Perf event writes metric

[geyslan]

1. Explain what the PR does

bae1532 chore(performance): update tracee dashboard
5a5c2e7 chore: add perfbuf metric per event (METRICS=1)
5fe36d1 chore(go.mod): bump api to latest 1f796cb

5a5c2e7 chore: add perfbuf metric per event (METRICS=1)

Enabled only when built with METRICS=1.

BPFPerfEventSubmitAttemptsCount and BPFPerfEventSubmitFailuresCount
count the number of events processed by the eBPF programs and written to
or attempted to be written to the perf buffer.

It is incremented right after the attempt of writing the event to the
perf buffer, making it possible to measure if the that event was
successfully written to the perf buffer or not.

This metric can be used to monitor the performance of individual eBPF
events and to detect potential bottlenecks.

2. Explain how to test it

METRICS=1 make tracee

sudo ./dist/tracee -e socket_dup,execve --metrics --pyroscope --pprof

make -f builder/Makefile.performance dashboard-start

3. Other comments

[geyslan]

I'm working on a stress tool on top of this branch. Gonna let you know soon.

[NDStrahilevitz]

After offline discussion, registering another way of approaching this: by enabling the metric with a flag instead of requiring a rebuild.

It's feasible considering an increment of program sizes, an additional branch check (metric enabled) and a dangling map.

I actually prefer having this in a debug build precisely so that we don't include these overheads in regular user runtimes.

[geyslan]

@NDStrahilevitz @yanivagman I would appreciate your +1.

[yanivagman]

LGTM, see one nit comment

[geyslan]

LGTM, see one nit comment

I'll add a new commit to this to expose the new metrics via grpc. Gonna ping you soon.

[geyslan]

/fast-forward