fix get_avail_memory for cgroups v1/v2 (AB#24354)

.github/workflows/linting.yml

@@ -15,6 +15,6 @@ jobs:
         run: rpmlint Lmod-config.spec
 
       - name: Run luacheck
-        uses: lunarmodules/luacheck@v0
+        uses: lunarmodules/luacheck@v1
         with:
           args: -g .

Lmod-config.spec

@@ -1,6 +1,6 @@
 Summary: Sitepackage and other config files for Lmod
 Name: Lmod-config
-Version: 1.9
+Version: 1.10
 Release: 1
 License: GPL
 Group: Applications/System
@@ -40,6 +40,8 @@ exit 0
 %{_libexecdir}/lmod/run_lmod_cache.py
 
 %changelog
+* Fri Jan 17 2025 Alex Domingo <[email protected]>
+- Fix get_avail_memory for cgroups v1/v2
 * Mon Jan 06 2025 Cintia Willemyns <[email protected]>
 - Hide modules older than module_age 6, instead of 5
 * Wed Sep 25 2024 Samuel Moors <[email protected]>

SitePackage.lua

@@ -257,13 +257,29 @@ end
 local function get_avail_memory()
     -- If a limit is set, return the maximum allowed memory, else nil
 
-    -- look for the memory cgroup (if any)
-    local cgroup = nil
+    -- look for the memory cgroup (if any):
+    -- for cgroupv2: find the hierarchegy and the memory controller
+    -- for cgroupv1: look for the memory controller
+    local lines = {}
     for line in io.lines("/proc/self/cgroup") do
-        cgroup = line:match("^[0-9]+:memory:(.*)$")
-        if cgroup then
-            break
+        table.insert(lines, line)
+    end
+
+    local cgroup = nil
+    local memory_filepath
+    -- if it's one line: cgroupv2
+    if #lines == 1 then
+        cgroup = lines[1]:match("^[0-9]+::(.*)$")
+        memory_filepath = "/sys/fs/cgroup/_CGROUP_SET_/memory.max"
+
+    else
+        for _, line in ipairs(lines) do
+            cgroup = line:match("^[0-9]+:memory:(.*)$")
+            if cgroup then
+                break
+            end
         end
+        memory_filepath = "/sys/fs/cgroup/memory/_CGROUP_SET_/memory.memsw.limit_in_bytes"
     end
 
     if not cgroup then
@@ -273,13 +289,12 @@ local function get_avail_memory()
     -- Slurm tasks are only limited by the job step that launched it
     cgroup = cgroup:gsub("/task_[%d]+$", "")
 
-    -- read the current maximum allowed memory usage (memory + swap)
-    local memory_file = io.open("/sys/fs/cgroup/memory/" .. cgroup .. "/memory.memsw.limit_in_bytes")
-
+    -- read the current maximum allowed memory usage (memory)
+    memory_filepath = memory_filepath:gsub("_CGROUP_SET_", cgroup)
+    local memory_file = io.open(memory_filepath)
     if not memory_file then
         return nil
     end
-
     local memory_value = tonumber(memory_file:read())
     memory_file:close()