Set a default of 6 hours, but allow users to tell us to wait a longer or shorter time for executors to gracefully stop. Change-Id: I12aada7a3c3aaff7b8bbe51744a8dd4d367b6e5a
17 KiB
Zuul Operator
This is a Kubernetes Operator for the Zuul Project Gating System.
Zuul has a number of components and depencencies, and this operator is designed to simplify creating and maintaining Zuul systems in Kubernetes.
Somewhat unusually, this operator offers the ability to completely manage Zuul's operational dependencies, to the point of even installing other operators upon which it relies. Be sure to read about deployment options if you want to perform some of these tasks yourself.
Simple Example
The quickest way to get a running Zuul is to allow the operator to manage all of the dependencies for you. In this case, the operator will:
- Install cert-manager and set up a self-signed cluster issuer
- Install the Percona XtraDB operator and create a three-node PXC database cluster
- Create a three-node ZooKeeper cluster
- And of course, create a Zuul system
Note
Installing other operators requires a high level of access, so when used in this manner, zuul-operator runs with cluster admin privileges. If you would like the operator to run with reduced privileges, see Managing Operator Dependencies.
From the root of the zuul-operator repo, run:
kubectl apply -f deploy/crds/zuul-ci_v1alpha1_zuul_crd.yaml
kubectl apply -f deploy/rbac-admin.yaml
kubectl apply -f deploy/operator.yaml
You probably want a namespace, so go ahead and create one with:
kubectl create namespace zuul
You will need to prepare two config files for Zuul: the Nodepool config file and the Zuul tenant config file. See the Zuul and Nodepool manuals for how to prepare those. When they are ready, add them to Kubernetes with commands like:
kubectl -n zuul create secret generic zuul-nodepool-config --from-file=nodepool.yaml
kubectl -n zuul create secret generic zuul-tenant-config --from-file=main.yaml
Then create a file called zuul.yaml
which looks
like:
---
apiVersion: operator.zuul-ci.org/v1alpha2
kind: Zuul
metadata:
name: zuul
spec:
executor:
count: 1
sshkey:
secretName: gerrit-secrets
scheduler:
config:
secretName: zuul-tenant-config
launcher:
config:
secretName: zuul-nodepool-config
web:
count: 1
connections:
opendev:
driver: git
baseurl: https://opendev.org
This will create the most basic of Zuul installations, with one each
of the zuul-executor, zuul-scheduler, and zuul-web processes. It will also create a
Nodepool launcher for each of the providers listed in your
nodepool.yaml
. If your Zuul tenant config file requires
more connections, be sure to add them here.
Managing Operator Dependencies
You may not want zuul-operator to install other operators (for example, if your cluster has other users and you don't want cert-manager or pxc-operator to be tied to a Zuul installation, or if you would prefer to avoid granting zuul-operator cluster admin privileges). In that case, you may install the other operators yourself and still allow zuul-operator to use those other operators. It can still create a PXC cluster for you as long as the pxc-operator is present.
To use this mode of operation, make sure the following dependencies are installed before using zuul-operator:
- Cert-manager (at least version 1.2.0)
- Percona-xtradb-cluster-operator (at least version 1.7.0)
With these installed, you may install zuul-operator with reduced privileges:
kubectl apply -f deploy/crds/zuul-ci_v1alpha1_zuul_crd.yaml
kubectl apply -f deploy/rbac.yaml
kubectl apply -f deploy/operator.yaml
After this point, usage is the same as other methods.
Externally Managed Zuul Dependencies
If you want zuul-operator to do even less work, you can have it avoid managing either ZooKeeper or the SQL database.
Externally Managed ZooKeeper
If you already have a ZooKeeper instance you would like Zuul to use, add the following to the Zuul spec:
---
apiVersion: operator.zuul-ci.org/v1alpha2
kind: Zuul
spec:
zookeeper:
connectionString: ...
secretName: ...
The connectionString
field should be a standard
ZooKeeper connection string, and the secretName
field
should be a Kubernetes TLS secret with the client cert for Zuul to use
when connecting to ZooKeeper. TLS is required.
Externally Managed Database
If you would like to use an existing database, add the following to the Zuul spec:
---
apiVersion: operator.zuul-ci.org/v1alpha2
kind: Zuul
spec:
database:
dburi: ...
The dburi
field should contain a Python db-api URI; it
corresponds to the dburi
entry in
zuul.conf
.
Secrets
The operator uses Kubernetes secrets as input values for several aspects of operation. There are some general rules about how secrets are used:
- For configuration files, secret keys are expected to be the typical
filename (for example,
nodepool.yaml
for the Nodepool config file). - For Zuul connection entries, secret keys correspond to the
configuration file attributes for that section (e.g.,
app_key
for the github driver).
See the reference documentation for the specific secretName entry for details.
Zuul Preview
The operator has optional support for deploying a zuul-preview
service. This is an experimental add-on for Zuul to serve Zuul artifacts
from the root of a domain (this can be useful for serving static
HTML/Javascript sites). If you enable this, the operator will configure
a zuul-preview
service to which you may route an Ingress or
LoadBalancer.
Zuul Registry
The operator has optional support for deploying a zuul-registry service. This is an experimental add-on for Zuul to act as an intermediate registry for the container image jobs in zuul-jobs.
If you enable this, the operator will, by default, configure a
zuul-registry
service in a manner appropriate for access
from within the cluster only. If you need to access the registry from
outside the cluster, you will need to additionally add an Ingress or
LoadBalancer, as well as provide TLS certs with the appropriate
hostname. Currently, zuul-registry performs its own TLS termination.
If you usue this, you will also need to provide a
registry.yaml
config file in a secret. You only need to
provide the users
and, if you are accessing the registry
outside the cluster, the public-url
setting (omit it if you
are accessing it from within the cluster only).
registry:
users:
- name: testuser
pass: testpass
access: write
Specification Reference
This is a fully populated example (with the exception of connection entries which can contain zuul.conf attributes passed through verbatim):
apiVersion: zuul-ci.org/v1alpha2
kind: Zuul
spec:
database:
secretName: mariadbSecret
zookeeper:
hosts: zk.example.com:2282
secretName: zookeeperTLS
merger:
count: 5
git_user_email: zuul@example.org
git_user_name: Example Zuul
executor:
count: 5
manage_ansible: false
web:
count: 1
status_url: https://zuul.example.org
fingergw:
count: 1
scheduler:
count: 1
connections:
gerrit:
driver: gerrit
server: gerrit.example.com
secretName: gerritSecrets
user: zuul
baseurl: http://gerrit.example.com:8080
github:
driver: github
secretName: githubSecrets
rate_limit_logging: false
app_id: 1234
jobVolumes:
- context: trusted
access: ro
path: /authdaemon/token
volume:
name: gcp-auth
hostPath:
path: /var/authdaemon/executor
type: DirectoryOrCreate
Zuul
The Zuul kind is currently the only resource directly handled by the operator. It holds a complete description of a Zuul system (though at least partly via secrets referenced by this resource).
spec
imagePrefix
The prefix to use for images. The image names are fixed
(zuul-executor
, etc). However, changing the prefix will
allow you to use custom images or private registries.
zuulImageVersion
The image tag to append to the Zuul images.
zuulPreviewImageVersion
The image tag to append to the Zuul Preview images.
nodepoolImageVersion
The image tag to append to the Nodepool images.
database
This is not required unless you want to manage the database yourself. If you omit this section, zuul-operator will create a Percona XtraDB cluster for you.
You may add any attribute corresponding to the database section of zuul.conf here. The
dburi
attribute will come from the secret below.
secretName
The name of a secret containing connection information for the database.
The key name in the secret should be dburi
.
allowUnsafeConfig
If you are running in a resource constrained environment (such as minikube), the requested resource values for the Percona XtraDB may be too large. Set this to True to override the default values and construct the cluster regardless. Only use this for testing.
zookeeper
This is not required unless you want to manage the ZooKeeper cluster yourself. If you omit this section, zuul-operator will create a ZooKeeper cluster for you
You may add any attribute corresponding to the zookeeper section of zuul.conf here. The
hosts
and TLS attributes will come from the secret
below.
hosts
A standard ZooKeeper connection string.
secretName
The name of a secret containing a TLS client certificate and key for ZooKeeper. This should be (or the format should match) a standard Kubernetes TLS secret.
The key names in the secret should be:
ca.crt
tls.crt
tls.key
env
A list of environment variables. This will be passed through to the Pod specifications for the scheduler, launcher, and web. This may be used to set http_proxy environment variables.
scheduler
config
secretName
The name of a secret containing the Zuul tenant config file.
The key name in the secret should be main.yaml
.
launcher
config
secretName
The name of a secret containing the Nodepool config file.
The key name in the secret should be nodepool.yaml
.
executor
count
How many executors to manage. This is a required component and should be at least 1.
sshkey
secretName
The name of a secret containing the SSH private key that executors should use when logging into Nodepool nodes. You will need to arrange for the public half of this key to be installed on those nodes via whatever mechanism provided by your cloud.
The key name in the secret should be sshkey
.
terminationGracePeriodSeconds
When performing a rolling restart of the executors, wait this long for jobs to finish normally. If an executor takes longer than this amount of time, it will be hard-stopped (and jobs will abort and retry). The default is 6 hours, but depending on the workload, a higher or lower value may be appropriate.
merger
count
How many mergers to manage. Executors also act as mergers so this is not required. They may be useful on a busy system.
web
count
How many Zuul webservers to manage. This is a required component and should be at least 1.
fingergw
count
How many Zuul finger gateway servers to manage.
connections
This is a mapping designed to match the connections entries in the main Zuul config file (zuul.conf). Each key in the mapping is the name of a connection (this is the name you will use in the tenant config file), and the values are key/value pairs that are directly added to that connectien entry in zuul.conf. In the case of keys which are typically files (for example, SSH keys), the values will be written to disk for you (so you should include the full values here and not the path).
You may provide any of these values directly in this resource or using the secret described below. You may use both, and the values will be combined (for example, you may include all the values here except a private key which you include in a secret).
Example:
connections:
opendev:
driver: git
baseurl: https://opendev.org
gerrit:
driver: git
baseurl: https://gerrit.examplec.mo
secretName: gerrit-secrets
<name>
The name of the connection. You will use this is the tenant config file. All of the attributes describing this connection should be included underneath this key.
secretName
The name of a secret describing this connection. All of the keys and
values in this secret will be merged with the keys and values described
in this connection entry. If you need to provide a file (for example,
the sshkey
attribute of a Gerrit connection), include the
contents as the value of the sshkey
attribute in the
secret.
jobVolumes
A list of Kubernetes volumes to be bind mounted into the executor's execution context. These correspond to the trusted_ro_paths, untrusted_ro_paths, trusted_rw_paths, and untrusted_ro_paths entries in zuul.conf.
The first part of each entry describes how the volume should appear in the executor, and the volume attribute describes the Kubernetes volume.
context
One of the following values:
trusted
To be mounted in the trusted execution context.
untrusted
To be mounted in the untrusted execution context.
access
One of the following values:
rw
To be mounted read/write.
ro
To be mounted read-only.
path
The mount point within the execution context.
volume
A mapping corresponding to a Kubernetes volume.
preview
count
How many Zuul Preview servers to manage.
registry
count
How many Zuul Registry servers to manage.
volumeSize
The requested size of the registry storage volume.
tls
secretName
The name of a secret containing a TLS client certificate and key for Zuul Registry. This should be (or the format should match) a standard Kubernetes TLS secret.
If you omit this, the operator will create a secret for you.
config
secretName
The name of a secret containing a registry configuration file. The
key in the secret should be registry.yaml
. Only provide the
users
and, if exposing the registry outside the cluster,
the public-url
entries.