Update to 2021-12-01 19:13

roles/pve/files/online_hook.pl

@@ -0,0 +1,111 @@
+#!/usr/bin/perl -w
+
+use strict;
+use warnings;
+use File::Basename;
+use JSON;
+use Logger::Syslog;
+
+$SIG{CHLD} = 'IGNORE';
+
+info( "GUEST HOOK: " . join( ' ', @ARGV ) );
+
+my $vmid = shift;
+my $phase = shift;
+
+if ( not -l '/etc/pve/local' ) {
+  error( "Can't find /etc/pve/local link" );
+  die;
+} elsif ( not defined $vmid or not defined $phase ) {
+  error( "Need to pass both vmid and phase arguments" );
+  die;
+} elsif ( $vmid !~ m/^\d+$/ ) {
+  error( "vmid must be only numerical" );
+  die;
+}
+
+# We must identify the local node
+my $local_node = basename( readlink( '/etc/pve/local' ) );
+
+if ( $phase eq 'pre-start' ) {
+  info( "Runing pre-start hook for guest $vmid" );
+} elsif ( $phase eq 'post-start' ) {
+  info( "Running post-start hook for guest $vmid" );
+  # A VM has just started. Let's check if it's already on the local node
+  # If it's an incoming live migration, it might still be running on another node
+  # All of this must run in the background because hooks are synchronous
+  my $pid = fork();
+  if ( $pid != 0 ) {
+    # main script can stop now, everything will run in a forked process
+    POSIX::_exit 0;
+  } elsif ( defined $pid ) {
+    # All those fh must be closed for the fork to be independant of its parent
+    close STDOUT;
+    close STDERR;
+    close STDIN;
+    POSIX::setsid();
+
+    my $i = 0;
+    my $guest_found = 0;
+
+    # We'll loop for up to 30min, which should be sufficient. If migration takes longer than that,
+    # something is probably wrong
+    LOOP: while ( $i lt 1800 ) {
+      # Here, we query the API for all the VM
+      my $resources = from_json( qx(pvesh get /cluster/resources --type=vm --output-format=json) );
+      # Then we loop through all the VM to find the one we're interested in
+      foreach my $vm ( @{$resources} ){
+        next if ( $vm->{id} !~ m{^(qemu|lxc)/$vmid$} );
+
+        # OK, we found the guest $vmid
+        info("Found guest $vmid, running on node " . $vm->{node});
+        $guest_found = 1;
+
+        # Is the guest running on local node ? If yes, it means migration is finished, and we
+        # can redirect IP failover and routing table
+        if ( $vm->{node} eq $local_node ) {
+
+          # pve-online use this env var to check if we must unplug/replug the WAN NIC
+          $ENV{PVE_GUEST_TYPE} = $1;
+
+          # And here we go !
+          qx(/bin/systemd-cat /usr/local/bin/pve-online --update-routes --update-gre --migrate-ipfo=$vmid);
+
+          # Update routing table of the other online nodes
+          my $nodes = from_json( qx(pvesh get /nodes --output-format=json) );
+          foreach my $node ( @{$nodes} ) {
+            if ( $node->{status} eq 'online' and $node->{node} ne $local_node ) {
+              info("Updating routing table of node $node->{node}");
+              qx(ssh -o ConnectTimeout=3 -l root $node->{node} /usr/local/bin/pve-online --update-routes);
+            }
+          }
+
+          # And we're done, stop looping
+          last LOOP;
+
+        # Guest is not running on the local node = migration is still running
+        # Wait a bit and start again
+        } else {
+          info( "Guest $vmid is still running on node " . $vm->{node} . " not yet on $local_node. Waiting a bit more for migration to finish" );
+          sleep 1;
+          next LOOP;
+        }
+      }
+
+      # We looped through all the guests and couldn't find the one we're looking for, nothing more we can do
+      if ( not $guest_found ) {
+        error( "No such guest with id $vmid" );
+        die;
+      }
+    }
+  }
+} elsif ( $phase eq 'pre-stop' ) {
+  info( "Running pre-stop hook" );
+} elsif ( $phase eq 'post-stop' ) {
+  info( "Running post-stop hook" );
+  # Just remove routes if needed
+  qx(/bin/systemd-cat /usr/local/bin/pve-online --update-routes)
+} else {
+  error( "Unknown hook phase : $phase" );
+  die;
+}