topo: handle offline cpus

Some CPUs can become offline for various reasons. Do not crash when
generating a DOT diagram from a sosreport with offline CPUs. Adjust
topology and interrupt counters parsing to take offline CPUs into
account.

Display offline CPUs in gray in the diagram.

Reported-by: Abhiram R N <[email protected]>
Signed-off-by: Robin Jarry <[email protected]>

sosviz/collect/irq.py

@@ -18,12 +18,20 @@ def parse_report(path: pathlib.Path, data: D):
     if not f.is_file():
         return
 
+    cpu_ids = []
+    for cpu in path.glob("sys/devices/system/cpu/cpu[0-9]*"):
+        cpu_ids.append(int(re.match(r"cpu(\d+)", cpu.name).group(1)))
+    counters_len = max(*cpu_ids) + 1
+
     for match in INTERRUPT_RE.finditer(f.read_text()):
-        irq, counters, desc = match.groups()
+        irq = match.group(1)
+        counters = [0] * counters_len
+        for i, c in enumerate(match.group(2).split()):
+            counters[cpu_ids[i]] = int(c)
         irqs[irq] = D(
             irq=irq,
-            desc=re.sub(r"\s+", " ", desc.strip()),
-            counters=[int(c) for c in counters.split()],
+            desc=re.sub(r"\s+", " ", match.group(3).strip()),
+            counters=counters,
         )
         try:
             irqs[irq].requested_affinity = parse_cpu_set(

sosviz/collect/topo.py

@@ -9,7 +9,7 @@
 
 
 def parse_report(path: pathlib.Path, data: D):
-    nodes = list(path.glob("sys/devices/system/node/node*"))
+    nodes = list(path.glob("sys/devices/system/node/node[0-9]*"))
     for node in nodes:
         match = re.match(r"^node(\d+)$", node.name)
         if not match:
@@ -38,25 +38,26 @@ def parse_report(path: pathlib.Path, data: D):
                 size = int(match.group(1)) * 1024
                 numa.setdefault("hugepages", D())[size] = int(huge.read_text())
 
-        for cpu in path.glob("sys/devices/system/cpu/cpu*/topology"):
+        offline_cpus = set()
+        for cpu in path.glob("sys/devices/system/cpu/cpu[0-9]*"):
+            if not (cpu / f"node{numa_id}").is_dir():
+                continue
+            cpu_id = int(re.match(r"cpu(\d+)", cpu.name).group(1))
+            online = cpu / "online"
+            if online.is_file() and online.read_text().strip() == "0":
+                offline_cpus.add(cpu_id)
+                continue
+            topo = cpu / "topology"
             try:
-                package_id = int((cpu / "physical_package_id").read_text())
-                if package_id != numa_id:
-                    continue
-            except FileNotFoundError:
-                cpu_id = int(re.match(r"cpu(\d+)", cpu.parent.name).group(1))
-                if cpu_id not in cpus:
-                    continue
-            try:
-                threads = parse_cpu_set((cpu / "thread_siblings_list").read_text())
+                threads = parse_cpu_set((topo / "thread_siblings_list").read_text())
             except FileNotFoundError:
                 try:
-                    threads = parse_cpu_set((cpu / "core_cpus_list").read_text())
+                    threads = parse_cpu_set((topo / "core_cpus_list").read_text())
                 except FileNotFoundError:
                     # hyperthreading disabled
                     continue
-            cpus.update(threads)
             for t in threads:
                 siblings[t] = threads - {t}
         numa.cpus = cpus
+        numa.offline_cpus = offline_cpus
         numa.thread_siblings = siblings

sosviz/output/dot.py

@@ -659,6 +659,16 @@ def phy_numa(self, numa: D):
                 tooltip=self.irq_counters_tooltip(housekeeping_cpus),
                 color="blue",
             )
+            if numa.offline_cpus:
+                self.node(
+                    f"phy_cpus_offline_{numa.id}",
+                    [
+                        '<font color="gray"><b>Offline</b></font>',
+                        f'<font color="gray">CPUs {bit_list(numa.offline_cpus)}</font>',
+                    ],
+                    tooltip=self.irq_counters_tooltip(numa.offline_cpus),
+                    color="gray",
+                )
 
         labels = [f"<b>memory {human_readable(numa.total_memory, 1024)}</b>"]
         for size, num in numa.get("hugepages", {}).items():