DG
collect a few millions of requests and you will see that actually the traffic split accurate enough 🙂
Size: a a a
2021 November 15
M
Ehm.... 😅😅😅😅 unfortunately I already have a million of requests and about 900k are directed to a "replica1", other 300k to a "replica2" and a few others to "replica3" that's why I'm asking
M
Because I need to solve this problem 😅
VR
all pods within a deployment are inherently equal and receive roughly the same amount of traffic.
See https://docs.openshift.com/container-platform/4.8/applications/deployments/route-based-deployment-strategies.html#deployments-blue-green-setting-up_route-based-deployment-strategies for a better way to archeive this
See https://docs.openshift.com/container-platform/4.8/applications/deployments/route-based-deployment-strategies.html#deployments-blue-green-setting-up_route-based-deployment-strategies for a better way to archeive this
M
Vadim unfortunately this is not what I've got right now. As I said I've got a replica 3 deploy and all the replicas are not well balanced because the first one replica has more or less 900k requests, the second one has more or less 300k requests and the third one has the other 200k requests
M
what I want to achieve is to have 500k requests on each pod
I
Is it possible that your service have some sort of sticky-ness on these requests? So that established connections will persist on the pod it first connected to. And the startup of service's replicas take some considerable time?
DG
well, is there any chance that we don’t know the whole schema? for example, who (or what) is the origin for the traffic? openshift route seem to be a sessian aware guy by default
VR
Version, cni, service/router settings, your test strategy?
M
Version: OCP 4.2
CNI: default
Service/Router settings: I have to come back to home in order to give you these information
Test strategy: yesterday a freeradius Deploy was deployed on my infrastructure and it was scaled on replica 3. Today I checked the numbers of calls directed on /var/log/freeradius/radius.log file of all replicas and I saw these differences
CNI: default
Service/Router settings: I have to come back to home in order to give you these information
Test strategy: yesterday a freeradius Deploy was deployed on my infrastructure and it was scaled on replica 3. Today I checked the numbers of calls directed on /var/log/freeradius/radius.log file of all replicas and I saw these differences
M
Replica 1: has more or less 900k calls
Replica 2: has more or less 300k calls
Replica 3: has more or less 200k calls
Replica 2: has more or less 300k calls
Replica 3: has more or less 200k calls
M
The service is just a simple load balancer
M
In front of the cluster there's a F5 load balancer that is able to balance the traffic to 3 Openshift infra-node
M
So the traffic flows from the F5 to one of the 3 infra nodes, comes into the cluster and will be redirected to the correct service. This service will select its endpoint using the selector but it will send the request to one of the 3 replicas in way the I don't understand
VR
Wait, so no router involved?
M
You mean the Openshift router?
M
Or you mean a route?
VR
Openshift router
M
Oh, I thought you mean a simple route. So, yep there is a routing when the packet comes into the cluster
M
Because the infra nodes needs to send the packet to the right service and the service will send the packet to its endpoints