Looking at an undercloud (in CI) while it is deploying the top 4 memory consumers are 21004 heat 20 0 808064 485444 2928 R 75.0 9.6 18:09.16 heat-engine 21005 heat 20 0 617492 294752 2808 S 13.2 5.8 15:08.34 heat-engine 15211 mysql 20 0 1575892 241784 4940 S 6.9 4.8 3:21.39 mysqld 19989 nova 20 0 505228 169188 2560 S 27.3 3.3 3:32.31 nova-api 19988 nova 20 0 506928 169168 2952 S 11.2 3.3 3:37.26 nova-api 20003 nova 20 0 481380 155520 1996 S 0.0 3.1 0:26.20 nova-api 20004 nova 20 0 481244 155508 1996 S 0.0 3.1 0:17.71 nova-api 19949 nova 20 0 453008 128524 4124 R 1.6 2.5 1:02.90 nova-api 22557 nova 20 0 437680 113972 4224 S 0.0 2.2 1:04.07 nova-compute 20082 nova 20 0 391476 100780 2100 S 0.0 2.0 2:52.78 nova-conductor 20083 nova 20 0 391256 100612 2100 S 0.7 2.0 2:53.96 nova-conductor 19902 nova 20 0 402068 98516 4144 S 0.0 1.9 0:13.49 nova-cert 16971 rabbitmq 20 0 1355748 94020 2020 S 2.3 1.9 1:35.68 beam.smp 20241 neutron 20 0 403212 89424 2004 S 0.0 1.8 2:03.29 neutron-server The 2 heat processes are using 780Mb in total Should we be expecting heat to need this much RAM? We have 4 options here 1. investigate if heat memory footprint can be reduced 2. Bump up the RAM in CI a little more (its at 5Gb) 3. Add some swap to the undercloud - testing a patch for this here https://review.openstack.org/270274 4. find something we can turn off that we don't need on the undercloud Swap would be easiest, I'm trying this today but it would be nice to figure out if heats memory footprint can be reduced.