zuul
/
zuul
Code Issues Proposed changes

Fix Multi Node Builds with k8s Build Nodes 

Multi-node builds are an explicit feature of Zuul, but are currently
broken with k8s/OpenShift Pod nodes.
Nodepool spawns each k8s pod in its own separate namespace, with
separate service accounts and contexts.
Ansible handles the connection to k8s pods via its kubectl connection
plugin and allows to play against pods in separate contexts via the
`ansible_kubectl_context` host var.
However, Zuul misses to set this host var, defaulting to the
`default-context` of the kubeconfig it prepares for the build.
This leads to a situation where Ansible tries to play each task against
the same pod instead of the ones specified by the respective task.
This change fixes this by setting the `ansible_kubectl_context` host
vars in the inventory accordingly.

Also, the preparation of the executors kubeconfig is error-prone as it
adds multiple clusters with the same name to the config file. This leads
to errors when kubectl evaluates its config.
This change therefore makes sure that a cluster is added only once.

Change-Id: I8e31f2ab209be5b3fef565901f7aa16b00ab93f6
This commit is contained in:
Benjamin Schanzel 2020-08-17 16:15:36 +02:00
parent 43f1ebeb82
commit f928fb2d15
1 changed files with 20 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
zuul/executor/server.py
View File

@ -1590,6 +1590,9 @@ class AnsibleJob(object):
if self.winrm_read_timeout is not None:
host_vars['ansible_winrm_read_timeout_sec'] = \
self.winrm_read_timeout
elif connection_type == "kubectl":
host_vars['ansible_kubectl_context'] = \
node.get('kubectl_context')
host_keys = []
for key in node.get('host_keys', []):
@ -1906,17 +1909,23 @@ class AnsibleJob(object):
}
# Add cluster
cluster_name = urlsplit(data['host']).netloc.replace('.', '-')
cluster = {
'server': data['host'],
}
if data.get('ca_crt'):
cluster['certificate-authority-data'] = data['ca_crt']
if data['skiptls']:
cluster['insecure-skip-tls-verify'] = True
kube_cfg['clusters'].append({
'name': cluster_name,
'cluster': cluster,
})
# Do not add a cluster/server that already exists in the kubeconfig
# because that leads to 'duplicate name' errors on multi-node builds.
# Also, as the cluster name directly corresponds to a server, there
# is no need to add it twice.
if cluster_name not in [c['name'] for c in kube_cfg['clusters']]:
cluster = {
'server': data['host'],
}
if data.get('ca_crt'):
cluster['certificate-authority-data'] = data['ca_crt']
if data['skiptls']:
cluster['insecure-skip-tls-verify'] = True
kube_cfg['clusters'].append({
'name': cluster_name,
'cluster': cluster,
})
# Add user
user_name = "%s:%s" % (data['namespace'], data['user'])