prometheus: Add S3 4xx and 5xx S3 monitoring (#15052)

Currently minio_s3_requests_errors_total covers 4xx and 
5xx S3 responses which can be confusing when s3 applications 
sent a lot of HEAD requests with obvious 404 responses or 
when the replication is enabled.

Add 
- minio_s3_requests_4xx_errors_total
- minio_s3_requests_5xx_errors_total

to help users monitor 4xx and 5xx HTTP status codes separately.
This commit is contained in:
Anis Elleuch
2022-06-08 19:22:34 +01:00
committed by GitHub
parent 2420f6c000
commit 5fb420c703
5 changed files with 75 additions and 23 deletions

View File

@@ -148,6 +148,8 @@ type HTTPStats struct {
currentS3Requests HTTPAPIStats
totalS3Requests HTTPAPIStats
totalS3Errors HTTPAPIStats
totalS34xxErrors HTTPAPIStats
totalS35xxErrors HTTPAPIStats
totalS3Canceled HTTPAPIStats
}
@@ -178,6 +180,12 @@ func (st *HTTPStats) toServerHTTPStats() ServerHTTPStats {
serverStats.TotalS3Errors = ServerHTTPAPIStats{
APIStats: st.totalS3Errors.Load(),
}
serverStats.TotalS34xxErrors = ServerHTTPAPIStats{
APIStats: st.totalS34xxErrors.Load(),
}
serverStats.TotalS35xxErrors = ServerHTTPAPIStats{
APIStats: st.totalS35xxErrors.Load(),
}
serverStats.TotalS3Canceled = ServerHTTPAPIStats{
APIStats: st.totalS3Canceled.Load(),
}
@@ -186,27 +194,29 @@ func (st *HTTPStats) toServerHTTPStats() ServerHTTPStats {
// Update statistics from http request and response data
func (st *HTTPStats) updateStats(api string, r *http.Request, w *logger.ResponseWriter) {
// A successful request has a 2xx response code or < 4xx response
successReq := w.StatusCode >= 200 && w.StatusCode < 400
if !strings.HasSuffix(r.URL.Path, prometheusMetricsPathLegacy) ||
!strings.HasSuffix(r.URL.Path, prometheusMetricsV2ClusterPath) ||
!strings.HasSuffix(r.URL.Path, prometheusMetricsV2NodePath) {
st.totalS3Requests.Inc(api)
if !successReq {
switch w.StatusCode {
case 0:
case 499:
// 499 is a good error, shall be counted as canceled.
st.totalS3Canceled.Inc(api)
default:
st.totalS3Errors.Inc(api)
}
}
// Ignore non S3 requests
if strings.HasSuffix(r.URL.Path, minioReservedBucketPathWithSlash) {
return
}
// Increment the prometheus http request response histogram with appropriate label
httpRequestsDuration.With(prometheus.Labels{"api": api}).Observe(w.TimeToFirstByte.Seconds())
code := w.StatusCode
switch {
case code == 0:
case code == 499:
// 499 is a good error, shall be counted as canceled.
st.totalS3Canceled.Inc(api)
case code >= http.StatusBadRequest:
st.totalS3Errors.Inc(api)
if code >= http.StatusInternalServerError {
st.totalS35xxErrors.Inc(api)
} else {
st.totalS34xxErrors.Inc(api)
}
}
}
// Prepare new HTTPStats structure