Introduction
This document describes the procedure to analyze the issue related to high memory usage in Cisco Virtualized Infrastructure Manager (CVIM) compute nodes.
Prerequisites
Requirements
Cisco recommends you have knowledge of memory management and HugePages in Linux.
What are HugePages?
Enabling HugePages allows the operating system to support memory pages larger than the default (usually 4 KB). The use of very large page sizes can enhance system performance by reducing the system resources needed to access page table entries. Consequently, HugePages are typically employed to mitigate memory latency.
Problem Description
High memory usage alerts on CVIM compute nodes while CVIM has not triggered any alerts. Alerts related to memory utilization could be via a third-party monitoring tool or monitoring dashboard.
Analysis
It is observed that high memory utilization in the OS as per the free and sar command output in Linux.
[root@cvim-computex ~]# free -m
total used free shared buff/cache available
Mem: 385410 365882 7602 3621 11925 8411
Swap: 2047 0 2047
[root@cvim-computex ~]# sar -r
Linux 4.18.0-193.81.1.el8_2.x86_64 (pod1-compute4.mx2) 08/24/2023 _x86_64_ (112 CPU)
12:00:46 AM kbmemfree kbavail kbmemused %memused kbbuffers kbcached kbcommit %commit kbactive kbinact kbdirty
12:10:34 AM 7493576 7871200 387166528 98.10 4240 9334356 12893752 3.25 4891940 6325076 68
12:20:11 AM 7503208 7883396 387156896 98.10 4240 9337364 12872708 3.24 4885008 6328096 16
12:30:34 AM 7485648 7869540 387174456 98.10 4240 9340556 12902748 3.25 4892948 6331276 36
12:40:46 AM 7494396 7880940 387165708 98.10 4240 9343636 12866964 3.24 4886908 6334364 20
12:50:34 AM 7479616 7869772 387180488 98.10 4240 9346720 12905156 3.25 4892408 6337444 56
01:00:46 AM 7490304 7883016 387169800 98.10 4240 9349832 12860152 3.24 4885308 6340500 56
01:10:34 AM 7472248 7868672 387187856 98.11 4240 9352836 12896932 3.25 4892604 6343556 28
01:20:46 AM 7484308 7883276 387175796 98.10 4240 9355948 12867972 3.24 4885172 6346676 16
01:30:34 AM 7475092 7869596 387185012 98.11 4240 9350840 12904328 3.25 4892448 6341556 44
01:40:46 AM 7485436 7882508 387174668 98.10 4240 9353932 12864252 3.24 4885148 6344660 56
01:50:34 AM 7468840 7869520 387191264 98.11 4240 9357036 12907464 3.25 4893552 6347752 164
02:00:46 AM 7479076 7882428 387181028 98.10 4240 9360124 12861892 3.24 4886044 6350844 68
Use the ps command to identify the processes with the highest memory usage.
[root@cvim-computex ~]# ps -aux --sort -rss
USER PID %CPU %MEM VSZ RSS TTY STAT START TIME COMMAND
root 328199 1207 0.2 541893584 ? RLl Mar12 2948779:31 /usr/bin/vpp -c /etc/vpp/vpp.conf
root 1829 0.0 0.0 379024 227692 ? Ss Mar12 14:21 /usr/lib/systemd/systemd-journald
Verify the container memory usage by checking the statistics using the podman or docker commands.
[root@cvim-computex ~]# podman stats
ID NAME CPU % MEM USAGE / LIMIT MEM % NET IO BLOCK IO PIDS
2f8fdc4b63a4 fluentd_31902 -- 301.2MB / 404.1GB 0.07% -- / -- 9.265MB / 89.68GB 75
34d806a30733 novalibvirt_31902 -- 42.16MB / 404.1GB 0.01% -- / -- 589.8kB / 22.13MB 44
48292d2fa956 novassh_31902 -- 5.882MB / 404.1GB 0.00% -- / -- 475.1kB / 167.3MB 2
7b2ce84e86b3 novacompute_31902 -- 231.8MB / 404.1GB 0.06% -- / -- 761.9kB / 2.43GB 49
89c01c14ef3f neutron_vpp_31902 -- 1.209GB / 404.1GB 0.30% -- / -- 0B / 7.66MB 35
Based on the provided output, it appears that no processes are exhibiting high memory usage. Additionally, the containers seem to be utilizing a low amount of memory.
The free command still shows high memory usage.
root@cvim-computex ~]# free -m
total used free shared buff/cache available
Mem: 385410 366751 7310 3496 11348 7696
Swap: 2047 5 2042
[root@cvim-computex ~]#
Troubleshoot
To comprehend this memory utilization, knowledge of HugePage memory is essential.
If the pod is enabled with HugePages, care must be taken to use the right flavor, to ensure that the system memory is not used to launch the VMs. The usage of system memory for VMs can lead to CVIM instability, as both the workload and the infrastructure are competing for the resources reserved for the infrastructure.
Check the HugePages:
[root@cvim-computex ~]# tail /sys/devices/system/node/node0/hugepages/hugepages-2048kB/nr_hugepages
90001
[root@cvim-computex ~]# tail /sys/devices/system/node/node0/hugepages/hugepages-1048576kB/nr_hugepages
0
[root@cvim-computex ~]# tail /sys/devices/system/node/node1/hugepages/hugepages-2048kB/nr_hugepages
90000
[root@cvim-computex ~]# tail /sys/devices/system/node/node1/hugepages/hugepages-1048576kB/nr_hugepages
0
[root@cvim-computex ~]#
nr_hugepages is the total number of HugePages.
(90001 + 90000) x 2M = 360GB is reserved for HugePage.
Also, note that 5% memory of total physical memory is reserved for normal memory pages (4KB) for OS usage even if 100% HugePage is configured.
385GB (free total) - 360GB (reserved for HugePage) = 25GB is reserved for Normal Pages.
So, high memory utilization, as observed in the sar and free commands is expected.
Use the mentioned command to check the actual memory usage.
[root@mgmt-node ~]# ip -br -4 a s br_api
br_api UP 10.x.x.x/24
[root@mgmt-node ~]# curl -sS -g -u admin:password --cacert /var/www/mercury/mercury-ca.crt https://10.x.x.x:9090/api/v1/query --data-urlencode 'query=100 * (mem_free + mem_buffered + mem_cached) / ((mem_total - sum without(NUMAnode, pagename, pagesize) (hugepages_nr)) or mem_total)' | python -mjson.tool
sample output:
{
"status": "success",
"data": {
"resultType": "vector",
"result": [
{
"metric": {
"host": "cvim-computex",
"instance": "10.x.x.x:9273",
"job": "telegraf",
"node_type": "compute"
},
"value": [
1693479719.383,
"76.16486394450624" --> Actual available memory percentage.
]
},
{
"metric": {
"host": "cvim-computey",
"instance": "10.x.x.x:9273",
"job": "telegraf",
"node_type": "compute"
},
"value": [
1693479719.383,
"76.63431887455388"
CVIM triggers an alert only when the available memory is less than 10%.
Alert Name - mem_available_percent
There is less than 10% of available system memory. Regular 4K pages memory is used by both the system and OpenStack infrastructure services and does not include huge pages. This alert can indicate either an insufficient amount of RAM or abnormal memory usage by the system or infrastructure.
Feedback