1386547 - Dead firefox56/ -contentproc -childID 1 -isForBrowser -intPrefs consuming loads of CPU time.

Status: RESOLVED INCOMPLETE

(Remote Protocol :: Marionette, defect, P5)

Description

[wrs310]

User Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/59.0.3071.115 Safari/537.36

Steps to reproduce:

Confirmed happening on: Ubuntu 17.04 64-bit and Debian 9. Firefox nightly without any add-ons.

From Firefox 54 and higher automation tests (marionette driver) leave dead child processes? behind that use large amounts of cpu time. I run about 20 concurrent browser sessions, and after a few hours these child processes that have no firefox window anymore accumulate and utlize 100% cpu together.

"/root/Desktop/firefox56/firefox -contentproc -childID 1 -isForBrowser" 


Actual results:

Below a ps auxf output. Note the cpu time for these processes, 355:02, 263:10, 255:52 etc. While a normal instance runs on max 0:05!! cpu time before finishing. These processes are also not shut down after the test is finished. They have no GUI anymore and I have no clue what they are doing, each eating 15% cpu on an i7 7700k. This is happening since firefox 54 onward, or atleast on firefox 53 this issue wasn't there.

root      8082 83.0  0.9 1977936 301492 ?      Rl   02:25 355:02 /root/Desktop/firefox56/firefox -contentproc -childID 1 -isForBrowser -intPrefs
root     21621 76.4  0.8 1895124 271092 ?      Sl   03:48 263:10 /root/Desktop/firefox56/firefox -contentproc -childID 1 -isForBrowser -intPrefs
root      6320 75.8  0.7 1938688 256492 ?      Sl   03:55 255:52 /root/Desktop/firefox56/firefox -contentproc -childID 1 -isForBrowser -intPrefs
root     23590 71.5  0.9 1911844 309656 ?      Sl   04:53 200:25 /root/Desktop/firefox56/firefox -contentproc -childID 1 -isForBrowser -intPrefs
root     18066 78.5  0.9 1897044 297888 ?      Rl   05:03 211:42 /root/Desktop/firefox56/firefox -contentproc -childID 1 -isForBrowser -intPrefs
root      2176 66.8  0.7 1903716 248028 ?      Sl   07:35  78:51 /root/Desktop/firefox56/firefox -contentproc -childID 1 -isForBrowser -intPrefs




Expected results:

Firefox child processes? (I dont know the architecture of firefox) should always be killed when the main firefox process is stopped, also when it crashes.

Comment 1

[Henrik Skupin [:whimboo][⌚️UTC+1]]

What kind of tests are you running? Maybe you can give us some more details in how to get this reproduced.

Also it means you cannot see it with Firefox 53? If that is the case, would you mind checking the available beta releases of Firefox 54 to find out which version is affected first? That would be a great help.

Comment 2

[wrs310]

(In reply to Henrik Skupin (:whimboo) from comment #1)
> What kind of tests are you running? Maybe you can give us some more details
> in how to get this reproduced.

The tests can be classified as normal browsing using, right now I do not have a realible way for other people to reproduce this. I will see if I can come up with something. It is however not trivial since the only knowledge I have is that one of these processes show up maybe every hour, which is roughly ~1000 brower sessions (browser start, run test, browser close). And this definitly did not happen using firefox 53 and before 47 (webdriver) and earlier. 

I now have a hotfix which is a script that kills any firefox related processes if the cputime > 10. Before after a day or two of running tests, it was impossible to even ssh to the server because these processes were totally freezing the system. 

(In reply to Henrik Skupin (:whimboo) from comment #1)
> would you mind checking the available beta releases of Firefox 54 to find out
> which version is affected first? That would be a great help.

I can do this but the results will take a while.

Comment 3

[David Burns :automatedtester]

It would be good if you could give steps to reproduce. At the moment there is not enough information to make this an actionable bug.

Comment 4

[Henrik Skupin [:whimboo][⌚️UTC+1]]

No response from the reporter.