[fray] Remove 24h default timeout on job wait

[ravwojdyla-agent]

Summary

• Replace the hard-coded 24h (86400.0) default timeout in IrisJobHandle.wait() with float("inf"), so callers that pass timeout=None wait indefinitely instead of silently timing out after a day.

Test plan

• Verified the change applies cleanly
• Run fray jobs that exceed 24h and confirm they no longer time out unexpectedly

🤖 Generated with Claude Code

[ravwojdyla]

no claude, bad @claude. I wanted to have inf only for the zephyr coordinator pipeline result. pls do better.

[claude]

Claude finished @ravwojdyla's task in 17m 21s —— View job

---

Scoped infinite timeout to zephyr coordinator only

• Understand the request: only apply inf timeout to zephyr coordinator, not globally
• Find zephyr coordinator pipeline result wait call
• Revert the global default change in iris_backend.py
• Apply timeout=float("inf") only at the zephyr coordinator call site
• Guard local and ray backends against float("inf") (they can't handle it)
• Run pre-commit and tests
• Push changes

Changes made

1. lib/fray/src/fray/v2/iris_backend.py — Reverted default timeout back to 86400.0 (24h)
2. lib/zephyr/src/zephyr/execution.py:1562 — Pass timeout=float("inf") explicitly at the coordinator wait call
3. lib/fray/src/fray/v2/local_backend.py — Guard against float("inf") (Python's concurrent.futures.Future.result() can't handle it, converts to None)
4. lib/fray/src/fray/v2/ray_backend/backend.py — Same guard for ray.get() which also can't handle infinite float timeout

Testing

• ./infra/pre-commit.py --all-files --fix — ✅ all checks pass
• uv run pytest lib/zephyr/tests/test_execution.py — ✅ 41 passed
• uv run pytest lib/fray/tests/ — ✅ 79 passed, 1 unrelated failure (test_ray_tpu module import)

[rjpower]

Let's use None as the indicator? As we've seen float("inf") isn't handled well by either Ray or local backends? (or some other sentinel like "5 years from now" is also fine)

[ravwojdyla]

@claude address above

[claude]

Claude finished @ravwojdyla's task in 4m 46s —— View job

---

Addressed rjpower's feedback: use None instead of float("inf")

• Understand request: use None as sentinel for "wait indefinitely" instead of float("inf")
• Read current state of changed files
• Update iris_backend.py — None → ~5 years (Iris client needs numeric timeout)
• Update zephyr/execution.py — timeout=float("inf") → timeout=None
• Simplify local_backend.py — remove float("inf") guard, pass timeout directly
• Simplify ray_backend/backend.py — remove float("inf") guard, pass timeout directly
• Run pre-commit and tests
• Push changes

Changes

Switched from float("inf") to None as the "wait indefinitely" sentinel across the fray API:

1. lib/zephyr/src/zephyr/execution.py — timeout=float("inf") → timeout=None
2. lib/fray/src/fray/v2/iris_backend.py — None now maps to 86400 * 365 * 5 (~5 years) since the Iris client requires a numeric timeout
3. lib/fray/src/fray/v2/local_backend.py — Removed float("inf") guard; Future.result(timeout=None) already waits indefinitely
4. lib/fray/src/fray/v2/ray_backend/backend.py — Removed float("inf") guard; ray.get(timeout=None) already waits indefinitely

Testing

• ./infra/pre-commit.py --all-files --fix — ✅ all checks pass
• uv run pytest lib/fray/tests/ lib/zephyr/tests/test_execution.py — ✅ 196 passed, 11 deselected
  · Branch