參數(shù)解釋

使用 Prometheus 配置 kubernetes 環(huán)境中 Container 的 CPU 使用率時(shí)，會(huì)經(jīng)常遇到 CPU 使用超出 100%，下面就來解釋一下：

1. container_spec_cpu_period

   當(dāng)對(duì)容器進(jìn)行 CPU 限制時(shí)，CFS 調(diào)度的時(shí)間窗口，又稱容器 CPU 的時(shí)鐘周期通常是 100，000 微秒

2. container_spec_cpu_quota

   是指容器的使用 CPU 時(shí)間周期總量，如果 quota 設(shè)置的是 700，000，就代表該容器可用的 CPU 時(shí)間是 7*100,000 微秒，通常對(duì)應(yīng) kubernetes 的 resource.cpu.limits 的值

3. container_spec_cpu_share

   是指 container 使用分配主機(jī) CPU 相對(duì)值，比如 share 設(shè)置的是 500m，代表窗口啟動(dòng)時(shí)向主機(jī)節(jié)點(diǎn)申請(qǐng) 0.5 個(gè) CPU，也就是 50，000 微秒，通常對(duì)應(yīng) kubernetes 的 resource.cpu.requests 的值

4. container_cpu_usage_seconds_total

   統(tǒng)計(jì)容器的 CPU 在一秒內(nèi)消耗使用率，應(yīng)注意的是該 container 所有的 CORE

5. container_cpu_system_seconds_total

   統(tǒng)計(jì)容器內(nèi)核態(tài)在一秒時(shí)間內(nèi)消耗的 CPU

6. container_cpu_user_seconds_total

   統(tǒng)計(jì)容器用戶態(tài)在一秒時(shí)間內(nèi)消耗的 CPU

   參考官方地址 https://docs.signalfx.com/en/latest/integrations/agent/monitors/cadvisor.html https://github.com/google/cadvisor/blob/master/docs/storage/prometheus.md

具體公式

1. 默認(rèn)如果直接使用 container_cpu_usage_seconds_total 的話，如下

   sum(irate(container_cpu_usage_seconds_total{container="$Container",instance="$Node",pod="$Pod"}[5m])*100)by(pod)
   

   默認(rèn)統(tǒng)計(jì)的數(shù)據(jù)是該容器所有的 CORE 的平均使用率

   

2. 如果要精確計(jì)算每個(gè)容器的 CPU 使用率，使用 % 呈現(xiàn)的形式，如下

   sum(irate(container_cpu_usage_seconds_total{container="$Container",instance="$Node",pod="$Pod"}[5m])*100)by(pod)/sum(container_spec_cpu_quota{container="$Container",instance="$Node",pod="$Pod"}/container_spec_cpu_period{container="$Container",instance="$Node",pod="$Pod"})by(pod)
   

   其中 container_spec_cpu_quota/container_spec_cpu_period，就代表該容器有多少個(gè) CORE

   

3. 參考官方 git issue

   https://github.com/google/cadvisor/issues/2026#issuecomment-415819667

docker stats

docker stats 輸出的指標(biāo)列是如何計(jì)算的，如下：

首先 docker stats 是通過 Docker API /containers/(id)/stats 接口來獲得 live data stream，再通過 docker stats 進(jìn)行整合。

在 Linux 中使用 docker stats 輸出的內(nèi)存使用率（MEM USAGE），實(shí)則該列的計(jì)算是不包含 Cache 的內(nèi)存。

cache usage 在 ≤ docker 19.03 版本的 API 接口輸出對(duì)應(yīng)的字段是 memory_stats.total_inactive_file，而 > docker 19.03 的版本對(duì)應(yīng)的字段是 memory_stats.cache。

docker stats 輸出的 PIDS 一列代表的是該容器創(chuàng)建的進(jìn)程或線程的數(shù)量，threads 是 Linux kernel 中的一個(gè)術(shù)語，又稱 lightweight process & kernel task。

1. 如何通過 Docker API 查看容器資源使用率，如下

   $ curl -s --unix-socket /var/run/docker.sock "http://localhost/v1.40/containers/10f2db238edc/stats" | jq -r
   {
     "read": "2022-01-05T06:14:47.705943252Z",
     "preread": "0001-01-01T00:00:00Z",
     "pids_stats": {
       "current": 240
     },
     "blkio_stats": {
       "io_service_bytes_recursive": [
         {
           "major": 253,
           "minor": 0,
           "op": "Read",
           "value": 0
         },
         {
           "major": 253,
           "minor": 0,
           "op": "Write",
           "value": 917504
         },
         {
           "major": 253,
           "minor": 0,
           "op": "Sync",
           "value": 0
         },
         {
           "major": 253,
           "minor": 0,
           "op": "Async",
           "value": 917504
         },
         {
           "major": 253,
           "minor": 0,
           "op": "Discard",
           "value": 0
         },
         {
           "major": 253,
           "minor": 0,
           "op": "Total",
           "value": 917504
         }
       ],
       "io_serviced_recursive": [
         {
           "major": 253,
           "minor": 0,
           "op": "Read",
           "value": 0
         },
         {
           "major": 253,
           "minor": 0,
           "op": "Write",
           "value": 32
         },
         {
           "major": 253,
           "minor": 0,
           "op": "Sync",
           "value": 0
         },
         {
           "major": 253,
           "minor": 0,
           "op": "Async",
           "value": 32
         },
         {
           "major": 253,
           "minor": 0,
           "op": "Discard",
           "value": 0
         },
         {
           "major": 253,
           "minor": 0,
           "op": "Total",
           "value": 32
         }
       ],
       "io_queue_recursive": [],
       "io_service_time_recursive": [],
       "io_wait_time_recursive": [],
       "io_merged_recursive": [],
       "io_time_recursive": [],
       "sectors_recursive": []
     },
     "num_procs": 0,
     "storage_stats": {},
     "cpu_stats": {
       "cpu_usage": {
         "total_usage": 251563853433744,
         "percpu_usage": [
           22988555937059,
           6049382848016,
           22411490707722,
           5362525449957,
           25004835766513,
           6165050456944,
           27740046633494,
           6245013152748,
           29404953317631,
           5960151933082,
           29169053441816,
           5894880727311,
           25772990860310,
           5398581194412,
           22856145246881,
           5140195759848
         ],
         "usage_in_kernelmode": 30692640000000,
         "usage_in_usermode": 213996900000000
       },
       "system_cpu_usage": 22058735930000000,
       "online_cpus": 16,
       "throttling_data": {
         "periods": 10673334,
         "throttled_periods": 1437,
         "throttled_time": 109134709435
       }
     },
     "precpu_stats": {
       "cpu_usage": {
         "total_usage": 0,
         "usage_in_kernelmode": 0,
         "usage_in_usermode": 0
       },
       "throttling_data": {
         "periods": 0,
         "throttled_periods": 0,
         "throttled_time": 0
       }
     },
     "memory_stats": {
       "usage": 8589447168,
       "max_usage": 8589926400,
       "stats": {
         "active_anon": 0,
         "active_file": 260198400,
         "cache": 1561460736,
         "dirty": 3514368,
         "hierarchical_memory_limit": 8589934592,
         "hierarchical_memsw_limit": 8589934592,
         "inactive_anon": 6947250176,
         "inactive_file": 1300377600,
         "mapped_file": 0,
         "pgfault": 3519153,
         "pgmajfault": 0,
         "pgpgin": 184508478,
         "pgpgout": 184052901,
         "rss": 6947373056,
         "rss_huge": 6090129408,
         "total_active_anon": 0,
         "total_active_file": 260198400,
         "total_cache": 1561460736,
         "total_dirty": 3514368,
         "total_inactive_anon": 6947250176,
         "total_inactive_file": 1300377600,
         "total_mapped_file": 0,
         "total_pgfault": 3519153,
         "total_pgmajfault": 0,
         "total_pgpgin": 184508478,
         "total_pgpgout": 184052901,
         "total_rss": 6947373056,
         "total_rss_huge": 6090129408,
         "total_unevictable": 0,
         "total_writeback": 0,
         "unevictable": 0,
         "writeback": 0
       },
       "limit": 8589934592
     },
     "name": "/k8s_prod-xc-fund_prod-xc-fund-646dfc657b-g4px4_prod_523dcf9d-6137-4abf-b4ad-bd3999abcf25_0",
     "id": "10f2db238edc13f538716952764d6c9751e5519224bcce83b72ea7c876cc0475"
   

2. 如何計(jì)算

   官方地址

   https://docs.docker.com/engine/api/v1.40/#operation/ContainerStats

   The precpu_stats is the CPU statistic of the previous read, and is used to calculate the CPU usage percentage. It is not an exact copy of the cpu_stats field.

   If either precpu_stats.online_cpus or cpu_stats.online_cpus is nil then for compatibility with older daemons the length of the corresponding cpu_usage.percpu_usage array should be used.

   To calculate the values shown by the stats command of the docker cli tool the following formulas can be used:

• used_memory = memory_stats.usage - memory_stats.stats.cache
• available_memory = memory_stats.limit
• Memory usage % = (used_memory / available_memory) * 100.0
• cpu_delta = cpu_stats.cpu_usage.total_usage - precpu_stats.cpu_usage.total_usage
• system_cpu_delta = cpu_stats.system_cpu_usage - precpu_stats.system_cpu_usage
• number_cpus = lenght(cpu_stats.cpu_usage.percpu_usage) or cpu_stats.online_cpus
• CPU usage % = (cpu_delta / system_cpu_delta) * number_cpus * 100.0