Adds SMP tests for idle & ascii (#1826)

### What does this PR do?

Introduces SMP (Single Machine Performance) regression detector experiments for lading. Adds an idle baseline test and four HTTP ASCII throughput tests (1, 10, 100, 1000 MiB/s) with memory and CPU quality gates.

### Motivation

Establishes continuous performance regression detection for lading using the SMP regression detector. The idle test provides a resource floor, while the ASCII tests cover the HTTP generator + blackhole path across a range of throughput levels. Each experiment enforces bounds on `total_pss_bytes` and `avg(total_cpu_usage_millicores)`.

### Related issues

N/A

### Additional Notes

- Each ASCII experiment is paired with the local benchmark in `lading_payload/benches/ascii.rs`; a cross-reference comment was added to the benchmark file.
- All experiments run lading in `--no-target --experiment-duration-infinite` mode (shutdown performance is out of scope).
- `experiments/regression/config.yaml` pins lading version to 0.31.2 with profiling replicas disabled.

Author: preinlein

experiments/regression/README.md

@@ -0,0 +1,9 @@
+# Regression Experiments
+
+This folder defines all the regression detector experiments. 
+
+The regression detector is a tool provided by Single Machine Performance thru the `smp` CLI.
+
+It allows us to performance test lading under different scenarios/loads.
+
+IMPORTANT: Any local benchmarks should be supplemented with SMP regression detector experiments that exercise that same local benchmark test.

experiments/regression/cases/http_ascii_1000mib/README.md

@@ -0,0 +1,21 @@
+# HTTP ASCII 1000 MiB/s
+
+Resource usage of lading under a steady 1000 MiB/s HTTP load with ASCII payloads.
+
+## What
+
+Runs lading as a target with an HTTP generator sending ASCII payloads at 1000 MiB/s to its own HTTP blackhole. This exercises the HTTP generator, ASCII payload construction, and HTTP blackhole within a single lading instance at maximum throughput.
+
+## Why
+
+Establishes a baseline for lading's resource consumption under an extreme HTTP workload. Regressions here indicate overhead in the HTTP generator or payload path at high throughput.
+
+## Paired Benchmark
+
+This experiment is paired with the local benchmark in `lading_payload/benches/ascii.rs`.
+If throughput sizes change in either place, update the other to match.
+
+## Enforcements
+
+Memory usage is enforced by bounding `total_pss_bytes`.
+CPU usage is enforced by bounding `avg(total_cpu_usage_millicores)`.

experiments/regression/cases/http_ascii_1000mib/experiment.yaml

@@ -0,0 +1,27 @@
+optimization_goal: memory
+erratic: false
+
+target:
+  name: lading-target
+  # Setting experiment duration infinite here implies that lading will not go thru the shutdown process
+  # This is acceptable as we are less concerned about the shutdown performance
+  command: "/usr/bin/lading --no-target --experiment-duration-infinite --config-path /etc/lading-target/lading.yaml"
+  cpu_allotment: 2
+  memory_allotment: 1250 MiB
+
+  environment:
+    DD_SERVICE: lading-target
+    RUST_LOG: info
+    RUST_BACKTRACE: 1
+
+checks:
+  - name: memory_usage
+    description: "Memory usage quality gate. Bounds total memory usage."
+    bounds:
+      series: total_pss_bytes
+      upper_bound: "1130 MiB"
+  - name: cpu_usage
+    description: "CPU usage quality gate. Bounds total average millicore usage."
+    bounds:
+      series: avg(total_cpu_usage_millicores)
+      upper_bound: 400

experiments/regression/cases/http_ascii_1000mib/lading-target/lading.yaml

@@ -0,0 +1,18 @@
+generator:
+  - http:
+      seed: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32]
+      headers: {}
+      target_uri: "http://127.0.0.1:8080/"
+      bytes_per_second: "1000 MiB"
+      parallel_connections: 1
+      method:
+        post:
+          maximum_prebuild_cache_size_bytes: "5 GiB"
+          variant: "ascii"
+
+blackhole:
+  - http:
+      binding_addr: "0.0.0.0:8080"
+
+telemetry:
+  addr: "0.0.0.0:9000"

experiments/regression/cases/http_ascii_1000mib/lading/lading.yaml

@@ -0,0 +1,9 @@
+generator: []
+
+blackhole:
+  - http:
+      binding_addr: "127.0.0.1:9091"
+
+target_metrics:
+  - prometheus:
+      uri: "http://127.0.0.1:9000/metrics"

experiments/regression/cases/http_ascii_100mib/README.md

@@ -0,0 +1,21 @@
+# HTTP ASCII 100 MiB/s
+
+Resource usage of lading under a steady 100 MiB/s HTTP load with ASCII payloads.
+
+## What
+
+Runs lading as a target with an HTTP generator sending ASCII payloads at 100 MiB/s to its own HTTP blackhole. This exercises the HTTP generator, ASCII payload construction, and HTTP blackhole within a single lading instance.
+
+## Why
+
+Establishes a baseline for lading's resource consumption under a high HTTP workload. Regressions here indicate overhead in the HTTP generator or payload path independent of payload complexity.
+
+## Paired Benchmark
+
+This experiment is paired with the local benchmark in `lading_payload/benches/ascii.rs`.
+If throughput sizes change in either place, update the other to match.
+
+## Enforcements
+
+Memory usage is enforced by bounding `total_pss_bytes`.
+CPU usage is enforced by bounding `avg(total_cpu_usage_millicores)`.

experiments/regression/cases/http_ascii_100mib/experiment.yaml

@@ -0,0 +1,27 @@
+optimization_goal: memory
+erratic: false
+
+target:
+  name: lading-target
+  # Setting experiment duration infinite here implies that lading will not go thru the shutdown process
+  # This is acceptable as we are less concerned about the shutdown performance
+  command: "/usr/bin/lading --no-target --experiment-duration-infinite --config-path /etc/lading-target/lading.yaml"
+  cpu_allotment: 2
+  memory_allotment: 630 MiB
+
+  environment:
+    DD_SERVICE: lading-target
+    RUST_LOG: info
+    RUST_BACKTRACE: 1
+
+checks:
+  - name: memory_usage
+    description: "Memory usage quality gate. Bounds total memory usage."
+    bounds:
+      series: total_pss_bytes
+      upper_bound: "575 MiB"
+  - name: cpu_usage
+    description: "CPU usage quality gate. Bounds total average millicore usage."
+    bounds:
+      series: avg(total_cpu_usage_millicores)
+      upper_bound: 155

experiments/regression/cases/http_ascii_100mib/lading-target/lading.yaml

@@ -0,0 +1,18 @@
+generator:
+  - http:
+      seed: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32]
+      headers: {}
+      target_uri: "http://127.0.0.1:8080/"
+      bytes_per_second: "100 MiB"
+      parallel_connections: 1
+      method:
+        post:
+          maximum_prebuild_cache_size_bytes: "500 MiB"
+          variant: "ascii"
+
+blackhole:
+  - http:
+      binding_addr: "0.0.0.0:8080"
+
+telemetry:
+  addr: "0.0.0.0:9000"

experiments/regression/cases/http_ascii_100mib/lading/lading.yaml

@@ -0,0 +1,9 @@
+generator: []
+
+blackhole:
+  - http:
+      binding_addr: "127.0.0.1:9091"
+
+target_metrics:
+  - prometheus:
+      uri: "http://127.0.0.1:9000/metrics"

experiments/regression/cases/http_ascii_10mib/README.md

@@ -0,0 +1,21 @@
+# HTTP ASCII 10 MiB/s
+
+Resource usage of lading under a steady 10 MiB/s HTTP load with ASCII payloads.
+
+## What
+
+Runs lading as a target with an HTTP generator sending ASCII payloads at 10 MiB/s to its own HTTP blackhole. This exercises the HTTP generator, ASCII payload construction, and HTTP blackhole within a single lading instance.
+
+## Why
+
+Establishes a baseline for lading's resource consumption under a moderate HTTP workload. Regressions here indicate overhead in the HTTP generator or payload path independent of payload complexity.
+
+## Paired Benchmark
+
+This experiment is paired with the local benchmark in `lading_payload/benches/ascii.rs`.
+If throughput sizes change in either place, update the other to match.
+
+## Enforcements
+
+Memory usage is enforced by bounding `total_pss_bytes`.
+CPU usage is enforced by bounding `avg(total_cpu_usage_millicores)`.