Bug 253862
| Summary: | prometheus_sysctl_exporter outputs invalid format | ||
|---|---|---|---|
| Product: | Base System | Reporter: | Devin Smith <devinrsmith> |
| Component: | bin | Assignee: | freebsd-bugs (Nobody) <bugs> |
| Status: | Closed DUPLICATE | ||
| Severity: | Affects Only Me | CC: | asomers, ed, freebsd, lapo, lwhsu, thomas |
| Priority: | --- | ||
| Version: | 12.2-RELEASE | ||
| Hardware: | amd64 | ||
| OS: | Any | ||
Same problem with 13.0. Actuall sysctl -a returns 2 entries. root@feeebsd:~ # sysctl -a | grep log.bucket.size: vm.uma.tcp_log_bucket.size: 176 vm.uma.tcp_log.bucket_size: 30 Confirmed on multiple hosts. Root cause is that default "sysctl" entries name conversion method when "." are replaced by "-" causes metric name clash for vm.uma.tcp_log_bucket.size vm.uma.tcp_log.bucket_size I do not see any good solution for it. Meanwhile workaround is to remove both metrics by running /usr/bin/sed -I "" '/sysctl_vm_uma_tcp_log_bucket_size .*$/d' filename. against prometheus_sysctl_exporter output file. *** Bug 258393 has been marked as a duplicate of this bug. *** This is not worked around by excluding "-d", it happens anyways on values (not only on #HELP lines). I'm using this as a workaround: prometheus_sysctl_exporter | sort -k1,1 -u (of course, ) (of course, this will drop any value after the first) This is still an issue on 13-STABLE as of a few days ago. It is outputting data which is not valid - it doesn't work. The workaround by Lapo Luchini in comment #5 works, but this behaviour should be built into the prometheus_system_exporter itself. The program could/should maintain a list of keys already output and skip if duplicates are found. I don't have time to look at creating a patch right now, hopefully someone else does :) Closing this as a duplicate of 259607, even though it was reported first, because the other bug is more general. *** This bug has been marked as a duplicate of bug 259607 *** |
It appears that /usr/sbin/prometheus_sysctl_exporter is outputting an invalid prometheus format when used with the "-d" option. $ prometheus_sysctl_exporter -d | grep HELP | awk -F " " '{print $3}' | sort | uniq -c | sort | grep -v " 1 " 2 sysctl_vfs_zfs_arc_max 2 sysctl_vfs_zfs_arc_min 2 sysctl_vfs_zfs_l2arc_feed_again 2 sysctl_vfs_zfs_l2arc_feed_min_ms 2 sysctl_vfs_zfs_l2arc_feed_secs 2 sysctl_vfs_zfs_l2arc_headroom 2 sysctl_vfs_zfs_l2arc_noprefetch 2 sysctl_vfs_zfs_l2arc_norw 2 sysctl_vfs_zfs_l2arc_write_boost 2 sysctl_vfs_zfs_l2arc_write_max "Only one HELP line may exist for any given metric name." - https://github.com/prometheus/docs/blob/33f891f/content/docs/instrumenting/exposition_formats.md#comments-help-text-and-type-information This is causing a downstream consumer to fail during parsing: $ grep node_exporter /var/log/messages Feb 25 20:43:51 prometheus node_exporter[46973]: level=error ts=2021-02-26T04:43:51.646Z caller=textfile.go:209 collector=textfile msg="failed to collect textfile data" file=sysctl.prom err="failed to parse textfile data from \"/var/tmp/node_exporter/sysctl.prom\": text format parsing error in line 1974: second HELP line for metric name \"sysctl_vfs_zfs_l2arc_norw\"" It's easy enough to workaround by just excluding the "-d" flag, but it would be nice for prometheus to gather the additional context that "-d" provides.