I just noticed a potential issue in fault message generation process: It looks like the Cartridge Agent is sending ports_not_open messages to CEP if the application ports are not open. I think this should happen other way around. Otherwise if cartridge agent goes down the system will think that the member is still active and another instance may not be spawned. WDYT?
Thanks On Tue, Dec 17, 2013 at 10:21 AM, Imesh Gunaratne <[email protected]> wrote: > Hi, > > I just wanted to document the inputs and outputs of the Complex Event > Processor. Once this is finalized we could add this information to the Wiki. > > > *Input Stream Payload Definitions* > Following payload definitions describe the messages sent to CEP by load > balancers and cartridge agents: > > *1. In-Flight Request Count:* > > <payloadData> > <property name="cluster_id" type="String"/> > <property name="network_partition_id" type="String"/> > <property name="in_flight_request_count" type="int"/> > </payloadData> > > These messages are sent by load balancers to the CEP. > > *2. Cartridge Agent Health Stats:* > > <payloadData> > <property name="cluster_id" type="String" /> > <property name="network_partition_id" type="String"/> > <property name="member_id" type="String" /> > <property name="health_description" type="String"/> > <property name="value" type="double"/> > </payloadData> > > These messages are sent by cartridge agents to CEP. > > health_description: 'port_not_open', 'load_average', 'memory_consumption' > Here I would like to suggest to rename "health_description" to something > like "parameter". > > > *Output Message Formats* > Following are the formats of the messages sent by the CEP to the message > broker: > > *1. Average In-Flight Requests:* > > {"average_in_flight_requests":{"cluster_id":"{{cluster_id}}","network_partition_id":"{{network_partition_id}}","value":"{{count}}"}} > > *2. Average Load Average:* > > {"average_load_average":{"cluster_id":"{{cluster_id}}","network_partition_id":"{{network_partition_id}}","value":"{{average_load_average}}"}} > > *3. Average Memory Consumption:* > > {"average_memory_consumption":{"cluster_id":"{{cluster_id}}","network_partition_id":"{{network_partition_id}}","value":"{{average_memory_consumption}}"}} > > *4. Fault Member:* > > {"member_fault":{"cluster_id":"{{cluster_id}}","network_partition_id":"{{network_partition_id}}","member_id":"{{member_id}}"}} > > *5. Gradient In-Flight Requests:* > > {"gradient_in_flight_requests":{"cluster_id":"{{cluster_id}}","network_partition_id":"{{network_partition_id}}","value":"{{count}}"}} > > *6. Gradient Load Average:* > > {"gradient_load_average":{"cluster_id":"{{cluster_id}}","network_partition_id":"{{network_partition_id}}","value":"{{gradient_load_average}}"}} > > *7. Gradient Memory Consumption:* > > {"gradient_memory_consumption":{"cluster_id":"{{cluster_id}}","network_partition_id":"{{network_partition_id}}","value":"{{gradient_memory_consumption}}"}} > > *8. Second Derivative In-Flight Requests:* > > {"second_derivative_in_flight_requests":{"cluster_id":"{{cluster_id}}","network_partition_id":"{{network_partition_id}}","value":"{{count}}"}} > > *9. Second Derivative Load Average:* > > {"second_derivative_load_average":{"cluster_id":"{{cluster_id}}","network_partition_id":"{{network_partition_id}}","value":"{{second_derivative_load_average}}"}} > > *10. Second Derivative Memory Consumption:* > > {"second_derivative_memory_consumption":{"cluster_id":"{{cluster_id}}","network_partition_id":"{{network_partition_id}}","value":"{{second_derivative_memory_consumption}}"}} > > I think we need to implement message processors and event listeners for > these messages in messaging component and Autoscaler could use those to > listen to required events. > > > Many Thanks > Imesh >
