Improvements on reassembling back annotations from InfluxDB.

[chris-ramon]

Details

Issue: #117

• [x] tl;dr - Improves how we reassemble InfluxDB fields to span annotations:
  • Updates func annotationsFromRow(...) to include common logic that we use to reassemble data back from InfluxDB fields to span annotations. Now it uses func annotationsFromEvents(..) which will discard those annotations that should no be included because were not explicitly save by Collect(...) but that are present because InfluxDB will return all the fields present on spans measurement:
  • Eg. A field named Client.Request.Method might contain values for sub-spans but not for root-spans. Since span annotations are non-deterministic, current strategy for query those annotations(fields) is to do select * from spans ...., therefore: Client.Request.Method will be included in the query result even for root-spans:

[chris-ramon]

tl:dr —There's not performance impact introduced to the InfluxDBStore.Trace(...) & InfluxDBStre.Traces(...) with this PR.

I went ahead and re-run benchmarks from #114 for InfluxDBStore.Trace(...) & InfluxDBStore.Traces(...) in order to confirm if there are big performance impact with changes introduced in this PR because were are using unmarshalling/marshalling via: UnmarshalEvents & MarshalEvent.

---

InfluxDBStore.Trace(...)

Before:

$ go test -run=XYZ -bench BenchmarkInfluxDBStoreTrace. -benchtime=40s > trace
$ more trace

PASS
BenchmarkInfluxDBStoreTrace100-4             200         411849069 ns/op
BenchmarkInfluxDBStoreTrace250-4              50        1051788921 ns/op
BenchmarkInfluxDBStoreTrace1000-4             10        4227713513 ns/op
ok      sourcegraph.com/sourcegraph/appdash     380.156s

After:

$ go test -run=XYZ -bench BenchmarkInfluxDBStoreTrace[0-9]+ -benchtime=40s > trace
$ more trace

PASS
BenchmarkInfluxDBStoreTrace100-4             100         412541990 ns/op
BenchmarkInfluxDBStoreTrace250-4              50        1079761891 ns/op
BenchmarkInfluxDBStoreTrace1000-4             10        4208731896 ns/op
ok      sourcegraph.com/sourcegraph/appdash     294.010s

---

InfluxDBStore.Traces(...)

Before:

$ go test -run=XYZ -bench BenchmarkInfluxDBStoreTracesDefaultPerPage > traces
$ more traces

PASS
BenchmarkInfluxDBStoreTracesDefaultPerPage-4         100          13433622 ns/op
ok      sourcegraph.com/sourcegraph/appdash     10.595s

After:

$ go test -run=XYZ -bench BenchmarkInfluxDBStoreTracesDefaultPerPage > traces
$ more traces
PASS
BenchmarkInfluxDBStoreTracesDefaultPerPage-4         100          13300171 ns/op
ok      sourcegraph.com/sourcegraph/appdash     11.540s

[slimsag]

I hope to test this change with our app soon to confirm, but this looks like a very good change. This seems much more correct than #119 and using using MarshalEvent/UnmarshalEvent seems correct to me (but also highlights why Appdash should eventually move to a more simplified storage modal).

I'll follow-up with more issues if I spot them in my testing. Thanks Chris! :+1: