mrleo1nid/gpu-rent
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Implement idle-killer enhancements and swarm management improvements

Browse Source 
- Marked critical bugs as resolved in the review documentation, including changes to the `arm_idle_killer` function to raise errors on credential creation failures and ensure proper file permissions for JSON credentials.
- Introduced a new `_try_arm_idle_killer` function in `provision.py` to manage idle-killer state more effectively, ensuring it arms correctly during provisioning.
- Updated the `swarm_busy` function in `remote/idle_killer.py` to allow idle state after a specified duration of Swarm unavailability, preventing unnecessary billing.
- Enhanced performance tuning logic in `tune_swarm_perf.py` to ensure proper handling of pip installation success before applying extra arguments.
- Added tests to validate the new idle-killer behavior and swarm management logic, ensuring robustness in handling idle states and error conditions.
This commit is contained in:
Leonid Pershin
2026-08-21 07:04:09 +03:00
parent d409e2e154
commit 1ec615c03e
7 changed files with 268 additions and 104 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/reviews/2026-08-21-review-3.md
+5 -5
View File
@@ -81,10 +81,10 @@ N/A for CLI critical pass (or no critical issues).
Critical only:
- [ ] 1. [Bug] `arm_idle_killer`: on cred create failure **raise** or return False; never set `notes.idle_killer=armed`; surface same ⚠ as `"failed"` — `src/gpu_rent/idle_killer.py` line 151, `src/gpu_rent/provision.py` line 516
- [ ] 2. [Bug] Idle-killer: if Swarm unreachable longer than N minutes (e.g. 2× idle or fixed 60m), treat as idle/allow delete (llm-only already bypasses) — `src/gpu_rent/remote/idle_killer.py` line 75
- [ ] 3. [Logic] On mid-`up` failure after server create, arm killer anyway or fail loudly and refuse to leave session without killer / document mandatory `stop` — `src/gpu_rent/provision.py` / `session.py`
- [ ] 4. [Bug] Perf tune: set `pip_ok` only if pip succeeded; do **not** write `--use-sage-attention` ExtraArgs unless install OK (or allow retry when `pip_ok` false) — `src/gpu_rent/remote/tune_swarm_perf.py` line 109
- [ ] 5. [Security] Write idle-killer creds JSON with `mode=0o600` before `mv` (same as GIT_TOKEN) — `src/gpu_rent/idle_killer.py` line 156
- [x] 1. [Bug] `arm_idle_killer`: on cred create failure **raise** or return False; never set `notes.idle_killer=armed`; surface same ⚠ as `"failed"` — `src/gpu_rent/idle_killer.py` line 151, `src/gpu_rent/provision.py` line 516
- [x] 2. [Bug] Idle-killer: if Swarm unreachable longer than N minutes (e.g. 2× idle or fixed 60m), treat as idle/allow delete (llm-only already bypasses) — `src/gpu_rent/remote/idle_killer.py` line 75
- [x] 3. [Logic] On mid-`up` failure after server create, arm killer anyway or fail loudly and refuse to leave session without killer / document mandatory `stop` — `src/gpu_rent/provision.py` / `session.py`
- [x] 4. [Bug] Perf tune: set `pip_ok` only if pip succeeded; do **not** write `--use-sage-attention` ExtraArgs unless install OK (or allow retry when `pip_ok` false) — `src/gpu_rent/remote/tune_swarm_perf.py` line 109
- [x] 5. [Security] Write idle-killer creds JSON with `mode=0o600` before `mv` (same as GIT_TOKEN) — `src/gpu_rent/idle_killer.py` line 156
**Verified OK (critical):** `cmd_stop` delete path, tunnel Ctrl+C detach, local-watchdog stop on stale lease, `requires: ollama` filter, balance notify (toast only), Ollama unit bind to data dir, 127 tests green at review time.