forceful-arp.md

• State: finished

Forceful ARP

As new VM comes up, ARP entries (on AWS for example) on other VMs are not cleared.

Proposal: send explicit messages to all agents to flush ARP entries for specific IPs.

Stories

• Investigate if arp -d is able to forcefully kill the connection while conn is in diff states

• Agent can flush ARP entries on request

  • asynchronous action
  • "flush_arp" action with one argument {"ips": ["10.10.0.2", ...]}
  • runs "arp -d" for each IP
  • support ubuntu and centos

• During bosh deploy when new VM comes up, send flush_arp request

  • make a configuration option (director.flush_arp=bool) and keep this disabled by default for now
  • find all VMs in all deployments and send flush_arp after Director finds out IPs for new machine
  • make sure to collect all new IPs

• During cck/ressurection when new VM comes up, send flush_arp

  • should be noop as it should go through VmCreator

TBD

• Optimization to only send it to VMs within necessary subnet