perf(cloud-agent): reduce idle DO wake-ups with 1-hour reaper interval

[eshurakov]

Summary

Reduces unnecessary Durable Object wake-ups by introducing a 1-hour idle reaper interval (REAPER_IDLE_INTERVAL_MS) when no execution is active, replacing the previous 5-minute default. When an execution starts, the alarm is immediately rescheduled to the 2-minute active interval so stale/hung execution detection remains responsive.

Key changes:

• New REAPER_IDLE_INTERVAL_MS (1 hour) used as the default interval after alarm runs when no execution is active.
• On execution start, setAlarm is called with REAPER_ACTIVE_INTERVAL_MS (2 min) to close the gap where the idle alarm could be up to an hour away.
• Error fallback in the alarm handler uses REAPER_INTERVAL_MS_DEFAULT (5 min) rather than the idle interval, so the reaper retries conservatively when state is unknown.
• Clarifying JSDoc on getReaperIntervalMs() documenting that the env override only affects the initial alarm scheduling.

Verification

• Code review performed in this session — identified and applied two improvements (error fallback safety, env override documentation)
• Typecheck passed (tsgo --noEmit for cloud-agent-next and dependencies, via pre-push hook)
• Lint passed (oxfmt and oxlint, via pre-push hook)
• Additional verification (tests, manual checks, etc.)

Visual Changes

N/A

Reviewer Notes

• The REAPER_INTERVAL_MS env var override now only controls the initial alarm via ensureAlarmScheduled(). Steady-state scheduling uses the new constants directly. This is documented in the added JSDoc.
• The error catch block in the alarm handler intentionally falls back to 5 min (REAPER_INTERVAL_MS_DEFAULT) rather than the 1-hour idle interval — if getActiveExecutionId() throws, we can't know whether an execution is active, so we err on the side of checking sooner.

[kilo-code-bot]

WARNING: Hourly idle cadence delays idle wrapper shutdown

cleanupIdleKiloServer() still relies on this alarm to enforce KILO_SERVER_IDLE_TIMEOUT_MS_DEFAULT (15 minutes). After a short execution completes, the next active alarm will usually observe idleMs < idleTimeoutMs, and then this reschedules the following check for 1 hour later. That means an otherwise idle wrapper can stay alive for roughly 60 minutes instead of 15, increasing sandbox usage until the next wake-up.

[kilo-code-bot]

Code Review Summary

Status: 1 Issues Found | Recommendation: Address before merge

Overview

Severity	Count
CRITICAL	0
WARNING	1
SUGGESTION	0

Fix these issues in Kilo Cloud

Issue Details (click to expand)

WARNING

File	Line	Issue
cloud-agent-next/src/persistence/CloudAgentSession.ts	1362	Idle sessions now wait up to an hour before cleanupIdleKiloServer() runs again, so wrappers can stay alive far past the intended 15-minute idle timeout.

Other Observations (not in diff)

Issues found in unchanged code that cannot receive inline comments:

File	Line	Issue
N/A	N/A	None.

Files Reviewed (1 files)

• cloud-agent-next/src/persistence/CloudAgentSession.ts - 1 issue

---

_{Reviewed by gpt-5.4-2026-03-05 · 901,764 tokens}