New issue
Advanced search Search tips
Note: Color blocks (like or ) mean that a user may not be available. Tooltip shows the reason.

Issue 791605 link

Starred by 8 users

Issue metadata

Status: Fixed
Owner:
Closed: Dec 2017
Cc:
EstimatedDays: ----
NextAction: ----
OS: Chrome
Pri: 1
Type: Bug



Sign in to add a comment

min_filelist_kbytes stuck at 100M

Project Member Reported by teravest@chromium.org, Dec 4 2017

Issue description

This was found while investigating  crbug.com/774341 .

On version 64.0.3273.0, /proc/sys/vm/min_filelist_kbytes is 100000, even after the android container is set up.

Interestingly, running "swap set_min_filelist default" adjusts the value correctly to 400000.

I'm not sure if this is a regression in 64 or if this has already existed for a while.
 
It appears that start-arc-instance isn't getting fired, but stop-arc-instance is. I've confirmed this locally by logging from both scripts.
Cc: yusukes@chromium.org
To reproduce: log out (double control-Q) and log back in.  stop-arc-instance runs, but start-arc-instance does not.

Attached the contents of /var/log, per yusuke's request.
logs.tar.gz
212 KB Download
This appears to be present in R62.
Labels: M-62 M-63

Comment 7 by igo@chromium.org, Dec 4 2017

Cc: igo@chromium.org josa...@chromium.org
Labels: OS-Chrome

Comment 8 by wpwoo...@gmail.com, Dec 4 2017

If you set min_filelist_kbytes to default, then it is incorrectly reset to 100000 only when logging out and back in, but not when rebooting.
#8 thanks.  For this bug (i.e. in the presence of ARC++) it's not necessary to set min_filelist_kbytes to default.  It happens independently of that.
Status: Started (was: Assigned)
On M62+, start-arc-instance is fired only when the user opts into ARC++ for the first time. When already-opted-in user signs into their Chrome OS user session, 'continue-arc-boot' Upstart signal is sent instead. I'm modifying arc-start-sysctl.conf's start stanza.
Wait... wouldn't it make more sense to pair "start-arc-instance" and "stop-arc-instance" so that they alternate?  You should instead have a separate signal for opt-into-arc (or similar).
Owner: yusukes@chromium.org
Assuming Yusuke is gonna own this since he marked it started.
On M62+, we use 4 signals (We're trying to reduce them to 3 but it's a separate topic):

* start-arc-instance-for-login-screen
* continue-arc-boot
* start-arc-instance
* stop-arc-instance

stop-arc-instance is either paired with start-arc-instance or start-arc-instance-for-login-screen. However, I'd use continue-arc-boot here instead of start-arc-instance-for-login-screen because start-arc-instance-for-login-screen is fired too early (before the user signs in.)

CL: https://chrome-internal-review.googlesource.com/c/chromeos/cheets-scripts/+/520743/1/arc-start-sysctl.conf


Question for Luigi/Justin:
Is it okay to run arc-stop-sysctl.conf without running arc-start-sysctl.conf? If it is okay, the fix I mentioned in comment #11 would work fine.


----------------------------------------------
Details on the Upstart signals.

When signing in, the following happens:

1. When Chrome OS login screen is shown, start-arc-instance-for-login-screen is sent.
2-a. If opt-out user (or guest user) signs into Chrome OS, stop-arc-instance is sent.
2-b. If opt-in user signs into Chrome OS, continue-arc-boot is sent.

When in user session, and ARC++ is not running, if the user opt into ARC++, start-arc-instance is sent.

When in user session, and ARC++ is running, stop-arc-instance is sent if 1) the user opts out from ARC++, or 2) the user signs out from the user session.
----------------------------------------------

Fix is in CQ (and auto test for preventing this from happening again is under review.)
Cc: elijahtaylor@chromium.org bccheng@chromium.org vovoy@chromium.org
+vovoy +bccheng who are working on jank/memory stuff.

Currently, M62+ Chrome OS doesn't have proper /proc/sys/vm/min_filelist_kbytes settings which was introduced in crbug.com/709660 .
"Question for Luigi/Justin:
Is it okay to run arc-stop-sysctl.conf without running arc-start-sysctl.conf? If it is okay, the fix I mentioned in comment #11 would work fine."

Yes, that should be fine.
Project Member

Comment 18 by bugdroid1@chromium.org, Dec 5 2017

The following revision refers to this bug:
  https://chrome-internal.googlesource.com/chromeos/cheets-scripts/+/edf51829d68343898bddebb72f7d4126408c544d

commit edf51829d68343898bddebb72f7d4126408c544d
Author: yusukes <yusukes@google.com>
Date: Tue Dec 05 04:02:52 2017

Cc: gkihumba@chromium.org
Labels: -M-62
Please add merge-request when CL is confirmed in ToT. We should evaluate merging back to M63/M64

ps. removing M-62 since no more releases planned for it


Cc: kbleicher@chromium.org
Labels: Merge-Request-64 Merge-Request-63
+Merge-Request-64 and +Merge-Request-63

I'd like to merge https://chrome-internal.googlesource.com/chromeos/cheets-scripts/+/edf51829d68343898bddebb72f7d4126408c544d# to Chrome OS M64 and M63. It's one-line fix to ARC++ specific Upstart script.

I've confirmed that xbuddy://remote/veyron_minnie/R65-10188.0.0/test (canary) works fine both manually and with auto test (CL:*521358). Luigi also tested the CL manually and it worked on his device.

Project Member

Comment 21 by sheriffbot@chromium.org, Dec 5 2017

Labels: -Merge-Request-63 Merge-Review-63 Hotlist-Merge-Review
This bug requires manual review: Request affecting a post-stable build
Please contact the milestone owner if you have questions.
Owners: cmasso@(Android), cmasso@(iOS), gkihumba@(ChromeOS), govind@(Desktop)

For more details visit https://www.chromium.org/issue-tracking/autotriage - Your friendly Sheriffbot
Project Member

Comment 22 by bugdroid1@chromium.org, Dec 5 2017

Project Member

Comment 23 by bugdroid1@chromium.org, Dec 5 2017

The following revision refers to this bug:
  https://chrome-internal.googlesource.com/chromeos/autotest-cheets/+/cbc8b876a7e449d7ce1cb5311a973c90d0d4a68b

commit cbc8b876a7e449d7ce1cb5311a973c90d0d4a68b
Author: yusukes <yusukes@google.com>
Date: Tue Dec 05 19:00:04 2017

(comment #22 and #23 are for adding auto tests. Please ignore.)
Labels: -Merge-Review-63 Merge-Approved-63
Project Member

Comment 26 by bugdroid1@chromium.org, Dec 5 2017

Labels: merge-merged-release-R63-10032.B
The following revision refers to this bug:
  https://chrome-internal.googlesource.com/chromeos/cheets-scripts/+/97b3a670e2ae432630c478e63c403b6c7f00881a

commit 97b3a670e2ae432630c478e63c403b6c7f00881a
Author: yusukes <yusukes@google.com>
Date: Tue Dec 05 21:59:16 2017

Labels: -Merge-Request-64 Merge-Approved-64
Approving merge to M64 Chrome OS.
Project Member

Comment 28 by bugdroid1@chromium.org, Dec 6 2017

Labels: merge-merged-release-R64-10176.B
The following revision refers to this bug:
  https://chrome-internal.googlesource.com/chromeos/cheets-scripts/+/54328961ccc2daeef9fa9f6a5bb1e3e71196e107

commit 54328961ccc2daeef9fa9f6a5bb1e3e71196e107
Author: yusukes <yusukes@google.com>
Date: Wed Dec 06 01:18:12 2017

Status: Fixed (was: Started)
Fixed. On M63+, min_filelist_kbytes will be 400k even if the (opt-in) user signs in, signs out, then signs in again.

Cc: osh...@chromium.org
CC'ing a few users who reported slowness after upgrading to R62. If the problem shows up again, please do "ctrl-alt-t" to bring up crosh and type "swap" to see the min_filelist_kbytes value. If it says 100000, then it is likely caused by the same issue.
Cc: abodenha@chromium.org cros-perf-detectives@google.com
 Issue 782016  has been merged into this issue.
Project Member

Comment 32 by sheriffbot@chromium.org, Dec 11 2017

Cc: kbleicher@google.com gkihumba@google.com
This issue has been approved for a merge. Please merge the fix to any appropriate branches as soon as possible!

If all merges have been completed, please remove any remaining Merge-Approved labels from this issue.

Thanks for your time! To disable nags, add the Disable-Nags label.

For more details visit https://www.chromium.org/issue-tracking/autotriage - Your friendly Sheriffbot
Labels: -Merge-Approved-63 -Merge-Approved-64

Sign in to add a comment