axcient / freebsd-nfs-exporter Goto Github PK

Hey!

I think I stumbled upon an interesting behavior in the NFS exporter.

To provide you with some background, I am collecting NFS duration stats on my NFS server. Then I use those duration stats to calculate the latency in Grafana.

What I see is that sometimes the reported latency is unbelievably large (minutes instead of milliseconds):

I calculate the latency like this:

(sum without (method) (rate(nfs_nfsd_total_duration[$__rate_interval])) / rate(nfs_nfsd_done_count[$__rate_interval])) / 1000 / 1000

which I modeled after the C code from nfsstat:

*ms_per_transfer = DELTA_T(srvduration[curop]);
*ms_per_transfer /= total_operations;
*ms_per_transfer *= 1000;

When I took a look at the actual values, I noticed that sometimes the duration counter decreases:

2023-06-06 07:50:45.000
28072724885384316
2023-06-06 07:50:50.000
28072724887013588
2023-06-06 07:50:55.000
28072725009863908         
          ^ 
2023-06-06 07:51:00.000
28072725000875584
          ^ Less than before
2023-06-06 07:51:05.000
28072725017900904
2023-06-06 07:51:10.000
28072725242506450
2023-06-06 07:51:15.000
28072725252723510

I am still tracking down what causes this behavior. This could be the exporter, nfsstat, or even the kernel.

If you have any ideas or pointers, I'm all ears. Thanks!

It doesn't need to say much, but this utility ought to have its own man page.

Hello, thank you for this project.

I'm looking to run this on an IPv6 address, but I get the following. May I request IPv6 support for this project please?

$ nfs-exporter -b "::0" -p 11111
thread 'main' panicked at 'can not parse listen addr: AddrParseError(())', src/main.rs:64:45
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace