NTOP is showing me about 70Mbps / sec out on a host. However, it’s showing -zero- of that traffic accounted for in Percona.
Pls advise.
2 Likes
@Matt_Westfall please could you confirm your version of PMM and the dashboard that you are referring to?
2 Likes
Yes Sir,
ProjectName: pmm-managed
Version: 2.28.0
PMMVersion: 2.28.0
Timestamp: 2022-04-28 14:23:37 (UTC)
FullCommit: 071a587e34bb34330eb24f6220b4650464bffbc3
-Client-
ProjectName: pmm-agent
Version: 2.28.0
PMMVersion: 2.28.0
Timestamp: 2022-05-10 16:55:33 (UTC)
FullCommit: 44d20e7bb2a0dc0e0439cdf93f1d622a55d3b5e9
System → Node Summary → Network
As of this post I’m seeing 60Mbps out via NTOP on this host.
Percona is only reporting ~~ 9Mbps
2 Likes
Thanks for the information. Right now, I don’t have an IPv6 server to hand, so can’t quickly test myself.
The expression used for the panel only excludes the loopback device (device!="lo") and the code for the node exporter supports IPv6. Are you sure that the extra traffic is not across the loopback device?
If you use the drop-down menu that is alongside “Network Traffic” (when you hover over it) and select “Explore” then you will see the expression in the explorer. If you then remove the exclusion of the loopback device (i.e. any occurrences of device!="lo") and “Run query”, does the traffic change on the chart and look closer to what you are seeing?
2 Likes
iptraf-ng - interface eno1
Outgoing rates: 110839.42 kbps |
| 1966 pps
Right now.
so 110 Mbps
Percona is reporting 11Mbps right now
Doing the query modification you noted, it goes up to 14,480,000 from about 12,000,000
But that’s no where near the 110,000,000 iptraf-ng is currently showing for eno1 right now.
Thanks,
Matt
1 Like
Ah OK, so not the loopback device.
Percona is reporting 11Mbps right now
Not quite… what you see is extrapolation between metric samples and so whilst you have live traffic in iptraf, you do not have live traffic in PMM. You have samples based upon the metric at a given point in time.
1 Like
OK well I guess I may not understand how Percona works.
So if I have a chart that’s supposed to show me network usage.
And the server is consistently using 80-120Mbps at all times with no drops.
Would I not expect the graph to indicate the same usage? The metric of the “network graph” should be “The network usage?”
I mean the graph is called “Network Traffic” and it’s values are in Bits per second.
Am I missing something?
1 Like
That might be what you are missing … bytes, not bits?
Are you getting a range of ~ 10-15?
Just to confirm, running a local agent whilst using fast.com I get:
- ~120000 in iptraf
- 120 Mbps from the speed test, and
- 14.88 MBps in the dashboard
1 Like
PMM is set up to use exporters to gather information at certain time intervals…we call those intervals the “resolution” and at the standard resolution we check the state of the system at 5, 10, or 60 seconds for high, medium and low resolution data), there’s also a Rare and Frequent default or you can set custom as well in settings (just be aware that ALL exporters will run at the intervals you define). more info here
Most exporters are not designed to “get all data since last check” instead just get “the current state” so there’s always a chance we poll right at the dip but the more samples taken over time tends to smooth things out a bit. Do be careful about tweaking resolutions as you can have unintended consequences (you query table stats every 1 second on large DB’s you can topple a system quickly as the last query is still running when the next is requested).
The notable exceptions to “the state when the exporter ran” is Query Analytics…queries are captured in buckets (60 seconds worth) and the exporters grab the completed buckets at the resolution intervals so you never miss a single one.
Hope this helps demystify how PMM works. Here are some architectural diagrams that may make more sense than my rambling!
1 Like