Use kubectl to administer Astronomer Software
Kubectl and Helm
Kubectl and helm are the two primary ways DevOps users will interact/administer the Astronomer Platform.
Setup
Both these tools are needed to deploy Astronomer onto a Kubernetes cluster. We also recommend using kubectx
to simplify commands (the rest of this guide will use kubectx
).
Base namespace and release
The initial helm install
command to deploy Astronomer requires a namespace to deploy the base platform pods into.
helm install -f values.yaml . datarouter
This deploys a randomly named release of our helm charts into the datarouter
namespace.
All pods can be listed in the namespace with kubectl.
root@orbiter: datarouter
Context "gke_astronomer-dev-190903_us-east4-a_astronomer-dev-lybjumoigv" modified.
Active namespace is "datarouter".
root@orbiter: kubectl get pods
NAME READY STATUS RESTARTS AGE
cautious-seal-postgresql-0 1/1 Running 0 2d
singed-chimp-alertmanager-0 1/1 Running 0 1h
singed-chimp-cli-install-9f97f8b-m6l7t 1/1 Running 0 1h
singed-chimp-commander-68b95959c5-xb9ht 1/1 Running 0 1h
singed-chimp-elasticsearch-client-84d85f87fb-lmbsf 1/1 Running 0 1h
singed-chimp-elasticsearch-client-84d85f87fb-wjpwc 1/1 Running 0 1h
singed-chimp-elasticsearch-data-0 1/1 Running 0 1h
singed-chimp-elasticsearch-data-1 1/1 Running 0 1h
singed-chimp-elasticsearch-exporter-5fc97b847f-52qsq 1/1 Running 0 1h
singed-chimp-elasticsearch-master-0 1/1 Running 0 1h
singed-chimp-elasticsearch-master-1 1/1 Running 0 1h
singed-chimp-elasticsearch-master-2 1/1 Running 0 1h
singed-chimp-elasticsearch-nginx-9c4bc78-f5zzz 1/1 Running 0 1h
singed-chimp-fluentd-ztb6k 1/1 Running 0 1h
singed-chimp-grafana-6df6859854-btdc6 1/1 Running 0 1h
singed-chimp-houston-86556ffd74-dkbr8 1/1 Running 0 1h
singed-chimp-kibana-5ccc7b98d4-gk7vb 1/1 Running 0 1h
singed-chimp-kube-state-78bd89bf47-m7nrm 1/1 Running 0 1h
singed-chimp-kubed-5699fc79b4-9dz28 1/1 Running 0 1h
singed-chimp-nginx-5cbdf7967f-2q4dr 1/1 Running 0 1h
singed-chimp-nginx-default-backend-55788c586-2s9nl 1/1 Running 0 1h
singed-chimp-astro-ui-748dfd5748-77v66 1/1 Running 0 1h
singed-chimp-prisma-6ddf8f6784-lfnfr 1/1 Running 0 1h
singed-chimp-prometheus-0 1/1 Running 0 1h
singed-chimp-registry-0 1/1 Running 0 1h
Run helm ls
to list all releases:
root@orbiter: helm ls
NAME REVISION UPDATED STATUS CHART NAMESPACE
cautious-seal 1 Mon May 6 14:37:34 2019 DEPLOYED postgresql-0.18.1 datarouter
singed-chimp 3 Thu May 9 12:08:36 2019 DEPLOYED astronomer-0.8.2 datarouter
There is a release for Astronomer (single-chimp
) and postgres (cautious-seal
) in the datarouter
namespace.
Creating new Airflow deployments
If you navigate to the Software UI and create a new deployment, it will create a new helm release in a new namespace.
root@orbiter helm ls
NAME REVISION UPDATED STATUS CHART NAMESPACE
accurate-bolide-9914 2 Thu May 9 12:06:06 2019 DEPLOYED airflow-0.8.2 datarouter-accurate-bolide-9914
cautious-seal 1 Mon May 6 14:37:34 2019 DEPLOYED postgresql-0.18.1 datarouter
singed-chimp 3 Thu May 9 12:08:36 2019 DEPLOYED astronomer-0.8.2 datarouter
The accurate-bolide-9914
helm release now lives in the a namespace generated by Astronomer named $basenamespace-release_name
(datarouter-accurate-bolide-9914) and lives on the URL $release-name-airflow-BASEDOMAIN
.
Switching into the datarouter-accurate-bolide-9914
namespace reveals the Airflow pods.
root@orbiter: kubectl get namespaces
NAME STATUS AGE
datarouter Active 2d
datarouter-accurate-bolide-9914 Active 1h
default Active 222d
root@orbiter: kubens datarouter-accurate-bolide-9914
root@orbiter: kubectl get pods
NAME READY STATUS RESTARTS AGE
accurate-bolide-9914-pgbouncer-57b46d67bb-b4tpw 2/2 Running 0 1h
accurate-bolide-9914-scheduler-7d4f5596b5-r457d 2/2 Running 0 1h
accurate-bolide-9914-statsd-84d76854fb-wz45c 1/1 Running 0 1h
accurate-bolide-9914-webserver-86bcbc48b-5xcwr 1/1 Running 0 1h
accurate-bolide-9914-worker-0 1/1 Running 0 1m
accurate-bolide-9914-redis-0 1/1 Running 0 1m
accurate-bolide-9914-flower-84f7d67fbd-q5sd7 1/1 Running 0 1m
Since this deployment is running the Celery executor with 1 worker, there are pods for Redis and Flower.
Managing Pods
Airflow logs can be fetched directly from the underlying pods:
root@orbiter: kubectl logs accurate-bolide-9914-scheduler-7d4f5596b5-r457d
Waiting for host: accurate-bolide-9914-pgbouncer 6543
Initializing airflow database...
[2019-05-09 17:53:22,865] {settings.py:182} INFO - settings.configure_orm(): Using pool settings. pool_size=5, pool_recycle=1800, pid=18
[2019-05-09 17:53:24,345] {__init__.py:51} INFO - Using executor KubernetesExecutor
DB: postgresql://accurate_bolide_9914_airflow:***@accurate-bolide-9914-pgbouncer:6543/accurate-bolide-9914-metadata
[2019-05-09 17:53:24,969] {db.py:350} INFO - Creating tables
INFO [alembic.runtime.migration] Context impl PostgresqlImpl.
INFO [alembic.runtime.migration] Will assume transactional DDL.
Done.
[2019-05-09 17:53:28,159] {settings.py:182} INFO - settings.configure_orm(): Using pool settings. pool_size=5, pool_recycle=1800, pid=8
[2019-05-09 17:53:29,663] {__init__.py:51} INFO - Using executor KubernetesExecutor
____________ _____________
____ |__( )_________ __/__ /________ __
____ /| |_ /__ ___/_ /_ __ /_ __ \_ | /| / /
___ ___ | / _ / _ __/ _ / / /_/ /_ |/ |/ /
_/_/ |_/_/ /_/ /_/ /_/ \____/____/|__/
[2019-05-09 17:53:30,338] {jobs.py:1500} INFO - Starting the scheduler
[2019-05-09 17:53:30,339] {jobs.py:1508} INFO - Running execute loop for -1 seconds
[2019-05-09 17:53:30,340] {jobs.py:1509} INFO - Processing each file at most -1 times
[2019-05-09 17:53:30,340] {jobs.py:1512} INFO - Searching for files in /usr/local/airflow/dags
[2019-05-09 17:53:30,341] {jobs.py:1514} INFO - There are 2 files in /usr/local/airflow/dags
[2019-05-09 17:53:30,343] {kubernetes_executor.py:691} INFO - Start Kubernetes executor
A full description of a pods status, resource request, and other data can be found with the describe
command
root@orbiter: kubectl describe po/accurate-bolide-9914-scheduler-7d4f5596b5-g72pg
Name: accurate-bolide-9914-scheduler-7d4f5596b5-g72pg
Namespace: datarouter-accurate-bolide-9914
Priority: 0
PriorityClassName: <none>
Node: gke-astronomer-dev-l-astronomer-dev-n-1c6fc689-s8pg/10.150.0.77
Start Time: Thu, 09 May 2019 13:53:20 -0400
Labels: component=scheduler
platform=singed-chimp
pod-template-hash=3809115261
release=accurate-bolide-9914
tier=airflow
workspace=cjvgu6b2d00110b785tfxyqp0
Pods can also be deleted as a way to restart any Airflow component.
root@orbiter: kubectl delete po/accurate-bolide-9914-scheduler-7d4f5596b5-r457d
This will delete that copy of the pod and spin up a new one. All pods in an Airflow deployment are meant to be stateless, so deleting one and letting it recreate should not cause any
Other Resources
For additional kubectl
commands, check out this Kubernetes Cheat Sheet.