platform_CryptohomeStress: suspends timing out and affected by fio parameters |
|||||
Issue descriptionWhat steps will reproduce the problem? (1) Run the edited platform_CryptohomeStress.surfing Changed files here: https://chromium-review.googlesource.com/#/c/399893/ to fix some of issue 646626 . What is the expected result? The tests will pass. What happens instead? The power_SuspendStress part of the test either fails with ERROR: Sanity check failed: did not try to suspend, or passes but gets a warning that most of the iterations had non-fatal suspend failures due to timeouts. This only seems to occur when the platform_CryptohomeFio part of the test is given 'crypto' for the disk_configs parameter. The actual mounting and unmounting of the cryptohome vault does not seem to be the problem, since changing the tmpdir from the user path to the tmpdir used with the other parameters seems to get around the problem. It seems like the platform_CryptohomeFio test is interfering with the power_SuspendStress test it uses the user path but the assumption is that it will just run between suspends regardless of the tmpdir.
,
Feb 25 2017
The wake even received is the alarm for the suspend, so it does appear that when using crypto it just takes too long to fully suspend, seeming to take a lot of time to sync. Increasing the timeout for the call to /usr/bin/powerd_dbus_suspend did not work, and it seems like this is because the increased timeout was not actually changing the alarm, so the wake event was still received before the suspend could complete and before the increased timeout time. I have tried increasing the actual suspend time, paired with an increased timeout and it seems promising. Even though changing the suspend time looks like it's working, it changes the timing of the overall stress test a bit since the fio test is only running when the device is not suspended. I'm trying to see how much the suspend actually needs to increase and trying to figure out the least intrusive way to make these adjustments.
,
Feb 28 2017
Did changing the vm diry ratio and associated knobs not help?
,
Feb 28 2017
It looked like it was helping at first, but as I tried to find the best place to set it, it stopped working. Even when I set it back down to 1000 the suspends all timed out. I will see if changing the associated values helps at all.
,
Feb 28 2017
I would suggest continuing to go down the tweaking vm dirty thresholds & pdflush times since then the test itself needs to be changed and nothing outside of it (such as powerd) needs to. My only worry would be whether in fact we do need to make changes to powerd's timeouts since something underlying has changed sufficiently. Gwendal, your thoughts?
,
Mar 28 2017
Note that for tpm 2.0, trunksd now performs actions on suspend (SuspendImminent from powerd), which are required for correct cryptohomed operation after resume. So, I believe, this test should allow enough time for suspend.
,
May 3 2017
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/third_party/autotest/+/e5b5fa849448a63cf1b2ccd55df8b078037d65d8 commit e5b5fa849448a63cf1b2ccd55df8b078037d65d8 Author: Alexis Savery <asavery@chromium.org> Date: Wed May 03 03:24:17 2017 platform_CryptohomeStress: fix fio and suspend stress interaction Increase min_suspend to provide additional time to sync and suspend. Make dirty_ratio 1 for the duration of the test to reduce the time needed to sync. BUG= chromium:668326 TEST=Ran platform_CryptohomeStress and platform_CryptohomeStress.surfing Change-Id: I5e199e1fca5c56fca6fc630482bb7e3f13e7e43d Reviewed-on: https://chromium-review.googlesource.com/479677 Commit-Ready: Alexis Savery <asavery@chromium.org> Tested-by: Alexis Savery <asavery@chromium.org> Reviewed-by: Gwendal Grignou <gwendal@chromium.org> [modify] https://crrev.com/e5b5fa849448a63cf1b2ccd55df8b078037d65d8/client/site_tests/platform_CryptohomeStress/control [modify] https://crrev.com/e5b5fa849448a63cf1b2ccd55df8b078037d65d8/client/site_tests/platform_CryptohomeStress/platform_CryptohomeStress.py [modify] https://crrev.com/e5b5fa849448a63cf1b2ccd55df8b078037d65d8/client/site_tests/platform_CryptohomeStress/control.surfing
,
Aug 1
,
Sep 28
Triage nag: This Chrome OS bug has an owner but no component. Please add a component so that this can be tracked by the relevant team.
,
Oct 12
,
Dec 10
The test is still a little flaky but appears to be for unrelated reasons - marking this fixed |
|||||
►
Sign in to add a comment |
|||||
Comment 1 by asavery@chromium.org
, Feb 1 20177.8 KB
7.8 KB Download
18.6 MB
18.6 MB Download
56.1 KB
56.1 KB Download