fix: remove IO PSI row — false CRITICAL on GB10 with VLLM idle (v0.2.3)

Author: parallelArchitect

CHANGELOG.md

@@ -1,3 +1,9 @@
+## v0.2.3 — 2026-04-23
+
+### Fixed
+- Removed IO PSI row — false CRITICAL on GB10 with VLLM idle
+- IO PSI references removed from anomaly logger and pressure detection
+
 ## v0.2.2 — 2026-04-19
 
 ### Added

README.md

@@ -108,6 +108,8 @@ Detection is load-gated — evaluation only occurs when GPU utilization confirms
 
 ## GB10 Power Rails (spark_hwmon)
 
+v0.2.3 removes the IO PSI row (false CRITICAL on GB10 with VLLM idle).
+
 v0.2.2 adds a PWR row for GB10 systems using the spark_hwmon kernel module (https://github.com/antheas/spark_hwmon).
 
 Install:

main.py

@@ -90,20 +90,17 @@ def build(term_height: int = 40) -> Table:
         t.append(f"  {temp}  {pw}  Mem {gi(g['mem_used'])}/{gi(g['mem_total'])}")
         if g["is_uma"]:
             mem_psi = psi.get("mem", psi)
-            io_psi = psi.get("io", {})
             pwr = power_rails.read() if power_rails.is_available() else None
             # red — critical: hardware brake, throttle, high pressure, high temp
             uma_red = (
                 mem_psi.get("level") in ("HIGH", "CRITICAL")
-                or io_psi.get("level") in ("HIGH", "CRITICAL")
                 or any(th.get("status") == "THROTTLED" for th in throttle)
                 or any(g.get("temperature", 0) > 80 for g in gpus)
                 or (pwr and pwr.prochot)
             )
             # yellow — warning: approaching limits
             uma_yellow = (
                 mem_psi.get("level") == "MOD"
-                or io_psi.get("level") == "MOD"
                 or any(th.get("status") == "LOCKED" for th in throttle)
                 or any(60 <= g.get("temperature", 0) <= 80 for g in gpus)
                 or (pwr and pwr.cap_exceeded)
@@ -171,7 +168,6 @@ def build(term_height: int = 40) -> Table:
 
     # ── PSI ──────────────────────────────────────────
     mem_psi = psi.get("mem", {})
-    io_psi = psi.get("io", {})
     if mem_psi.get("available"):
         level = mem_psi["level"]
         color = psi_color(level)
@@ -183,18 +179,7 @@ def build(term_height: int = 40) -> Table:
         t.append(f"{level:8s}", style=color + " bold")
         t.append(f"  some {mem_psi['some_avg10']:.2f}  full {mem_psi['full_avg10']:.2f}")
         grid.add_row(t)
-    sep(grid)
-    if io_psi.get("available"):
-        level = io_psi["level"]
-        color = psi_color(level)
-        score = min(io_psi["some_avg10"] * 100 / 0.30, 100)
-        t = Text()
-        t.append(" IO    ", style="bold cyan")
-        t.append(f"{bar(score)} ", style=color)
-        t.append(f"{level:8s}", style=color + " bold")
-        t.append(f"  some {io_psi['some_avg10']:.2f}  full {io_psi['full_avg10']:.2f}")
-        grid.add_row(t)
-    if mem_psi.get("available") or io_psi.get("available"):
+    if mem_psi.get("available"):
         sep(grid)
 
     # ── TEMP ────────────────────────────────────────
@@ -303,10 +288,10 @@ def fmt_rate(b: float) -> str:
 
     footer = Text()
     if is_logging():
-        footer.append("  [dim]Ctrl+C to quit  sparkview v0.2.2[/dim]")
+        footer.append("  [dim]Ctrl+C to quit  sparkview v0.2.3[/dim]")
         footer.append("  ● LOGGING", style="bold red")
     else:
-        footer.append("  [dim]Ctrl+C to quit  sparkview v0.2.2[/dim]")
+        footer.append("  [dim]Ctrl+C to quit  sparkview v0.2.3[/dim]")
     grid.add_row(footer)
     return grid
 

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "sparkview"
-version = "0.2.0"
+version = "0.2.3"
 description = "GB10 Grace Blackwell unified memory GPU monitor"
 requires-python = ">=3.10"
 dependencies = [

sparkview/layers/logger.py

@@ -33,11 +33,8 @@ def _detect_trigger(psi: dict, throttle: list, mem: dict, gpus: list, cpu: dict
     if pwr and pwr.prochot:
         return "PROCHOT"
     mem_psi = psi.get("mem", psi)
-    io_psi = psi.get("io", {})
     if mem_psi.get("level") in ("MOD", "HIGH", "CRITICAL"):
         return f"PSI_{mem_psi.get('level')}"
-    if io_psi.get("level") in ("MOD", "HIGH", "CRITICAL"):
-        return f"IO_PSI_{io_psi.get('level')}"
     for th in throttle:
         if th.get("status") in ("THROTTLED", "LOCKED"):
             return f"CLOCK_{th.get('status')}"
@@ -56,11 +53,8 @@ def should_log(psi: dict, throttle: list, mem: dict, gpus: list, cpu: dict | Non
     if pwr and pwr.prochot:
         return True
     mem_psi = psi.get("mem", psi)
-    io_psi = psi.get("io", {})
     if mem_psi.get("level") in ("MOD", "HIGH", "CRITICAL"):
         return True
-    if io_psi.get("level") in ("MOD", "HIGH", "CRITICAL"):
-        return True
     for th in throttle:
         if th.get("status") in ("THROTTLED", "LOCKED"):
             return True
@@ -152,16 +146,10 @@ def write_log(
         _log_file.write(f"CLOCK: {status}  {clk:.0f}MHz / {clk_max:.0f}MHz  {pstate}\n")
 
     mem_psi = psi.get("mem", psi)
-    io_psi = psi.get("io", {})
     level = mem_psi.get("level", "?")
     some = mem_psi.get("some_avg10", 0)
     full = mem_psi.get("full_avg10", 0)
     _log_file.write(f"PSI:   {level}  some {some:.2f}  full {full:.2f}\n")
-    if io_psi.get("available"):
-        io_level = io_psi.get("level", "?")
-        io_some = io_psi.get("some_avg10", 0)
-        io_full = io_psi.get("full_avg10", 0)
-        _log_file.write(f"IO:    {io_level}  some {io_some:.2f}  full {io_full:.2f}\n")
 
     pwr = power_rails.read()
     if pwr is not None:

sparkview/layers/pressure.py

@@ -3,7 +3,6 @@
 import pathlib
 
 PSI_MEM_PATH = pathlib.Path("/proc/pressure/memory")
-PSI_IO_PATH = pathlib.Path("/proc/pressure/io")
 
 
 def _parse_psi(path: pathlib.Path) -> dict:
@@ -39,5 +38,4 @@ def _parse_psi(path: pathlib.Path) -> dict:
 
 def get_pressure() -> dict:
     mem = _parse_psi(PSI_MEM_PATH)
-    io = _parse_psi(PSI_IO_PATH)
-    return {"mem": mem, "io": io}
+    return {"mem": mem}