Supporting design doc for the "e2e: add upgrade test framework" issue. Covers the concrete types, package layout, and scheduling semantics; the issue itself stays at the requirements/acceptance-criteria level and links here.
Test authors declare what to check, not how to schedule it. A Suite
type hides all t.Run/t.Parallel/channel wiring behind a declarative
| $ mkdir -p /tmp/opencode/e2e-artifacts2 && chmod 777 /tmp/opencode/e2e-artifacts2 && podman run --rm \ | |
| -e KUBECONFIG=/tmp/kubeconfig \ | |
| -v /tmp/opencode/e2e-kubeconfig:/tmp/kubeconfig:Z \ | |
| -v /tmp/opencode/e2e-artifacts2:/artifacts:Z \ | |
| quay.io/opendatahub/odh-mcp-lifecycle-operator-e2e:odh-pr-67 \ | |
| 2>&1; echo "EXIT: $?" | |
| [e2e-run] running: gotestsum --raw-command --junitfile /artifacts/e2e-results/junit.xml --junitfile-project-name mcp-lifecycle-operator --jsonfile /artifacts/e2e-results/log.jsonl --format testname -- /e2e/e2e-run exec | |
| [e2e-run] running: /usr/local/bin/test2json -t -p e2e /e2e/e2e-tests -test.count=1 -test.v=test2json | |
| 2026/08/13 15:30:15 [e2e-dsc] DataScienceCluster "default-dsc" mcplifecycleoperator.managementState = "Managed" |
debug_py_step.py suspended at line 235 of rk4.py (velocity_vector += line), step 2 of calc.fire().DebugOneStep.kt suspended at line 321 of TrajectorySolver.kt (return State( line), step 2 of computeTrajectory().Both breakpoints capture all k1-k4 values AFTER they are computed but BEFORE the state update.
Conversion factor: 1/3.2808399 (= Kotlin's FPS_PER_MPS inverse)
| #!/usr/bin/env python3 | |
| from py_ballisticcalc import ( | |
| Ammo, Angular, Atmo, Calculator, Distance, DragModel, | |
| Pressure, Shot, TableG7, Temperature, Velocity, Weapon, Weight, Wind, | |
| ) | |
| bc = 0.209 | |
| dm = DragModel(bc=bc, drag_table=TableG7, | |
| weight=Weight.Grain(77.0), | |
| diameter=Distance.Millimeter(5.56), |
| #!/usr/bin/env python3 | |
| """PR #3381 — Single-command outlet test. | |
| Usage: python3 .ai/bug4-outlet-toggle-test.py <test> [port] | |
| Tests: | |
| og1.off.le SOG_0 (OG1) OFF, LE word order | |
| og1.on.le SOG_0 (OG1) ON, LE word order | |
| og1.off.be SOG_0 (OG1) OFF, BE word order | |
| og1.on.be SOG_0 (OG1) ON, BE word order |
| #!/usr/bin/env python3 | |
| """Test EchterAgo's probe idea: write CANCEL|OUTPUT_ON|OUTPUT_OFF to OutletCommand_BF. | |
| On spec-compliant firmware: EMBXILVAL (invalid command combo detected) | |
| On LE-buggy firmware: different error or no error (combo lands in reserved bits) | |
| Must run with NUT driver stopped: systemctl stop nut-driver@apc | |
| """ | |
| import sys |
| #!/usr/bin/env python3 | |
| """Test: Modbus/TCP word order via APC NMC. | |
| Reads raw register values to verify read word order, then tests write | |
| word order for OutletCommand (load.on) in both big-endian and little-endian. | |
| Safe: load.on while already on is a NOOP. | |
| Usage: | |
| python3 bug4-test-modbus-tcp.py <NMC_HOST> [NMC_PORT] |
| #!/usr/bin/env python3 | |
| """Bug 4 — Test: Confirm word-order bug in apc_modbus OutletCommand writes. | |
| The NUT driver's _apc_modbus_from_uint64() stores 32-bit values in big-endian | |
| word order: [MSW, LSW]. But APC UPS expects little-endian: [LSW, MSW]. | |
| For command value 0x0102 (load.on + MOG): | |
| - NUT sends: [0x0000, 0x0102] → FAILS (value in wrong register) | |
| - Correct: [0x0102, 0x0000] → WORKS (value in register 1538) |
| #!/usr/bin/env python3 | |
| """Bug 4 — Test: Does setting SOURCE_RJ45_PORT bit fix OutletCommand writes? | |
| This script bypasses the NUT driver and sends a raw Modbus write to register | |
| 1538 (OutletCommand_BF) with the SOURCE_RJ45_PORT bit (bit 14) set. | |
| Command: load.on targeting MOG (Main Outlet Group) | |
| - Without source bit: 0x0102 = CMD_OUTPUT_ON | TARGET_MOG | |
| - With source bit: 0x4102 = CMD_OUTPUT_ON | TARGET_MOG | SOURCE_RJ45_PORT |