Add parallel FYRE OpenShift benchmarks with 40 GB disks

CHANGELOG.md

@@ -7,6 +7,27 @@ and this project uses [Semantic Versioning](https://semver.org/spec/v2.0.0.html)
 
 ## [Unreleased]
 
+### Added
+
+- Add a FYRE OpenShift comparison profile that provisions 40 GB nodes and runs
+  the built-in and external dataplane lanes concurrently on separate target,
+  Locust, and Fast Time workers. Package the pinned `oc` runtime, collect
+  gateway-side memory, and produce a self-contained Markdown report alongside
+  the existing CSV, JSON, and Slack-ready PNG artifacts.
+- Add a quota-sized 2 vCPU / 2 GiB OpenShift profile that runs all eight
+  built-in/external and 125/250/500/1,000-user measurements concurrently with
+  exact pod limits on three dedicated-role workers. Reserve 1.75 vCPU / 1.75
+  GiB per 2 vCPU / 2 GiB target so all eight targets schedule with host
+  headroom, set 40 GB master and worker root disks, disable FYRE's automatic
+  worker data disks, and preserve structured FYRE API validation errors.
+
+### Changed
+
+- Generalize `load fyre run|status|destroy` ownership and cleanup to support
+  run-owned OpenShift clusters as well as standalone Terraform VMs.
+- Resume an interrupted OpenShift campaign when `load fyre run` is invoked
+  again with the same run ID and saved configuration.
+
 ## [0.5.0] - 2026-09-16
 
 ### Added

benchmarks/fyre/README.md

@@ -44,6 +44,11 @@ disk and expose no create-time root-disk setting. The default three-VM run
 therefore needs 750 GB of FYRE disk quota. The CLI checks CPU, memory, disk, and
 public-IP quota before it creates any benchmark VM.
 
+The OpenShift profile requires `FYRE_PRODUCT_GROUP_ID`, Docker on the
+orchestration host, and access to the FYRE OpenShift API and cluster DNS. The
+CLI runs a digest-pinned OpenShift client container, so a host `oc` installation
+is not required.
+
 ## Run the complete comparison
 
 The bare command is the CI entrypoint for the complete eight-run comparison:
@@ -137,3 +142,98 @@ cf-integration load fyre run \
 
 Comparison reports derive the target allocation from the selected profile; both
 lanes always run sequentially on that same VM.
+
+## Parallel OpenShift profile with 40 GB disks
+
+`openshift.yaml` runs the same eight comparison measurements on a FYRE
+OpenShift cluster while reducing every master and worker root disk to 40 GB
+and explicitly disabling FYRE's default worker data disks.
+The built-in and external lanes run concurrently and remain isolated on six
+dedicated workers:
+
+| Lane role | Workers | Pod allocation on each worker |
+| --- | ---: | ---: |
+| Built-in and external Locust | 2 | 4 vCPU / 16 GiB each |
+| Built-in and external target | 2 | 4 vCPU / 4 GiB each |
+| Built-in and external Fast Time | 2 | 8 vCPU / 32 GiB each |
+
+Each Locust pod contains one master and three workers. Each target allocation
+includes its supporting PostgreSQL/Redis or Redis/JWKS containers. Each lane
+has a separate zero-delay Fast Time service. No measured target, load generator,
+or backend shares a worker node with the other lane.
+
+Run the complete parallel comparison with:
+
+```bash
+cf-integration load fyre run \
+  --file benchmarks/fyre/openshift.yaml \
+  --run-id openshift-builtin-external
+```
+
+The short form is:
+
+```bash
+cf-integration l f r -f benchmarks/fyre/openshift.yaml -i openshift-builtin-external
+```
+
+The command creates the cluster through the FYRE OpenShift API, authenticates
+with the generated kubeadmin credential, assigns the six workers by their
+configured CPU and memory, runs both lanes in parallel at 125, 250, 500, and
+1,000 users, downloads every phase before deleting its pods, writes the final
+report, deletes the benchmark namespace, and deletes only the run-owned
+cluster. A failed or interrupted campaign retains its local run state and
+retries cluster cleanup three times. Run the same command with the same run ID
+to resume an interrupted OpenShift campaign from its saved manifest.
+
+OpenShift artifacts use the same
+`$CF_INTEGRATION_DIR/fyre/<run-id>/results/` layout and add `report.md`, a
+self-contained report with the result table, memory averages and peaks,
+request mix, pinned images, and Mermaid architecture. The cluster record and
+manifest recursively omit passwords, tokens, pull secrets, API keys, and
+kubeconfig data.
+
+The built-in lane remains pinned to the MCP SDK v2 fixture image until that SDK
+change is available in the main gateway image. The profile must not be changed
+back to the main image before that merge because both lanes use the same modern
+`2026-07-28` client.
+
+### Fully parallel 2 vCPU / 2 GiB comparison
+
+`openshift-2v2-parallel.yaml` runs all eight measurements at the same time:
+built-in and external dataplane lanes at 125, 250, 500, and 1,000 users. Each
+measurement gets its own target, Locust, and Fast Time pods with identical
+requests and limits. Pods share only with pods serving the same role.
+
+| Dedicated worker role | Worker size | Pods | Reserved per measurement |
+| --- | ---: | ---: | ---: |
+| Target | 16 vCPU / 16 GiB | 8 | 2 vCPU / 2 GiB limits; 1.75 vCPU / 1.75 GiB requests |
+| Locust | 14 vCPU / 12 GiB | 8 | 1.5 vCPU / 1.25 GiB |
+| Fast Time | 14 vCPU / 12 GiB | 8 | 1.5 vCPU / 1.375 GiB |
+
+Each target reservation includes its supporting PostgreSQL and Redis
+containers for the built-in dataplane, or Redis and loopback JWKS containers
+for the external dataplane. Each Locust pod has one master and three workers.
+The helper pressure gate rejects the campaign if the shared helper workers or
+individual helper pods become the bottleneck.
+
+Run the full comparison with one command:
+
+```bash
+cf-integration load fyre run \
+  --file benchmarks/fyre/openshift-2v2-parallel.yaml \
+  --run-id openshift-2v2-parallel
+```
+
+The short form is:
+
+```bash
+cf-integration l f r -f benchmarks/fyre/openshift-2v2-parallel.yaml -i openshift-2v2-parallel
+```
+
+The three OpenShift masters and all three workers use 40 GB root disks, the
+FYRE API VM uses its fixed 500 GB root disk, and workers have no additional
+data disks. The cluster therefore requests 740 GB of disk in total. The
+orchestration command may run on a persistent VM or CI worker; the benchmark
+continues if the developer laptop sleeps. Artifacts are downloaded to
+`$CF_INTEGRATION_DIR/fyre/<run-id>/results/` before the run-owned cluster is
+deleted.

benchmarks/fyre/openshift-2v2-parallel.yaml

@@ -0,0 +1,54 @@
+schema_version: 1
+benchmark_kind: comparison
+infrastructure:
+  kind: openshift
+  os: Red Hat CoreOS
+  expiry_hours: 24
+  openshift:
+    version: "4.21.27"
+    base_disk_gb: 40
+    oc_image: quay.io/openshift/origin-cli@sha256:f3fc63fc3d2fd1c3a2c3311bf517b776f6596244e697dad32d62ca9fd6482fe5
+    master: { cpu: 4, memory_gb: 16 }
+    api: { cpu: 4, memory_gb: 8 }
+    target_pod: { cpu_millicores: 1750, memory_mib: 1792 }
+    load_pod: { cpu_millicores: 1500, memory_mib: 1280 }
+    backend_pod: { cpu_millicores: 1500, memory_mib: 1408 }
+    worker_pools:
+      - { role: target, cpu: 16, memory_gb: 16, count: 1 }
+      - { role: locust, cpu: 14, memory_gb: 12, count: 1 }
+      - { role: fast-time, cpu: 14, memory_gb: 12, count: 1 }
+images:
+  dataplane: ghcr.io/contextforge-org/contextforge-data-plane@sha256:0b3026b21659ce6c494142aee2c5866b44122232e73fcf21102da1f210692a96
+  controlplane: ghcr.io/contextforge-org/cf-integration-fixture@sha256:5b206e6f863cea9f8cabea6451392428fbe67b16bdf10fefd1f7ac8032d95b52
+  fast_time: ghcr.io/ibm/cfex-mcp-fast-time-server@sha256:110e1826f5d763e5afadba770b731dac93e0819c1bbadb68671b0124260603cf
+  helpers: ghcr.io/contextforge-org/cf-integration-helpers@sha256:f7e557e263737328225d827e73b519d0b40facb2698c55ee3d2de0ed82f00fa3
+  locust: mirror.gcr.io/locustio/locust@sha256:fd39232c31971fe7509582717a95592f0fbf9a917b705f355bd7471e03bdd649
+  postgres: postgres@sha256:4ef4dbc939d61acea57712655ddb4b4ab27419c913f94cca0cd57cb3ea3c2280
+  redis: mirror.gcr.io/library/redis@sha256:a7859ed111db3c1f5404a973a4747505d559fb5ca32d37e447afc0ef845a2103
+workload:
+  protocol_version: 2026-07-28
+  user_levels: [125, 250, 500, 1000]
+  first_users: 125
+  maximum_users: 1000
+  ramp_seconds: 30
+  warmup_seconds: 30
+  measure_seconds: 3600
+  repetitions: 1
+  maximum_campaign_seconds: 7200
+  plateau_improvement_percent: 5.0
+  boundary_percent: 12.5
+  config_cache_seconds: 60
+  helper_cpu_percent: 70.0
+  helper_memory_percent: 80.0
+  worker_core_percent: 85.0
+  parallel_lanes: true
+  parallel_user_levels: true
+  tools:
+    - convert_time
+    - echo
+    - get_stats
+    - get_system_time
+    - schema_success
+    - verify-protocol
+scenarios:
+  - { id: comparison, label: Built-in dataplane vs external dataplane, replicas: 1, cpu: 2, memory_gb: 2, multiplier: 1 }

benchmarks/fyre/openshift.yaml

@@ -0,0 +1,57 @@
+schema_version: 1
+benchmark_kind: comparison
+infrastructure:
+  kind: openshift
+  os: Red Hat CoreOS
+  expiry_hours: 12
+  openshift:
+    version: "4.21.27"
+    base_disk_gb: 40
+    oc_image: quay.io/openshift/origin-cli@sha256:f3fc63fc3d2fd1c3a2c3311bf517b776f6596244e697dad32d62ca9fd6482fe5
+    master: { cpu: 4, memory_gb: 16 }
+    api: { cpu: 4, memory_gb: 8 }
+    target_pod: { cpu_millicores: 3500, memory_mib: 3584 }
+    load_pod: { cpu_millicores: 4000, memory_mib: 16384 }
+    backend_pod: { cpu_millicores: 8000, memory_mib: 32768 }
+    worker_pools:
+      - { role: target-builtin, cpu: 6, memory_gb: 12, count: 1 }
+      - { role: target-external, cpu: 6, memory_gb: 12, count: 1 }
+      - { role: locust-builtin, cpu: 6, memory_gb: 20, count: 1 }
+      - { role: locust-external, cpu: 6, memory_gb: 20, count: 1 }
+      - { role: fast-time-builtin, cpu: 10, memory_gb: 40, count: 1 }
+      - { role: fast-time-external, cpu: 10, memory_gb: 40, count: 1 }
+images:
+  dataplane: ghcr.io/contextforge-org/contextforge-data-plane@sha256:0b3026b21659ce6c494142aee2c5866b44122232e73fcf21102da1f210692a96
+  controlplane: ghcr.io/contextforge-org/cf-integration-fixture@sha256:5b206e6f863cea9f8cabea6451392428fbe67b16bdf10fefd1f7ac8032d95b52
+  fast_time: ghcr.io/ibm/cfex-mcp-fast-time-server@sha256:110e1826f5d763e5afadba770b731dac93e0819c1bbadb68671b0124260603cf
+  helpers: ghcr.io/contextforge-org/cf-integration-helpers@sha256:f7e557e263737328225d827e73b519d0b40facb2698c55ee3d2de0ed82f00fa3
+  locust: mirror.gcr.io/locustio/locust@sha256:fd39232c31971fe7509582717a95592f0fbf9a917b705f355bd7471e03bdd649
+  postgres: postgres@sha256:4ef4dbc939d61acea57712655ddb4b4ab27419c913f94cca0cd57cb3ea3c2280
+  redis: mirror.gcr.io/library/redis@sha256:a7859ed111db3c1f5404a973a4747505d559fb5ca32d37e447afc0ef845a2103
+workload:
+  protocol_version: 2026-07-28
+  user_levels: [125, 250, 500, 1000]
+  first_users: 125
+  maximum_users: 1000
+  ramp_seconds: 30
+  warmup_seconds: 30
+  measure_seconds: 3600
+  repetitions: 1
+  maximum_campaign_seconds: 18000
+  plateau_improvement_percent: 5.0
+  boundary_percent: 12.5
+  config_cache_seconds: 60
+  helper_cpu_percent: 70.0
+  helper_memory_percent: 80.0
+  worker_core_percent: 85.0
+  parallel_lanes: true
+  parallel_user_levels: false
+  tools:
+    - convert_time
+    - echo
+    - get_stats
+    - get_system_time
+    - schema_success
+    - verify-protocol
+scenarios:
+  - { id: comparison, label: Built-in dataplane vs external dataplane, replicas: 1, cpu: 4, memory_gb: 4, multiplier: 1 }