jiazhizhong/higress
1
0
Fork 0
mirror of https://github.com/alibaba/higress.git synced 2026-05-29 23:27:28 +08:00
Code Issues Packages Projects Releases Wiki Activity

feat: support retry on http status code (#1817)

Browse Source 
Co-authored-by: Kent Dong <ch3cho@qq.com>
This commit is contained in:
Se7en
2025-03-11 13:38:02 +08:00
committed by GitHub
parent beb60fcacd
commit 19496e5759
7 changed files with 72 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
plugins/wasm-go/extensions/ai-proxy/provider/failover.go
View File

@@ -30,6 +30,8 @@ type failover struct {
healthCheckTimeout int64 `required:"false" yaml:"healthCheckTimeout" json:"healthCheckTimeout"`
// @Title zh-CN 健康检测使用的模型
healthCheckModel string `required:"false" yaml:"healthCheckModel" json:"healthCheckModel"`
// @Title zh-CN 需要进行 failover 的原始请求的状态码，支持正则表达式匹配
failoverOnStatus []string `required:"false" yaml:"failoverOnStatus" json:"failoverOnStatus"`
// @Title zh-CN 本次请求使用的 apiToken
ctxApiTokenInUse string
// @Title zh-CN 记录本次请求时所有可用的 apiToken
@@ -92,6 +94,14 @@ func (f *failover) FromJson(json gjson.Result) {
f.healthCheckTimeout = 5000
}
f.healthCheckModel = json.Get("healthCheckModel").String()
for _, status := range json.Get("failoverOnStatus").Array() {
f.failoverOnStatus = append(f.failoverOnStatus, status.String())
}
// If failoverOnStatus is empty, default to retry on 4xx and 5xx
if len(f.failoverOnStatus) == 0 {
f.failoverOnStatus = []string{"4.*", "5.*"}
}
}
func (f *failover) Validate() error {
@@ -558,17 +568,21 @@ func (c *ProviderConfig) resetSharedData() {
_ = proxywasm.SetSharedData(c.failover.ctxApiTokenRequestFailureCount, nil, 0)
}
func (c *ProviderConfig) OnRequestFailed(activeProvider Provider, ctx wrapper.HttpContext, apiTokenInUse string, apiTokens []string, log wrapper.Log) types.Action {
if c.isFailoverEnabled() {
func (c *ProviderConfig) OnRequestFailed(activeProvider Provider, ctx wrapper.HttpContext, apiTokenInUse string, apiTokens []string, status string, log wrapper.Log) types.Action {
if c.isFailoverEnabled() && util.MatchStatus(status, c.failover.failoverOnStatus) {
c.handleUnavailableApiToken(ctx, apiTokenInUse, log)
}
if c.isRetryOnFailureEnabled() && ctx.GetContext(ctxKeyIsStreaming) != nil && !ctx.GetContext(ctxKeyIsStreaming).(bool) {
if c.isRetryOnFailureEnabled() && util.MatchStatus(status, c.retryOnFailure.retryOnStatus) && isNotStreamingResponse(ctx) {
c.retryFailedRequest(activeProvider, ctx, apiTokenInUse, apiTokens, log)
return types.HeaderStopAllIterationAndWatermark
}
return types.ActionContinue
}
func isNotStreamingResponse(ctx wrapper.HttpContext) bool {
return ctx.GetContext(ctxKeyIsStreaming) != nil && !ctx.GetContext(ctxKeyIsStreaming).(bool)
}
func (c *ProviderConfig) GetApiTokenInUse(ctx wrapper.HttpContext) string {
token, _ := ctx.GetContext(c.failover.ctxApiTokenInUse).(string)
return token