Browser crash files are generated in system crash directory when/after Autotest test is being ran |
|||||
Issue description
Some time ago in some long tests we noticed chrome(process) crash files show under system crash directory /var/spool/crash - issue 590258.
Looking into reproducing browser crash on test image I see same generated crash files at the same location. URLs to generate the crash:
- chrome://crash
- chrome://inducebrowsercrashforrealz
1) The intended behavior should be to generate the following crash files in the user crash dir /home/chronos/u*/crash:
- chrome.*.dmp
- chrome.*.meta
- chrome.*.chrome.txt.gz
- chrome.*.i915_error_state.log.xz(for intel gfx boards)
and it is true for RECOVERY build imaged devices.
2) The observed behavior for TEST build image is crash files are generated in the system crash directory /var/spool/crash, and following system process crash file types:
- chrome.*.core
- chrome.*.dmp
- chrome.*.log
- chrome.*.meta
Few questions:
- Is this WAI?
- Can we assume correct crash files generation is expected for recovery image 1), having the chrome process crash files are generated for test image 2) devices?
- Can the intended browser crash files 1) be generated for test image build too?
Some logs seen from testing with cyan board:
2016-09-13T11:24:13.267728-07:00 WARNING crash_reporter[13341]: [user] Received crash notification for chrome[13025] sig 11, user 1000 (developer build - not testing - always dumping)
2016-09-13T11:24:13.270696-07:00 WARNING crash_reporter[13343]: Received crash notification for chrome[13025] user 1000 (called directly)
2016-09-13T11:24:13.271099-07:00 INFO crash_reporter[13341]: State of crashed process [13025]: S (sleeping)
2016-09-13T11:24:13.277354-07:00 ERR crash_reporter[13347]: Could not create /var/spool/crash/chrome.20160913.112413.13025.chrome.txt: Permission denied
2016-09-13T11:24:13.278390-07:00 INFO crash_reporter[13343]: Log command " for f in $(ls -1rt /var/log/chrome/chrome_[0-9]* | tail -2) $(ls -1rt /home/chronos/u-*/log/chrome_[0-9]* 2>/dev/null | tail -2); do echo "===$f (tail)==="; tail -20 $f; echo EOF; echo; done; echo "===session_manager (tail)==="; awk '$3 ~ "^session_manager\[" { print }' /var/log/messages | tail -20; echo EOF" exited with 127
2016-09-13T11:24:13.320749-07:00 ERR crash_reporter[13343]: Could not write file /var/spool/crash/chrome.20160913.112413.13025.i915_error_state.log.xz Written: -1 Len: 84
2016-09-13T11:24:13.320792-07:00 ERR crash_reporter[13343]: Unable to write /var/spool/crash/chrome.20160913.112413.13025.meta
2016-09-13T11:24:15.834937-07:00 INFO crash_reporter[13341]: Stored minidump to /var/spool/crash/chrome.20160913.112413.13025.dmp
2016-09-13T11:24:15.835243-07:00 INFO crash_reporter[13341]: Leaving core file at /var/spool/crash/chrome.20160913.112413.13025.core due to developer image
-------------------------------
crosh> shell
chronos@localhost / $ ls ~/crash
ls: cannot access /home/chronos/user/crash: No such file or directory
chronos@localhost / $ ls /var/spool/crash
chrome.20160913.112413.13025.core chrome.20160913.112413.13025.dmp chrome.20160913.112413.13025.log chrome.20160913.112413.13025.meta
,
Sep 13 2016
no, this isn't WAI. do you only see this on test images ? are you running as guest ? did you run autotests against the system which might have messed with the settings ? can you reproduce on a normal/dev image ? i just verified over here that it's working correctly: Version 53.0.2785.36 dev Platform 8530.35.0 (Official Build) dev-channel veyron_minnie-cheets 2016-09-13T16:14:16.654575-04:00 WARNING crash_reporter[28907]: [user] Received crash notification for chrome[28878] sig 11, user 1000 (ignoring call by kernel - chrome crash; waiting for chrome to call us directly) 2016-09-13T16:14:16.668314-04:00 WARNING crash_reporter[28909]: Received crash notification for chrome[28878] user 1000 (called directly) 2016-09-13T16:14:23.575203-04:00 NOTICE crash_sender[29021]: Considering metadata /home/chronos/u-68c0446c02d481dc89f9f7fce552fe7697bc3ae5/crash/chrome.20160913.161416.28878.meta. 2016-09-13T16:14:46.932112-04:00 NOTICE crash_sender[29252]: Sending crash: 2016-09-13T16:14:46.934186-04:00 NOTICE crash_sender[29253]: Sending crash report on behalf of Chrome_ChromeOS 2016-09-13T16:14:46.936299-04:00 NOTICE crash_sender[29254]: Metadata: /home/chronos/u-68c0446c02d481dc89f9f7fce552fe7697bc3ae5/crash/chrome.20160913.161416.28878.meta (minidump) 2016-09-13T16:14:46.937955-04:00 NOTICE crash_sender[29255]: Payload: /home/user/68c0446c02d481dc89f9f7fce552fe7697bc3ae5/crash/chrome.20160913.161416.28878.dmp 2016-09-13T16:14:46.939397-04:00 NOTICE crash_sender[29256]: Version: 53.0.2785.36 2016-09-13T16:14:46.941062-04:00 NOTICE crash_sender[29257]: Exec name: chrome 2016-09-13T16:14:46.960349-04:00 INFO list_proxies[29258]: Resolving proxies for URL: https://clients2.google.com/cr/report 2016-09-13T16:14:46.966955-04:00 INFO list_proxies[29258]: Found proxies via browser signal: direct:// 2016-09-13T16:14:48.506579-04:00 NOTICE crash_sender[29267]: Crash report receipt ID 3e791f9500000000 2016-09-13T16:14:48.512803-04:00 NOTICE crash_sender[29269]: Successfully sent crash /home/chronos/u-68c0446c02d481dc89f9f7fce552fe7697bc3ae5/crash/chrome.20160913.161416.28878.meta and removing.
,
Sep 13 2016
No guest mode. Reproduced ONLY on test image for normal and dev mode, with and without autotest being invoked by tests. For the default test account, and for any consumer gmail account, with enabled or disabled crash reports collection. It might be the case where the consistent behavior started after Autotest has been invoked by tests being run on the device at my desk.I am re-imaging with a USB stick will check again.
,
Sep 13 2016
Fact is that after it takes place, I can't get the device to generate the right files, even with transition between modes(dev and normal) on the test image build.
,
Sep 13 2016
After reimaging with USB stick, I get to see the correct files generated. chronos@localhost / $ ls /var/spool/crash /home/chronos/u*/crash ls: cannot access /var/spool/crash: No such file or directory /home/chronos/u-5332aedf3dbe904a30533cd2b0f83a075cd9132c/crash: chrome.20160913.141814.1351.chrome.txt.gz chrome.20160913.141814.1351.i915_error_state.log.xz chrome.20160913.141814.1351.dmp chrome.20160913.141814.1351.meta Can anything be done to get these generated for browser crash while/after Autotest is being ran?
,
Sep 13 2016
probably. you'd have to narrow down what steps you performed before things started failing. if autotest modifies the rootfs during its tests, then it's not really surprising that things start misbehaving after that, and it's not directly recoverable.
,
Sep 13 2016
Yes, it is the case of when a device runs tests with Autotest, browser crash files generation changes to process file generation. +infra folks who might have idea about this. In short - Running Autotest on a device with test image makes crash_reporter output crash files(and the not intended ones) in system crash folder, instead of user crash folder. Afterwards this is consistent behavior, and does not change for the device no matter under test or not.
,
Sep 20 2016
Checked with my team, and we're not sure this is an infra issue. Can you clarify?
,
Sep 21 2016
It is undesired effect Autotest having over DUT behavior. Is there component or label to track, and respectively owner for such issue?
,
Sep 21 2016
AFAIK, this sort of problem belongs to "OS>Systems". I'm less certain of the ownership details, but I understand that that charter belongs to adlr@.
,
Dec 6 2016
,
Aug 17
|
|||||
►
Sign in to add a comment |
|||||
Comment 1 by ka...@chromium.org
, Sep 13 2016