Combine global/local logic into one unified flush and parallelize sinks. #292
Conversation
flusher.go
Outdated
| span := tracer.StartSpan("flush").(*trace.Span) | ||
| defer span.Finish() | ||
|
|
||
| // TODO Move this to an independent ticker routine or something? |
There was a problem hiding this comment.
So, I don't think we can do this in a separate goroutine, because we want this to be reporting the memory at the time of maximum usage (right before we clear and flush everything).
The alternate way of looking at it: we need to do it in a separate ticker, and choose our interval for that ticker such that we are sampling the underlying, sinusoidal data accurately.
Either way, this would also be a great place to report softnet data, once this is merged: #291
flusher.go
Outdated
|
|
||
| go s.metricSinks[0].FlushEventsChecks(span.Attach(ctx), events, checks) // we can do all of this separately | ||
| go s.flushTraces(span.Attach(ctx)) | ||
| go s.flushTraces(span.Attach(ctx)) // this too! |
There was a problem hiding this comment.
I think this comment no longer makes sense?
flusher_test.go
Outdated
|
|
||
| server.HandleTracePacket(packet) | ||
| server.Flush() | ||
| server.Flush(context.TODO()) |
There was a problem hiding this comment.
I think we actually want context.Background() for all of these, no?
There was a problem hiding this comment.
Ha! Yes, I was doing TODO out of habit for a temporary fix when in fact Background was the permanent answer here. :) Thanks!
flusher.go
Outdated
| go func(ms metricSink) { | ||
| err := ms.Flush(span.Attach(ctx), finalMetrics) | ||
| if err != nil { | ||
| log.WithError(err).WithField("sink", ms.Name()).Warn("Error flushin sink") |
| for _, sink := range s.metricSinks { | ||
| wg.Add(1) | ||
| go func(ms metricSink) { | ||
| err := ms.Flush(span.Attach(ctx), finalMetrics) |
There was a problem hiding this comment.
Now that this is happening asynchronously, we'll want to make doubly-sure that our interface definition specifies that the Flush method is not allowed to mutate the slice it receives.
There was a problem hiding this comment.
Definitely! How do we do that? :)
There was a problem hiding this comment.
I'd add it to the interface definition in metric_sink.go.
Gerald Rule: Copy Observability on Veneur and Unilog pull requestscc @stripe/observability |
cory-stripe
force-pushed
the
cory-metric-sink-cleanup
branch
from
b70830c to
979f3bb
Compare
|
re-r? @aditya-stripe |
plugin_test.go
Outdated
| } | ||
|
|
||
| f.server.Flush() | ||
| f.server.Flush(context.TODO()) |
There was a problem hiding this comment.
nit: we also want Background() here too. context.TODO is intended for functions that are not yet fully traced, and for static analysis tools that can track whether a context originates with a TODO or a BACKGROUND.
|
One comment. lgtm after that, so feel free to self-approve :) |
cory-stripe
force-pushed
the
cory-metric-sink-cleanup
branch
from
4262f3c to
142613d
Compare
cory-stripe
force-pushed
the
cory-metric-sink-cleanup
branch
from
142613d to
a137e5c
Compare
|
Self-approval post rebase |
Summary
Improve the Server's
Flushwork from callingFlushLocalandFlushGlobal, which was a poorly compartmentalized implementation that led to code duplication and confusion.Motivation
At the end of the Datadog sink PR we are in a world with a single, hardcoded sink. This isn't the intent! We want to convert plugins to sinks and execute all of them.
This patch aims to do just that, but also takes an opportunity to refactor away the
FlushLocalandFlushGlobalfunctions. This was done by first inlining them, removing duplicate code, and then refactoring the special cases away. Notably, thetallyMetricssize calculation.Notes
context.TODO()sprinkled around because we want to implement a deadline for sinks. I will fix this.Test plan
Existing tests.
r? @aditya-stripe