diff options
Diffstat (limited to 'health/guides/cpu/10min_cpu_iowait.md')
| -rw-r--r-- | health/guides/cpu/10min_cpu_iowait.md | 36 |
1 files changed, 36 insertions, 0 deletions
diff --git a/health/guides/cpu/10min_cpu_iowait.md b/health/guides/cpu/10min_cpu_iowait.md new file mode 100644 index 00000000..b05530e8 --- /dev/null +++ b/health/guides/cpu/10min_cpu_iowait.md @@ -0,0 +1,36 @@ +### Understand the alert + +This alarm calculates the average time of `iowait` through 10 minute interval periods. `iowait` is the percentage of time where there has been at least one I/O request in progress while the CPU has been idle. + +I/O -at a process level- is the use of the read and write services, such as reading data from a physical drive. + +It's important to note that during the time a process waits on I/O, the system can schedule other processes, but `iowait` is measured specifically while the CPU is idle. + +A common example of when this alert might be triggered would be when your CPU requests some data and the device responsible for it can't deliver it fast enough. As a result the CPU (in the next clock interrupt) is idle, so you +encounter `iowait`. If this persists for some time and the average from the metrics we gather exceeds the value that is being checked in the `.conf` file, then the alert is raised because the CPU is being bottlenecked by your system’s disks. + +### Troubleshooting Section + +- Check for main I/O related processes and hardware issues + +Generally, this issue is caused by having slow hard drives that cannot keep up with the speed of your CPU. You can see the percentage of `iowait` by going to your node on Netdata Cloud and clicking the `iowait` dimension under the Total CPU Utilization chart. + +- You can use `vmstat` (or `vmstat 1`, to set a delay between updates in seconds) + +The `procs` column, shows the number of processes blocked waiting for I/O to complete. + +After that, you can use `ps` and specifically `ps -eo s,user,cmd | grep ^[D]` to fetch the processes that their state code starts with `D` which means uninterruptible sleep (usually IO). + +- It could be helpful to close any of the main consumer processes, but Netdata strongly suggests knowing exactly what processes you are closing and being certain that they are not necessary. + +- If you see that you don't have a lot of processes that you can terminate (or you need them for your workflow), then you would have to upgrade your system’s drives; if you have an HDD, upgrading to an SSD or an NVME drive would make a great impact on this metric. + +### Are you operating a database? + +In a database environment, you would want to optimize your operations. Check for potential inserts on large data sets, keeping in mind that `write` operations take more time than `read`. You should also search for + complex requests, like large joins and queries over a big data set. These can introduce `iowait` and need to be optimized. + +### Useful resources + +- [What exactly is "iowait"?](https://serverfault.com/questions/12679/can-anyone-explain-precisely-what-iowait-is) + |