查询表达式:sum by (instance) (rate(http_requests_total{handler="receive",method="post",instance="$instance",code="200"}[1m]))*60
查询表达式:sum by () (rate(http_request_size_bytes_sum{instance="$instance"}[1m])/1024/1024)
查询表达式:sum by (code) (rate(http_requests_total{handler="receive",method="post",instance="$instance",code!="200"}[1m]))*60
查询表达式:sum by (instance) (rate(process_cpu_seconds_total{instance="$instance"}[1m]))*60
查询表达式:
物理内存:sum by (instance) (max_over_time(process_resident_memory_bytes{instance="$instance"}[1m]))/1024/1024/1024
虚拟内存:sum by (instance) (max_over_time(process_virtual_memory_bytes{instance="$instance"}[1m]))/1024/1024/1024
thanos receiver的内存,并没有因为切换tsdb而明显的降低,没有请求后,内存释放得非常缓慢:
查询表达式:prometheus_tsdb_head_samples_appended_total{instance="$instance"}
新增data point 85490118,平均 2590610/min, 43177/s
查询表达式:sum by () (rate(prometheus_tsdb_head_samples_appended_total{instance="$instance"}[1m]))*60
峰值:4807544/min, 80126/s,可以认为thanos receiver单核的极限的处理data point的能力为8万每秒。
查询表达式:prometheus_tsdb_head_series{instance="$instance"}
查询表达式:sum by () (rate(prometheus_tsdb_head_series_created_total{instance="$instance"}[1m]))*60
因为同样的数据文件发送了两次,所以time series的总数是一致的;time series总数达到 3934198
查询表达式:prometheus_tsdb_head_chunks{instance="$instance"}
tsdb的两个参数:"--tsdb.min-block-duration=30m","--tsdb.max-block-duration=30m"
,因此14:29~14:49达到了tsdb切换的周期
查询表达式:sum by () (rate(prometheus_tsdb_head_chunks_created_total{instance="$instance"}[1m]))*60
查询表达式:prometheus_tsdb_compactions_triggered_total{instance="$instance"}
为什么只在没有请求的时候触发 compaction ?
查询表达式:prometheus_tsdb_head_truncations_total{instance="$instance"}
truncate动作的执行规律,还没搞明白
thanos receiver单核处理data point的能力大约在 43177/s ~ 80126/s
粗略的认为:存储 3934198 time series + 85490118个data point花费了12.1GB内存
手机扫一扫
移动阅读更方便
你可能感兴趣的文章