-
Notifications
You must be signed in to change notification settings - Fork 462
openstack: Add coredns, mdns-publisher and haproxy static pods #740
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
cgwalters
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
/approve
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This will need to be added to image-references and come from the payload.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Ditto.
templates/master/00-master/openstack/files/openstack-mdns-publisher.yaml
Outdated
Show resolved
Hide resolved
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think the more correct approach is for the image to install the Python it wants.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This feels a bit hacky...
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I agree. I added a TODO to come up with a better method of getting these.
|
/hold need to properly publish the coredns and mdns-publisher images this patch uses |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
nit: remove kni
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
These requests are pretty high.
This is part of the work to remove the service VM from the openstack architecture. This relies on the coredns/mdns static pods setup in: openshift/machine-config-operator/pull/740
This is part of the work to remove the service VM from the openstack architecture. This relies on the coredns/mdns static pods setup in: openshift/machine-config-operator/pull/740
This is part of the work to remove the service VM from the openstack architecture. This relies on the coredns/mdns static pods setup in: openshift/machine-config-operator/pull/740
This is part of the work to remove the service VM from the openstack architecture. This relies on the coredns/mdns static pods setup in: openshift/machine-config-operator/pull/740
This is part of the work to remove the service VM from the openstack architecture. This relies on the coredns/mdns static pods setup in: openshift/machine-config-operator/pull/740
openshift#740 and openshift/installer#1959 depend on each other. We need to break this dependency to allow them to merge while keeping the e2e-openstack job green. This patch disables the static pods added for the openstack platform in openshift#740 unless the installer provided the needed info set via openshift/installer#1959. It can be safely reverted once openshift/installer#1959 merges.
… pods In order to have a more fault tolerant networking architecture, we are replacing the functionality of the service vm with a number of static pod resources that are run on master and worker nodes.
We use Baremetal-RuntimeCfg to clean up our code, as well as align our architecture much more closely with that of the Baremetal Team.
We need the wildcard record on the master nodes to resolve the Ingress IP. Since the `hosts` plugin doesn't support wildcards, it is replaced with the RFC 1035-style zone DB file which serves the API and wildcard records via the `file` plugin. This also adds the API record to both the bootstrap and master nodes.
…tches openshift#740 and openshift/installer#1959 depend on each other. We need to break this dependency to allow them to merge while keeping the e2e-openstack job green. This patch disables the static pods added for the openstack platform in openshift#740 unless the installer provided the needed info set via openshift/installer#1959. It can be safely reverted once openshift/installer#1959 merges.
|
/test e2e-openstack |
1 similar comment
|
/test e2e-openstack |
|
/retest |
|
/retest This worked for me manually on the master. Looks more like a timeout on the openstack job. |
|
/retest |
|
@cgwalters @runcom would you mind having a look at this PR? It's the OpenStack equivalent to: #795 We've tested it locally (with and without the corresponding installer PR openshift/installer#1959), it passes the CI and OpenStack has a FFE for this work. Is there anything else you'd like us to do? |
| // Loop over templates/common which applies everywhere | ||
| for _, dir := range []string{platformBase, platform} { | ||
| // Bypass OpenStack template rendering until | ||
| // https://github.com/openshift/installer/pull/1959 merges |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Hum...so if I understand this correctly, this PR is adding all of the code, and then we need to get the installer PR in first, then we'll do a PR to drop these conditionals?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yes that's the intention. The installer PR depends on the changes there, but this PR depends on the changes in the installer PR so this let's us break that cycle.
The conditionals are isolated in a single commit we can revert afterwards.
|
/lgtm |
|
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: cgwalters, trown The full list of commands accepted by this bot can be found here. The pull request process is described here DetailsNeeds approval from an approver in each of these files:
Approvers can indicate their approval by writing |
|
I find this part encouraging:
Bigger picture, I wonder whether we should consider having the cluster own DNS by default across the board in a uniform way. That's another topic though. The commit message in openshift/installer#1959 is great, very helpful! |
|
/retest Please review the full test history for this PR and help us cut down flakes. |
The experimental OpenStack backend used to create an extra server running DNS and load balancer services that the cluster needed. OpenStack does not always come with DNSaaS or LBaaS so we had to provide the functionality the OpenShift cluster depends on (e.g. the etcd SRV records, the api-int records & load balancing, etc.). This approach is undesirable for two reasons: first, it adds an extra node that the other IPI platforms do not need. Second, this node is a single point of failure. The Baremetal platform has faced the same issues and they have solved them with a few virtual IP addresses managed by keepalived in combination with coredns static pod running on every node using the mDNS protocol to update records as new nodes are added or removed and a similar static pod haproxy to load balance the control plane internally. The VIPs are defined here in the installer and they use the PlatformStatus field to be passed to the necessary machine-config-operator fields: openshift/api#374 The Bare Metal IPI Networking Infrastructure document is applicable here as well: https://github.com/openshift/installer/blob/master/docs/design/baremetal/networking-infrastructure.md There is also a great opportunity to share some of the configuration files and scripts here. This change needs several other pull requests: Keepalived plus the coredns & haproxy static pods in the MCO: openshift/machine-config-operator#740 Passing the API and DNS VIPs through the installer: openshift#1998 Co-authored-by: Emilio Garcia <[email protected]> Co-authored-by: John Trowbridge <[email protected]> Co-authored-by: Martin Andre <[email protected]> Co-authored-by: Tomas Sedovic <[email protected]> Massive thanks to the Bare Metal and oVirt people!
The experimental OpenStack backend used to create an extra server running DNS and load balancer services that the cluster needed. OpenStack does not always come with DNSaaS or LBaaS so we had to provide the functionality the OpenShift cluster depends on (e.g. the etcd SRV records, the api-int records & load balancing, etc.). This approach is undesirable for two reasons: first, it adds an extra node that the other IPI platforms do not need. Second, this node is a single point of failure. The Baremetal platform has faced the same issues and they have solved them with a few virtual IP addresses managed by keepalived in combination with coredns static pod running on every node using the mDNS protocol to update records as new nodes are added or removed and a similar static pod haproxy to load balance the control plane internally. The VIPs are defined here in the installer and they use the PlatformStatus field to be passed to the necessary machine-config-operator fields: openshift/api#374 The Bare Metal IPI Networking Infrastructure document is applicable here as well: https://github.com/openshift/installer/blob/master/docs/design/baremetal/networking-infrastructure.md There is also a great opportunity to share some of the configuration files and scripts here. This change needs several other pull requests: Keepalived plus the coredns & haproxy static pods in the MCO: openshift/machine-config-operator#740 Co-authored-by: Emilio Garcia <[email protected]> Co-authored-by: John Trowbridge <[email protected]> Co-authored-by: Martin Andre <[email protected]> Co-authored-by: Tomas Sedovic <[email protected]> Massive thanks to the Bare Metal and oVirt people!
This is part of the work to remove the "service" VM from the OpenStack architecture. Instead of running coredns on this extra VM, we will run it in a static pod on all masters. We are using the mdns work from metal3.io in order to coordinate the DNS records on the masters.