pkg/trace/api: limit simultaneous otlp requests, do not drop payloads

[knusbaum]

What does this PR do?

This commit causes the Trace Agent OTLP Receiver to not drop traces when the payload channel is full.

This commit also limits the number of simultaneous RPC requests, forwarding the backpressure from the pipeline to the client, and limiting our memory to O(MaxConnections * MaxStreams * MaxPayloadSize) in the receiver, with MaxStreams being 1.

Finally, it eliminates the trace buffer in the collector component exporter, preventing the core agent from buffering GiBs of traces under load.

All of these changes reduce trace drops while preventing uncontrolled memory growth.

Motivation

Previously, the OTLP Receiver in the Trace Agent dropped traces when the processor channel was full. This means that the receiver was dropping traces unnecessarily, and a better strategy is to block on the send. This will cause backpressure up through the otlp pipeline.

By default, a gRPC server will allow practically unlimited concurrent RPC requests, spawning a goroutine for each. The routine will read and deserialize the payload before calling the handler, which is now blocking.

This means that our memory usage is not reasonably bounded, since we could be holding hundreds of payloads in memory waiting to be processed. Limiting the number of concurrent RPC requests is required.

Finally, the backpressure goes to the core agent collector component. Here we need to remove the trace buffer to prevent GiBs of trace payloads from being buffered when it feels the backpressure from the Trace Agent.

Additional Notes

Possible Drawbacks / Trade-offs

Eliminating the trace buffer in the collector component could in theory result in higher trace loss in certain bursty scenarios. In practice, we don't see such scenarios and so this is a fair trade-off.

Describe how to test/QA your changes

We will use the new OTel load testing environment to exercise this change.

We should compare:

throughput (are traces being dropped)
memory usage (is memory usage better or worse)
See @knusbaum for details about more specifics regarding infrastructure.

[pr-commenter]

Bloop Bleep... Dogbot Here

Regression Detector Results

Run ID: c0508a00-2993-4501-bd7c-688aaba42bf2
Baseline: a82a8f9
Comparison: 31698d2
Total CPUs: 7

Performance changes are noted in the perf column of each table:

• ✅ = significantly better comparison variant performance
• ❌ = significantly worse comparison variant performance
• ➖ = no significant change in performance

Experiments with missing or malformed data

• basic_py_check

Usually, this warning means that there is no usable optimization goal data for that experiment, which could be a result of misconfiguration.

No significant changes in experiment optimization goals

Confidence level: 90.00%
Effect size tolerance: |Δ mean %| ≥ 5.00%

There were no significant changes in experiment optimization goals at this confidence level and effect size tolerance.

Experiments ignored for regressions

Regressions in experiments with settings containing erratic: true are ignored.

perf	experiment	goal	Δ mean %	Δ mean % CI
➖	file_to_blackhole	% cpu utilization	-1.98	[-8.47, +4.52]

Fine details of change detection per experiment

perf	experiment	goal	Δ mean %	Δ mean % CI
➖	otel_to_otel_logs	ingress throughput	+1.38	[+0.76, +2.00]
➖	process_agent_real_time_mode	memory utilization	+0.33	[+0.29, +0.37]
➖	uds_dogstatsd_to_api_cpu	% cpu utilization	+0.27	[-1.19, +1.72]
➖	idle	memory utilization	+0.07	[+0.04, +0.10]
➖	tcp_syslog_to_blackhole	ingress throughput	+0.06	[+0.00, +0.12]
➖	file_tree	memory utilization	+0.03	[-0.06, +0.12]
➖	trace_agent_json	ingress throughput	+0.02	[-0.02, +0.06]
➖	process_agent_standard_check	memory utilization	+0.02	[-0.02, +0.06]
➖	tcp_dd_logs_filter_exclude	ingress throughput	+0.00	[-0.00, +0.00]
➖	trace_agent_msgpack	ingress throughput	+0.00	[-0.00, +0.00]
➖	uds_dogstatsd_to_api	ingress throughput	+0.00	[-0.00, +0.00]
➖	process_agent_standard_check_with_stats	memory utilization	-0.40	[-0.43, -0.37]
➖	file_to_blackhole	% cpu utilization	-1.98	[-8.47, +4.52]

Explanation

A regression test is an A/B test of target performance in a repeatable rig, where "performance" is measured as "comparison variant minus baseline variant" for an optimization goal (e.g., ingress throughput). Due to intrinsic variability in measuring that goal, we can only estimate its mean value for each experiment; we report uncertainty in that value as a 90.00% confidence interval denoted "Δ mean % CI".

For each experiment, we decide whether a change in performance is a "regression" -- a change worth investigating further -- if all of the following criteria are true:

1. Its estimated |Δ mean %| ≥ 5.00%, indicating the change is big enough to merit a closer look.

2. Its 90.00% confidence interval "Δ mean % CI" does not contain zero, indicating that if our statistical model is accurate, there is at least a 90.00% chance there is a difference in performance between baseline and comparison variants.

3. Its configuration does not mark it "erratic".

[knusbaum]

Re: regression detector,
This is a false positive, as we're touching nothing remotely related to UDS or dogstatsd.

[hestonhoffman]

Small edit for consistency

[dineshg13]

/merge

[dd-devflow]

🚂 MergeQueue

Pull request added to the queue.

This build is going to start soon! (estimated merge in less than 49m)

Use /merge -c to cancel this operation!

[dd-devflow]

❌ MergeQueue

This PR is rejected because it was updated

If you need support, contact us on Slack #ci-interfaces with those details!

[dineshg13]

/merge

[dd-devflow]

🚂 MergeQueue

Pull request added to the queue.

This build is going to start soon! (estimated merge in less than 49m)

Use /merge -c to cancel this operation!