PVE/CLI/pve5to6.pm

   1 package PVE::CLI::pve5to6;
   2
   3 use strict;
   4 use warnings;
   5
   6 use PVE::API2::APT;
   7 use PVE::API2::Ceph;
   8 use PVE::API2::LXC;
   9 use PVE::API2::Qemu;
  10
  11 use PVE::Ceph::Tools;
  12 use PVE::Cluster;
  13 use PVE::Corosync;
  14 use PVE::INotify;
  15 use PVE::JSONSchema;
  16 use PVE::RPCEnvironment;
  17 use PVE::Storage;
  18 use PVE::Tools;
  19
  20 use Term::ANSIColor;
  21
  22 use PVE::CLIHandler;
  23
  24 use base qw(PVE::CLIHandler);
  25
  26 my $nodename = PVE::INotify::nodename();
  27
  28 sub setup_environment {
  29     PVE::RPCEnvironment->setup_default_cli_env();
  30 }
  31
  32 my $min_pve_rel = '5.4';
  33 my $min_pve_pkgrel = 2;
  34
  35 my $counters = {
  36     pass => 0,
  37     skip => 0,
  38     warn => 0,
  39     fail => 0,
  40 };
  41
  42 my $log_line = sub {
  43     my ($level, $line) = @_;
  44
  45     $counters->{$level}++ if defined($level) && defined($counters->{$level});
  46
  47     print uc($level), ': ' if defined($level);
  48     print "$line\n";
  49 };
  50
  51 sub log_pass {
  52     print color('green');
  53     $log_line->('pass', @_);
  54     print color('reset');
  55 }
  56
  57 sub log_info {
  58     $log_line->('info', @_);
  59 }
  60 sub log_skip {
  61     $log_line->('skip', @_);
  62 }
  63 sub log_warn {
  64     print color('yellow');
  65     $log_line->('warn', @_);
  66     print color('reset');
  67 }
  68 sub log_fail {
  69     print color('red');
  70     $log_line->('fail', @_);
  71     print color('reset');
  72 }
  73
  74 my $get_pkg = sub {
  75     my ($pkg) = @_;
  76
  77     my $versions = eval { PVE::API2::APT->versions({ node => $nodename }); };
  78
  79     if (!defined($versions)) {
  80         my $msg = "unable to retrieve package version information";
  81         $msg .= "- $@" if $@;
  82         log_fail("$msg");
  83         return undef;
  84     }
  85
  86     my $pkgs = [ grep { $_->{Package} eq $pkg } @$versions ];
  87     if (!defined $pkgs || $pkgs == 0) {
  88         log_fail("unable to determine installed $pkg version.");
  89         return undef;
  90     } else {
  91         return $pkgs->[0];
  92     }
  93 };
  94
  95 sub check_pve_packages {
  96     print "CHECKING VERSION INFORMATION FOR PVE PACKAGES\n";
  97
  98     print "\nChecking for package updates..\n";
  99     my $updates = eval { PVE::API2::APT->list_updates({ node => $nodename }); };
 100     if (!defined($updates)) {
 101         log_warn("$@") if $@;
 102         log_fail("unable to retrieve list of package updates!");
 103     } elsif (@$updates > 0) {
 104         my $pkgs = join(', ', map { $_->{Package} } @$updates);
 105         log_warn("updates for the following packages are available: $pkgs");
 106     } else {
 107         log_pass("all packages uptodate");
 108     }
 109
 110     print "\nChecking proxmox-ve package version..\n";
 111     if (defined(my $proxmox_ve = $get_pkg->('proxmox-ve'))) {
 112         my $min_pve_ver = "$min_pve_rel-$min_pve_pkgrel";
 113
 114         if ($proxmox_ve->{OldVersion} =~ m/^$min_pve_rel-(\d+)/ && $1 >= $min_pve_pkgrel) {
 115             log_pass("proxmox-ve package has version >= $min_pve_ver");
 116         } else {
 117             log_fail("proxmox-ve package is too old, please upgrade to >= $min_pve_ver!");
 118         }
 119     }
 120 }
 121
 122 sub check_storage_health {
 123     print "\nCHECKING CONFIGURED STORAGES\n\n";
 124     my $cfg = PVE::Storage::config();
 125
 126     my $ctime = time();
 127
 128     my $info = PVE::Storage::storage_info($cfg);
 129
 130     foreach my $storeid (keys %$info) {
 131         my $d = $info->{$storeid};
 132         if ($d->{enabled}) {
 133             if ($d->{active}) {
 134                 log_pass("storage '$storeid' enabled and active.");
 135             } else {
 136                 log_warn("storage '$storeid' enabled but not active!");
 137             }
 138         } else {
 139             log_skip("storage '$storeid' disabled.");
 140         }
 141     }
 142 }
 143
 144 sub check_cluster_corosync {
 145     print "\nCHECKING CLUSTER HEALTH/SETTINGS\n\n";
 146
 147     if (!PVE::Corosync::check_conf_exists(1)) {
 148         log_skip("standalone node.");
 149         return;
 150     }
 151
 152     if (PVE::Cluster::check_cfs_quorum(1)) {
 153         log_pass("Cluster is quorate.");
 154     } else {
 155         log_fail("Cluster lost quorum!");
 156     }
 157
 158     my $conf = PVE::Cluster::cfs_read_file('corosync.conf');
 159     my $conf_nodelist = PVE::Corosync::nodelist($conf);
 160
 161     if (!defined($conf_nodelist)) {
 162         log_fail("unable to retrieve nodelist from corosync.conf");
 163     } elsif (grep { $conf_nodelist->{$_}->{quorum_votes} != 1 } keys %$conf_nodelist) {
 164         log_warn("non-default quorum_votes distribution detected!");
 165     }
 166
 167     my $cfs_nodelist = PVE::Cluster::get_clinfo()->{nodelist};
 168     my $offline_nodes = grep { $cfs_nodelist->{$_}->{online} != 1 } keys %$cfs_nodelist;
 169     if ($offline_nodes > 0) {
 170         log_fail("$offline_nodes nodes are offline!");
 171     }
 172
 173     my $conf_nodelist_count = scalar(keys %$conf_nodelist);
 174     my $cfs_nodelist_count = scalar(keys %$cfs_nodelist);
 175     log_warn("cluster consists of less than three nodes!")
 176         if $conf_nodelist_count < 3;
 177
 178     log_fail("corosync.conf ($conf_nodelist_count) and pmxcfs ($cfs_nodelist_count) don't agree about size of nodelist.")
 179         if $conf_nodelist_count != $cfs_nodelist_count;
 180
 181     foreach my $cs_node (keys %$conf_nodelist) {
 182         my $entry = $conf_nodelist->{$cs_node};
 183         log_fail("No name entry for node '$cs_node' in corosync.conf.")
 184             if !defined($entry->{name});
 185         log_fail("No nodeid configured for node '$cs_node' in corosync.conf.")
 186             if !defined($entry->{nodeid});
 187
 188         my $verify_ring_ip = sub {
 189             my $key = shift;
 190             my $ring = $entry->{$key};
 191             if (defined($ring) && !PVE::JSONSchema::pve_verify_ip($ring, 1)) {
 192                 log_fail("$key '$ring' of node '$cs_node' is not an IP address, consider replacing it with the currently resolved IP address.");
 193             }
 194         };
 195         $verify_ring_ip->('ring0_addr');
 196         $verify_ring_ip->('ring1_addr');
 197     }
 198
 199     my $totem = $conf->{main}->{totem};
 200
 201     my $transport = $totem->{transport};
 202     if (defined($transport)) {
 203         log_fail("Corosync transport expliclitly set to '$transport' instead of implicit default!");
 204     }
 205
 206     if ((!defined($totem->{secauth}) || $totem->{secauth} ne 'on') && (!defined($totem->{crypto_cipher}) || $totem->{crypto_cipher} eq 'none')) {
 207         log_fail("Corosync authentication/encryption is not explicitly enabled (secauth / crypto_cipher / crypto_hash)!");
 208     }
 209
 210     if (defined($totem->{crypto_cipher}) && $totem->{crypto_cipher} eq '3des') {
 211         log_fail("Corosync encryption cipher set to '3des', no longer supported in Corosync 3.x!");
 212     }
 213
 214     my $prefix_info = sub { my $line = shift; log_info("$line"); };
 215     eval {
 216         print "\n";
 217         log_info("Printing detailed cluster status..");
 218         PVE::Tools::run_command(['corosync-quorumtool', '-siH'], outfunc => $prefix_info, errfunc => $prefix_info);
 219     };
 220
 221     print "\nCHECKING INSTALLED COROSYNC VERSION\n\n";
 222     if (defined(my $corosync = $get_pkg->('corosync'))) {
 223         if ($corosync->{OldVersion} =~ m/^2\./) {
 224             log_fail("corosync 2.x installed, cluster-wide upgrade to 3.x needed!");
 225         } elsif ($corosync->{OldVersion} =~ m/^3\./) {
 226             log_pass("corosync 3.x installed.");
 227         } else {
 228             log_fail("unexpected corosync version installed: $corosync->{OldVersion}!");
 229         }
 230     }
 231 }
 232
 233 sub check_ceph {
 234     print "\nCHECKING HYPER-CONVERGED CEPH STATUS\n\n";
 235
 236     if (PVE::Ceph::Tools::check_ceph_inited(1)) {
 237         log_info("hyper-converged ceph setup detected!");
 238     } else {
 239         log_skip("no hyper-converged ceph setup detected!");
 240         return;
 241     }
 242
 243     log_info("getting Ceph status/health information..");
 244     my $ceph_status = eval { PVE::API2::Ceph->status({ node => $nodename }); };
 245     my $osd_flags = eval { PVE::API2::Ceph->get_flags({ node => $nodename }); };
 246     my $noout;
 247     $noout = $osd_flags =~ m/noout/ if $osd_flags;
 248
 249     if (!$ceph_status || !$ceph_status->{health}) {
 250         log_fail("unable to determine Ceph status!");
 251     } else {
 252         my $ceph_health = $ceph_status->{health}->{status};
 253         if (!$ceph_health) {
 254             log_fail("unable to determine Ceph health!");
 255         } elsif ($ceph_health eq 'HEALTH_OK') {
 256             log_pass("Ceph health reported as 'HEALTH_OK'.");
 257         } elsif ($ceph_health eq 'HEALTH_WARN' && $noout && (keys %{$ceph_status->{health}->{checks}} == 1)) {
 258                 log_pass("Ceph health reported as 'HEALTH_WARN' with a single failing check and 'noout' flag set.");
 259         } else {
 260                 log_warn("Ceph health reported as '$ceph_health'");
 261         }
 262     }
 263
 264     log_info("getting Ceph OSD flags..");
 265     eval {
 266         if (!$osd_flags) {
 267             log_fail("unable to get Ceph OSD flags!");
 268         } else {
 269             if ($osd_flags =~ m/recovery_deletes/ && $osd_flags =~ m/purged_snapdirs/) {
 270                 log_pass("all PGs have been scrubbed at least once while running Ceph Luminous.");
 271             } else {
 272                 log_fail("missing 'recovery_deletes' and/or 'purged_snapdirs' flag, scrub of all PGs required before upgrading to Nautilus!");
 273             }
 274             if ($noout) {
 275                 log_pass("noout flag set to prevent rebalancing during cluster-wide upgrades.");
 276             }  else {
 277                 log_warn("noout flag not set - recommended to prevent rebalancing during upgrades.");
 278             }
 279         }
 280     };
 281
 282     log_info("getting Ceph daemon versions..");
 283     my $ceph_versions = eval { PVE::Ceph::Tools::get_cluster_versions(undef, 1); };
 284     if (!$ceph_versions) {
 285         log_fail("unable to determine Ceph daemon versions!");
 286     } else {
 287         my $services = [
 288             { 'key' => 'mon', 'name' => 'monitor' },
 289             { 'key' => 'mgr', 'name' => 'manager' },
 290             { 'key' => 'mds', 'name' => 'MDS' },
 291             { 'key' => 'osd', 'name' => 'OSD' },
 292         ];
 293
 294         foreach my $service (@$services) {
 295             my $name = $service->{name};
 296             if (my $service_versions = $ceph_versions->{$service->{key}}) {
 297                 if (keys %$service_versions == 0) {
 298                     log_skip("no running instances detected for daemon type $name.");
 299                 } elsif (keys %$service_versions == 1) {
 300                     log_pass("single running version detected for daemon type $name.");
 301                 } else {
 302                     log_warn("multiple running versions detected for daemon type $name!");
 303                 }
 304             } else {
 305                 log_skip("unable to determine versions of running Ceph $name instances.");
 306             }
 307         }
 308
 309         my $overall_versions = $ceph_versions->{overall};
 310         if (!$overall_versions) {
 311             log_warn("unable to determine overall Ceph daemon versions!");
 312         } elsif (keys %$overall_versions == 1) {
 313             log_pass("single running overall version detected for all Ceph daemon types.");
 314         } else {
 315             log_warn("overall version mismatch detected, check 'ceph versions' output for details!");
 316         }
 317     }
 318 }
 319
 320 sub check_misc {
 321     print "\nMISCELLANEOUS CHECKS\n\n";
 322     my $ssh_config = eval { PVE::Tools::file_get_contents('/root/.ssh/config') };
 323     if (defined($ssh_config)) {
 324         log_fail("Unsupported SSH Cipher configured for root in /root/.ssh/config: $1")
 325             if $ssh_config =~ /^Ciphers .*(blowfish|arcfour|3des).*$/m;
 326     } else {
 327         log_skip("No SSH config file found.");
 328     }
 329
 330     my $root_free = PVE::Tools::df('/', 10);
 331     log_warn("Less than 2G free space on root file system.")
 332         if defined($root_free) && $root_free->{avail} < 2*1024*1024*1024;
 333
 334     my $running_guests = 0;
 335     my $vms = eval { PVE::API2::Qemu->vmlist({ node => $nodename }) };
 336     log_warn("Failed to retrieve information about this node's VMs - $@") if $@;
 337     $running_guests += grep { $_->{status} eq 'running' } @$vms
 338         if defined($vms);
 339     my $cts = eval { PVE::API2::LXC->vmlist({ node => $nodename }) };
 340     log_warn("Failed to retrieve information about this node's CTs - $@") if $@;
 341     $running_guests += grep { $_->{status} eq 'running' } @$cts
 342         if defined($cts);
 343     log_warn("$running_guests running guests detected - consider migrating/stopping them.")
 344         if $running_guests > 0;
 345 }
 346
 347 __PACKAGE__->register_method ({
 348     name => 'checklist',
 349     path => 'checklist',
 350     method => 'GET',
 351     description => 'Check (pre-/post-)upgrade conditions.',
 352     parameters => {
 353         additionalProperties => 0,
 354         properties => {
 355         },
 356     },
 357     returns => { type => 'null' },
 358     code => sub {
 359         my ($param) = @_;
 360
 361         check_pve_packages();
 362         check_cluster_corosync();
 363         check_ceph();
 364         check_storage_health();
 365         check_misc();
 366
 367         print "\n\nSUMMARY:\n";
 368         print colored("PASSED: $counters->{pass}\n", 'green');
 369         print "SKIPPED: $counters->{skip}\n";
 370         print colored("WARNINGS: $counters->{warn}\n", 'yellow');
 371         print colored("FAILURES: $counters->{fail}\n", 'red');
 372
 373         print colored("\nATTENTION: Please check the output for detailed information!\n", 'red')
 374             if ($counters->{warn} > 0 || $counters->{fail} > 0);
 375
 376         return undef;
 377     }});
 378
 379 our $cmddef = {
 380     checklist => [ __PACKAGE__, 'checklist', [], {}],
 381 };
 382
 383 1;