openclaw - 💡(How to fix) Fix [Bug]: Session cleanup per #75707 workaround has no effect on ~18s prep-stage overhead (macOS ARM64, v2026.4.29) [1 comments, 2 participants]

kingkong9817 · 2026-05-02T08:47:37Z

[openclaw] Follow-up to 75861 closed as duplicate of 75707 and 75688 closed, superseded by narrower trackers . This report tracks what was tried, measured resu… Follow-up to #75861 (closed as duplicate of #75707) and #75688 (closed, superseded by narrower trackers). This report tracks **what was tried, measured results, and whether any improvement occurred** on an affected macOS ARM64 machine after applying one of the #75707 workarounds. ## Fix / Workaround Follow-up to #75861 (closed as duplicate of #75707) and #75688 (closed, superseded by narrower trackers). This report tracks **what was tried, measured results, and whether any improvement occurred** on an affected macOS ARM64 machine after applying one of the #75707 workarounds. Following the #75707 workaround list, the **most impactful item** was attempted: **The zombie-session workaround from #75707 did NOT resolve the pre-flight overhead on this macOS configuration.** The `core-plugin-tools` (6.2s), `system-prompt` (5.0s), and `stream-setup` (5.0s) stages are completely unaffected by session file count. This confirms @adminsystemease's observation on #75861 that the bottleneck is synchronous JS work in the embedded runtime, not I/O or session state. ## Summary Follow-up to #75861 (closed as duplicate of #75707) and #75688 (closed, superseded by narrower trackers). This report tracks **what was tried, measured results, and whether any improvement occurred** on an affected macOS ARM64 machine after applying one of the #75707 workarounds. ## Environment (unchanged) - **OpenClaw:** 2026.4.29 (a448042) - **Node:** v25.9.0 - **OS:** macOS 26.4.1 (ARM64, Mac Studio M1 Max, 64GB) - **Channel:** Feishu WebSocket - **Model:** DeepSeek V4 Flash / V4 Pro (remote), Ollama qwen30-a3b-128k (local) - **Gateway PID:** 38057, running ~13h - **Memory:** 941MB RSS (peak ~960MB) - **Plugins:** feishu, llm-task, memclaw - **MCP:** mempalace (Python), context-mode (npx) ## What Was Tried Following the #75707 workaround list, the **most impactful item** was attempted: > "Delete zombie sessions (`rm ~/.openclaw/agents/*/sessions/*.jsonl`) — CPU 100%+ → 30%" **Action taken at ~2026-05-02 06:45 UTC:** - Identified 53 session files (26 sessions + trajectories), 41MB total under `agents/main/sessions/` - Classified by content: - ~20 sessions: cron auto-tasks (news capture/clean/briefing, token reports, WeChat drafts, GitHub issue checks, NAS backup) - ~15 sessions: subagent child tasks - ~3 sessions: human direct chat (including the active session) - Archived 46 files (23 sessions) to `sessions-backup/` — all auto-task and subagent sessions - Retained 2 human-chat sessions (1 active: `d66a10df`, 1 recent: `323ed82f`) - Sessions directory reduced from 41MB / 53 files → 18MB / ~6 files ## Measured Results — Prep Stages (pre-model call overhead) **Before cleanup (#75861 data, ~03:56 UTC):** ``` prep stages totalMs=18325: core-plugin-tools: 6216ms system-prompt: 4986ms stream-setup: 4939ms bundle-tools: 1104ms session-resource-loader: 1075ms bootstrap-context: 5ms workspace-sandbox: 0ms ``` **After cleanup (latest 5 runs, 08:20–08:35 UTC):** ``` Run 1: totalMs=18266 core-plugin-tools:6142 system-prompt:4999 stream-setup:4938 bundle-tools:1110 session-resource-loader:1074 Run 2: totalMs=18221 core-plugin-tools:6170 system-prompt:4994 stream-setup:4918 bundle-tools:1068 session-resource-loader:1067 Run 3: totalMs=18869 core-plugin-tools:6147 system-prompt:4972 stream-setup:4953 bundle-tools:1100 session-resource-loader:1690 Run 4: totalMs=18209 core-plugin-tools:6151 system-prompt:4983 stream-setup:4925 bundle-tools:1069 session-resource-loader:1079 Run 5: totalMs=18320 core-plugin-tools:6213 system-prompt:4991 stream-setup:4954 bundle-tools:1071 session-resource-loader:1088 ``` **No measurable improvement in prep stages. All values within ±2% of pre-cleanup baseline.** ## Measured Results — Event Loop Blocking (liveness warnings) **Before:** ``` eventLoopDelayMaxMs: 7017–10502ms eventLoopUtilization: 0.78–0.91 interval: 30–36s ``` **After:** ``` eventLoopDelayMaxMs: 5763–8015ms (marginal ~20% reduction, still 10× above normal <500ms) eventLoopUtilization: 0.23–0.97 (still widely fluctuating) interval: 31–36s (unchanged) ``` **Marginal improvement, far from resolved.** ## Measured Results — Gateway Process | Metric | Before | After | |--------|--------|-------| | CPU | 53.7% (under load) | 8.4% (idle moment) | | RSS | 940MB | 941MB | | node.list latency | 67ms–8428ms | unchanged pattern | ## Persistent Issues (still present after cleanup) 1. **agent.wait peaks:** 315,808ms (5min 16s) observed 2. **trajectory-flush timeouts:** ≥10s, occurred twice 3. **API timeouts (AxiosError 10000ms):** continuing 4. **Startup `auth` stage:** 4.5–5.0s (tracked at #75782, no fix yet) ## Correlation with Other Reports | Issue | Overlap | Status | |-------|---------|--------| | #75707 | Same symptoms, different OS (Ubuntu) | OPEN | | #75290 | core-plugin-tools latency | OPEN, PR by @hexsprite pending |

openclaw2026-05-02 08:47:37

ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

GitHub issue URL

Helpful · Quick feedback

GitHub stats

openclaw/openclaw#76009•Fetched 2026-05-03 04:43:21

View on GitHub

Comments

Participants

Timeline

Reactions

Author

kingkong9817

Participants

clawsweeper[bot]

kingkong9817

Timeline (top)

mentioned ×2subscribed ×2closed ×1commented ×1

Follow-up to #75861 (closed as duplicate of #75707) and #75688 (closed, superseded by narrower trackers). This report tracks what was tried, measured results, and whether any improvement occurred on an affected macOS ARM64 machine after applying one of the #75707 workarounds.

Root Cause

Fix Action

Fix / Workaround

Following the #75707 workaround list, the most impactful item was attempted:

The zombie-session workaround from #75707 did NOT resolve the pre-flight overhead on this macOS configuration. The core-plugin-tools (6.2s), system-prompt (5.0s), and stream-setup (5.0s) stages are completely unaffected by session file count. This confirms @adminsystemease's observation on #75861 that the bottleneck is synchronous JS work in the embedded runtime, not I/O or session state.

Code Example

prep stages totalMs=18325:
  core-plugin-tools:       6216ms
  system-prompt:           4986ms
  stream-setup:            4939ms
  bundle-tools:            1104ms
  session-resource-loader: 1075ms
  bootstrap-context:          5ms
  workspace-sandbox:          0ms

---

Run 1: totalMs=18266  core-plugin-tools:6142  system-prompt:4999  stream-setup:4938  bundle-tools:1110  session-resource-loader:1074
Run 2: totalMs=18221  core-plugin-tools:6170  system-prompt:4994  stream-setup:4918  bundle-tools:1068  session-resource-loader:1067
Run 3: totalMs=18869  core-plugin-tools:6147  system-prompt:4972  stream-setup:4953  bundle-tools:1100  session-resource-loader:1690
Run 4: totalMs=18209  core-plugin-tools:6151  system-prompt:4983  stream-setup:4925  bundle-tools:1069  session-resource-loader:1079
Run 5: totalMs=18320  core-plugin-tools:6213  system-prompt:4991  stream-setup:4954  bundle-tools:1071  session-resource-loader:1088

---

eventLoopDelayMaxMs: 7017–10502ms
eventLoopUtilization: 0.78–0.91
interval: 30–36s

---

eventLoopDelayMaxMs: 5763–8015ms  (marginal ~20% reduction, still 10× above normal <500ms)
eventLoopUtilization: 0.23–0.97  (still widely fluctuating)
interval: 31–36s  (unchanged)

---

2026-05-02T16:20:43.829+08:00 [trace:embedded-run] prep stages totalMs=18266 stages=workspace-sandbox:1ms,skills:0ms,core-plugin-tools:6142ms,bootstrap-context:2ms,bundle-tools:1110ms,system-prompt:4999ms,session-resource-loader:1074ms,agent-session:0ms,stream-setup:4938ms
2026-05-02T16:23:49.343+08:00 [trace:embedded-run] prep stages totalMs=18221 stages=workspace-sandbox:1ms,skills:0ms,core-plugin-tools:6170ms,bootstrap-context:3ms,bundle-tools:1068ms,system-prompt:4994ms,session-resource-loader:1067ms,agent-session:0ms,stream-setup:4918ms
2026-05-02T16:28:58.692+08:00 [trace:embedded-run] prep stages totalMs=18869 stages=workspace-sandbox:0ms,skills:0ms,core-plugin-tools:6147ms,bootstrap-context:7ms,bundle-tools:1100ms,system-prompt:4972ms,session-resource-loader:1690ms,agent-session:0ms,stream-setup:4953ms
2026-05-02T16:31:54.072+08:00 [trace:embedded-run] prep stages totalMs=18209 stages=workspace-sandbox:0ms,skills:0ms,core-plugin-tools:6151ms,bootstrap-context:1ms,bundle-tools:1069ms,system-prompt:4983ms,session-resource-loader:1079ms,agent-session:1ms,stream-setup:4925ms
2026-05-02T16:35:51.065+08:00 [trace:embedded-run] prep stages totalMs=18320 stages=workspace-sandbox:1ms,skills:0ms,core-plugin-tools:6213ms,bootstrap-context:2ms,bundle-tools:1071ms,system-prompt:4991ms,session-resource-loader:1088ms,agent-session:0ms,stream-setup:4954ms

---

2026-05-02T16:20:45.871+08:00 [diagnostic] liveness warning: reasons=event_loop_delay,event_loop_utilization,cpu interval=32s eventLoopDelayP99Ms=6727.7 eventLoopDelayMaxMs=7017.1 eventLoopUtilization=0.966 cpuCoreRatio=0.99 active=1
2026-05-02T16:31:56.369+08:00 [diagnostic] liveness warning: reasons=event_loop_delay,cpu interval=34s eventLoopDelayP99Ms=6752.8 eventLoopDelayMaxMs=7021.3 eventLoopUtilization=0.935 cpuCoreRatio=0.959 active=1

RAW_BUFFERClick to expand / collapse

Summary

Environment (unchanged)

OpenClaw: 2026.4.29 (a448042)
Node: v25.9.0
OS: macOS 26.4.1 (ARM64, Mac Studio M1 Max, 64GB)
Channel: Feishu WebSocket
Model: DeepSeek V4 Flash / V4 Pro (remote), Ollama qwen30-a3b-128k (local)
Gateway PID: 38057, running ~13h
Memory: 941MB RSS (peak ~960MB)
Plugins: feishu, llm-task, memclaw
MCP: mempalace (Python), context-mode (npx)

What Was Tried

Following the #75707 workaround list, the most impactful item was attempted:

"Delete zombie sessions (rm ~/.openclaw/agents/*/sessions/*.jsonl) — CPU 100%+ → 30%"

Action taken at ~2026-05-02 06:45 UTC:

Identified 53 session files (26 sessions + trajectories), 41MB total under agents/main/sessions/
Classified by content:
- ~20 sessions: cron auto-tasks (news capture/clean/briefing, token reports, WeChat drafts, GitHub issue checks, NAS backup)
- ~15 sessions: subagent child tasks
- ~3 sessions: human direct chat (including the active session)
Archived 46 files (23 sessions) to sessions-backup/ — all auto-task and subagent sessions
Retained 2 human-chat sessions (1 active: d66a10df, 1 recent: 323ed82f)
Sessions directory reduced from 41MB / 53 files → 18MB / ~6 files

Measured Results — Prep Stages (pre-model call overhead)

Before cleanup (#75861 data, ~03:56 UTC):

prep stages totalMs=18325:
  core-plugin-tools:       6216ms
  system-prompt:           4986ms
  stream-setup:            4939ms
  bundle-tools:            1104ms
  session-resource-loader: 1075ms
  bootstrap-context:          5ms
  workspace-sandbox:          0ms

After cleanup (latest 5 runs, 08:20–08:35 UTC):

Run 1: totalMs=18266  core-plugin-tools:6142  system-prompt:4999  stream-setup:4938  bundle-tools:1110  session-resource-loader:1074
Run 2: totalMs=18221  core-plugin-tools:6170  system-prompt:4994  stream-setup:4918  bundle-tools:1068  session-resource-loader:1067
Run 3: totalMs=18869  core-plugin-tools:6147  system-prompt:4972  stream-setup:4953  bundle-tools:1100  session-resource-loader:1690
Run 4: totalMs=18209  core-plugin-tools:6151  system-prompt:4983  stream-setup:4925  bundle-tools:1069  session-resource-loader:1079
Run 5: totalMs=18320  core-plugin-tools:6213  system-prompt:4991  stream-setup:4954  bundle-tools:1071  session-resource-loader:1088

No measurable improvement in prep stages. All values within ±2% of pre-cleanup baseline.

Measured Results — Event Loop Blocking (liveness warnings)

Before:

eventLoopDelayMaxMs: 7017–10502ms
eventLoopUtilization: 0.78–0.91
interval: 30–36s

After:

eventLoopDelayMaxMs: 5763–8015ms  (marginal ~20% reduction, still 10× above normal <500ms)
eventLoopUtilization: 0.23–0.97  (still widely fluctuating)
interval: 31–36s  (unchanged)

Marginal improvement, far from resolved.

Measured Results — Gateway Process

Metric	Before	After
CPU	53.7% (under load)	8.4% (idle moment)
RSS	940MB	941MB
node.list latency	67ms–8428ms	unchanged pattern

Persistent Issues (still present after cleanup)

agent.wait peaks: 315,808ms (5min 16s) observed
trajectory-flush timeouts: ≥10s, occurred twice
API timeouts (AxiosError 10000ms): continuing
Startup auth stage: 4.5–5.0s (tracked at #75782, no fix yet)

Correlation with Other Reports

Issue	Overlap	Status
#75707	Same symptoms, different OS (Ubuntu)	OPEN
#75290	core-plugin-tools latency	OPEN, PR by @hexsprite pending
#75782	auth stage 10–15s	OPEN, diagnosed only
#75999	4.29 regression, ~73s prep stages	OPEN, created today
#75439	heartbeat exec-event runaway	OPEN, PR by @hexsprite pending

Conclusion

The machine remains on 2026.4.29. The two pending PRs from @hexsprite (manifest scan cache + heartbeat cooldown) are the most promising near-term fixes, but neither addresses system-prompt or stream-setup directly.

Raw log excerpts

2026-05-02T16:20:43.829+08:00 [trace:embedded-run] prep stages totalMs=18266 stages=workspace-sandbox:1ms,skills:0ms,core-plugin-tools:6142ms,bootstrap-context:2ms,bundle-tools:1110ms,system-prompt:4999ms,session-resource-loader:1074ms,agent-session:0ms,stream-setup:4938ms
2026-05-02T16:23:49.343+08:00 [trace:embedded-run] prep stages totalMs=18221 stages=workspace-sandbox:1ms,skills:0ms,core-plugin-tools:6170ms,bootstrap-context:3ms,bundle-tools:1068ms,system-prompt:4994ms,session-resource-loader:1067ms,agent-session:0ms,stream-setup:4918ms
2026-05-02T16:28:58.692+08:00 [trace:embedded-run] prep stages totalMs=18869 stages=workspace-sandbox:0ms,skills:0ms,core-plugin-tools:6147ms,bootstrap-context:7ms,bundle-tools:1100ms,system-prompt:4972ms,session-resource-loader:1690ms,agent-session:0ms,stream-setup:4953ms
2026-05-02T16:31:54.072+08:00 [trace:embedded-run] prep stages totalMs=18209 stages=workspace-sandbox:0ms,skills:0ms,core-plugin-tools:6151ms,bootstrap-context:1ms,bundle-tools:1069ms,system-prompt:4983ms,session-resource-loader:1079ms,agent-session:1ms,stream-setup:4925ms
2026-05-02T16:35:51.065+08:00 [trace:embedded-run] prep stages totalMs=18320 stages=workspace-sandbox:1ms,skills:0ms,core-plugin-tools:6213ms,bootstrap-context:2ms,bundle-tools:1071ms,system-prompt:4991ms,session-resource-loader:1088ms,agent-session:0ms,stream-setup:4954ms

2026-05-02T16:20:45.871+08:00 [diagnostic] liveness warning: reasons=event_loop_delay,event_loop_utilization,cpu interval=32s eventLoopDelayP99Ms=6727.7 eventLoopDelayMaxMs=7017.1 eventLoopUtilization=0.966 cpuCoreRatio=0.99 active=1
2026-05-02T16:31:56.369+08:00 [diagnostic] liveness warning: reasons=event_loop_delay,cpu interval=34s eventLoopDelayP99Ms=6752.8 eventLoopDelayMaxMs=7021.3 eventLoopUtilization=0.935 cpuCoreRatio=0.959 active=1

extent analysis

TL;DR

The zombie-session workaround from #75707 did not resolve the pre-flight overhead issue on the macOS configuration, suggesting that the bottleneck is synchronous JS work in the embedded runtime.

Guidance

Review the core-plugin-tools, system-prompt, and stream-setup stages to identify potential synchronous JS work that may be causing the bottleneck.
Investigate the pending PRs from @hexsprite (manifest scan cache + heartbeat cooldown) as potential near-term fixes.
Analyze the raw log excerpts to gain further insight into the prep stages and liveness warning issues.
Consider testing the system with a reduced workload or in a controlled environment to isolate the root cause of the issue.

Example

No code snippet is provided as the issue does not contain specific code-related information.

Notes

The issue is complex and may require a multi-faceted approach to resolve. The provided guidance is based on the information available in the issue and may not be exhaustive.

Recommendation

Apply the pending PRs from @hexsprite (manifest scan cache + heartbeat cooldown) as they may address some of the underlying issues, although they do not directly address the system-prompt or stream-setup stages.

Vote matrix · Quick signals

Works

Did the solution work? Tap to confirm.

Easy Fix

Was it a quick fix?

Time Saver

Did it save you time?

Blocking

Was it severely blocking?

Common Issue

Are others likely hitting this too?

Flaky / Intermittent

Is it intermittent?

Verified / Reproducible

Can you reproduce it reliably?

#api #agent execution #callback error #memory management #API rate limit

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Data

Security

Network

Code

UI/UX

Text

System

Multimedia

Protocol

API

Engineering

openclaw - 💡(How to fix) Fix [Bug]: Session cleanup per #75707 workaround has no effect on ~18s prep-stage overhead (macOS ARM64, v2026.4.29) [1 comments, 2 participants]

Recommended Tools

GitHub issue graph ai analysis

Root Cause

Fix Action

Fix / Workaround

Code Example

Summary

Environment (unchanged)

What Was Tried

Measured Results — Prep Stages (pre-model call overhead)

Measured Results — Event Loop Blocking (liveness warnings)

Measured Results — Gateway Process

Persistent Issues (still present after cleanup)

Correlation with Other Reports

Conclusion

Raw log excerpts

extent analysis

TL;DR

Guidance

Example

Notes

Recommendation

Still need to ship something?

TRENDING

openclaw - 💡(How to fix) Fix [Bug]: Session cleanup per #75707 workaround has no effect on ~18s prep-stage overhead (macOS ARM64, v2026.4.29) [1 comments, 2 participants]

Recommended Tools

GitHub issue graph ai analysis

Root Cause

Fix Action

Fix / Workaround

Code Example

Summary

Environment (unchanged)

What Was Tried

Measured Results — Prep Stages (pre-model call overhead)

Measured Results — Event Loop Blocking (liveness warnings)

Measured Results — Gateway Process

Persistent Issues (still present after cleanup)

Correlation with Other Reports

Conclusion

Raw log excerpts

extent analysis

TL;DR

Guidance

Example

Notes

Recommendation

Still need to ship something?

RELATED_DISCOVERY

TRENDING