Adds Prometheus export (requests and variant eval) #221
Conversation
Codecov Report
@@ Coverage Diff @@
## master #221 +/- ##
==========================================
- Coverage 85.43% 83.87% -1.56%
==========================================
Files 24 24
Lines 1387 1433 +46
==========================================
+ Hits 1185 1202 +17
- Misses 146 174 +28
- Partials 56 57 +1
Continue to review full report at Codecov.
|
| return | ||
| } | ||
| config.Global.Prometheus.EvalCounter.WithLabelValues( | ||
| util.SafeStringWithDefault(r.EvalContext.EntityType, "null"), |
There was a problem hiding this comment.
not very familiar with the convention of prometheus lable values, does it need to have a name in the value? For example,
fmt.Sprintf("flag_id:%s", util.SafeStringWithDefault(r.FlagID, "null"))
Otherwise, how can we know if it's variantID or flagID, or something else?
There was a problem hiding this comment.
@zhouzhuojie that's a bit confusing, yeah - the labels are coded in when the Counter is created here: https://github.com/checkr/flagr/pull/221/files#diff-091ace02b49a46f55f2586cfec89a160R96
|
|
||
| func TestSetupPrometheusWithLatencies(t *testing.T) { | ||
| prometheus.DefaultRegisterer = prometheus.NewRegistry() | ||
| Config.PrometheusEnabled = true |
There was a problem hiding this comment.
Can you reset the configuration after the test?
There was a problem hiding this comment.
sure, no problem
| Config.PrometheusEnabled = false | ||
| setupPrometheus() | ||
| assert.Nil(t, Global.Prometheus.EvalCounter) | ||
| Config.PrometheusEnabled = true |
There was a problem hiding this comment.
done, just pushed
| PrometheusEnabled bool `env:"FLAGR_PROMETHEUS_ENABLED" envDefault:"false"` | ||
| // PrometheusPath - set the path on which prometheus metrics are available to scrape | ||
| PrometheusPath string `env:"FLAGR_PROMETHEUS_PATH" envDefault:"/metrics"` | ||
| // PrometheusIncludeLatencyHistogram - set whether Prometheus should also export a histogram of request latencies (this increases cardinality significantly) |
|
@zhouzhuojie thanks! |
Description
This PR adds a Prometheus exporter based on the prometheus client-go package. It exports metrics regarding the requests (a count of requests as well as optionally a histogram of request latencies), and a count of variant allocations per flag (similar to datadog statsd).
Motivation and Context
Closes #196
We are trialing flagr at Ecosia, and we use Prometheus for monitoring all our systems running Kubernetes.
How Has This Been Tested?
Added unit tests for config, including creating the prometheus metrics (which will panic if they cannot be created). I also built the docker image and am running it in our dev cluster.
Types of changes
Checklist: