Skip to content

Latest commit

 

History

History
188 lines (141 loc) · 6.8 KB

README.md

File metadata and controls

188 lines (141 loc) · 6.8 KB
Raw

dbaas-poc

playground for dbaas releated PoCs

Build

How to create config package:

cd packages/eks
kubectl crossplane build configuration
kubectl crossplane push configuration ghcr.io/denisok/dbaas-platform:v0.0.12

Use

Setup and Install: script that deploys it in minikube

Use: kubectl apply -f DBClaim.yaml

Check out DBClaim.yaml to see how you can use the claim:

apiVersion: dbaas.percona.com/v1alpha1
kind: DBclaim
metadata:
  name: psm-db-1
spec:
  compositionSelector:
    matchLabels:
      engine: psmdb
      storage: default
      network: default
  clusterSize: 1
  engineVersion: ""
  engineConfig: ""

Troubleshoot:

kubectl get event --all-namespaces --sort-by='.metadata.managedFields[0].time'

PoC results

DevOps/PET could define both infrastructure (EKS, VMs) and services, like deploying PMM on AWS Instance, deploying operators as well as define templates for DBClaims.

DBClaims could be later consumed either directly when deploying app, just adding DBClaim yaml and consume DB connection and secrets from k8s Secrets.

Or UI could just create/list DBClaim in more general way or any DB types.

One of the result is the way to template DBs that will be deployed is to use compositions: PSMDB composition

DevOps/PET could create number of those compositions to deploy CR with different parameters (like storage, networking and etc), so user could consume it by selecting in claims:

  compositionSelector:
    matchLabels:
      engine: psmdb
      storage: fast
      network: public

Another result is testing access to API.

Currently dbaas-controller implements proxy api to k8s Percona operators thus dbaas-controller needs to call k8s and gets data from k8s operators. That could be expensive call to k8s as api returns full spec and in case of 100 or 1000 of server it could be quite large and time consuming response.

Also if there is no good connection to the k8s cluster situation would be much worse.

Here I measured API request time done in different environments.

PMM in the same k8s cluster

TIME="%e\n" time curl -k -s -o file.out --request POST --url https://172.105.146.244/v1/management/DBaaS/DBClust
ers/List --header 'accept: application/json' --header 'authorization: Basic YWRtaW46VyUrMS5bVCszMV5LemQ7Nw==' --header 'content-type: application
/json' --data '
{
     "kubernetes_cluster_name": "default-pmm-cluster"
}
'
7.42

TIME="%e\n" time curl -k -s -o file.out --request POST --url https://172.105.146.244/v1/management/DBaaS/DBClusters/List --header 'accept: application/json' --header 'authorization: Basic YWRtaW46VyUrMS5bVCszMV5LemQ7Nw==' --header 'content-type: application/json' --data '
{
     "kubernetes_cluster_name": "default-pmm-cluster"
}
'
7.90

PMM in the same k8s cluster

TIME="%e\n" time curl -k -s -o file1.out --request POST --url https://localhost:8443/v1/management/DBaaS/DBClusters/List --header 'accept: application/json' --header 'authorization: Basic YWRtaW46YWRtaW4=' --header 'content-type: application/json' --data '
{
     "kubernetes_cluster_name": "lke75856"
}
'

TIME="%e\n" time curl -k -s -o file1.out --request POST --url https://localhost:8443/v1/management/DBaaS/DBClusters/List --header 'accept: application/json' --header 'authorization: Basic YWRtaW46YWRtaW4=' --header 'content-type: application/json' --data '
{
     "kubernetes_cluster_name": "lke75856"
}
'
30.66

TIME="%e\n" time curl -k -s -o file1.out --request POST --url https://localhost:8443/v1/management/DBaaS/DBClusters/List --header 'accept: application/json' --header 'authorization: Basic YWRtaW46YWRtaW4=' --header 'content-type: application/json' --data '
{
     "kubernetes_cluster_name": "lke75856"
}
'
26.60

TIME="%e\n" time curl -k -s --request POST --url https://localhost:8443/v1/management/DBaaS/DBClusters/List --header 'accept: application/json' --header 'authorization: Basic YWRtaW46YWRtaW4=' --header 'content-type: application/json' --data '
{
     "kubernetes_cluster_name": "lke75856"
}
'
<html>
<head><title>504 Gateway Time-out</title></head>
<body>
<center><h1>504 Gateway Time-out</h1></center>
<hr><center>nginx</center>
</body>
</html>
60.04

##$ Request to the k8s UI directly and test limit

TIME="%e\n" time curl -s -k --header "Authorization: Bearer $JWT_TOKEN_KUBESYSTEM_DEFAULT" $KUBE_API/apis/dbaas.percona.com/v1alpha1/dbclaims?limit=5 | jq '.items | length'
0.47
5

TIME="%e\n" time curl -s -k --header "Authorization: Bearer $JWT_TOKEN_KUBESYSTEM_DEFAULT" $KUBE_API/apis/dbaas.percona.com/v1alpha1/dbclaims?limit=5 | jq '.items | length'
0.32
5

TIME="%e\n" time curl -s -k --header "Authorization: Bearer $JWT_TOKEN_KUBESYSTEM_DEFAULT" $KUBE_API/apis/dbaas.percona.com/v1alpha1/dbclaims | jq '.items | length'
1.46
20

TIME="%e\n" time curl -s -k --header "Authorization: Bearer $JWT_TOKEN_KUBESYSTEM_DEFAULT" $KUBE_API/apis/dbaas.percona.com/v1alpha1/dbclaims | jq '.items | length'
0.56
20

TIME="%e\n" time curl -s -k --header "Authorization: Bearer $JWT_TOKEN_KUBESYSTEM_DEFAULT" $KUBE_API/apis/dbaas.percona.com/v1alpha1/dbclaims | jq '.items | length'
0.57
20

TIME="%e\n" time curl -s -k --header "Authorization: Bearer $JWT_TOKEN_KUBESYSTEM_DEFAULT" $KUBE_API/apis/dbaas.percona.com/v1alpha1/dbclaims | jq '.items | length'
0.53
20

TIME="%e\n" time curl -s -k --header "Authorization: Bearer $JWT_TOKEN_KUBESYSTEM_DEFAULT" $KUBE_API/apis/pxc.percona.com/v1/perconaxtradbclusters | jq '.items | length'
0.54
10

TIME="%e\n" time curl -s -k --header "Authorization: Bearer $JWT_TOKEN_KUBESYSTEM_DEFAULT" $KUBE_API/apis/pxc.percona.com/v1/perconaxtradbclusters | jq '.items | length'
1.09
10

TIME="%e\n" time curl -s -k --header "Authorization: Bearer $JWT_TOKEN_KUBESYSTEM_DEFAULT" $KUBE_API/apis/pxc.percona.com/v1/perconaxtradbclusters | jq '.items | length'
2.03
10

TIME="%e\n" time curl -s -k --header "Authorization: Bearer $JWT_TOKEN_KUBESYSTEM_DEFAULT" $KUBE_API/apis/pxc.percona.com/v1/perconaxtradbclusters | jq '.items | length'
1.36
10

So k8s mitigating such kind of issues be breaking response in chunks.

Another point is that for listing clusters we don't need full CR response.

The way to mitigate this issue:

  1. use k8s directly from UI
  2. use limit
  3. implement proxy layer (operator, crossplane) in k8s

The point #1 is proven by calling apis/pxc.percona.com/v1/perconaxtradbclusters - it takes much less time than going through PMM.

Point #2 is also shown, using that from UI would further improve performance, as we don't need full list for UI.

Point #3 is apis/dbaas.percona.com/v1alpha1/dbclaims it suppose to return much less info (only that is in DBClaim) and thus further improved bandwidth.