Since blueprint placement-claims in Pike, the Nova FilterScheduler
uses the placement service to make resource allocation 'claims'
before sending the build request to the chosen compute host to
perform the legacy style resource claim. This allows us to safely
scale out the number of scheduler workers when using the FilterScheduler.
The [scheduler]workers option defaults to ncpu if using the
FilterScheduler (which is the default scheduler driver) so to avoid
out of memory issues, we need to set $API_WORKERS scheduler workers
if using the FilterScheduler in devstack.
Depends-On: Ifdcd363d7bc22e73d76d69777483e5aaff4036e3
Change-Id: Ieae234eb5388560b3f66bf60c156a91a8e831bc4
... | ... |
@@ -413,6 +413,9 @@ function create_nova_conf { |
413 | 413 |
iniset $NOVA_CONF DEFAULT rootwrap_config "$NOVA_CONF_DIR/rootwrap.conf" |
414 | 414 |
iniset $NOVA_CONF scheduler driver "$SCHEDULER" |
415 | 415 |
iniset $NOVA_CONF filter_scheduler enabled_filters "$FILTERS" |
416 |
+ if [[ $SCHEDULER == "filter_scheduler" ]]; then |
|
417 |
+ iniset $NOVA_CONF scheduler workers "$API_WORKERS" |
|
418 |
+ fi |
|
416 | 419 |
iniset $NOVA_CONF DEFAULT default_floating_pool "$PUBLIC_NETWORK_NAME" |
417 | 420 |
if [[ $SERVICE_IP_VERSION == 6 ]]; then |
418 | 421 |
iniset $NOVA_CONF DEFAULT my_ip "$HOST_IPV6" |