enhancement(core): Erlang/OTP-style supervisor system

[tobz]

Summary

This PR introduces a new runtime system in saluki-core modeled after Erlang/OTP's supervision trees, a powerful design pattern for building fault-tolerant systems composed of many interdependent "processes" in a way that they can recover from failure.

Supervision trees are explained succinctly by the Erlang/OTP documentation, but generally: we have a collection of processes (asynchronous tasks) that have some level of fallibility, and we want to -- where possible -- maximize their resilience by allowing them to be restarted if they fail. Supervision trees approach this from the perspective of declaring a "specification" that describes how to create the process (in our case, building the Future that is spawned) and then adding those specifications to a supervisor, which is responsible for managing those processes, restarting them when die, giving up if certain restart limits are hit, and so on.

While we lack a lot of the power of Erlang itself, in terms of what Erlang/OTP's supervisors can build on top of, even having the basics provides us a lot of leeway for building things in a more fault-tolerant way:

• restarting crashed topology components, or arbitrary background tasks, in a uniform and consistent way
• providing the scaffolding to track all active tasks and how they're nested
• ordered shutdown for everything instead of just the topology

We're intentionally only implementing many of the basics here -- worker specification, restart strategy, nested supervisors, orderly shutdown, etc -- but will add enhanced capabilities in future iterations, such as runtime telemetry/introspection, dynamic workers, and more.

Change Type

• Bug fix
• New feature
• Non-functional (chore, refactoring, docs)
• Performance

How did you test this PR?

N/A

References

AGTMETRICS-233

[pr-commenter]

Regression Detector (Agent Data Plane)

Regression Detector Results

Run ID: 0546c77b-62eb-48c2-825b-d04173cd8996

Baseline: 0957bb3
Comparison: e09fddb
Diff

Optimization Goals: ✅ No significant changes detected

Fine details of change detection per experiment

perf	experiment	goal	Δ mean %	Δ mean % CI	trials	links
➖	quality_gates_rss_dsd_medium	memory utilization	+0.30	[+0.12, +0.47]	1
➖	quality_gates_rss_dsd_heavy	memory utilization	+0.29	[+0.19, +0.38]	1
➖	quality_gates_rss_idle	memory utilization	+0.17	[+0.15, +0.19]	1
➖	dsd_uds_10mb_3k_contexts_throughput	ingress throughput	+0.03	[-0.02, +0.08]	1
➖	dsd_uds_512kb_3k_contexts_throughput	ingress throughput	+0.00	[-0.01, +0.02]	1
➖	quality_gates_rss_dsd_ultraheavy	ingress throughput	+0.00	[-0.04, +0.04]	1
➖	dsd_uds_1mb_3k_contexts_throughput	ingress throughput	-0.00	[-0.02, +0.01]	1
➖	dsd_uds_100mb_3k_contexts_throughput	ingress throughput	-0.01	[-0.08, +0.06]	1
➖	quality_gates_rss_dsd_low	memory utilization	-0.07	[-0.22, +0.07]	1
➖	dsd_uds_500mb_3k_contexts_throughput	ingress throughput	-3.41	[-3.54, -3.27]	1

Bounds Checks: ✅ Passed

perf	experiment	bounds_check_name	replicates_passed	links
✅	quality_gates_rss_dsd_heavy	memory_usage	10/10
✅	quality_gates_rss_dsd_low	memory_usage	10/10
✅	quality_gates_rss_dsd_medium	memory_usage	10/10
✅	quality_gates_rss_dsd_ultraheavy	memory_usage	10/10
✅	quality_gates_rss_idle	memory_usage	10/10

Explanation

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

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

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".

[pr-commenter]

Regression Detector (Agent Data Plane w/ Checks)

Regression Detector Results

Run ID: 9136ff2d-887d-4d14-b93e-b9227939eed7

Baseline: 0957bb3
Comparison: e09fddb
Diff

Optimization Goals: ✅ No significant changes detected

Fine details of change detection per experiment

perf	experiment	goal	Δ mean %	Δ mean % CI	trials	links
➖	quality_gates_rss_idle	memory utilization	+0.02	[+0.01, +0.04]	1
➖	quality_gates_rss_basic	memory utilization	-0.14	[-0.16, -0.13]	1

Bounds Checks: ✅ Passed

perf	experiment	bounds_check_name	replicates_passed	links
✅	quality_gates_rss_basic	memory_usage	10/10
✅	quality_gates_rss_idle	memory_usage	10/10

Explanation

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

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

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".

[pr-commenter]

Regression Detector Links

ADP Experiment Result Links

experiment	link(s)
dsd_uds_100mb_3k_contexts_throughput	[Profiling] [SMP Dashboard]
dsd_uds_10mb_3k_contexts_throughput	[Profiling] [SMP Dashboard]
dsd_uds_1mb_3k_contexts_throughput	[Profiling] [SMP Dashboard]
dsd_uds_500mb_3k_contexts_throughput	[Profiling] [SMP Dashboard]
dsd_uds_512kb_3k_contexts_throughput	[Profiling] [SMP Dashboard]
quality_gates_rss_dsd_heavy	[Profiling] [SMP Dashboard]
quality_gates_rss_dsd_low	[Profiling] [SMP Dashboard]
quality_gates_rss_dsd_medium	[Profiling] [SMP Dashboard]
quality_gates_rss_dsd_ultraheavy	[Profiling] [SMP Dashboard]
quality_gates_rss_idle	[Profiling] [SMP Dashboard]

ADP && Checks Experiment Result Links

experiment	link(s)
quality_gates_rss_basic	[Profiling] [SMP Dashboard]
quality_gates_rss_idle	[Profiling] [SMP Dashboard]

[blt]

Confirming, unlike Erlang processes saluki-core processes do not have a mailbox / input queue associated for message passing?

[blt]

Confirmed later, no mailbox. Probably worth adding this to the documentation in case someone -- like myself -- comes along with an Erlang mental model inappropriately.

[blt]

I'm not opposed, just curious why we don't reject invalid names outright? Haven't confirmed yet, it's possible the intention here is for process names to arrive via user config, in which case this does make some sense.

[blt]

+1

[blt]

I like how the API works. Probably worth developing a lint or something to confirm that processes actually check their shutdown signal properly. Shame we don't have the ability to preempt.