OpenClaw 2026.3.4 - provider-pending lifecycle guardrails

Release highlights (2026.3.4)

• Lifecycle reliability: suspend, resume, and reboot operations now return 202 for temporary

provider pending/locked states instead of surfacing false 500 errors.

• Retry metadata: lifecycle pending responses now include pending, retryable, provider,

and operation fields so clients can handle retries cleanly.

• Reconcile hardening: docker host reconcile now defers fresh destroying hosts, retries stale

ones, and reports provider-pending issue counts.

• Ops visibility: provider-pending lifecycle events are now recorded as deployment events for

faster incident triage.

Why this matters

Provider control planes are eventually consistent. Returning 500 on transient lock states makes healthy systems look broken and causes unnecessary support load. This release converts those transients into explicit retryable states and improves operational diagnostics.

Operator notes

• Treat 202 on lifecycle actions as expected retryable behavior.
• Use deployment events where type=provisioningstep and metadata.step=providerpending to

track impacted hosts.

• Monitor reconcile counters:

- deferredDestroyingHosts - retriedDestroyingHosts - providerPendingIssues

Runbook links

• Admin runbook: docs/admin-runbook.md
• Provider transient lifecycle runbook: docs/ops/provider-transient-lifecycle.md