Created attachment 79274 [details]
used for git bisect run
I have recently noticed issues with the use of `journalctl -a` (Arch Linux on Raspberry Pi) and `journalctl --since today` (Arch Linux x86_64). Where the command should stop (on end of log file), it continues printing older entries and repeats that until the journalctl is killed.
I suspect it is related to corrupted logs or to the timestamps, the Raspberry Pi boots with a timestamp in 1970, my laptop's RTC somehow jumped back 7 hours.
results of bisect:
a3e6f050de81a9830e52af09d5d38dad9a356e3b is the first bad commit
Author: Marius Vollmer <email@example.com>
Date: Thu Apr 18 22:34:36 2013 +0200
journal: when iterating through a file we might lose messages when changing direction.
:100644 100644 1221b39cd419325010e331d66068c1fd44039816 cd66914b5c6da29db002723e4eb7487f1070b35c M TODO
:040000 040000 a0196f64399d332d2fd43455f3bc0575146d3806 a0bf7285eddc892d51642be205983bc11f85ac9c M src
Reverting this commit on master (v204-25-g32821c7) fixes the wrap-around issue.
I prefer not to share the logs in public, if you need them please send me a email. Attached is a quick-n-dirty bisect script which I used to isolate the faulty commit.
Can you check if http://cgit.freedesktop.org/systemd/systemd/commit/?id=87011c25d9 fixes things for you? Also,
http://cgit.freedesktop.org/systemd/systemd/commit/?id=cbd671772 should fix one the ways that looped journals are produced, but it won't immediately help if you have such journals currently.
Using v204-144-g7699b6e, `time journalctl -a | head -n100000 | wc -l` terminates properly (63294 entries in 2.7 real time).
Thanks for fixing!
Tested v204-138-g8d98da3 too which is still affected by the bug. v204-139-g87011c2 (the commit you pointed to) is no longer affected by the bug.