needs-restarting: fix for RTC not in UTC #518
Conversation
For systems where the RTC is not in UTC, the boot time is computed in the future if the timezone is +something. The correct time can be determined by btime in /proc/stat. Such systems are usually configured with `timedatectl set-local-rtc 1` which changes the 3rd line of /etc/adjtime from UTC to LOCAL. See man timedatectl.
| with open('/proc/stat', 'r') as f: | ||
| btime = [line for line in f if 'btime' in line] | ||
| proc_stat_boot_time = int(btime[0].strip().split(' ')[1]) | ||
| return max(proc_1_boot_time, proc_stat_boot_time) |
There was a problem hiding this comment.
Won't doing the max here not work if (1) proc_1_boot_time is X hours ahead of the correct boot time, due to set-local-rtc 1 and (2) proc_stat_boot_time is correct?
There was a problem hiding this comment.
right. actually, returning proc_stat_boot_time should be enough since it should work every time local rtc is on... if not in a container, like you mention here (rpm-software-management/dnf5#1184)
There was a problem hiding this comment.
Does the btime from /proc/stat give us any more info than /proc/uptime? My current understanding is:
- When we are NOT in a container:
- mtime of
/proc/1is unreliable because an RTC may be unavailable, or it may be incorrect due toset-local-rtc 1. - current time minus
/proc/uptimeis correct /proc/statbtimeis correct
- mtime of
- When we ARE in a container:
- mtime of
/proc/1is correct /proc/statbtimeis incorrect, has uptime of host system/proc/uptimeis incorrect, has uptime of the host system
- mtime of
I propose getting UserspaceTimestampMonotonic, which is essentially uptime, from systemd (I mentioned here also). If systemd is available, this number will definitely be correct. If systemd is not available, we are probably in a container, and we fall back to the old method of max(proc_1_boot_time, proc_uptime_boot_time).
EDIT: I misunderstood UserspaceTimestampMonotonic: rpm-software-management/dnf5#1184 (comment)
this would definitely improve the situation, and it makes sense to see it implemented in dnf4/5. from my pessimistic point of view, I believe we'll hit more cases in the future that will not fall in this generalization, but I would say yours is a solid approach. |
|
See rpm-software-management/dnf5#1184 (comment) and rpm-software-management/dnf5#1198, I think the same fix could be applied here to DNF 4. |
|
Closing in favor of this #527 |
For systems where the RTC is not in UTC, the boot time is computed in the future if the timezone is +something.
The correct time can be determined by btime in /proc/stat. Such systems are usually configured with
timedatectl set-local-rtc 1which changes the 3rd line of /etc/adjtime from UTC to LOCAL. See man timedatectl.