Crostini: improve logout time when containers are enabled |
|||
Issue descriptionChrome Version: 72.0.3612.0 Chrome OS Version: 11276.0.0 Chrome OS Platform: Eve Network info: WiFi Please specify Cr-* of the system to which this bug/feature applies (add the label below). Steps To Reproduce: (1) Install linux and observe terminal opened. (2) Logout device (3) Observe time taking for logout Expected Result: - Logout time same as without linux(around 5 or less). Actual Result: - It is taking too long. Around 15 to 18 seconds. How frequently does this problem reproduce? (Always, sometimes, hard to reproduce?) Always Please check below observation video. Feedback report :https://listnr.corp.google.com/report/85796875073
,
Nov 19
there's always going to be overhead when we have containers/VMs running. expecting it to be zero overhead is unreasonable (unless you don't care about losing data?).
,
Nov 19
True, a performance test will help to ensure that we don't regress.
,
Nov 20
I'd guess there is a lot of low hanging fruit here. If there has been a lot of disk IO it's going to take a while to sync it all. However the common case feels like it should be much faster than it currently is. chirantan, any ideas about the concierge/maitred or ciccerone/garcon shutdown process that might be taking a while to complete?
,
Dec 1
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/platform/tremplin/+/a8e30c1592f745d08c2347da2d9c517e1862aeaa commit a8e30c1592f745d08c2347da2d9c517e1862aeaa Author: Chirantan Ekbote <chirantan@chromium.org> Date: Sat Dec 01 09:08:40 2018 Change shutdown timeout to 9 seconds The systemd version in debian stretch appears to have a bug where it doesn't exit even after being sent a signal to shut down the container. As a result, lxd waits for the full default timeout before force killing every process left. We can't do much about systemd but we can improve things somewhat by lowering lxd's timeout to 9 seconds. maitred's shut down timeout is 10 seconds so making lxd's timeout 9 seconds ensures that it will exit in time. BUG=chromium:906804 TEST=concierge_client --stop_all Change-Id: I1eaf5cadb91ae83e506ba5765a10e8d82d2be101 Signed-off-by: Chirantan Ekbote <chirantan@chromium.org> Reviewed-on: https://chromium-review.googlesource.com/1357249 Reviewed-by: Stephen Barber <smbarber@chromium.org> [modify] https://crrev.com/a8e30c1592f745d08c2347da2d9c517e1862aeaa/src/chromiumos/tremplin/main.go
,
Dec 2
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/platform2/+/d79ed8b4d7f4b29d65e2a96bc422fe943636c5f3 commit d79ed8b4d7f4b29d65e2a96bc422fe943636c5f3 Author: Chirantan Ekbote <chirantan@chromium.org> Date: Sun Dec 02 05:19:16 2018 vm_tools: maitred: Send SIGPWR to lxd on shutdown lxd treats SIGPWR as a signal that the system is going to lose power and that it should shut down all containers and exit. Send SIGPWR to lxd before sending SIGTERM to all other processes during shutdown. This will not improve things much for the stretch container due to an issue with systemd but it should make shutdown for all other types of containers much faster. BUG=chromium:906804 TEST=concierge_client --stop_all Change-Id: I79d0f618c7c0dc24a16d9d5fc7e7259acaa71f4a Signed-off-by: Chirantan Ekbote <chirantan@chromium.org> Reviewed-on: https://chromium-review.googlesource.com/1357643 Reviewed-by: Stephen Barber <smbarber@chromium.org> Reviewed-by: Mike Frysinger <vapier@chromium.org> [modify] https://crrev.com/d79ed8b4d7f4b29d65e2a96bc422fe943636c5f3/vm_tools/maitred/init.cc
,
Dec 3
Due to a bug in the systemd version in debian stretch we will have to wait for the full lxd shutdown timeout every time we log out. The change in #5 lowered this timeout to 9 seconds. The upgrade to buster should make things much faster but I'm not sure there is a lot more we can do about this right now. |
|||
►
Sign in to add a comment |
|||
Comment 1 by rohi...@google.com
, Nov 19