AML-1804 Asynchronous dogstatsd flush

[vickenty]

What does this PR do?

Add an option (off by default) to perform the bulk of the dogstatsd flushing without blocking the rest of the pipeline.

Motivation

Dogstatsd server has a queue that buffers incoming packets if they can not be processed immediately. Flushing large number of metrics can block the pipeline for a long time (around 1-2 seconds) leading to queue to buffer a lot of packets.

Additional Notes

Processing new metrics needs to write to the context map when it encounters a new metric context. At the same time, flush process reads the same map to extract context information for metrics being flushed. To avoid lock overhead, asynchronous flush instead copies the context map, which is fast (~100ms) and needs less space than the buffer queue. Then we extract and remove buckets that need to be flushed from the working set. After this, metrics processing can resume, while the flush process works in background.

Possible Drawbacks / Trade-offs

Describe how to test/QA your changes

[github-actions[bot]]

Serverless Benchmark Results

BenchmarkStartEndInvocation comparison between 0e3a2a61f13e4a208b0b0fa40cd3d2626b938054 and 61ce86493b56372d3cc670f401c847817ba463bb.

tl;dr

Use these benchmarks as an insight tool during development.

1. Skim down the vs base column in each chart. If there is a ~, then there was no statistically significant change to the benchmark. Otherwise, ensure the estimated percent change is either negative or very small.

2. The last row of each chart is the geomean. Ensure this percentage is either negative or very small.

What is this benchmarking?

The BenchmarkStartEndInvocation compares the amount of time it takes to call the start-invocation and end-invocation endpoints. For universal instrumentation languages (Dotnet, Golang, Java, Ruby), this represents the majority of the duration overhead added by our tracing layer.

The benchmark is run using a large variety of lambda request payloads. In the charts below, there is one row for each event payload type.

How do I interpret these charts?

The charts below comes from benchstat. They represent the statistical change in duration (sec/op), memory overhead (B/op), and allocations (allocs/op).

The benchstat docs explain how to interpret these charts.

Before the comparison table, we see common file-level configuration. If there are benchmarks with different configuration (for example, from different packages), benchstat will print separate tables for each configuration.

The table then compares the two input files for each benchmark. It shows the median and 95% confidence interval summaries for each benchmark before and after the change, and an A/B comparison under "vs base". ... The p-value measures how likely it is that any differences were due to random chance (i.e., noise). The "~" means benchstat did not detect a statistically significant difference between the two inputs. ...

Note that "statistically significant" is not the same as "large": with enough low-noise data, even very small changes can be distinguished from noise and considered statistically significant. It is, of course, generally easier to distinguish large changes from noise.

Finally, the last row of the table shows the geometric mean of each column, giving an overall picture of how the benchmarks changed. Proportional changes in the geomean reflect proportional changes in the benchmarks. For example, given n benchmarks, if sec/op for one of them increases by a factor of 2, then the sec/op geomean will increase by a factor of ⁿ√2.

I need more help

First off, do not worry if the benchmarks are failing. They are not tests. The intention is for them to be a tool for you to use during development.

If you would like a hand interpreting the results come chat with us in #serverless-agent in the internal DataDog slack or in #serverless in the public DataDog slack. We're happy to help!

Benchmark stats

goos: linux
goarch: amd64
pkg: github.com/DataDog/datadog-agent/pkg/serverless/daemon
cpu: AMD EPYC 7763 64-Core Processor                
                                      │ baseline/benchmark.log │       current/benchmark.log        │
                                      │         sec/op         │   sec/op     vs base               │
api-gateway-appsec.json                            86.56µ ± 4%   87.65µ ± 2%       ~ (p=0.247 n=10)
api-gateway-kong-appsec.json                       68.09µ ± 2%   68.09µ ± 1%       ~ (p=0.670 n=10)
api-gateway-kong.json                              67.47µ ± 2%   67.22µ ± 2%       ~ (p=0.436 n=10)
api-gateway-non-proxy-async.json                   105.7µ ± 1%   106.2µ ± 1%       ~ (p=0.481 n=10)
api-gateway-non-proxy.json                         107.1µ ± 1%   106.6µ ± 1%       ~ (p=0.280 n=10)
api-gateway-websocket-connect.json                 70.63µ ± 1%   70.09µ ± 1%       ~ (p=0.101 n=10)
api-gateway-websocket-default.json                 63.51µ ± 1%   62.98µ ± 2%       ~ (p=0.149 n=10)
api-gateway-websocket-disconnect.json              63.97µ ± 2%   63.31µ ± 1%  -1.02% (p=0.029 n=10)
api-gateway.json                                   117.3µ ± 3%   115.3µ ± 2%       ~ (p=0.063 n=10)
application-load-balancer.json                     64.57µ ± 1%   64.07µ ± 1%       ~ (p=0.165 n=10)
cloudfront.json                                    48.35µ ± 2%   47.27µ ± 1%  -2.23% (p=0.000 n=10)
cloudwatch-events.json                             39.32µ ± 2%   37.36µ ± 3%  -4.99% (p=0.000 n=10)
cloudwatch-logs.json                               67.94µ ± 2%   66.42µ ± 1%  -2.24% (p=0.000 n=10)
custom.json                                        30.92µ ± 3%   31.36µ ± 2%       ~ (p=0.353 n=10)
dynamodb.json                                      95.16µ ± 2%   95.00µ ± 1%       ~ (p=0.529 n=10)
empty.json                                         29.59µ ± 2%   29.43µ ± 2%       ~ (p=0.247 n=10)
eventbridge-custom.json                            42.12µ ± 2%   42.14µ ± 2%       ~ (p=0.927 n=10)
http-api.json                                      72.47µ ± 1%   72.56µ ± 2%       ~ (p=0.631 n=10)
kinesis-batch.json                                 70.53µ ± 1%   70.53µ ± 1%       ~ (p=0.971 n=10)
kinesis.json                                       54.48µ ± 2%   53.46µ ± 2%  -1.86% (p=0.011 n=10)
s3.json                                            60.88µ ± 2%   59.74µ ± 1%       ~ (p=0.143 n=10)
sns-batch.json                                     89.53µ ± 1%   89.81µ ± 1%       ~ (p=0.481 n=10)
sns.json                                           64.41µ ± 1%   65.26µ ± 1%  +1.32% (p=0.012 n=10)
snssqs.json                                        111.5µ ± 3%   114.0µ ± 2%  +2.23% (p=0.023 n=10)
snssqs_no_dd_context.json                          97.15µ ± 2%   99.26µ ± 2%  +2.17% (p=0.009 n=10)
sqs-aws-header.json                                54.59µ ± 3%   56.15µ ± 3%  +2.86% (p=0.015 n=10)
sqs-batch.json                                     92.47µ ± 2%   94.96µ ± 1%  +2.70% (p=0.001 n=10)
sqs.json                                           67.98µ ± 2%   70.38µ ± 2%  +3.53% (p=0.000 n=10)
sqs_no_dd_context.json                             62.02µ ± 5%   63.86µ ± 1%  +2.96% (p=0.015 n=10)
geomean                                            67.33µ        67.36µ       +0.04%

                                      │ baseline/benchmark.log │        current/benchmark.log        │
                                      │          B/op          │     B/op      vs base               │
api-gateway-appsec.json                           37.26Ki ± 0%   37.32Ki ± 0%  +0.18% (p=0.000 n=10)
api-gateway-kong-appsec.json                      26.91Ki ± 0%   26.92Ki ± 0%       ~ (p=0.591 n=10)
api-gateway-kong.json                             24.41Ki ± 0%   24.42Ki ± 0%       ~ (p=0.223 n=10)
api-gateway-non-proxy-async.json                  48.01Ki ± 0%   48.07Ki ± 0%  +0.13% (p=0.000 n=10)
api-gateway-non-proxy.json                        47.24Ki ± 0%   47.30Ki ± 0%  +0.12% (p=0.000 n=10)
api-gateway-websocket-connect.json                25.44Ki ± 0%   25.47Ki ± 0%  +0.11% (p=0.000 n=10)
api-gateway-websocket-default.json                21.35Ki ± 0%   21.39Ki ± 0%  +0.17% (p=0.000 n=10)
api-gateway-websocket-disconnect.json             21.14Ki ± 0%   21.17Ki ± 0%  +0.17% (p=0.000 n=10)
api-gateway.json                                  49.53Ki ± 0%   49.54Ki ± 0%       ~ (p=0.138 n=10)
application-load-balancer.json                    22.32Ki ± 0%   23.26Ki ± 0%  +4.21% (p=0.000 n=10)
cloudfront.json                                   17.64Ki ± 0%   17.66Ki ± 0%  +0.09% (p=0.049 n=10)
cloudwatch-events.json                            11.69Ki ± 0%   11.71Ki ± 0%  +0.19% (p=0.000 n=10)
cloudwatch-logs.json                              53.38Ki ± 0%   53.37Ki ± 0%       ~ (p=0.516 n=10)
custom.json                                       9.718Ki ± 0%   9.730Ki ± 0%  +0.13% (p=0.009 n=10)
dynamodb.json                                     40.69Ki ± 0%   40.69Ki ± 0%       ~ (p=0.896 n=10)
empty.json                                        9.277Ki ± 0%   9.296Ki ± 0%       ~ (p=0.063 n=10)
eventbridge-custom.json                           13.41Ki ± 0%   13.44Ki ± 0%  +0.24% (p=0.014 n=10)
http-api.json                                     23.70Ki ± 0%   23.80Ki ± 0%  +0.41% (p=0.000 n=10)
kinesis-batch.json                                27.00Ki ± 0%   27.04Ki ± 0%  +0.12% (p=0.009 n=10)
kinesis.json                                      17.80Ki ± 0%   17.83Ki ± 0%  +0.15% (p=0.011 n=10)
s3.json                                           20.33Ki ± 0%   20.35Ki ± 0%       ~ (p=0.239 n=10)
sns-batch.json                                    38.65Ki ± 0%   38.63Ki ± 0%       ~ (p=0.353 n=10)
sns.json                                          23.95Ki ± 0%   23.99Ki ± 0%  +0.17% (p=0.035 n=10)
snssqs.json                                       50.75Ki ± 0%   50.81Ki ± 0%  +0.11% (p=0.007 n=10)
snssqs_no_dd_context.json                         44.78Ki ± 0%   44.89Ki ± 0%  +0.24% (p=0.000 n=10)
sqs-aws-header.json                               18.79Ki ± 0%   18.83Ki ± 0%       ~ (p=0.143 n=10)
sqs-batch.json                                    41.64Ki ± 0%   41.66Ki ± 0%       ~ (p=0.239 n=10)
sqs.json                                          25.54Ki ± 0%   25.59Ki ± 0%       ~ (p=0.755 n=10)
sqs_no_dd_context.json                            20.69Ki ± 1%   20.68Ki ± 0%       ~ (p=0.839 n=10)
geomean                                           25.70Ki        25.76Ki       +0.26%

                                      │ baseline/benchmark.log │        current/benchmark.log        │
                                      │       allocs/op        │ allocs/op   vs base                 │
api-gateway-appsec.json                             629.5 ± 0%   630.0 ± 0%       ~ (p=1.000 n=10)
api-gateway-kong-appsec.json                        488.0 ± 0%   488.0 ± 0%       ~ (p=1.000 n=10) ¹
api-gateway-kong.json                               466.0 ± 0%   466.0 ± 0%       ~ (p=1.000 n=10)
api-gateway-non-proxy-async.json                    725.0 ± 0%   726.0 ± 0%       ~ (p=0.370 n=10)
api-gateway-non-proxy.json                          716.0 ± 0%   716.0 ± 0%       ~ (p=1.000 n=10)
api-gateway-websocket-connect.json                  453.0 ± 0%   453.0 ± 0%       ~ (p=1.000 n=10)
api-gateway-websocket-default.json                  379.0 ± 0%   379.0 ± 0%       ~ (p=0.303 n=10)
api-gateway-websocket-disconnect.json               370.0 ± 0%   370.0 ± 0%       ~ (p=0.582 n=10)
api-gateway.json                                    791.0 ± 0%   791.0 ± 0%       ~ (p=0.582 n=10)
application-load-balancer.json                      352.0 ± 0%   353.0 ± 0%  +0.28% (p=0.000 n=10)
cloudfront.json                                     284.0 ± 0%   284.0 ± 0%       ~ (p=0.211 n=10)
cloudwatch-events.json                              220.0 ± 0%   220.0 ± 0%       ~ (p=1.000 n=10)
cloudwatch-logs.json                                216.0 ± 0%   216.0 ± 0%       ~ (p=1.000 n=10)
custom.json                                         168.0 ± 0%   168.0 ± 0%       ~ (p=1.000 n=10)
dynamodb.json                                       589.0 ± 0%   589.0 ± 0%       ~ (p=0.210 n=10)
empty.json                                          160.0 ± 1%   160.0 ± 1%       ~ (p=1.000 n=10)
eventbridge-custom.json                             254.0 ± 0%   254.0 ± 0%       ~ (p=1.000 n=10)
http-api.json                                       432.5 ± 0%   433.0 ± 0%       ~ (p=0.470 n=10)
kinesis-batch.json                                  390.0 ± 0%   390.0 ± 0%       ~ (p=1.000 n=10)
kinesis.json                                        285.0 ± 0%   285.0 ± 0%       ~ (p=0.474 n=10)
s3.json                                             358.0 ± 0%   358.0 ± 1%       ~ (p=1.000 n=10)
sns-batch.json                                      455.0 ± 0%   455.0 ± 0%       ~ (p=0.520 n=10)
sns.json                                            322.5 ± 0%   323.0 ± 0%       ~ (p=0.081 n=10)
snssqs.json                                         450.5 ± 0%   450.5 ± 0%       ~ (p=0.977 n=10)
snssqs_no_dd_context.json                           399.0 ± 1%   400.0 ± 0%  +0.25% (p=0.008 n=10)
sqs-aws-header.json                                 273.5 ± 0%   274.0 ± 0%       ~ (p=0.206 n=10)
sqs-batch.json                                      503.5 ± 0%   504.0 ± 0%       ~ (p=0.331 n=10)
sqs.json                                            351.0 ± 0%   351.0 ± 1%       ~ (p=0.952 n=10)
sqs_no_dd_context.json                              324.5 ± 0%   324.0 ± 1%       ~ (p=0.490 n=10)
geomean                                             376.8        376.9       +0.04%
¹ all samples are equal

[pr-commenter[bot]]

Test changes on VM

Use this command from test-infra-definitions to manually test this PR changes on a VM:

inv create-vm --pipeline-id=38165569 --os-family=ubuntu

Note: This applies to commit 77646f64

[pr-commenter[bot]]

Regression Detector

Regression Detector Results

Run ID: 82e10bdc-dbf6-46d0-9ca5-49a277e3b470 Metrics dashboard Target profiles

Baseline: b3e5cab1c83d5156e6619e2d470431afc0306879 Comparison: 77646f648fd98fd163cc61f683da6697340d93bd

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

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.

Fine details of change detection per experiment

perf	experiment	goal	Δ mean %	Δ mean % CI	links
➖	tcp_syslog_to_blackhole	ingress throughput	+8.95	[-4.60, +22.50]	Logs
➖	pycheck_1000_100byte_tags	% cpu utilization	+0.82	[-3.97, +5.61]	Logs
➖	uds_dogstatsd_to_api	ingress throughput	+0.00	[-0.00, +0.00]	Logs
➖	tcp_dd_logs_filter_exclude	ingress throughput	-0.00	[-0.01, +0.01]	Logs
➖	idle	memory utilization	-0.14	[-0.18, -0.10]	Logs
➖	basic_py_check	% cpu utilization	-0.48	[-3.09, +2.13]	Logs
➖	otel_to_otel_logs	ingress throughput	-0.63	[-1.44, +0.18]	Logs
➖	uds_dogstatsd_to_api_cpu	% cpu utilization	-0.65	[-1.54, +0.24]	Logs
➖	file_tree	memory utilization	-0.81	[-0.90, -0.72]	Logs

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".