Question

我在cassandra节点上执行sjk（swiss java kill）命令来调查DC上的高负载：

./ nodetool sjk ttop -o CPU -n 20

获得以下输出：

Process summary
  process cpu=177.48%
  application cpu=154.86% (user=126.50% sys=28.35%)
  other: cpu=22.62%
  thread count: 568
  heap allocation rate 224mb/s
[1052346] user=48.79% sys= 1.73% alloc=  162mb/s - CompactionExecutor:1351
[1054022] user=10.61% sys= 0.47% alloc=   31mb/s - MemtableFlushWriter:40483
[000496] user= 7.70% sys= 2.66% alloc= 2592kb/s - SharedPool-Worker-1
[000498] user= 7.49% sys= 2.13% alloc= 2184kb/s - SharedPool-Worker-2
[000665] user= 6.45% sys= 2.02% alloc= 2165kb/s - SharedPool-Worker-3
[000666] user= 5.72% sys= 1.93% alloc= 1887kb/s - SharedPool-Worker-4
[1054020] user= 4.47% sys= 2.46% alloc= 3909kb/s - RMI TCP Connection(90985)-xx.xx.xx.xx
[000670] user= 5.31% sys= 1.47% alloc= 1611kb/s - SharedPool-Worker-5
[000669] user= 3.85% sys= 1.08% alloc= 1324kb/s - SharedPool-Worker-6

此行

[1052346]用户= 48.79％sys = 1.73％分配= 162mb / s- CompactionExecutor：1351

显示压缩正在使节点麻烦。

实际上，我可以在OpsCenter.rollup_state表上看到频繁的压缩。

如何进一步进行，我应该进行哪些调查才能找出根本原因？

还增加 concurrent_compactors 个计数可以有所帮助吗？

如何调查直流上的高负载？

0 个答案: