52abc18f1c
Take another VM stop reason to account, add aditional setting for cpu QOS and add a job to sync VM pids to resctrl. Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
97 lines
3.9 KiB
Diff
97 lines
3.9 KiB
Diff
From bdd805eec082062e042acda6caf38ca17dbaec50 Mon Sep 17 00:00:00 2001
|
|
From: Keqian Zhu <zhukeqian1@huawei.com>
|
|
Date: Thu, 4 Aug 2022 14:46:33 -0400
|
|
Subject: [PATCH 1/3] guestinfo: Take another VM stop reason to account
|
|
|
|
When VM is closed by OpenStack, the exception code is not
|
|
VIR_ERR_NO_DOMAIN but exception message contains "domain
|
|
is not running".
|
|
|
|
And refactor code to make it more readable.
|
|
|
|
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
|
|
---
|
|
data_collector/guestinfo.py | 48 +++++++++++++++++--------------------
|
|
1 file changed, 22 insertions(+), 26 deletions(-)
|
|
|
|
diff --git a/data_collector/guestinfo.py b/data_collector/guestinfo.py
|
|
index 415b3f6..38fa827 100644
|
|
--- a/data_collector/guestinfo.py
|
|
+++ b/data_collector/guestinfo.py
|
|
@@ -28,6 +28,7 @@ DEFAULT_PRIORITY = "machine"
|
|
HIGH_PRIORITY = "high_prio_machine"
|
|
LOW_PRIORITY = "low_prio_machine"
|
|
PIDS_CGRP_PATH = "/sys/fs/cgroup/pids"
|
|
+DOMAIN_STOP_MSG = "domain is not running"
|
|
|
|
|
|
class DomainInfo:
|
|
@@ -141,43 +142,38 @@ class GuestInfo:
|
|
self.running_domain_in_cpus.append([])
|
|
|
|
self.get_all_active_domain(conn)
|
|
-
|
|
for dom in self.domain_online:
|
|
self.vm_online_dict[dom.ID()] = dom
|
|
+ # Remove ever see but now stopped domains
|
|
+ for vm_id in list(self.vm_dict):
|
|
+ if vm_id not in self.vm_online_dict:
|
|
+ del self.vm_dict[vm_id]
|
|
+
|
|
for vm_id in self.vm_online_dict:
|
|
- ret = -1
|
|
- if vm_id in self.vm_dict:
|
|
- try:
|
|
+ try:
|
|
+ if vm_id in self.vm_dict:
|
|
ret = self.vm_dict.get(vm_id).update_domain_info(self.vm_online_dict.get(vm_id), host_topo)
|
|
- except libvirt.libvirtError as e:
|
|
- if e.get_error_code() != libvirt.VIR_ERR_NO_DOMAIN:
|
|
- raise
|
|
- if ret < 0:
|
|
- del self.vm_dict[vm_id]
|
|
- continue
|
|
- else:
|
|
- try:
|
|
- vm_info = DomainInfo()
|
|
- ret = vm_info.set_domain_attribute(self.vm_online_dict.get(vm_id), host_topo)
|
|
- except libvirt.libvirtError as e:
|
|
- if e.get_error_code() != libvirt.VIR_ERR_NO_DOMAIN:
|
|
- raise
|
|
- if ret < 0:
|
|
- continue
|
|
- self.vm_dict[vm_id] = vm_info
|
|
+ else:
|
|
+ self.vm_dict[vm_id] = DomainInfo()
|
|
+ ret = self.vm_dict.get(vm_id).set_domain_attribute(self.vm_online_dict.get(vm_id), host_topo)
|
|
+ except libvirt.libvirtError as e:
|
|
+ ret = -1
|
|
+ # If domain doesn't stop, raise exception
|
|
+ if e.get_error_code() != libvirt.VIR_ERR_NO_DOMAIN and \
|
|
+ DOMAIN_STOP_MSG not in e.get_error_message():
|
|
+ raise
|
|
+ if ret < 0:
|
|
+ del self.vm_dict[vm_id]
|
|
+ continue
|
|
|
|
+ if self.vm_dict.get(vm_id).priority == 1:
|
|
+ self.low_prio_vm_dict[vm_id] = self.vm_dict.get(vm_id)
|
|
for cpu in range(host_topo.max_cpu_nums):
|
|
self.running_domain_in_cpus[cpu].append((self.vm_dict.get(vm_id).cpu_usage[cpu],
|
|
self.vm_dict.get(vm_id).domain_id,
|
|
self.vm_dict.get(vm_id).domain_name,
|
|
self.vm_dict.get(vm_id).priority))
|
|
|
|
- for vm_id in list(self.vm_dict):
|
|
- if vm_id not in self.vm_online_dict:
|
|
- del self.vm_dict[vm_id]
|
|
- elif vm_id not in self.low_prio_vm_dict and self.vm_dict.get(vm_id).priority == 1:
|
|
- self.low_prio_vm_dict[vm_id] = self.vm_dict.get(vm_id)
|
|
-
|
|
def get_all_active_domain(self, conn):
|
|
try:
|
|
self.domain_online = conn.listAllDomains(flags=libvirt.VIR_CONNECT_LIST_DOMAINS_ACTIVE)
|
|
--
|
|
2.33.0
|
|
|