Skip to content

Commit d47b87d

Browse files
author
LoCoBench Bot
committed
chore: mark US-006 as passing, update progress log
1 parent e1ad33b commit d47b87d

File tree

2 files changed

+32
-2
lines changed

2 files changed

+32
-2
lines changed

prd.json

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -90,8 +90,8 @@
9090
"Both tasks registered in configs/selected_benchmark_tasks.json and nlqa_2config.sh"
9191
],
9292
"priority": 6,
93-
"passes": false,
94-
"notes": "VS Code: src/vs/workbench/services/extensions/electron-sandbox/nativeExtensionService.ts, ExtensionHostProcess. Cilium: pkg/datapath/loader/, pkg/policy/, pkg/maps/. Both require understanding process/component isolation boundaries."
93+
"passes": true,
94+
"notes": "VS Code v1.99.3 (17baf841). Cilium v1.17.9 (a2f97aa8). 4 questions each: process isolation, IPC/deployment, crash handling/policy distribution, isolation mechanisms/map scoping. Ground truth: 8 findings, 5 file refs, 2 causal chains, 2 negative checks per task."
9595
}
9696
]
9797
}

progress.txt

Lines changed: 30 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -128,3 +128,33 @@
128128
- argoproj/argo-cd repo uses Apache-2.0 license (not BUSL like Terraform)
129129
- Used exploratory agent to research Argo CD architecture efficiently — comprehensive findings in ~5 minutes
130130
---
131+
132+
## 2026-02-16 - US-006
133+
- Created nlqa-debug-001: VS Code extension host crash isolation Q&A task
134+
- Created nlqa-debug-002: Cilium eBPF fault isolation Q&A task
135+
- Files created:
136+
- benchmarks/ccb_nlqa/nlqa-debug-001/ (task.toml, instruction.md, Dockerfile, ground_truth.json, test.sh)
137+
- benchmarks/ccb_nlqa/nlqa-debug-002/ (task.toml, instruction.md, Dockerfile, ground_truth.json, test.sh)
138+
- Files modified:
139+
- configs/nlqa_2config.sh (added nlqa-debug-001 and nlqa-debug-002 to ALL_TASK_IDS and TASK_SG_REPO_NAMES)
140+
- configs/selected_benchmark_tasks.json (added nlqa-debug-001 and nlqa-debug-002 entries, updated total_selected from 192 to 194)
141+
- prd.json (marked US-006 as passes=true)
142+
- **Learnings for future iterations:**
143+
- VS Code v1.99.3 commit SHA: 17baf841131aa23349f217ca7c570c76ee87b957
144+
- Key VS Code extension host isolation files: nativeExtensionService.ts, abstractExtensionService.ts, extensionHostManager.ts, utilityProcess.ts, extensionHostStarter.ts, localProcessExtensionHost.ts
145+
- VS Code extension host runs as Electron utility process (separate PID), not same-process thread or web worker
146+
- ExtensionHostCrashTracker implements 5-minute / 3-crash auto-restart policy
147+
- IPC via MessagePort/MessageChannelMain, crash detection via onExit event listener
148+
- Detached process on Windows (detached: !!platform.isWindows) to prevent cascading termination
149+
- Cilium v1.17.9 commit SHA: a2f97aa8d2de4bb360bee1e295e20556ce4166ce
150+
- Key Cilium eBPF isolation files: pkg/datapath/loader/base.go, pkg/datapath/linux/config/config.go, pkg/endpoint/endpoint.go, pkg/maps/, pkg/bpf/map_linux.go, bpf/node_config.h
151+
- Cilium deployed as DaemonSet (one cilium-agent per node), each agent independently compiles/loads eBPF programs
152+
- Per-node eBPF compilation uses node_config.h with node-specific #defines (kernel features, enabled options)
153+
- eBPF programs attach to kernel-local TC hooks via netlink, no inter-node coordination
154+
- BPF maps pinned to /sys/fs/bpf (node-local filesystem), either PIN_GLOBAL_NS or PIN_OBJECT_NS scoping
155+
- CiliumNetworkPolicy CRDs distributed cluster-wide via Kubernetes API, but enforcement is per-node (each agent independently regenerates BPF programs)
156+
- Datapath forwarding/policy enforcement doesn't require kvstore (etcd) availability — policies materialized into local BPF
157+
- Debugging Q&A tasks require "why does this happen?" framing, not "how do I fix this?" — focus on architectural isolation mechanisms, not bugs
158+
- Negative checks should prevent blaming wrong abstraction layers (e.g., same-process vs separate process, centralized vs distributed compilation)
159+
- Used parallel exploratory agents (2 concurrent tasks) to research VS Code and Cilium simultaneously — saved significant time
160+
---

0 commit comments

Comments
 (0)