注意
本文档适用于 Ceph 的开发版本。
Perf Histograms
Perf Histograms 建立在 perf counters 基础设施之上。直方图针对许多计数器构建，并简化了收集数据，以便了解哪些组的计数器值在一段时间内最常出现。Perf Histograms 目前是无符号 64 位整数计数器，因此它们主要用于时间和大小。通过 perf histogram 转储的数据可以馈送到其他分析工具/脚本中。
访问
perf histogram 数据通过 admin socket 访问。例如
ceph daemon osd.0 perf histogram schema
ceph daemon osd.0 perf histogram dump
集合
直方图被分组到命名集合中，通常代表一个子系统或一个子系统的实例。例如，内部 throttle 机制报告有关其如何进行节流的统计信息，每个实例的命名类似于
op_r_latency_out_bytes_histogram
op_rw_latency_in_bytes_histogram
op_rw_latency_out_bytes_histogram
...
模式
perf histogram schema 命令转储一个 json 描述，说明哪些值可用，以及它们的类型。每个命名值都有一个 type 位域，第 5 位始终设置，后续位已定义。
位 |
含义 |
|---|---|
1 |
浮点值 |
2 |
无符号 64 位整数值 |
4 |
平均值（总和 + 计数对） |
8 |
计数器（与 gauge 相比） |
换句话说，类型为“18”的直方图是无符号 64 位整数值的直方图 (16 + 2)。
这是一个 schema 输出的示例
{
"AsyncMessenger::Worker-0": {},
"AsyncMessenger::Worker-1": {},
"AsyncMessenger::Worker-2": {},
"mutex-WBThrottle::lock": {},
"objecter": {},
"osd": {
"op_r_latency_out_bytes_histogram": {
"type": 18,
"description": "Histogram of operation latency (including queue time) + da ta read",
"nick": ""
},
"op_w_latency_in_bytes_histogram": {
"type": 18,
"description": "Histogram of operation latency (including queue time) + da ta written",
"nick": ""
},
"op_rw_latency_in_bytes_histogram": {
"type": 18,
"description": "Histogram of rw operation latency (including queue time) + data written",
"nick": ""
},
"op_rw_latency_out_bytes_histogram": {
"type": 18,
"description": "Histogram of rw operation latency (including queue time) + data read",
"nick": ""
}
}
}
转储
实际转储与模式类似，只是存在实际的值组。例如
"osd": {
"op_r_latency_out_bytes_histogram": {
"axes": [
{
"name": "Latency (usec)",
"min": 0,
"quant_size": 100000,
"buckets": 32,
"scale_type": "log2",
"ranges": [
{
"max": -1
},
{
"min": 0,
"max": 99999
},
{
"min": 100000,
"max": 199999
},
{
"min": 200000,
"max": 399999
},
{
"min": 400000,
"max": 799999
},
{
"min": 800000,
"max": 1599999
},
{
"min": 1600000,
"max": 3199999
},
{
"min": 3200000,
"max": 6399999
},
{
"min": 6400000,
"max": 12799999
},
{
"min": 12800000,
"max": 25599999
},
{
"min": 25600000,
"max": 51199999
},
{
"min": 51200000,
"max": 102399999
},
{
"min": 102400000,
"max": 204799999
},
{
"min": 204800000,
"max": 409599999
},
{
"min": 409600000,
"max": 819199999
},
{
"min": 819200000,
"max": 1638399999
},
{
"min": 1638400000,
"max": 3276799999
},
{
"min": 3276800000,
"max": 6553599999
},
{
"min": 6553600000,
"max": 13107199999
},
{
"min": 13107200000,
"max": 26214399999
},
{
"min": 26214400000,
"max": 52428799999
},
{
"min": 52428800000,
"max": 104857599999
},
{
"min": 104857600000,
"max": 209715199999
},
{
"min": 209715200000,
"max": 419430399999
},
{
"min": 419430400000,
"max": 838860799999
},
{
"min": 838860800000,
"max": 1677721599999
},
{
"min": 1677721600000,
"max": 3355443199999
},
{
"min": 3355443200000,
"max": 6710886399999
},
{
"min": 6710886400000,
"max": 13421772799999
},
{
"min": 13421772800000,
"max": 26843545599999
},
{
"min": 26843545600000,
"max": 53687091199999
},
},
{
"min": 53687091200000
}
]
},
{
"name": "Request size (bytes)",
"min": 0,
"quant_size": 512,
"buckets": 32,
"scale_type": "log2",
"ranges": [
{
"max": -1
},
{
"min": 0,
"max": 511
},
{
"min": 512,
"max": 1023
},
{
"min": 1024,
"max": 2047
},
{
"min": 2048,
"max": 4095
},
{
"min": 4096,
"max": 8191
},
{
"min": 8192,
"max": 16383
},
{
"min": 16384,
"max": 32767
},
{
"min": 32768,
"max": 65535
},
{
"min": 65536,
"max": 131071
},
{
"min": 131072,
"max": 262143
},
{
"min": 262144,
"max": 524287
},
{
"min": 524288,
"max": 1048575
},
{
"min": 1048576,
"max": 2097151
},
{
"min": 2097152,
"max": 4194303
},
{
"min": 4194304,
"max": 8388607
},
{
"min": 8388608,
"max": 16777215
},
{
"min": 16777216,
"max": 33554431
},
{
"min": 33554432,
"max": 67108863
},
{
"min": 67108864,
"max": 134217727
},
{
"min": 134217728,
"max": 268435455
},
{
"min": 268435456,
"max": 536870911
},
{
"min": 536870912,
"max": 1073741823
},
{
"min": 1073741824,
"max": 2147483647
},
{
"min": 2147483648,
"max": 4294967295
},
{
"min": 4294967296,
"max": 8589934591
},
{
"min": 8589934592,
"max": 17179869183
},
{
"min": 17179869184,
"max": 34359738367
},
{
"min": 34359738368,
"max": 68719476735
},
{
"min": 68719476736,
"max": 137438953471
},
{
"min": 137438953472,
"max": 274877906943
},
{
"min": 274877906944
}
]
}
],
"values": [
[
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
],
[
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
],
[
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
],
[
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
],
[
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
],
[
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
],
[
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
],
[
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
],
[
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
]
]
}
},
这代表 2D 直方图，包含 9 个历史条目和每个历史条目 32 个值组。“Ranges”元素表示每个值组的值边界。“Buckets”表示值组的数量（“buckets”），“Min”是最小接受值，“quant_size”是量化单位，“scale_type”是“log2”（对数刻度）或“linear”（线性刻度）。您可以使用 histogram_dump.py 工具（参见 src/tools/histogram_dump.py）快速可视化现有直方图数据。