Looking for a good reason why the Puppet provider decides to remove the start and stop operations.

provider code.rb

    if @property_hash[:existing_resource] == :false || force_reinstall == :true
       [...]
    else
      if @property_hash[:promotable] == :false && @property_hash[:existing_promotable] == :true
        self.class.run_command_in_cib([command(:pcs), pcs_subcommand, 'delete', '--force', "ms_#{@property_hash[:name]}"], @resource[:cib])
      end
      @property_hash[:existing_operations].reject { |op| @property_hash[:operations].include?(op) }.each do |o|
        cmd = [command(:pcs), pcs_subcommand, 'op', 'remove', (@property_hash[:name]).to_s]
        cmd << o.keys.first.to_s
        o.values.first.each_pair do |k, v|
          cmd << "#{k}=#{v}"
        end
        self.class.run_command_in_cib(cmd, @resource[:cib])
      end
      cmd = [command(:pcs), pcs_subcommand, 'update', (@property_hash[:name]).to_s]
      cmd += parameters unless parameters.nil?
      cmd += operations unless operations.nil?
      cmd += utilization unless utilization.nil?
      cmd += metadatas unless metadatas.nil?
      self.class.run_command_in_cib(cmd, @resource[:cib])
      if @property_hash[:promotable] == :true
        cmd = [command(:pcs), pcs_subcommand, 'update', "ms_#{@property_hash[:name]}"]
        unless @property_hash[:ms_metadata].empty? && @property_hash[:existing_ms_metadata].empty?
          cmd << 'meta'
          @property_hash[:ms_metadata].each_pair do |k, v|
            cmd << "#{k}=#{v}"
          end
          @property_hash[:existing_ms_metadata].keys.reject { |key| @property_hash[:ms_metadata].key?(key) }.each do |k|
            cmd << "#{k}="
          end
        end
        self.class.run_command_in_cib(cmd, @resource[:cib])
      end
    end

Puppet debug

Debug: Cs_primitive[duncan_vip](provider=pcs): {
  :primitive_class=>"ocf",
  :primitive_type=>"IPaddr2",
  :provided_by=>"heartbeat",
  :name=>:duncan_vip,
  :ensure=>:present,
  :provider=>:pcs,
  :parameters=>{
    "cidr_netmask"=>"24",
    "ip"=>"172.16.210.101"},
  :operations=>[
    {"monitor"=>{"interval"=>"10s"}},
    {"start"=>{"interval"=>"0s", "timeout"=>"20s"}},
    {"stop"=>{"interval"=>"0s", "timeout"=>"20s"}}
  ],
  :utilization=>{},
  :metadata=>{},
  :ms_metadata=>{},
  :promotable=>:false,
  :existing_resource=>:true,
  :existing_primitive_class=>"ocf",
  :existing_primitive_type=>"IPaddr2",
  :existing_promotable=>:false,
  :existing_provided_by=>"heartbeat",
  :existing_metadata=>{},
  :existing_ms_metadata=>{},
  :existing_operations=>[
    {"monitor"=>{"interval"=>"10s"}},
    {"start"=>{"interval"=>"0s", "timeout"=>"20s"}},
    {"stop"=>{"interval"=>"0s", "timeout"=>"20s"}}
  ]
}
Notice: /Stage[main]/Main/Cs_primitive[duncan_vip]/operations: 2 removed: start (interval=0s timeout=20s) stop (interval=0s timeout=20s) / 1 kept
Debug: Puppet::Type::Cs_primitive::ProviderPcs: Executing ["/usr/sbin/pcs", "resource", "op", "remove", "duncan_vip", "start", "interval=0s", "timeout=20s"] in the CIB
Debug: Executing: '/usr/sbin/pcs resource op remove duncan_vip start interval=0s timeout=20s'
Debug: Puppet::Type::Cs_primitive::ProviderPcs: Executing ["/usr/sbin/pcs", "resource", "op", "remove", "duncan_vip", "stop", "interval=0s", "timeout=20s"] in the CIB
Debug: Executing: '/usr/sbin/pcs resource op remove duncan_vip stop interval=0s timeout=20s'
Debug: Puppet::Type::Cs_primitive::ProviderPcs: Executing ["/usr/sbin/pcs", "resource", "update", "duncan_vip", "cidr_netmask=24", "ip=172.16.210.101", "op", "monitor", "interval=10s"] in the CIB
Debug: Executing: '/usr/sbin/pcs resource update duncan_vip cidr_netmask=24 ip=172.16.210.101 op monitor interval=10s'
Debug: /Stage[main]/Main/Cs_primitive[duncan_vip]: The container Class[Main] will propagate my refresh event

test.rb

# frozen_string_literal: true

ops = [
  { 'monitor' => { 'interval' => '10s' } },
  { 'start'   => { 'interval' => '0s', 'timeout' => '20s' } },
  { 'stop'    => { 'interval' => '0s', 'timeout' => '20s' } }
]

exops = [
  { 'monitor' => { 'interval' => '10s' } },
  { 'start'   => { 'interval' => '0s', 'timeout' => '20s' } },
  { 'stop'    => { 'interval' => '0s', 'timeout' => '20s' } }
]

puts(exops.reject { |op| ops.include?(op) })

1. test.rb gives no returns, but apparently, the above code sample still pushes the start and stop ops to the removal stack.
2. force_reinstall is only true if the tuple (primitive_type, primitive_class, provided_by) is a mismatch.
3. It's the only occurence of the resource removal, especially when restricting it to the following resource removal.