    unlink $pve_ceph_cfgpath;
-    unlink $pve_ckeyring_path;
-    unlink $pve_mon_key_path;
-
-    unlink $ceph_bootstrap_osd_keyring;
-    unlink $ceph_bootstrap_mds_keyring;
+    my ($services) = @_;
+    my $is_local_mon;
+    my $monlist = [ split(',', PVE::CephConfig::get_monaddr_list($pve_ceph_cfgpath)) ];
+
+    foreach my $service (keys %$services) {
+	my $type = $services->{$service};
+	next if (!%$type);
+
+	foreach my $name (keys %$type) {
+	    my $dir_exists = $type->{$name}->{direxists};
+
+	    $is_local_mon = grep($type->{$name}->{addr}, @$monlist)
+		if $service eq 'mon';
+
+	    my $path = "/var/lib/ceph/$service";
+	    $path = '/var/log/ceph' if $service eq 'logs';
+	    if ($dir_exists) {
+		my $err;
+		File::Path::remove_tree($path, {
+			keep_root => 1,
+			error => \$err,
+		    });
+		warn "Error removing path, '$path'\n" if @$err;
+	    }
+	}
+    }
 
-    system("rm -rf /var/lib/ceph/mon/ceph-*");
+    if (scalar @$monlist > 0 && !$is_local_mon) {
+	warn "Foreign MON address in ceph.conf. Keeping config & keyrings\n"
+    } else {
+	print "Removing config & keyring files\n";
+	foreach my $file (%$config_hash) {
+	    unlink $file if (-e $file);
+	}
+    }
+}
 
-    # remove osd?
+sub purge_all_ceph_services {
+    my ($services) = @_;
+
+    foreach my $service (keys %$services) {
+	my $type = $services->{$service};
+	next if (!%$type);
+
+	foreach my $name (keys %$type) {
+	    my $service_exists = $type->{$name}->{service};
+
+	    if ($service_exists) {
+		eval {
+		    PVE::Ceph::Services::ceph_service_cmd('disable', "$service.$name");
+		    PVE::Ceph::Services::ceph_service_cmd('stop', "$service.$name");
+		};
+		my $err = $@ if $@;
+		warn "Could not disable/stop ceph-$service\@$name, error: $err\n"
+		if $err;
+	    }
+	}
+    }
 }
 
 sub check_ceph_installed {
-- 
2.26.2