Invalid node_network_speed_bytes value -125000 for disconnected or virtualized NICs #1967
Comments
|
Interesting. Do you happen to know if -1 documented in the kernel as a special value? Are there other special values? |
|
Good question, I haven't had much luck with the kernel source or documentation, the only thing I found was Documentation/ABI/testing/sysfs-class-net, which reads: It does not state what the attribute could read for "invalid" interfaces, though. However, the if (tb[ETHTOOL_A_LINKMODES_SPEED]) {
uint32_t val = mnl_attr_get_u32(tb[ETHTOOL_A_LINKMODES_SPEED]);
print_banner(nlctx);
if (val == 0 || val == (uint16_t)(-1) || val == (uint32_t)(-1))
printf("\tSpeed: Unknown!\n");
else
printf("\tSpeed: %uMb/s\n", val);
}Since ethtool is closely related to the kernels network subsystem (and thus should know what could be returned) and treats anything else as valid speed value, I guess it would make sense to just copy its behavior:
For reference, here are the outputs of ethtool and node-exporter: laptop without cable plugged in
virtio-based VM interface, connected and working
|
|
Thanks for the report, I'm just catching up on some of these issues. For "unknown" like this, we tend to not expose data, rather than report negative numbers. |
SuperQ
added a commit
that referenced
this issue
Mar 4, 2021
Some devices (ex virtual) don't have a speed and report `-1` as the speed value. Skip reporting speed for these devices. Fixes: #1967 Signed-off-by: Ben Kochie <superq@gmail.com>
SuperQ
added a commit
that referenced
this issue
Mar 17, 2021
Some devices (ex virtual) don't have a speed and report `-1` as the speed value. Add a flag to allow ignoring speed on these devices. Fixes: #1967 Signed-off-by: Ben Kochie <superq@gmail.com>
SuperQ
added a commit
that referenced
this issue
Mar 18, 2021
Some devices (ex virtual) don't have a speed and report `-1` as the speed value. Add a flag to allow ignoring speed on these devices. Fixes: #1967 Signed-off-by: Ben Kochie <superq@gmail.com>
oblitorum
pushed a commit
to shatteredsilicon/node_exporter
that referenced
this issue
Apr 9, 2024
Some devices (ex virtual) don't have a speed and report `-1` as the speed value. Add a flag to allow ignoring speed on these devices. Fixes: prometheus#1967 Signed-off-by: Ben Kochie <superq@gmail.com>
oblitorum
pushed a commit
to shatteredsilicon/node_exporter
that referenced
this issue
Apr 9, 2024
Some devices (ex virtual) don't have a speed and report `-1` as the speed value. Add a flag to allow ignoring speed on these devices. Fixes: prometheus#1967 Signed-off-by: Ben Kochie <superq@gmail.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
After upgrading to version
1.1.0from0.18.xor older in an oVirt / KVM virtual environment, we noticed that the metricnode_network_speed_bytesis reported as-125000when the file/sys/class/net/<nic>/speedcontains the value-1, while previously the metric returned0. The same behavior can be observed on physical wired connections when there is no cable plugged in, as I've just verified with my Arch Linux laptop.I'd understand if it returned
-1since that's whats in the file to indicate that this interface "has no speed" (essentially a sentinel value) because it is either a complicated implementation ofmemcpy(VM), or has no connection (physical interface w/o cable) and this indicates the absence of information about the speed of the interface.I've bisected this to commit 3ddc82c in PR #1580, included from version
1.0.0-rc.0onward. I'm not fluent in golang, but judging from the commit message, it seems that the early division caused it to essentially become0due to loss of precision, and thus it stayed at0when the multiplications were applied. After the PR was applied, the-1is brought to a high negative number before the division and so stays distinctly different from the raw input value or the value of0returned in previous versions.IMHO, the value
0before #1580 was merged was not entirely correct, but nevertheless adequate to indicate "no speed information for whatever reason". The new value of-125000most certainly is not correct, as it is neither a well-known sentinel value like-1nor an accurate indication of the speed of any network interface.Thus, I propose special-casing the value of
-1to return either:-1(absence of information) or0to reinstate the previous behavior.Host operating system: output of
uname -aLinux hostname 4.18.0-240.10.1.el8_3.x86_64 #1 SMP Mon Jan 18 17:05:51 UTC 2021 x86_64 x86_64 x86_64 GNU/LinuxCentOS 8.3 running in oVirt (KVM), also observed on Debian Buster on the same hypervisor
node_exporter version: output of
node_exporter --versionnode_exporter command line flags
None
Are you running node_exporter in Docker?
No
What did you do that produced an error?
Update to
1.1.0from0.18.3What did you expect to see?
Either the value returned by
0.18.3:Or ideally the raw value to indicate the absence of information, which would be:
What did you see instead?
The text was updated successfully, but these errors were encountered: