You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: packer/README.md
+6-6Lines changed: 6 additions & 6 deletions
Original file line number
Diff line number
Diff line change
@@ -7,9 +7,9 @@ The Packer configuration defined here builds "fat images" which contain binaries
7
7
- Ensures re-deployment of the cluster or deployment of additional nodes can be completed even if packages are changed in upstream repositories (e.g. due to RockyLinux or OpenHPC updates).
8
8
- Improves deployment speed by reducing the number of package downloads to improve deployment speed.
9
9
10
-
By default, a fat image build starts from a RockyLinux GenericCloud image and updates all DNF packages already present.
10
+
By default, a fat image build starts from a nightly image build containing Mellanox OFED, and updates all DNF packages already present. The 'latest' nightly build itself is from a RockyLinux GenericCloud image.
11
11
12
-
The fat images StackHPC builds and test in CI are available from [GitHub releases](https://github.com/stackhpc/ansible-slurm-appliance/releases). However with some additional configuration it is also possible to:
12
+
The fat images StackHPC builds and test in CI are available from [GitHub releases](https://github.com/stackhpc/ansible-slurm-appliance/releases). However with some additional configuration it is also possible to:
13
13
1. Build site-specific fat images from scratch.
14
14
2. Extend an existing fat image with additional software.
15
15
@@ -39,9 +39,9 @@ The steps for building site-specific fat images or extending an existing fat ima
Note that the `-only` flag here restricts the build to the non-OFED fat image "source" (in Packer terminology). Other
42
+
Note that the `-only` flag here restricts the build to the non-CUDA fat image "source" (in Packer terminology). Other
43
43
source options are:
44
-
- `-only=openstack.openhpc-ofed`: Build a fat image including Mellanox OFED
44
+
- `-only=openstack.openhpc-cuda`: Build a fat image including CUDA packages.
45
45
- `-only=openstack.openhpc-extra`: Build an image which extends an existing fat image - in this case the variable `source_image` or `source_image_name}` must also be set in the Packer variables file.
46
46
47
47
5. The built image will be automatically uploaded to OpenStack with a name prefixed `openhpc-` and including a timestamp and a shortened git hash.
@@ -70,7 +70,7 @@ What is Slurm Appliance-specific are the details of how Ansible is run:
70
70
openhpc-extra = ["foo"]
71
71
}
72
72
73
-
the build VM uses an existing "fat image" (rather than a RockyLinyux GenericCloud one) and is added to the `builder` and `foo` groups. This means only code targeting `builder` and `foo` groups runs. In this way an existing image can be extended with site-specific code, without modifying the part of the image which has already been tested in the StackHPC CI.
73
+
the build VM uses an existing "fat image" (rather than a 'latest' nightly one) and is added to the `builder` and `foo` groups. This means only code targeting `builder` and `foo` groups runs. In this way an existing image can be extended with site-specific code, without modifying the part of the image which has already been tested in the StackHPC CI.
74
74
75
75
- The playbook `ansible/fatimage.yml` is run which is only a subset of `ansible/site.yml`. This allows restricting the code
76
76
which runs during build for cases where setting `builder` groupvars is not sufficient (e.g. a role always attempts to configure or start services). This may eventually be removed.
@@ -82,5 +82,5 @@ There are some things to be aware of when developing Ansible to run in a Packer
82
82
- Build VM hostnames are not the same as for equivalent "real" hosts and do not contain `login`, `control` etc. Therefore variables used by the build VM must be defined as groupvars not hostvars.
83
83
- Ansible may need to proxy to real compute nodes. If Packer should not use the same proxy to connect to the
84
84
build VMs (e.g. build happens on a different network), proxy configuration should not be added to the `all` group.
85
-
- Currently two fat image "sources" are defined, with and without OFED. This simplifies CI configuration by allowing the
85
+
- Currently two fat image "sources" are defined, with and without CUDA. This simplifies CI configuration by allowing the
86
86
default source images to be defined in the `openstack.pkr.hcl` definition.
0 commit comments