Description
According to member blake of consul, the best way to implement consul is to register it on each node, but apisix starts from the api: v1/health/checks/ takes serviceID to get the data of each node, so there will be duplicate data here, for example
[{"Node":"consul-01","CheckID":"service:192.168.11.2:8001","Name":"Service 'test' check","Status":"passing","Notes": "","Output":"TCP connect 192.168.11.2:8001: Success","ServiceID":"192.168.11.2:8001","ServiceName":"test","ServiceTags":[],"Type": "tcp","Interval":"5s","Timeout":"5s","ExposedPort":0,"Definition":{},"CreateIndex":12014,"ModifyIndex":12015},{"Node" :"consul-03","CheckID":"service:192.168.11.2:8001","Name":"Service 'test' check","Status":"passing","Notes":"","Output ":"TCP connect 192.168.11.2:8001: Success","ServiceID":"192.168.11.2:8001","ServiceName":"test","ServiceTags":[],"Type":"tcp"," Interval":"5s","Timeout":"5s","ExposedPort":0,"Definition":{},"CreateIndex":10366,"ModifyIndex":12020}]If node A is registered but node B is not, then the serviceID obtained by apisix will not be balanced, and the weight of the serviceID will be offset, but the user does not know it.
So, I want to deduplicate serviceID
By the way, I want to add a scheduling based on the minimum IP load. I still use roundrobin, but when polling the upstream, the polling is based on the minimum IP schedule, because we currently encounter a scenario where one machine has multiple service, but at a certain point in time, many requests hit the machine at the same time, causing the CPU to skyrocket. We hope to design an optional plug-in parameter. If this parameter is selected, the IP will be sorted first, and then the port will be sorted.
Description
According to member blake of consul, the best way to implement consul is to register it on each node, but apisix starts from the api: v1/health/checks/ takes serviceID to get the data of each node, so there will be duplicate data here, for example
[{"Node":"consul-01","CheckID":"service:192.168.11.2:8001","Name":"Service 'test' check","Status":"passing","Notes": "","Output":"TCP connect 192.168.11.2:8001: Success","ServiceID":"192.168.11.2:8001","ServiceName":"test","ServiceTags":[],"Type": "tcp","Interval":"5s","Timeout":"5s","ExposedPort":0,"Definition":{},"CreateIndex":12014,"ModifyIndex":12015},{"Node" :"consul-03","CheckID":"service:192.168.11.2:8001","Name":"Service 'test' check","Status":"passing","Notes":"","Output ":"TCP connect 192.168.11.2:8001: Success","ServiceID":"192.168.11.2:8001","ServiceName":"test","ServiceTags":[],"Type":"tcp"," Interval":"5s","Timeout":"5s","ExposedPort":0,"Definition":{},"CreateIndex":10366,"ModifyIndex":12020}]If node A is registered but node B is not, then the serviceID obtained by apisix will not be balanced, and the weight of the serviceID will be offset, but the user does not know it.
So, I want to deduplicate serviceID
By the way, I want to add a scheduling based on the minimum IP load. I still use roundrobin, but when polling the upstream, the polling is based on the minimum IP schedule, because we currently encounter a scenario where one machine has multiple service, but at a certain point in time, many requests hit the machine at the same time, causing the CPU to skyrocket. We hope to design an optional plug-in parameter. If this parameter is selected, the IP will be sorted first, and then the port will be sorted.