chore(ci): Use free OSS 4 vCPU runners

[jszwedko]

https://github.blog/2024-01-17-github-hosted-runners-double-the-power-for-open-source/ announced
larger default runners for OSS projects. This should avoid our need to use the paid 4 vCPU runners
(previously they were 2 vCPU).

Signed-off-by: Jesse Szwedko [email protected]

[jszwedko]

/ci-run-all

[github-actions]

Regression Detector Results

Run ID: 63c30b16-28c7-4f89-a197-4e0ec301c0b6
Baseline: c119951
Comparison: bcf4c63
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

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
➖	syslog_log2metric_splunk_hec_metrics	ingress throughput	+2.43	[+2.29, +2.58]
➖	datadog_agent_remap_datadog_logs	ingress throughput	+1.66	[+1.57, +1.76]
➖	syslog_regex_logs2metric_ddmetrics	ingress throughput	+1.49	[+1.41, +1.57]
➖	http_to_http_acks	ingress throughput	+1.35	[+0.03, +2.66]
➖	syslog_humio_logs	ingress throughput	+1.13	[+1.02, +1.23]
➖	file_to_blackhole	egress throughput	+0.86	[-1.75, +3.48]
➖	otlp_http_to_blackhole	ingress throughput	+0.86	[+0.72, +1.00]
➖	syslog_loki	ingress throughput	+0.48	[+0.42, +0.53]
➖	socket_to_socket_blackhole	ingress throughput	+0.41	[+0.33, +0.49]
➖	syslog_splunk_hec_logs	ingress throughput	+0.23	[+0.17, +0.30]
➖	datadog_agent_remap_blackhole	ingress throughput	+0.16	[+0.07, +0.25]
➖	http_to_http_noack	ingress throughput	+0.16	[+0.08, +0.24]
➖	http_to_s3	ingress throughput	+0.12	[-0.17, +0.40]
➖	http_to_http_json	ingress throughput	+0.04	[-0.04, +0.12]
➖	splunk_hec_to_splunk_hec_logs_acks	ingress throughput	+0.00	[-0.16, +0.16]
➖	splunk_hec_indexer_ack_blackhole	ingress throughput	-0.00	[-0.14, +0.14]
➖	syslog_log2metric_humio_metrics	ingress throughput	-0.02	[-0.14, +0.11]
➖	splunk_hec_to_splunk_hec_logs_noack	ingress throughput	-0.06	[-0.18, +0.05]
➖	enterprise_http_to_http	ingress throughput	-0.07	[-0.15, -0.00]
➖	otlp_grpc_to_blackhole	ingress throughput	-0.30	[-0.38, -0.22]
➖	http_elasticsearch	ingress throughput	-0.34	[-0.41, -0.27]
➖	datadog_agent_remap_blackhole_acks	ingress throughput	-0.61	[-0.71, -0.50]
➖	http_text_to_http_json	ingress throughput	-0.71	[-0.85, -0.57]
➖	datadog_agent_remap_datadog_logs_acks	ingress throughput	-0.79	[-0.89, -0.70]
➖	fluent_elasticsearch	ingress throughput	-0.84	[-1.30, -0.39]
➖	syslog_log2metric_tag_cardinality_limit_blackhole	ingress throughput	-1.04	[-1.17, -0.90]
➖	splunk_hec_route_s3	ingress throughput	-2.66	[-3.16, -2.16]

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".

[datadog-vectordotdev]

Datadog Report

Branch report: jszwedko/use-free-4-core-runners
Commit report: d684b0e
Test service: vector

✅ 0 Failed, 2094 Passed, 0 Skipped, 1m 24.67s Wall Time

[github-actions]

Regression Detector Results

Run ID: 556a96a1-7d07-4a52-b7f6-eab5b249e145
Baseline: 857d049
Comparison: 10d237e
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

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
➖	datadog_agent_remap_blackhole_acks	ingress throughput	+0.57	[+0.46, +0.67]
➖	socket_to_socket_blackhole	ingress throughput	+0.40	[+0.33, +0.47]
➖	file_to_blackhole	egress throughput	+0.30	[-2.20, +2.80]
➖	http_elasticsearch	ingress throughput	+0.25	[+0.18, +0.32]
➖	syslog_loki	ingress throughput	+0.22	[+0.15, +0.29]
➖	datadog_agent_remap_datadog_logs	ingress throughput	+0.22	[+0.13, +0.30]
➖	http_to_http_acks	ingress throughput	+0.20	[-1.12, +1.53]
➖	http_to_http_noack	ingress throughput	+0.09	[-0.00, +0.19]
➖	http_to_s3	ingress throughput	+0.07	[-0.21, +0.35]
➖	http_to_http_json	ingress throughput	+0.06	[-0.02, +0.14]
➖	syslog_humio_logs	ingress throughput	+0.03	[-0.08, +0.14]
➖	splunk_hec_to_splunk_hec_logs_acks	ingress throughput	+0.00	[-0.16, +0.16]
➖	splunk_hec_indexer_ack_blackhole	ingress throughput	-0.00	[-0.14, +0.14]
➖	splunk_hec_to_splunk_hec_logs_noack	ingress throughput	-0.06	[-0.17, +0.05]
➖	enterprise_http_to_http	ingress throughput	-0.10	[-0.18, -0.02]
➖	otlp_http_to_blackhole	ingress throughput	-0.29	[-0.46, -0.13]
➖	syslog_log2metric_tag_cardinality_limit_blackhole	ingress throughput	-0.31	[-0.44, -0.18]
➖	datadog_agent_remap_blackhole	ingress throughput	-0.43	[-0.52, -0.33]
➖	fluent_elasticsearch	ingress throughput	-0.45	[-0.91, +0.02]
➖	otlp_grpc_to_blackhole	ingress throughput	-0.49	[-0.58, -0.40]
➖	http_text_to_http_json	ingress throughput	-0.66	[-0.78, -0.54]
➖	splunk_hec_route_s3	ingress throughput	-0.97	[-1.45, -0.48]
➖	datadog_agent_remap_datadog_logs_acks	ingress throughput	-1.37	[-1.46, -1.28]
➖	syslog_log2metric_splunk_hec_metrics	ingress throughput	-1.47	[-1.62, -1.33]
➖	syslog_splunk_hec_logs	ingress throughput	-1.50	[-1.58, -1.42]
➖	syslog_regex_logs2metric_ddmetrics	ingress throughput	-2.31	[-2.41, -2.21]
➖	syslog_log2metric_humio_metrics	ingress throughput	-3.53	[-3.65, -3.41]

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".

[github-actions]

Regression Detector Results

Run ID: 835a8304-1c8e-456c-ba26-bf5f4c5e5ab4
Baseline: a4aff31
Comparison: ce1fada
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

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
➖	syslog_log2metric_humio_metrics	ingress throughput	+4.56	[+4.42, +4.71]
➖	http_text_to_http_json	ingress throughput	+1.19	[+1.06, +1.32]
➖	syslog_log2metric_splunk_hec_metrics	ingress throughput	+0.98	[+0.83, +1.12]
➖	syslog_loki	ingress throughput	+0.87	[+0.79, +0.94]
➖	otlp_grpc_to_blackhole	ingress throughput	+0.61	[+0.52, +0.69]
➖	syslog_log2metric_tag_cardinality_limit_blackhole	ingress throughput	+0.58	[+0.45, +0.71]
➖	http_elasticsearch	ingress throughput	+0.40	[+0.33, +0.48]
➖	socket_to_socket_blackhole	ingress throughput	+0.39	[+0.32, +0.46]
➖	syslog_humio_logs	ingress throughput	+0.23	[+0.13, +0.34]
➖	http_to_s3	ingress throughput	+0.20	[-0.08, +0.49]
➖	http_to_http_noack	ingress throughput	+0.12	[+0.03, +0.22]
➖	http_to_http_json	ingress throughput	+0.04	[-0.04, +0.11]
➖	splunk_hec_indexer_ack_blackhole	ingress throughput	+0.00	[-0.14, +0.15]
➖	splunk_hec_to_splunk_hec_logs_acks	ingress throughput	-0.00	[-0.16, +0.16]
➖	splunk_hec_to_splunk_hec_logs_noack	ingress throughput	-0.04	[-0.15, +0.08]
➖	datadog_agent_remap_blackhole	ingress throughput	-0.10	[-0.19, -0.00]
➖	enterprise_http_to_http	ingress throughput	-0.11	[-0.19, -0.03]
➖	syslog_splunk_hec_logs	ingress throughput	-0.19	[-0.26, -0.12]
➖	splunk_hec_route_s3	ingress throughput	-0.21	[-0.71, +0.29]
➖	datadog_agent_remap_blackhole_acks	ingress throughput	-0.32	[-0.42, -0.22]
➖	syslog_regex_logs2metric_ddmetrics	ingress throughput	-0.60	[-0.71, -0.48]
➖	datadog_agent_remap_datadog_logs_acks	ingress throughput	-0.78	[-0.87, -0.69]
➖	datadog_agent_remap_datadog_logs	ingress throughput	-0.89	[-0.97, -0.80]
➖	fluent_elasticsearch	ingress throughput	-0.96	[-1.42, -0.51]
➖	http_to_http_acks	ingress throughput	-1.10	[-2.41, +0.21]
➖	otlp_http_to_blackhole	ingress throughput	-1.17	[-1.34, -1.01]
➖	file_to_blackhole	egress throughput	-1.51	[-3.98, +0.96]

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".

[kapilsinha]

Noticed that most builds (certainly all nightly, and even custom builds e.g. https://github.com/vectordotdev/vector/actions/runs/7833275067/job/21374004966) take over an hour to complete. Curious, how sizable a speed-up would there be by increasing CPUs?
Also noticed that there is no caching step, so the initial setup seems to take the lion's share of the time. Is caching something the team has tried before?

[jszwedko]

Noticed that most builds (certainly all nightly, and even custom builds e.g. https://github.com/vectordotdev/vector/actions/runs/7833275067/job/21374004966) take over an hour to complete. Curious, how sizable a speed-up would there be by increasing CPUs? Also noticed that there is no caching step, so the initial setup seems to take the lion's share of the time. Is caching something the team has tried before?

Thanks for the thoughts @kapilsinha ! We have looked into caching before using sccache but removed it because it was causing issues. We'd also like to look into building "build images" with cached build dependencies since we spend a good amount of time running apt update and friends.

Also, we aren't as sensitive to the build times of the nightly and custom build workflows because those run infrequently ("nightly" for the nightly build and on-demand for custom builds) rather than as part of CI on pull requests. The bulk of the time of those builds, though I haven't looked closely recently, is in the linking stage so increasing CPU count wouldn't help with that

[kapilsinha]

@jszwedko +1 on the build images, that makes a lot of sense
If linking eats up a lot of time, does using a faster linker (I know the mold linker is now MIT licensed and works well on Linux) help?

Frustrated with local build times (especially with proc macro heavy crates like async-graphql), I have modified the compiler itself and built some light tooling around it to speed up local incremental builds. 12% speed-up for vector.dev ‘cargo check’ out of the box:

I understand this may not be the right channel — so I had made a discussion post — but please free to redirect me elsewhere!

[jszwedko]

/ci-run-all

[github-actions]

Regression Detector Results

Run ID: a1257c4c-3337-43d1-98fc-7eb40a708df1 Metrics dashboard

Baseline: 88d6fe7
Comparison: 5de0820

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.

Experiments ignored for regressions

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

perf	experiment	goal	Δ mean %	Δ mean % CI	links
➖	file_to_blackhole	egress throughput	+6.03	[-1.25, +13.31]

Fine details of change detection per experiment

perf	experiment	goal	Δ mean %	Δ mean % CI	links
➖	file_to_blackhole	egress throughput	+6.03	[-1.25, +13.31]
➖	syslog_log2metric_splunk_hec_metrics	ingress throughput	+1.82	[+1.71, +1.92]
➖	datadog_agent_remap_blackhole_acks	ingress throughput	+1.52	[+1.41, +1.63]
➖	syslog_splunk_hec_logs	ingress throughput	+0.99	[+0.90, +1.08]
➖	http_to_s3	ingress throughput	+0.64	[+0.36, +0.92]
➖	datadog_agent_remap_blackhole	ingress throughput	+0.59	[+0.48, +0.71]
➖	http_to_http_acks	ingress throughput	+0.53	[-0.70, +1.76]
➖	syslog_humio_logs	ingress throughput	+0.52	[+0.42, +0.63]
➖	datadog_agent_remap_datadog_logs	ingress throughput	+0.32	[+0.14, +0.51]
➖	syslog_log2metric_humio_metrics	ingress throughput	+0.22	[+0.10, +0.33]
➖	otlp_grpc_to_blackhole	ingress throughput	+0.17	[+0.05, +0.28]
➖	http_to_http_noack	ingress throughput	+0.16	[+0.08, +0.25]
➖	http_to_http_json	ingress throughput	+0.06	[+0.00, +0.11]
➖	splunk_hec_to_splunk_hec_logs_acks	ingress throughput	+0.00	[-0.09, +0.10]
➖	splunk_hec_indexer_ack_blackhole	ingress throughput	+0.00	[-0.08, +0.09]
➖	splunk_hec_to_splunk_hec_logs_noack	ingress throughput	-0.01	[-0.10, +0.08]
➖	datadog_agent_remap_datadog_logs_acks	ingress throughput	-0.01	[-0.17, +0.15]
➖	syslog_log2metric_tag_cardinality_limit_blackhole	ingress throughput	-0.05	[-0.15, +0.05]
➖	syslog_loki	ingress throughput	-0.07	[-0.15, +0.01]
➖	otlp_http_to_blackhole	ingress throughput	-0.51	[-0.67, -0.35]
➖	http_elasticsearch	ingress throughput	-1.10	[-1.27, -0.93]
➖	socket_to_socket_blackhole	ingress throughput	-1.17	[-1.23, -1.11]
➖	syslog_regex_logs2metric_ddmetrics	ingress throughput	-1.30	[-1.49, -1.12]
➖	splunk_hec_route_s3	ingress throughput	-1.43	[-1.75, -1.10]
➖	fluent_elasticsearch	ingress throughput	-1.68	[-2.16, -1.19]
➖	http_text_to_http_json	ingress throughput	-2.38	[-2.50, -2.26]

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".