Set up : 3 openstack nodes(1 control node & 2 compute nodes) and 3 odls
ODL for L2 and l3-agent for L3

code : Legacy netvirt
branch : master (carbon)

Steps followed:
After stacking up all 3 openstack nodes, created networks,subnets and instances

Observation:

Ping from network1 dhcp server to vm instances created in compute node2 within network1 failed.
Ping succeeded for network2 across all the nodes.

For network1 dhcp the following like entry should be there in all the openstack nodes dump flows.
cookie=0x0, duration=84.981s, table=20, n_packets=3, n_bytes=126, priority=1024,arp,tun_id=0x439,arp_tpa=30.0.0.2,arp_op=1 actions=move:NXM_OF_ETH_SRC[]>NXM_OF_ETH_DST[],set_field:fa:16:3e:26:05:33>eth_src,load:0x2->NXM_OF_ARP_OP[],move:NXM_NX_ARP_SHA[]>NXM_NX_ARP_THA[],move:NXM_OF_ARP_SPA[]>NXM_OF_ARP_TPA[],load:0xfa163e260533->NXM_NX_ARP_SHA[],load:0x1e000002->NXM_OF_ARP_SPA[],IN_PORT

cookie=0x0, duration=292.095s, table=110, n_packets=14, n_bytes=1294, tun_id=0x439,dl_dst=fa:16:3e:26:05:33 actions=output:1

The above entry exists only in control node and compute1 node dump flows.
The entry is missing in compute2 node dump flows.

This is not always compute2 sometimes compute1 dump flows would have missed the entry.

To summarize: Either of the compute nodes does not have the network1 dhcp entry for table 110.

Due to which ping for vm instances in that particular compute node failed from network1 dhcp server.
Ssh also failed for these vm instances.

Attached the karaf logs for reference.