Attachment 109734 Details for Bug 86465 – #systemd IRC log discussing issue

#systemd IRC log discussing issue

irclog-systemd.2014-11-19.txt (text/plain), 2.24 KB, created by SW on 2014-11-19 14:57:38 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: SW

Created: 2014-11-19 14:57:38 UTC

Size: 2.24 KB

patch

obsolete

>stuart-warren: Is there a limit to the length of a SystemD journal entry of 2048? I'm having troubles with long Java stacktraces in docker logs being split over multiple entries. https://github.com/stuart-warren/testbigdockerlogs
>sztanpet: that is not how its spelled, just fyi
>ohsix: i don't think there's any hard limit for anything, let me look
>sztanpet: and there is a limit, but its not that small
>or at least shouldnt be
>ohsix: is it 4096 or a multiple of the page size?
>sztanpet: ohsix #define DEFAULT_MAX_SIZE_UPPER (128ULL*1024ULL*1024ULL) /* 128 MiB */
>ohsix: thanks
>i wouldn't be surprised if a trip through syslog or something would actually be limiting it
>stuart-warren: ohsix: my messages are being split at 2048.
>syslog normally splits at 1024 I thought
>sztanpet: #define ENTRY_SIZE_MAX (1024*1024*770u)
>theres that too
>stuart-warren: I've seen that you can send in big logs https://github.com/stuart-warren/bigjournallogs
>But something with getting from docker into the journal is splitting at that threshold
>This is coreos by the way
>sztanpet: for the record where did you get to spell systemd like that stuart-warren? only interested because of the fact that it is the first thing the wiki tells you to not do
>stuart-warren: Sorry my bad, guess it's just a bad habit I have picked up
>grawity: so I just tried `perl -e "print 'a'x4096, 'b'" | systemd-cat`
>which seems to be a rather direct route to the journal
>and it got split into maximum of 2048 each
>stuart-warren: great! i'm not crazy :)
>dreisner: is this a matter of how the log collection is done? i.e., collection from stdout/stderr is limited to 2048 bytes, but sd_journal_sendv is not?
>grawity: apparently it is
>tried `systemd.journal.send("a"*4096 + "b")` from Python, it came through as a single message
>dreisner: yeah, the stream fd based collection is kind of lackluster
>ohsix: i bet you knew the whole time, so clever
>dreisner: i wonder how long it'll take before the RHEL7 crowd discovers that restarting journald will drop all of your stdout/stderr logging...
>grawity: has the desktop crowd discovered yet that doing so will also sigpipe half of your desktop apps
>dreisner: at least stopping the socket no longer results in an orderly shutdown
>stuart-warren: Shall I raise a bug on the fd based collection?

Actions: View

Attachments on bug 86465: 109734

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.