[Bug malloc/21273] insufficient memory with more then 10 GB free memory

["adhemerval.zanella at linaro dot org" <sourceware-bugzilla at sourceware dot org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=21273

--- Comment #7 from Adhemerval Zanella <adhemerval.zanella at linaro dot org> ---
(In reply to dieter ferdinand from comment #6)
> hello,
> i make some tests on my testsystem and use 5 gb with chromium and this
> system has 7 gb usable memory and no swap.
> 
> on my master server, i make the same test and i get some out of memory
> messages and the tool i need reports max 4,5 GB used ram. i restart firefox
> and it crashed.
> and same other programs to.
> this system have 16GB ram and 8 GB swap.
> if i try to reserve 5gb memory with memtester i get out of memory and i have
> more then 10gb available ram and mor the 7gb free swapspace.
> 
> i can't find out why i can use on my masterserver only 25-30% of the
> physical ram.
> 
> i make the tests on both systems with 64bit kernel 4.9.23 and on the
> test-system with a 32bit kernel 3.18.47. i use on both system chromium 32
> bit.
> 
> i have opened on my master-system many tabs in chromium, exit it and start
> it again and it is crashed.
> 
> if i open chromium and open many tabs, no problem. but if i start chromium
> and it restore the tabs -> out of memory, can't fork -> chromium crash!
> 
> only on my testsystem chromium don't crash. but i can make some tests again.
> 
> i try to reserve memory with memtester and i can reserve 4,5 GB on my server
> so i use a little more as 8 gb. but this is impossible with other
> applications. i get always out of memory or resource not available, can't
> fork. i have less then 650 processes running. on testsystem less then 250
> processes.
> 
> but i can't make such tests on my server because it can kill importent
> programms.
> 
> i make the tests on identical systems, i copy the system from my server to
> the system. only in etc are differences because i don't want it to start the
> same processes as my server and other processes can't be started with the
> configuration of the server.
> 
> if i was not crazy, this will drive me crazy!
> 
> why can i use on a system with 8 gb physical ram and 7 gb usable ram more
> memory as on a system with 16GB physical and usable ram ?
> 
> goodby

No idea based on the information you provided so far.  As Carlos stated on
comment #6 we need some more technical data on how to actually link these
issues with glibc itself.  It could be:

  * A dump analysis by either attaching gdb and dumping the core.
  * A coredump due the aforementioned memory exhaustion along with the binary.
  * A testcase that stress such issue linking it to a glibc interface.

Otherwise it can countless issues with the vague technical information you
provided.  Also keep in mind we do not have much developers working on glibc
side, so it would be unlikely if someone will actually pursuit to check out
this issue if you do not provide a way to *actually* reproduce it.  

Just dumping random information about some experiments is not the correct way
to open a bug against glibc.  Check out other glibc bug reports to have a grasp
what kind of information you will need to provide us.

-- 
You are receiving this mail because:
You are on the CC list for the bug.