Merge pull request galaxyproject#17169 from natefoo/cgroupsv2

[23.2] Add support for Cgroupsv2

lib/galaxy/job_metrics/instrumenters/cgroup.py

@@ -1,4 +1,5 @@
 """The module describes the ``cgroup`` job metrics plugin."""
+import decimal
 import logging
 import numbers
 from collections import namedtuple
@@ -17,7 +18,27 @@
 
 log = logging.getLogger(__name__)
 
+VALID_VERSIONS = ("auto", "1", "2")
+DEFAULT_PARAMS = (
+    # cgroupsv1 - this is probably more params than are useful to collect, but don't remove any for legacy reasons
+    "memory.memsw.max_usage_in_bytes",
+    "memory.max_usage_in_bytes",
+    "memory.limit_in_bytes",
+    "memory.memsw.limit_in_bytes",
+    "memory.soft_limit_in_bytes",
+    "memory.failcnt",
+    "memory.oom_control.oom_kill_disable",
+    "memory.oom_control.under_oom",
+    "cpuacct.usage",
+    # cgroupsv2
+    "memory.events.oom_kill",
+    "memory.peak",
+    "cpu.stat.system_usec",
+    "cpu.stat.usage_usec",
+    "cpu.stat.user_usec",
+)
 TITLES = {
+    # cgroupsv1
     "memory.memsw.max_usage_in_bytes": "Max memory usage (MEM+SWP)",
     "memory.max_usage_in_bytes": "Max memory usage (MEM)",
     "memory.limit_in_bytes": "Memory limit on cgroup (MEM)",
@@ -27,14 +48,35 @@
     "memory.oom_control.oom_kill_disable": "OOM Control enabled",
     "memory.oom_control.under_oom": "Was OOM Killer active?",
     "cpuacct.usage": "CPU Time",
+    # cgroupsv2
+    "memory.events.low": "Number of times the cgroup was reclaimed due to high memory pressure even though its usage is under the low "
+    "boundary",
+    "memory.events.high": "Number of times processes of the cgroup were throttled and routed to perform direct memory reclaim because "
+    "the high memory boundary was exceeded",
+    "memory.events.max": "Number of times the cgroup's memory usage was about to go over the max boundary",
+    "memory.events.oom": "Number of time the cgroup's memory usage reached the limit and allocation was about to fail",
+    "memory.events.oom_kill": "Number of processes belonging to this cgroup killed by any kind of OOM killer",
+    "memory.events.oom_group_kill": "Number of times a group OOM has occurred",
+    "memory.high": "Memory usage throttle limit",
+    "memory.low": "Best-effort memory protection",
+    "memory.max": "Memory usage hard limit",
+    "memory.min": "Hard memory protection",
+    "memory.peak": "Max memory usage recorded",
+    "cpu.stat.system_usec": "CPU system time",
+    "cpu.stat.usage_usec": "CPU usage time",
+    "cpu.stat.user_usec": "CPU user time",
 }
 CONVERSION = {
     "memory.oom_control.oom_kill_disable": lambda x: "No" if x == 1 else "Yes",
     "memory.oom_control.under_oom": lambda x: "Yes" if x == 1 else "No",
+    "memory.peak": lambda x: nice_size(x),
     "cpuacct.usage": lambda x: formatting.seconds_to_str(x / 10**9),  # convert nanoseconds
+    "cpu.stat.system_usec": lambda x: formatting.seconds_to_str(x / 10**6),  # convert microseconds
+    "cpu.stat.usage_usec": lambda x: formatting.seconds_to_str(x / 10**6),  # convert microseconds
+    "cpu.stat.user_usec": lambda x: formatting.seconds_to_str(x / 10**6),  # convert microseconds
 }
-CPU_USAGE_TEMPLATE = r"""
-if [ -e "/proc/$$/cgroup" -a -d "{cgroup_mount}" ]; then
+CGROUPSV1_TEMPLATE = r"""
+if [ -e "/proc/$$/cgroup" -a -d "{cgroup_mount}" -a ! -f "{cgroup_mount}/cgroup.controllers" ]; then
     cgroup_path=$(cat "/proc/$$/cgroup" | awk -F':' '($2=="cpuacct,cpu") || ($2=="cpu,cpuacct") {{print $3}}');
     if [ ! -e "{cgroup_mount}/cpu$cgroup_path/cpuacct.usage" ]; then
         cgroup_path="";
@@ -44,12 +86,6 @@
             echo "__$(basename $f)__" >> {metrics}; cat "$f" >> {metrics} 2>/dev/null;
         fi;
     done;
-fi
-""".replace(
-    "\n", " "
-).strip()
-MEMORY_USAGE_TEMPLATE = """
-if [ -e "/proc/$$/cgroup" -a -d "{cgroup_mount}" ]; then
     cgroup_path=$(cat "/proc/$$/cgroup" | awk -F':' '$2=="memory"{{print $3}}');
     if [ ! -e "{cgroup_mount}/memory$cgroup_path/memory.max_usage_in_bytes" ]; then
         cgroup_path="";
@@ -61,6 +97,16 @@
 """.replace(
     "\n", " "
 ).strip()
+CGROUPSV2_TEMPLATE = r"""
+if [ -e "/proc/$$/cgroup" -a -f "{cgroup_mount}/cgroup.controllers" ]; then
+    cgroup_path=$(cat "/proc/$$/cgroup" | awk -F':' '($1=="0") {{print $3}}');
+    for f in {cgroup_mount}/${{cgroup_path}}/{{cpu,memory}}.*; do
+        echo "__$(basename $f)__" >> {metrics}; cat "$f" >> {metrics} 2>/dev/null;
+    done;
+fi
+""".replace(
+    "\n", " "
+).strip()
 
 
 Metric = namedtuple("Metric", ("key", "subkey", "value"))
@@ -76,7 +122,7 @@ def format(self, key, value):
                 return title, nice_size(value)
             except ValueError:
                 pass
-        elif isinstance(value, (numbers.Integral, numbers.Real)) and value == int(value):
+        elif isinstance(value, (decimal.Decimal, numbers.Integral, numbers.Real)) and value == int(value):
             value = int(value)
         return title, value
 
@@ -90,33 +136,36 @@ class CgroupPlugin(InstrumentPlugin):
     def __init__(self, **kwargs):
         self.verbose = asbool(kwargs.get("verbose", False))
         self.cgroup_mount = kwargs.get("cgroup_mount", "/sys/fs/cgroup")
+        self.version = str(kwargs.get("version", "auto"))
+        assert self.version in VALID_VERSIONS, f"cgroup metric version option must be one of {VALID_VERSIONS}"
         params_str = kwargs.get("params", None)
         if isinstance(params_str, list):
             params = params_str
         elif params_str:
             params = [v.strip() for v in params_str.split(",")]
         else:
-            params = list(TITLES.keys())
+            params = list(DEFAULT_PARAMS)
         self.params = params
 
     def post_execute_instrument(self, job_directory: str) -> List[str]:
         commands: List[str] = []
-        commands.append(self.__record_cgroup_cpu_usage(job_directory))
-        commands.append(self.__record_cgroup_memory_usage(job_directory))
+        if self.version in ("auto", "1"):
+            commands.append(self.__record_cgroup_v1_usage(job_directory))
+        if self.version in ("auto", "2"):
+            commands.append(self.__record_cgroup_v2_usage(job_directory))
         return commands
 
     def job_properties(self, job_id, job_directory: str) -> Dict[str, Any]:
         metrics = self.__read_metrics(self.__cgroup_metrics_file(job_directory))
         return metrics
 
-    def __record_cgroup_cpu_usage(self, job_directory: str) -> str:
-        # comounted cgroups (which cpu and cpuacct are on the supported Linux distros) can appear in any order (cpu,cpuacct or cpuacct,cpu)
-        return CPU_USAGE_TEMPLATE.format(
+    def __record_cgroup_v1_usage(self, job_directory: str) -> str:
+        return CGROUPSV1_TEMPLATE.format(
             metrics=self.__cgroup_metrics_file(job_directory), cgroup_mount=self.cgroup_mount
         )
 
-    def __record_cgroup_memory_usage(self, job_directory: str) -> str:
-        return MEMORY_USAGE_TEMPLATE.format(
+    def __record_cgroup_v2_usage(self, job_directory: str) -> str:
+        return CGROUPSV2_TEMPLATE.format(
             metrics=self.__cgroup_metrics_file(job_directory), cgroup_mount=self.cgroup_mount
         )
 

test/unit/job_metrics/test_cgroups.py

@@ -1,6 +1,6 @@
 from galaxy.job_metrics.instrumenters.cgroup import CgroupPlugin
 
-CGROUP_PRODUCTION_EXAMPLE_2201 = """__cpu.cfs_period_us__
+CGROUPV1_PRODUCTION_EXAMPLE_2201 = """__cpu.cfs_period_us__
 100000
 __cpu.cfs_quota_us__
 -1
@@ -105,18 +105,174 @@
 1
 """
 
+CGROUPV2_PRODUCTION_EXAMPLE_232 = """__cpu.idle__
+0
+__cpu.max__
+max 100000
+__cpu.max.burst__
+0
+__cpu.stat__
+usage_usec 8992210
+user_usec 6139150
+system_usec 2853059
+core_sched.force_idle_usec 0
+nr_periods 0
+nr_throttled 0
+throttled_usec 0
+nr_bursts 0
+burst_usec 0
+__cpu.weight__
+100
+__cpu.weight.nice__
+0
+__memory.current__
+139350016
+__memory.events__
+low 0
+high 0
+max 0
+oom 0
+oom_kill 0
+oom_group_kill 0
+__memory.events.local__
+low 0
+high 0
+max 0
+oom 0
+oom_kill 0
+oom_group_kill 0
+__memory.high__
+max
+__memory.low__
+0
+__memory.max__
+max
+__memory.min__
+0
+__memory.numa_stat__
+anon N0=864256
+file N0=129146880
+kernel_stack N0=32768
+pagetables N0=131072
+sec_pagetables N0=0
+shmem N0=0
+file_mapped N0=0
+file_dirty N0=0
+file_writeback N0=0
+swapcached N0=0
+anon_thp N0=0
+file_thp N0=0
+shmem_thp N0=0
+inactive_anon N0=819200
+active_anon N0=20480
+inactive_file N0=51507200
+active_file N0=77639680
+unevictable N0=0
+slab_reclaimable N0=8638552
+slab_unreclaimable N0=340136
+workingset_refault_anon N0=0
+workingset_refault_file N0=77
+workingset_activate_anon N0=0
+workingset_activate_file N0=0
+workingset_restore_anon N0=0
+workingset_restore_file N0=0
+workingset_nodereclaim N0=0
+__memory.oom.group__
+0
+__memory.peak__
+339906560
+__memory.reclaim__
+__memory.stat__
+anon 860160
+file 129146880
+kernel 9211904
+kernel_stack 32768
+pagetables 126976
+sec_pagetables 0
+percpu 0
+sock 0
+vmalloc 0
+shmem 0
+zswap 0
+zswapped 0
+file_mapped 0
+file_dirty 0
+file_writeback 0
+swapcached 0
+anon_thp 0
+file_thp 0
+shmem_thp 0
+inactive_anon 815104
+active_anon 20480
+inactive_file 51507200
+active_file 77639680
+unevictable 0
+slab_reclaimable 8642480
+slab_unreclaimable 340904
+slab 8983384
+workingset_refault_anon 0
+workingset_refault_file 77
+workingset_activate_anon 0
+workingset_activate_file 0
+workingset_restore_anon 0
+workingset_restore_file 0
+workingset_nodereclaim 0
+pgscan 0
+pgsteal 0
+pgscan_kswapd 0
+pgscan_direct 0
+pgsteal_kswapd 0
+pgsteal_direct 0
+pgfault 132306
+pgmajfault 524
+pgrefill 0
+pgactivate 18958
+pgdeactivate 0
+pglazyfree 0
+pglazyfreed 0
+zswpin 0
+zswpout 0
+thp_fault_alloc 19
+thp_collapse_alloc 0
+__memory.swap.current__
+0
+__memory.swap.events__
+high 0
+max 0
+fail 0
+__memory.swap.high__
+max
+__memory.swap.max__
+max
+__memory.zswap.current__
+0
+__memory.zswap.max__
+max
+"""
+
 
-def test_cgroup_collection(tmpdir):
+def test_cgroupv1_collection(tmpdir):
     plugin = CgroupPlugin()
     job_dir = tmpdir.mkdir("job")
-    job_dir.join("__instrument_cgroup__metrics").write(CGROUP_PRODUCTION_EXAMPLE_2201)
+    job_dir.join("__instrument_cgroup__metrics").write(CGROUPV1_PRODUCTION_EXAMPLE_2201)
     properties = plugin.job_properties(1, job_dir)
     assert "cpuacct.usage" in properties
     assert properties["cpuacct.usage"] == 7265342042
     assert "memory.limit_in_bytes" in properties
     assert properties["memory.limit_in_bytes"] == 9223372036854771712
 
 
+def test_cgroupv2_collection(tmpdir):
+    plugin = CgroupPlugin()
+    job_dir = tmpdir.mkdir("job")
+    job_dir.join("__instrument_cgroup__metrics").write(CGROUPV2_PRODUCTION_EXAMPLE_232)
+    properties = plugin.job_properties(1, job_dir)
+    assert "cpu.stat.usage_usec" in properties
+    assert properties["cpu.stat.usage_usec"] == 8992210
+    assert "memory.peak" in properties
+    assert properties["memory.peak"] == 339906560
+
+
 def test_instrumentation(tmpdir):
     # don't actually run the instrumentation but at least exercise the code the and make
     # sure templating includes cgroup_mount override.

test/unit/job_metrics/test_job_metrics.py

@@ -49,6 +49,20 @@ def test_job_metrics_format_cgroup():
         assert_title="Memory limit on cgroup (MEM)",
         assert_value="8.0 EB",
     )
+    _assert_format(
+        "cgroup",
+        "cpu.stat.usage_usec",
+        7982357892.000000,
+        assert_title="CPU usage time",
+        assert_value="2.0 hours and 13.0 minutes",
+    )
+    _assert_format(
+        "cgroup",
+        "memory.peak",
+        45097156608,
+        assert_title="Max memory usage recorded",
+        assert_value="42.0 GB",
+    )
 
 
 def test_job_metrics_uname():