Edit

[EDIT] FILE: analyzers.py

# coding=utf-8
#
# Copyright © Cloud Linux GmbH & Cloud Linux Software, Inc 2010-2019 All Rights Reserved
#
# Licensed under CLOUD LINUX LICENSE AGREEMENT
# http://cloudlinux.com/docs/LICENSE.TXT

import logging

from lvestat import LVEStat
from lvestats.core.plugin import LveStatsPlugin

KB = 1024
FAULTS = ("cpu_fault", "memphy_fault", "mem_fault", "io_fault", "iops_fault", "mep_fault", "nproc_fault")
LIMITS = ("lcpu", "lmemphy", "lmem", "io", "liops", "lep", "lnproc")
STATS = ("cpu_usage", "memphy", "mem_usage", "io_usage", "iops", "mep", "nproc")
LVEUSAGESLOTS = ("lve_version", "time", "has_changed_limits", "has_changed_nproc") + STATS + LIMITS + FAULTS

__author__ = "iseletsk"

log = logging.getLogger("LVEUsage")

class LVEUsageAnalyzer(LveStatsPlugin):
    """
    Calculates lve usage and updates it in lve_data, in field 'lve_usage'
    """

def set_config(self, config):
        self.config = config  # pylint: disable=attribute-defined-outside-init

def execute(self, lve_data):
        """
        :param dict lve_data:
        """
        # from /proc/lve
        new_stats = lve_data["stats"]
        old_stats = lve_data["old_stats"]

del lve_data["old_stats"]

old_now = lve_data.get("old_now", None)
        lve_data["old_now"] = self.now

total_hz = lve_data["totalHz"]
        procs = lve_data["procs"]

result = {}

for _id, lve_stat in new_stats.items():
            u = LVEUsage(lve_version=lve_data["LVE_VERSION"])
            lve_stat_old = old_stats.get(_id, None)
            u.calc(lve_stat_old, lve_stat, total_hz, self.now - old_now if old_now else None, procs)
            # LVES-210: clear all data if EP and NPROC == 0 and all other data not changed
            if (
                _id == 0
                or lve_stat.nproc > 0
                or lve_stat.mep > 0
                or (lve_stat_old is not None and u.is_lve_data_changed(lve_stat_old, lve_stat))
            ):
                result[_id] = u
        lve_usages = lve_data.get("lve_usages", [])
        lve_usages.append(result)
        lve_data["lve_usages_5s"] = [result]
        lve_data["lve_active_ids"] = list(result.keys())
        lve_data["lve_usages"] = lve_usages

class LVEUsage(object):
    __slots__ = LVEUSAGESLOTS

def __init__(self, lve_version):
        self.lve_version = lve_version
        self.time = None
        self.has_changed_limits = False
        self.has_changed_nproc = False

# if LVE_VERSION >= 4
        self.cpu_usage = 0
        self.cpu_fault = 0
        self.mep = 0
        self.mem_fault = 0
        self.mep_fault = 0
        self.mem_usage = 0
        self.lep = 0
        self.lcpu = 0
        self.lmem = 0

# if LVE_VERSION >=6
        self.io = 0
        self.io_usage = 0
        self.io_fault = 0
        self.lmemphy = 0
        self.memphy = 0
        self.memphy_fault = 0
        self.lnproc = 0
        self.nproc = 0
        self.nproc_fault = 0

# if LVE_VERSION >= 8
        self.liops = 0
        self.iops = 0
        self.iops_fault = 0

def init_limits_from_lvestat(self, lvestat: LVEStat, procs):
        """
        Prepares LVEUsage obj taking limits from LVEStat
        """
        self.lep = lvestat.lep
        self.lcpu = self.convert_lcpu_to_new_kernel_format(lvestat.cpu, self.lve_version, procs)
        # CMT-509: the incorrect LVE I/O units on CMT dashboard
        # Proc saves this value is in KBps,
        # but we should send it is in Bps,
        # because for an active user this value is sent in Bps
        self.io = lvestat.io * KB
        self.lmemphy = lvestat.lmemphy
        self.lnproc = lvestat.lnproc

if self.lve_version >= 8:
            self.liops = lvestat.liops
        return self

def __repr__(self):
        representation = "<"
        for k in (k for k in dir(self) if not k.startswith("_")):
            v = getattr(self, k)
            if isinstance(v, (str, int, float, bool, type(None))):
                representation += f"{k}:{v}, "
        representation += ">"
        return representation

def has_interesting_values(self):
        if any((self.cpu_usage, self.nproc, self.io_usage, self.iops)):
            return True
        if any((getattr(self, k) for k in FAULTS)):
            return True
        return False

def _has_changed_limits(self, new, old):
        """
        :type old: lvestat.LVEStat
        :type new: lvestat.LVEStat
        """
        for attr in LIMITS:
            if getattr(new, attr, None) != getattr(old, attr, None):
                return True
        return False

def _has_changed_nproc(self, new, old):
        """
        :type old: lvestat.LVEStat
        :type new: lvestat.LVEStat
        """
        return old.nproc != new.nproc

def convert_lcpu_to_new_kernel_format(self, lcpu, lve_ver, procs):
        """
        New kernel format lcpu is 100 times bigger than previous. So that it's 0-10000
        """
        if lve_ver >= 8:
            return lcpu
        else:
            # new_kernel_format = (old_format * 10000 * procs / 100) = old_format * 100 * procs
            return lcpu * procs * 100

def is_lve_data_changed(self, old, new):
        """
        Check if LVE data is changed
        :param old: Previous data
        :param new: Current data
        :return: True - changes, false - not
        """
        # Common data
        if (
            old.cpu_usage != new.cpu_usage
            or old.mep != new.mep
            or old.mem_fault != new.mem_fault
            or old.mep_fault != new.mep_fault
            or old.mem_usage != new.mem_usage
            or old.lep != new.lep
            or old.io != new.io
            or old.io_usage != new.io_usage
        ):
            return True

# LVE6+ data
        if (
            old.memphy != new.memphy
            or old.memphy_fault != new.memphy_fault
            or old.nproc != new.nproc
            or old.nproc_fault != new.nproc_fault
        ):
            return True

# LVE8+ data
        if self.lve_version >= 8 and old.iops != new.iops:
            return True

return False

def calc(self, old, new, hz, time, procs):
        self.time = time

# LCPU 100% now is 100*100 = 10 000
        # LCPU 0% is 0
        self.lcpu = self.convert_lcpu_to_new_kernel_format(new.cpu, self.lve_version, procs)

if bool(old):
            # Previous data present
            self.has_changed_limits = self._has_changed_limits(new, old)
            self.has_changed_nproc = self._has_changed_nproc(new, old)

# CPU USAGE is 0-10000*procs
            # new_kernel_format = (old_format * 10000 * procs / 100) = old_format * 100 * procs;
            # (100 * (new.cpu_usage - self.cpu_usage) / (hz * time)) = old_format
            assert time is not None, "oldstats is not None, but time is unknown"
            assert time != 0, "oldstats is not None, but time is 0"
            if new.cpu_usage > old.cpu_usage:
                self.cpu_usage = int(round(10000 * (new.cpu_usage - old.cpu_usage) * procs / (hz * time)))

if new.mem_fault > old.mem_fault:
                self.mem_fault = new.mem_fault - old.mem_fault
            if new.mep_fault > old.mep_fault:
                self.mep_fault = new.mep_fault - old.mep_fault
            if new.memphy_fault > old.memphy_fault:
                self.memphy_fault = new.memphy_fault - old.memphy_fault
            if new.nproc_fault > old.nproc_fault:
                self.nproc_fault = new.nproc_fault - old.nproc_fault
            if new.io_usage > old.io_usage:
                # bytes per second
                self.io_usage = int(round(KB * (new.io_usage - old.io_usage) / time))

if self.lve_version >= 8:
                if new.iops > old.iops:
                    self.iops = int(round((new.iops - old.iops) / time))

# Fault-counter convention (applies to every *_fault assignment below):
        # each is a per-interval boolean — 1 if usage was >= cap during this interval, otherwise 0.
        # `LVEUsage` is instantiated fresh per snapshot pair and `calc()` runs exactly once on it,
        # so each fault field is written at most once here.
        # Aggregators (AggregatedLveUsage.add) sum these per-interval 1s
        # across all sub-intervals in the reporting window,
        # yielding "number of intervals at-or-above the cap".
        if self.cpu_usage < 0:
            self.cpu_usage = 0
        elif self.cpu_usage >= self.lcpu > 0:
            self.cpu_usage = self.lcpu
            self.cpu_fault = 1

# CLOS-4011: synthesise fault=1 when usage is at-or-above the limit but the
        # /proc delta produced no fault. The kernel does not propagate end-user fault
        # events into a reseller LVP entry's mem/mep/memphy/nproc fault counters, so
        # without this fallback those fields always read 0 in reseller statistics
        # even when end-users are actively hitting their limits.
        #
        # `>=` (not `>`) matches the cpu_fault/io_fault semantics above and covers
        # the at-limit boundary: under normal kernel enforcement reseller memphy/nproc
        # are pinned at-or-below the cap.
        # The `not self.X_fault` guard prevents double-counting on regular-user LVEs
        # whose /proc-delta path already set X_fault > 0 earlier in this method.

self.lmem = new.lmem
        if self.mem_fault > 0 or new.mem_usage >= new.lmem > 0:
            self.mem_usage = new.lmem
            if not self.mem_fault and new.mem_usage >= new.lmem > 0:
                self.mem_fault = 1
        else:
            self.mem_fault = 0
            self.mem_usage = new.mem_usage

self.lep = new.lep
        if self.mep_fault > 0 or new.mep >= new.lep > 0:
            self.mep = new.lep
            if not self.mep_fault and new.mep >= new.lep > 0:
                self.mep_fault = 1
        else:
            self.mep_fault = 0
            self.mep = new.mep

if self.lve_version >= 8:
            self.liops = new.liops
            if self.iops >= self.liops > 0:
                self.iops = self.liops
                self.iops_fault = 1

self.io = KB * new.io  # limit in bytes per second
        if self.io_usage < 0:
            self.io_usage = 0
        elif self.io_usage >= self.io > 0:
            self.io_usage = self.io
            self.io_fault = 1

self.lmemphy = new.lmemphy
        self.lnproc = new.lnproc

# CLOS-4011: synthesise memphy_fault for reseller LVPs.
        # This is the field surfaced as PMemF in `lveinfo --reseller-name` and was the
        # primary symptom in the customer report (PMemF stayed at 0 on the reseller view
        # even when end-users were OOM-killed by the reseller PMEM cap).
        if self.memphy_fault > 0 or new.memphy >= new.lmemphy > 0:
            self.memphy = new.lmemphy
            if not self.memphy_fault and new.memphy >= new.lmemphy > 0:
                self.memphy_fault = 1
        else:
            self.memphy_fault = 0
            self.memphy = new.memphy

if self.nproc_fault > 0 or new.nproc >= new.lnproc > 0:
            self.nproc = new.lnproc
            if not self.nproc_fault and new.nproc >= new.lnproc > 0:
                self.nproc_fault = 1
        else:
            self.nproc_fault = 0
            self.nproc = new.nproc

def __str__(self):
        return "(cpu_usage=" + str(self.cpu_usage) + ", mep=" + str(self.mep) + ")"

def __getitem__(self, key):
        return getattr(self, key)

CANCEL