New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Large jumps in MAIN.s_resp_bodybytes output #4055
Comments
We have an optimization in varnish-cache to push statistics of worker threads only when they are idle. Thus, the behavior you describe is expected if they are (mostly) fully loaded. The statistics should even out if the rate is calculated over longer time frames. There is the |
OK, will try that. Thanks for the quick reply. |
Issue not fixed after |
bugwash: @walid-git noticed that @delthas confirmed that the "second server" showing the big jumps might actually be serving mostly H2. @delthas also asked about the performance impact of |
@delthas can you give #4059 a try ? When working on this, I found an interesting case for h1, where even having
This will generate traffic with a single connection for 30s, and if you observe varnishstat during that time, no counter will update until the 30s are done, regardless of |
Expected Behavior
MAIN.s_resp_bodybytes is somewhat stable, corresponding to the output seen from the NIC.
Current Behavior
MAIN.s_resp_bodybytes shows large jumps in its output, up to +10GB in a single second.
Context
I have three servers running Varnish 7.3.0 with a similar configuration.
I have plotted the output of
varnishstat -1
run every second on each server, grepping forMAIN.s_resp_bodybytes
. First graph: total value (subtracted from the value at t0). Second graph: Derive by taking the difference between each datapoint and the previous one (in other words, variation in that second).On server 1, the output looks stable, and corresponds to what I see on the NIC.
Graph 1: MAIN.s_resp_bodybytes - t0. Graph 2 : rate of MAIN.s_resp_bodybytes.
On server 2 & 3 (I've just shown server 2 here, but the behavior is similar on server 3), the output has huge jumps from one line to the next (as seen from the spikes in the second graph, up to 1e10 ie +10GB from one second to the next, way more than the average throughput of ~0.3GB/s). This does not reflect what the NIC sees. The NIC sees a somewhat stable throughput with no jumps. As an example, from one second to the next I got 52766514824 -> 61136417222, ie a 8GB difference.
Graph 1: MAIN.s_resp_bodybytes - t0. Graph 2 : rate of MAIN.s_resp_bodybytes.
I'm serving small files (~<1MB), no pipes. So I see no particular reason for suddenly accounting for a huge change in the throughput.
Is this a known issue?
Varnish Cache version
Varnish 7.3.0
The text was updated successfully, but these errors were encountered: