#Mem leak after upgrading to 2026.4.14.

1 messages · Page 1 of 1 (latest)

coral spear
#

Hey everyone — flagging something we noticed after upgrading to 2026.4.14.

The gateway (openclaw-gateway) has been climbing steadily in RSS since the upgrade — starting clean after a reboot and growing ~18MB+ every 30 min, sitting at ~950MB after ~7 hours. On an 8GB VM this gets noticeable fast.

Likely culprits from the changelog: new idle-aware background context maintenance (#65233) and active-memory recall buffering (#66144).

What we did:
Nightly 04:00 restart cron: 0 4 * * * /bin/systemctl restart openclaw-gateway,
Bumped VM RAM 8GB → 12GB,

Anyone else seeing this? 🙏

pliant horizonBOT
mellow pilotBOT
#

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

flint peakBOT
#

I haven’t seen a broad RSS-specific thread yet, but you’re probably not imagining it.

I did find a fresh adjacent report for 2026.4.14: #66708, where Active Memory caused major slowdown/terminated runs and improved immediately when disabled:
https://github.com/openclaw/openclaw/issues/66708

Also worth noting, 2026.4.15-beta.1 already has a follow-up fix in the new deferred context-engine maintenance path, #66820, which is pretty close to the change you called out.

A few quick checks I’d do:

openclaw --version
openclaw status --all
openclaw gateway status
openclaw logs --follow

And for isolation, try an A/B run with Active Memory off for a few hours:

# 
/active-memory off --global

If RSS flattens after that, I’d add your numbers to #66708, or open a separate GitHub issue if the main symptom is clearly steady RSS growth rather than latency.

My read:

  • steady RSS climb is suspicious
  • 2026.4.14 introduced two nearby changes
  • there is already at least one real regression report in the same area

Docs:
https://docs.openclaw.ai/help/faq
https://docs.openclaw.ai/gateway/troubleshooting
https://docs.openclaw.ai/concepts/active-memory

If you paste openclaw --version and whether Active Memory is enabled, I can help narrow which path looks hottest.

coral spear
#

OpenClaw 2026.4.14 (323493f)

flint peakBOT
#

Thanks, that helps. 2026.4.14 (323493f) is the exact release I was worried about, so this does not look like a version-mismatch problem.

Also, yes, there are at least a couple nearby reports now:

My best next split is:

  1. Check whether Active Memory is on:

    /active-memory status --global

  2. If it is on, disable it temporarily and watch RSS for a few hours:

# 
   /active-memory off --global

  1. If RSS growth mostly stops, that points hard at the Active Memory / recall path.

  2. If Active Memory is already off and RSS still climbs, then I’d suspect the deferred context-engine maintenance change more strongly.

If you can test on a non-prod box, I’d also try 2026.4.15-beta.1, because it already contains a follow-up fix in that area, #66820.

Docs:

If you want, paste the result of /active-memory status --global, and I can help narrow which side this is on.

coral spear
#

Active Memory status across all 13 agents: Provider: none, Dreaming: off on every single one. No vector search running at all. So Active Memory is definitively not the cause in our case.

That points the finger squarely at the deferred context-engine maintenance path (#65233). We'll upgrade to 2026.4.15 and report back. Thanks for the quick triage — really helpful.