openclaw - 💡(How to fix) Fix [Bug]: 2026.4.29 embedded agent reply latency ~40-47s before stream-ready; packaged channel deps missing [9 comments, 8 participants]

Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

After upgrading to OpenClaw 2026.4.29, Telegram and Discord replies consistently wait ~39-47s before stream-ready while logs show repeated embedded-run prep delays and event-loop/cpu liveness warnings.

Steps to reproduce

1. Upgrade to OpenClaw 2026.4.29.
2. Run Gateway via systemd user service.
3. Enable Telegram and Discord channels.
4. Send normal Telegram/Discord direct messages.
5. Observe user-visible delay and logs showing trace:embedded-run prep stages around 39-47s before stream-ready.
6. Optionally run openclaw status --deep; before manual hotfix it may print missing grammy / discord-api-types/v10 warnings.

Expected behavior

After a user message is received, the agent should begin streaming/replying without a consistent 40+ second OpenClaw-side preparation delay. 2026.4.26 did not show the same user-visible delay in this setup.

Actual behavior

After upgrade, replies frequently wait tens of seconds before the model response begins. Representative logs repeatedly show embedded-run prep taking about 39-47s before phase=stream-ready, with recurrent Gateway event_loop_delay / cpu liveness warnings. Telegram direct and Discord direct/channel conversations are both affected.

OpenClaw version

2026.4.29 (a448042)

Operating system

Linux 6.17.0-22-generic x64; Node v22.22.2

Install method

Stable channel, pnpm/global package; Gateway running as systemd user service

Model

Default model openai-codex/gpt-5.5; Discord test also observed slow behavior with x-ai/grok-4.1-fast

Provider / routing chain

Local OpenClaw Gateway on 127.0.0.1:18789; Telegram + Discord channels enabled; no external reverse proxy in this test

Additional provider/model setup details

Channels enabled: Telegram + Discord. Telegram network config already uses autoSelectFamily=false and dnsResultOrder=verbatim from a prior Telegram IPv6/IPv4 workaround. The latency appears in both Telegram and Discord, so it does not look channel-specific.

Related public reports/PRs found while triaging:

• #70615 (grammy missing after update)
• #62224 (discord-api-types/v10 missing after upgrade)
• #70857 (startup/reply latency, OpenClaw-side)
• #68327 (PR: reduce startup latency by ~36s; plugin/auth loading overhead)

Logs, screenshots, and evidence

Representative logs:


[trace:embedded-run] prep stages: phase=stream-ready totalMs=44202 stages=workspace-sandbox:2ms@2ms,skills:0ms@2ms,core-plugin-tools:15451ms@15453ms,bootstrap-context:11ms@15464ms,bundle-tools:2132ms@17596ms,system-prompt:10914ms@28510ms,session-resource-loader:5737ms@34247ms,agent-session:3ms@34250ms,stream-setup:9952ms@44202ms

[trace:embedded-run] prep stages: phase=stream-ready totalMs=47121 stages=workspace-sandbox:836ms@836ms,skills:0ms@836ms,core-plugin-tools:15702ms@16538ms,bootstrap-context:2278ms@18816ms,bundle-tools:2059ms@20875ms,system-prompt:10532ms@31407ms,session-resource-loader:5447ms@36854ms,agent-session:1ms@36855ms,stream-setup:10266ms@47121ms

liveness warning: reasons=event_loop_delay,event_loop_utilization,cpu interval=30s eventLoopDelayP99Ms=14294.2 eventLoopDelayMaxMs=14294.2 eventLoopUtilization=0.999 cpuCoreRatio=1.035 active=1 waiting=0 queued=0


Recent sample summary:


prep_total: n=28 min=39169 median=43098 max=47322 avg=43106
startup_total: n=27 min=14318 median=17234 max=34723 avg=20519
core_plugin_tools: n=28 min=15018 median=15405 max=15880 avg=15414
system_prompt: n=28 min=7250 median=10068 max=11549 avg=9499
stream_setup: n=28 min=8679 median=9718 max=10266 avg=9517
model_resolution: n=27 min=3648 median=4164 max=14914 avg=7156
auth: n=27 min=5333 median=6805 max=7381 avg=6720


Packaged dependency warnings observed before manual hotfix:


[channels] failed to load bundled channel discord: Cannot find module 'discord-api-types/v10'
Require stack:
- $GLOBAL_OPENCLAW/dist/extensions/discord/discord-*.js

[channels] failed to load bundled channel telegram: Cannot find module 'grammy'
Require stack:
- $GLOBAL_OPENCLAW/dist/extensions/telegram/send-*.js


Manual hotfix applied:


cd $GLOBAL_OPENCLAW
npm install --no-save [email protected] [email protected]


After the hotfix, module resolution succeeds:


grammy OK -> $GLOBAL_OPENCLAW/node_modules/grammy/out/mod.js
discord-api-types/v10 OK -> $GLOBAL_OPENCLAW/node_modules/discord-api-types/v10.js
json5 OK -> $GLOBAL_OPENCLAW/node_modules/json5/lib/index.js


However, reply latency persisted afterward; post-hotfix samples still show prep stages such as `totalMs=40597`, `44202`, and `47121`.

Impact and severity

Affected users/systems/channels: this setup's Telegram direct messages and Discord direct/channel messages.

Severity: high usability impact; normal chat replies feel delayed by ~40s before model output begins.

Frequency: repeated across many recent runs after the 2026.4.29 upgrade.

Consequence: messaging channels feel unreliable/slow, and the delay appears to be in OpenClaw/Gateway prep before provider streaming rather than only model latency.

Additional information

Other observations:

• Gateway CPU is often high during/after runs. A status snapshot showed the Gateway node process around 40% CPU while idle-ish.
• Recurrent liveness warnings include high event loop delay and event loop utilization.
• There are Discord API fetch timeout / native slash command deploy 429 warnings, but the same slow response is visible on Telegram too, so this does not look like a single Discord-only bottleneck.
• No session-write-lock / sessions.json.lock warnings were observed locally, so this may be adjacent to #70857 but not the exact same lock symptom.
• Local sanitized evidence was collected as status.txt and relevant-log-summary.txt; I can provide additional redacted lines if useful.