2.2 KiB
TODO
-
we are reporting the size and time in
io.instruments.goas both a counter and a histogram (count can be extracted from histogram in all "backends": prometheus, datadog, new relic, etc..?) -
allow different formats for the propagation of the trace (the
TextMapPropagator) in endpoint.go and in state.go. -
review how we pass the state.
-
we cannot use
skipPathswith the same value that we have to define the endpoints at the global layer, because we cannot know the matched pattern -
clean shutdown
-
add configuration options to allow to use grpc credentials
-
allow to tweak the
bucketlimits for different histograms (like latency and size)http/client/transport_metrics.go:timeBucketsOpt,sizeBucketsOpthttp/server/metrics.go:timeBucketsOpt,sizeBucketsOptio/instrumens.go:timeBucketsOpt,sizeBucketsOptlura/proxy.go: timeBucketsOpt, sizeBucketsOpt
-
review all metric instruments and procive the correct
metric.WithUnitvalue.
TO DECIDE
-
do we want to have the
service_nameto override the global ServiceConfig name for the KrakenD gateway ? -
in
exporter/prometheus/prometheus.gowe could add a config option that will add a prefix to all reported metrics (using theWithNamespaceoption). -
in
lura/backend.goandlura/proxy.gowe are setting the static attrs, and usingsemconv.ServerAddress, to set the concatenated list of hosts. Is that something that we want ?
TO CHECK
-
in
exporter/otelcollector/otelcollector.go, theWithEndpointstates that no http schema should be in cluded (nor path). To usehttpreporting methods, theWithInsecureoption should be used (and to user a pathWithURLPath). Check that current implementation works as expected, or fix it. -
in
http/client/transport.gowe have commented out theStartOptionsfor trace: review if would be useful to expose that in the config.
Known Issues
There is an issue, that we might have already started an span at the global layer, because we do not know if that path had to be ignored. So, at the global layer, there is no way to skip the paths