how to implement fault-tolerant warm-up forwarding for routing-Scalable Open Financial Architecture(SOFAStack)-阿里云帮助中心

When a machine in a cluster starts, it enters a warm-up period. During this time, an influx of requests can degrade the machine's performance and disrupt services. The RPC framework uses the warm-up forwarding feature to forward requests from a machine in its warm-up period to other machines in the cluster. After the warm-up period ends, the machine resumes normal operation.

You can specify the proportion of requests to forward during the warm-up period. For example, you can configure the system to forward 80% of requests and process the remaining 20% locally.

Unlike RPC stress test forwarding, warm-up forwarding is only active for a set period after the application starts. In contrast, stress test forwarding is a long-term configuration.

Precautions

Note the following when you use the warm-up forwarding feature:

Warm-up forwarding does not function if the server-side application starts on the same host.
Requests are not forwarded to another service that is also in its warm-up period.
If a service is not configured for prefetch forwarding, subsequent services will not forward traffic to the machine that hosts it.
All machines that participate in warm-up forwarding must use the same BOLT port.

RPC forwarding configuration

You can configure this feature in two ways:

Forward directly to an IP address
Add the following parameter to the application.properties file:
```
rpc.transmit.url = 10.**.**.2
```
Requests are always forwarded directly to the specified IP address. This configuration has the same effect as setting rpc.transmit.url=address:10.**.**.2.
Configure warm-up and weight
Add the following parameters to the application.properties file:
```
rpc.transmit.url=weightStarting:0.3,during:60,weightStarted:0.2,address:10.**.**.2,uniqueId:core_unique
```
- weightStarting: The forwarding weight during the warm-up period. Based on this weight, the RPC framework randomly decides whether to forward a request to another machine in the cluster or to process it locally.
- during: The duration of the warm-up period, in seconds.
- weightStarted: The forwarding weight after the warm-up period. This setting remains in effect permanently.
- address: The forwarding address after the warm-up period. This setting remains in effect permanently.
- uniqueId: In a multi-cluster deployment where clusters share the same appName, use this parameter to distinguish between the clusters. You can specify a custom system variable to ensure cluster uniqueness. For example, core_unique is a configuration in application.properties that can be dynamically replaced.

Note

You can configure the request forwarding timeout in the application.properties file. For example, rpc_transmit_url_timeout_tr=8000. The unit is milliseconds (ms), and the default value is 10,000 ms.