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): bump cold-cache timeouts (iter 131)

ruv95dc43be20f0
diff --git a/plugins/ruflo-metaharness/scripts/test-mcp-tools.mjs b/plugins/ruflo-metaharness/scripts/test-mcp-tools.mjs
index 0b4aacd70f..44aa474c79 100755
--- a/plugins/ruflo-metaharness/scripts/test-mcp-tools.mjs
+++ b/plugins/ruflo-metaharness/scripts/test-mcp-tools.mjs
@@ -142,19 +142,19 @@ async function main() {
input = { dryRun: true };
}
- // iter 124 β€” bumped default 30s β†’ 60s. The first MCP tool that shells
- // out via npx pays the cold-cache warmup cost (~25s observed), which
- // pushed metaharness_audit_list past the prior 30s budget intermittently.
- // iter 128 β€” oia_audit composes 5 sub-audits; same 90s budget as
- // drift_from_history (also a chain-tool).
- // iter 130 β€” also bump audit_list to 90s. In CI cold-cache, even after
- // oia_audit warmed up the npx cache, audit_list still timed out at 60s
- // (likely the memory `npx claude-flow memory list` invocation hits its
- // own cold path). audit_list is in the chain-tool bracket too.
+ // iter 124 β†’ 130 β€” timeouts have crept up as CI cold-cache npx
+ // warmup costs got measured. Final budgets:
+ // default : 60s
+ // chain-tools : 180s (drift_from_history + oia_audit + audit_list)
+ // iter 131 β€” bumped chain-tool budget 90s β†’ 180s. audit_list still
+ // timed out at 90s in CI; locally it runs in ~4s, but CI's
+ // `npx @claude-flow/cli@latest memory list` invocation pays both
+ // the npx fetch AND a full CLI startup (which loads agentic-flow +
+ // ONNX). 180s gives 30x headroom over the local cost.
const isChainTool = tool.name === 'metaharness_drift_from_history'
|| tool.name === 'metaharness_oia_audit'
|| tool.name === 'metaharness_audit_list';
- const timeoutMs = isChainTool ? 90_000 : 60_000;
+ const timeoutMs = isChainTool ? 180_000 : 60_000;
const handlerPromise = tool.handler(input);
const timeoutPromise = new Promise((_, reject) =>
setTimeout(() => reject(new Error(`${timeoutMs / 1000}s handler timeout`)), timeoutMs));
diff --git a/scripts/smoke-windows-hook-execution.mjs b/scripts/smoke-windows-hook-execution.mjs
index 283fc90d6d..f56c1a66d3 100644
--- a/scripts/smoke-windows-hook-execution.mjs
+++ b/scripts/smoke-windows-hook-execution.mjs
@@ -58,11 +58,13 @@ const initResult = spawnSync(
cwd: tmpDir,
env: { ...process.env, CI: 'true', FORCE_COLOR: '0' },
encoding: 'utf8',
- // iter 123 β€” bumped from 60_000 to 180_000 (same reason as
- // scripts/smoke-windows-init-hooks.mjs). Cold-start `ruflo init` on
- // Windows runners exceeds 60s; init exits with status=null (timer
- // fired) and no settings.json is written. 180s gives 3x headroom.
- timeout: 180_000,
+ // iter 131 β€” bumped from 180s to 300s (5 min). macos-latest in this
+ // matrix observed `init` running exactly 180s before the timer fired
+ // (CI cold ONNX download + agentic-flow init + MCP server spawn).
+ // The CI job's timeout-minutes is 10, so 300s leaves room for the
+ // smoke's assertion phase to also run. The default GH timeout-minutes
+ // is 360s; we're now well under that on all platforms.
+ timeout: 300_000,
}
);
diff --git a/scripts/smoke-windows-init-hooks.mjs b/scripts/smoke-windows-init-hooks.mjs
index dd4887b200..32be43dfe6 100644
--- a/scripts/smoke-windows-init-hooks.mjs
+++ b/scripts/smoke-windows-init-hooks.mjs
@@ -59,13 +59,11 @@ const result = spawnSync(
FORCE_COLOR: '0',
},
encoding: 'utf8',
- // iter 123 β€” bumped from 60_000 to 180_000. Windows cold-start `ruflo
- // init` consistently exceeds 60s (pnpm cache cold, hnswlib-node gyp
- // probe overhead, antivirus stat). The init exited with status=null
- // (= the spawn timer fired) which means it hadn't even finished the
- // init handler, never mind written settings.json. 180s gives 3x
- // headroom while still catching a genuinely-hung init.
- timeout: 180_000,
+ // iter 123 β†’ 131 β€” bumped from 180s to 300s. macos-latest observed
+ // running exactly 180s before the timer fired (CI cold ONNX download
+ // + agentic-flow init + MCP server spawn). 300s leaves room for the
+ // smoke's assertion phase under the job's 10-min cap.
+ timeout: 300_000,
// Don't fail if the command exits non-zero β€” init may partially succeed
}
);