Skip to content

armgit / ruflo

Public Imported from GitHub
0

🌊 The original agent harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversational AI systems. Features adaptive memory, self-learning intelligence, federation, vector RAG integration, and native Claude Code / Codex / Hermes and many more Integrated

fix(ci): bypass ruvllm forwarding in graph-trajectory smoke (#2312)

ruvnet60964995367b
diff --git a/.github/workflows/v3-ci.yml b/.github/workflows/v3-ci.yml
index b236867d9e..50a0cac01f 100644
--- a/.github/workflows/v3-ci.yml
+++ b/.github/workflows/v3-ci.yml
@@ -2322,13 +2322,17 @@ jobs:
pnpm install --frozen-lockfile
pnpm -r build
- name: Run trajectory graph edges smoke
- # #2312 — TEST 2's post-task chain (intelligence.recordTrajectory →
- # @ruvector/ruvllm SonaCoordinator) blows past the default 4 GB heap.
- # Bump the runner ceiling to 6 GB so CI can complete; the underlying
- # ruvllm allocation profile is tracked separately and will fix the
- # cause, not the symptom.
+ # #2312 — TEST 2's post-task chain forwards into the @ruvector/ruvllm
+ # SonaCoordinator, whose first-call SONA adapt step allocates GB-scale
+ # and OOMs even at a 6 GB heap. The graph-edge contract this smoke is
+ # validating doesn't depend on that path, so we bypass it explicitly
+ # via the env-var guard added to intelligence.recordTrajectory. The
+ # underlying ruvllm allocation profile is still tracked as the real
+ # follow-up. Keep 4 GB heap so a future regression in the graph path
+ # itself can't silently absorb growth.
env:
- NODE_OPTIONS: '--max-old-space-size=6144'
+ CLAUDE_FLOW_SKIP_RUVLLM_TRAJECTORY: '1'
+ NODE_OPTIONS: '--max-old-space-size=4096'
run: node scripts/smoke-trajectory-graph-edges.mjs
graph-plugin-adapter-smoke:
diff --git a/v3/@claude-flow/cli/src/memory/intelligence.ts b/v3/@claude-flow/cli/src/memory/intelligence.ts
index eec8814681..886e99d319 100644
--- a/v3/@claude-flow/cli/src/memory/intelligence.ts
+++ b/v3/@claude-flow/cli/src/memory/intelligence.ts
@@ -1148,23 +1148,31 @@ export async function recordTrajectory(
}
}
- // Forward trajectory to @ruvector/ruvllm SonaCoordinator if available
- const ruvllmCoord = await loadRuvllmCoordinator();
- if (ruvllmCoord) {
- try {
- const avgQuality = verdict === 'success' ? 1.0 : verdict === 'partial' ? 0.5 : 0.0;
- ruvllmCoord.recordTrajectory({
- steps: enrichedSteps.map(s => ({
- state: s.content,
- action: s.type,
- reward: avgQuality,
- embedding: s.embedding || []
- })),
- totalReward: avgQuality,
- success: verdict === 'success'
- });
- } catch {
- // ruvllm recording failed silently
+ // Forward trajectory to @ruvector/ruvllm SonaCoordinator if available.
+ // #2312 — set CLAUDE_FLOW_SKIP_RUVLLM_TRAJECTORY=1 to bypass this branch.
+ // The ruvllm SonaCoordinator's recordTrajectory currently allocates GB-scale
+ // during its first SONA adapt step, which OOMs CI smokes (heap 6 GB+). The
+ // env-var lets graph-edge smokes assert their contract without dragging in
+ // the heavy ML path; production keeps the default path. A try/catch can't
+ // rescue this because JS heap OOM aborts the process — guard before the call.
+ if (!process.env.CLAUDE_FLOW_SKIP_RUVLLM_TRAJECTORY) {
+ const ruvllmCoord = await loadRuvllmCoordinator();
+ if (ruvllmCoord) {
+ try {
+ const avgQuality = verdict === 'success' ? 1.0 : verdict === 'partial' ? 0.5 : 0.0;
+ ruvllmCoord.recordTrajectory({
+ steps: enrichedSteps.map(s => ({
+ state: s.content,
+ action: s.type,
+ reward: avgQuality,
+ embedding: s.embedding || []
+ })),
+ totalReward: avgQuality,
+ success: verdict === 'success'
+ });
+ } catch {
+ // ruvllm recording failed silently
+ }
}
}