Slowdown in create stack request between Havana and Icehouse
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
OpenStack Heat |
Fix Released
|
Medium
|
Unassigned |
Bug Description
The time to respond to a heat stack create request, for a moderate size pattern with 10 VMs, increased 2-3X from Havana stable to Icehouse stable in devstack configurations. (Timings may have improved slightly post-icehouse). The delay for a response to a stack create can increase to 10s of seconds under load.
Config: single machine devstacks, with fake hypervisor driver, 16 fake hypervisors, running in VM with 8 vcpus, 16GB memory, rabbitmq/mysql, flat networking
Template:
heat_template_
description: Start 10 VMs
resources:
VM_0:
type: OS::Nova::Server
properties:
key_name: mykey
flavor: m1.micro
image: F17-i386-cfntools
security_
VM_1:
type: OS::Nova::Server
...
through VM_9
Test did sequential stack create requests using CLI, with no (except cli and test script overhead) delay between requests.
The timings are representative. They changed very little (several percent) between test runs.
havana/stable (2014/05/27)
+------
| id | stack_name | stack_status | creation_time |
+------
| e5f6f965-
+------
real 0m2.488s
user 0m0.280s
sys 0m0.096s
+------
| id | stack_name | stack_status | creation_time |
+------
| e5f6f965-
| 959320a8-
+------
real 0m2.554s
user 0m0.256s
sys 0m0.120s
icehouse/stable 2014/05/27
+------
| id | stack_name | stack_status | creation_time |
+------
| 0e502e52-
+------
real 0m7.305s
user 0m0.308s
sys 0m0.076s
+------
| id | stack_name | stack_status | creation_time |
+------
| 0e502e52-
| a00ecb2a-
+------
real 0m8.492s
user 0m0.208s
sys 0m0.152s
Master 2014/05/27
+------
| id | stack_name | stack_status | creation_time |
+------
| 3d104172-
+------
real 0m6.554s
user 0m0.296s
sys 0m0.116s
+------
| id | stack_name | stack_status | creation_time |
+------
| 3d104172-
| 37b681c4-
+------
real 0m7.464s
user 0m0.328s
sys 0m0.068s
Changed in heat: | |
status: | Incomplete → New |
Changed in heat: | |
assignee: | nobody → Thomas Spatzier (thomas-spatzier) |
status: | New → In Progress |
Changed in heat: | |
assignee: | Thomas Spatzier (thomas-spatzier) → nobody |
status: | In Progress → Triaged |
Changed in heat: | |
status: | Triaged → Fix Released |
The fake hypervisor driver would help, but it is still hard to know if the performance regression came from heat or another openstack component.
Can you try running stable/icehouse heat against stable/havana everything-else?