36ba1cc319
Prior to this patch, the metadata agent was writing into SB database when a network had been provisioned with metadata on a particular chassis. Then, neutron-server would wait for that event to happen with a 15s timeout before sending the vif-plugged event to Nova. By removing this mechanism: 1) We'll save writes to OVN SB database which, in highly loaded systems and at scale reduces significantly the load on ovsdb-server. 2) Ignoring healthchecks (that still requires write to the SB DB), we can make OVN metadata agent to connect to slave instances when using active-backup OVN databases since writes are not needed. 3) There's a chance that the VM boots very fast and requests metadata before the service is ready but since the timeout was 15 seconds, we can safely rely on the the cloud-init retries. Signed-off-by: Daniel Alvarez Sanchez <dalvarez@redhat.com> Change-Id: Ia6cd7a9a3b9662a9a8ce106e01a93c357c255956 |
||
---|---|---|
.. | ||
data_model.rst | ||
database_consistency.rst | ||
distributed_ovsdb_events.rst | ||
index.rst | ||
l3_ha_rescheduling.rst | ||
loadbalancer.rst | ||
metadata_api.rst | ||
native_dhcp.rst | ||
ovn_network_logging.rst | ||
ovn_worker.rst | ||
port_forwarding.rst |