A demo setup of Elasticsearch (ES) on Openshift
After struggling for quite some time getting ES to run smoothly on Openshift here is a working setup.
It consist of three parts:
- an example Openshift configuration
- a small Java program indexing generated random data
- an Angular web frontend with a search-as-you-type input field.
- Openshift 3.9 instead of now Openshift 3.11
- NSF4 Storage instead of now OpenStack Cinder
- Using a Minishift Image (oops) ignoring memory settings and using only 512 MB
- Using supervisord to run a NewRelic service parallel prevents seeing ES restarts caused by OOM kills
- Incorrect interpretation of Openshifts resource setting ‘requests’ and ‘limit’
Use headless mode:
clusterIP: none
It is quite important to use a statefulset.
The value of
discovery.zen.ping.unicast.hosts
is a list of elasticsearch node names. TODO: is there a better way?
Without setting the
cluster.initial_master_nodes
value the eligible master nodes do not choose any elected master node. TODO: is this correct? Till now nowhere elsee found any hint on this.
We use a template
volumeClaimTemplates
specifiying hwo to create dynamically new volumes. The storageClassName is specific according to our openshift setup.
ES_JAVA_OPTS
should be always below 32GB. First best guess is to have 1/16th of the storage size. Add shards if you reach the 32 GB limit.
containers.resources
The request must be above the ES_JAVA_OPTS. Elasticsearch needs for caching a sufficient amount of unused RAM.
To achieve this on Openshift use a high value for containers.resources.request (TODO: verify).
Example for a generated ticket:
{
"pointOfSale": "47",
"time": "1970-01-01T07:49:16.923Z",
"productInfo": {
"id": "136",
"from": "Hamburg",
"to": "Milano"
},
"utilizationTime": {
"from": "2019-09-19",
"to": "2019-09-26"
},
"travelerInfo": {
"lastname": "Plate",
"firstname": "Abdu",
"dateOfBirth": "1993-01-25"
},
"price": {
"value": "09.4",
"currency": "CHF"
}
}