path: root/open_issues/ext2fs_page_cache_swapping_leak.mdwn

[[!meta copyright="Copyright © 2011, 2012, 2013 Free Software Foundation,
Inc."]]

[[!meta license="""[[!toggle id="license" text="GFDL 1.2+"]][[!toggleable
id="license" text="Permission is granted to copy, distribute and/or modify this
document under the terms of the GNU Free Documentation License, Version 1.2 or
any later version published by the Free Software Foundation; with no Invariant
Sections, no Front-Cover Texts, and no Back-Cover Texts.  A copy of the license
is included in the section entitled [[GNU Free Documentation
License|/fdl]]."]]"""]]

[[!tag open_issue_gnumach open_issue_hurd]]

There is a [[!FF_project 272]][[!tag bounty]] on this task.

[[!toc]]


# IRC, OFTC, #debian-hurd, 2011-03-24

    <youpi> I still believe we have an ext2fs page cache swapping leak, however
    <youpi> as the 1.8GiB swap was full, yet the ld process was only 1.5GiB big
    <pinotree> a leak at swapping time, you mean?
    <youpi> I mean the ext2fs page cache being swapped out instead of simply
      dropped
    <pinotree> ah
    <pinotree> so the swap tends to accumulate unuseful stuff, i see
    <youpi> yes
    <youpi> the disk content, basicallyt :)


# IRC, freenode, #hurd, 2011-04-18

    <antrik> damn, a cp -a simply gobbles down swap space...
    <braunr> really ?
    <braunr> that's weird
    <braunr> why would a copy use so much anonymous memory ?
    <braunr> unless the external pager is so busy that the kernel falls back to
      its default pager
    <youpi> that's what I suggested some time ago
    <braunr> maybe this case should be traced in the kernel
    <braunr> a simple message in the kernel buffer to warn that this condition
      happened may help
    <youpi> I'm seeing swap space being kept used on buildds for no real reason
      except possibly backing ext2fs pages
    <youpi> that could help, yes
    <antrik> youpi: I think it was actually slpz who suggested that...
    <youpi> I think we're generally missing feedback from memory behavior
    <antrik> youpi: do you think andrei's kernel instrumentation work might be
      helpful with analyzing such things?
    <youpi> antrik: I think I suggested it too, but never mind
    <youpi> antrik: no, because it's not a trace of events that you want
    <youpi> some specific events would be useful
    <youpi> but then we don't really need a whole framework for that
    <antrik> apt-get upgrade eats swap too
    <youpi> the upgrade itself, or the computation of the ugprade?
    <youpi> apt is a memory eater nowadays
    <antrik> installing the packages
    <antrik> seems to have stabilized though after a while...
    <antrik> so perhaps it's not a leak in this case
    <youpi> ideally we should have a way to know what was put in the swap
    <braunr> how would you represent what's in the swap ?
    <antrik> the apt-get process has 46M of virtual memory above the 128 M
      baseline
    <braunr> mostly libraries i guess
    <braunr> are trheads stacks 8 MiB like on Linux ?
    <youpi> braunr: at least knowing how much of each process is in the swap
    <youpi> braunr: 2MiB
    <braunr> ok
    <youpi> vminfo could also report which parts of the address space are in
      the swap
    <antrik> youpi: would be nice to have some simple utility reporting how
      much of a process' address space is anonymous
    <antrik> (in fact, I wonder why it's not reported by standard tools such as
      ps or top... this shouldn't be too difficult I would think?)
    <antrik> it would be much more useful information than the total virt size,
      which includes rather meaningless disk and device mappings...
    <youpi> agreed
    <braunr> well
    <braunr> there are tools like pmap for this
    <braunr> unfortunately, it's difficult in mach to know what backs a
      non-anonymous mapping
    <braunr> pagers should be able to name their mappings
    <youpi> that'd be helpful for debugging yes
    <braunr> there is almost no overhead in doing that, and it would be very
      useful
    <youpi> and could lead to /proc/pid/maps
    <braunr> yes
    <braunr> isn't there a maps already ?
    <youpi> nope
    <braunr> ok
    <youpi> (probably not very useful without the names)
    <braunr>  ithought i remembered maps without names, and guessed it might
      have been on the hurd for that reason
    <braunr> but i'm not sure
    <youpi> there's the vminfo command, yes
    <braunr> 14:06 < youpi> braunr: at least knowing how much of each process
      is in the swap
    <braunr> wouldn't it be clearer to do it the other way around ?
    <braunr> like a swapinfo tool indicating what it contains ?
    <youpi> sure, but it's a lot more difficult
    <braunr> really ?
    <braunr> why ?
    <youpi> because you have to traverse all the mappings
    <youpi> etc
    <youpi> (in all processes, I mean)
    <youpi> and you have to name what is waht
    <braunr> there are other ways
    <braunr> the swap is a central structure
    <youpi> while simply introducing the swap %  in vminfo
    <youpi> for a given process you know what is what
    <braunr> right
    <youpi> and doing that introduction is  probably very simple
    <braunr> that's a good point
    <braunr> top-down is effectively easier than bottom-up resolution in Mach
      VM
    <antrik> hm... the memory use caused by cp doesn't seem to be reflected in
      the virtual size of any particular process
    <antrik> ghost memory
    <braunr> what's cp vmsize at the time of the problem ?
    <antrik> it's at 134 M right now... so considering the 128 M baseline,
      nothing worth speaking of
    <braunr> right
    <braunr> maybe a copy map during I/O
    <braunr> but I don't know Mach copy maps in detail, as they have been
      eliminated from UVM
    <antrik> BTW, the memory eatup happens even before swap comes into
      play... swapping seems to be a result of the problem, not the cause
    <braunr> what do you mean ?
    <braunr> I thought swapping was the issue
    <braunr> you mean RAM is full before swapping ?
    <antrik> well, I don't know what the actual problem is... I just don't
      understand why the memory use increases without any particular process
      seeing an increase in size
    <antrik> the "free" size in vmstat decreses
    <antrik> once it's eatun up, swap space use increases
    <braunr> well it doesn't change much of it
    <braunr> the anonymous memory pager will use RAM before resorting to the
      external default-pager
    <antrik> I would suspect normal block caching... but then, shouldn't this
      show up in the memory info of the ext2 process?
    <braunr> although, again, I'm not sure of the behaviour of the anonymous
      memory pager
    <braunr> antrik: I don't know how block caching behaves
    <antrik> BTW, is it a know problem that doing ^C on a "cp -a" seems to hang
      the whole system?...
    <antrik> (the whole hurd instance that is... the other instance is not
      affected)
    <youpi> not that I know of
    <braunr> seems like a deadlock in the anonymous memory handling
    <youpi> (and I've never seen that)
    <antrik> happens both in my main system (using ancient hurd/libc) and in my
      subhurd (recently upgraded to current stuff)
    <antrik> this make testing this stuff quite a lot harder... [sigh]
    <antrik> any suggestions how to debug this hang?
    <braunr> antrik: no :/

2011-04-28: [[!taglink open_issue_documentation]]

    <antrik> hm... is it normal that "swap free" doesn't increase as a process'
      memory is paged back in?
    <youpi> yes
    <youpi> there's no real use cleaning swap
    <youpi> on the contrary, it makes paging the process out again longer
    <antrik> hm... so essentially, after swapping back and forth a bit, a part
      of the swap equal to the size of physical RAM will be occupied with stuff
      that is actually in RAM?
    <youpi> yes
    <youpi> so that that RAM can be freed immediately if needed
    <antrik> hm... that means my effective swap size is only like 300 MB... no
      wonder I see crashes under load
    <antrik> err... make that 230 actually
    <antrik> indeed, quitting the application freed both the physical RAM and
      swap space
    <braunr> 02:28 < antrik> hm... is it normal that "swap free" doesn't
      increase as a process' memory is paged back in?
    <braunr> swap is the backing store of anonymous memory, like ext2fs is the
      backing store of memory objects created from its pager
    <braunr> so you can view swap as the file system for everything that isn't
      an external memory object


# IRC, freenode, #hurd, 2011-11-15

    <braunr> hm, now my system got unstable
    <braunr> swap is increasing, without any apparent reason
    <antrik> you mean without any load?
    <braunr> with load, yes
    <braunr> :)
    <antrik> well, with load is "normal"...
    <antrik> at least for some loads
    <braunr> i can't create memory pressure to stress reclaiming without any
      load
    <antrik> what load are you using?
    <braunr> ftp mirrorring
    <antrik> hm... never tried that; but I guess it's similar to apt-get
    <antrik> so yes, that's "normal". I talked about it several times, and also
      wrote to the ML
    <braunr> antrik: ok
    <antrik> if you find out how to fix this, you are my hero ;-)
    <braunr> arg :)
    <antrik> I suspect it's the infamous double swapping problem; but that's
      just a guess
    <braunr> looks like this
    <antrik> BTW, if you give me the exact command, I could check if I see it
      too
    <braunr> i use lftp (mirror -Re) from a linux git repository
    <braunr> through sftp
    <braunr> (lots of small files, big content)
    <antrik> can't you just give me the exact command? I don't feel like
      figuring it out myself
    <braunr> antrik: cd linux-stable; lftp sftp://hurd_addr/
    <braunr> inside lftp: mkdir linux-stable; cd linux-stable; mirror -Re
    <braunr> hm, half of physical memory just got freed
    <braunr> our page cache is really weird :/
    <braunr> (i didn't delete any file when that happened)
    <antrik> hurd_addr?
    <braunr> ssh server ip address
    <braunr> or name
    <braunr> of your hurd :)
    <antrik> I'm confused. you are mirroring *from* the Hurd box?
    <braunr> no, to it
    <antrik> ah, so you login via sftp and then push to it?
    <braunr> yes
    <braunr> fragmentation looks very fine
    <braunr> even for the huge pv_entry cache and its 60k+ entries
    <braunr> (and i'm running a kernel with the cpu layer enabled)
    <braunr> git reset/status/diff/log/grep all work correctly
    <braunr> anyway, mcsim's branch looks quite stable to me
    <antrik> braunr: I can't reproduce the swap leak with ftp. free memory
      idles around 6.5 k (seems to be the threshold where paging starts), and
      swap use is constant
    <antrik> might be because everything swappable is already present in swap
      from previous load I guess...
    <antrik> err... scratch that. was connected to the wrong host, silly me
    <antrik> indeed swap gets eaten away, as expected
    <antrik> but only if free memory actually falls below the
      threshold. otherwise it just oscillates around a constant value, and
      never touches swap
    <antrik> so this seems to confirm the double swapping theory
    <youpi> antrik: is that "double swap" theory written somewhere?
    <youpi> (no, a quick google didn't tell me)


## IRC, freenode, #hurd, 2011-11-16

    <antrik> youpi:
      http://lists.gnu.org/archive/html/l4-hurd/2002-06/msg00001.html talks
      about "double paging". probably it's also the term others used for it;
      however, the term is generally used in a completely different meaning, so
      I guess it's not really suitable for googling either ;-)
    <antrik> IIRC slpz (or perhaps someone else?) proposed a solution to this,
      but I don't remember any details
    <youpi> ok so it's the same thing I was thinking about with swap getting
      filled
    <youpi> my question was: is there something to release the double swap,
      once the ext2fs pager managed to recover?
    <antrik> apparently not
    <antrik> the only way to free the memory seems to be terminating the FS
      server
    <youpi> uh :/


# IRC, freenode, #hurd, 2011-11-30

    <antrik> slpz: basically, whenever free memory goes below the paging
      threshold (which seems to be around 6 MiB) while there is other I/O
      happening, swap usage begins to increase continuously; and only gets
      freed again when the filesystem translator in question exits
    <antrik> so it sounds *very* much like pages go to swap because the
      filesystem isn't quick enough to properly page them out
    <antrik> slpz: I think it was you who talked about double paging a while
      back?
    <slpz> antrik: probably, sounds like me :-)
    <antrik> slpz: I have some indication that the degenerating performance and
      ultimate hang issues I'm seeing are partially or entirely caused by
      double paging...
    <antrik> slpz: I don't remember, did you propose some possible fix?
    <slpz> antrik: hmm... perhaps it wasn't me, because I don't remember trying
      to fix that problem...
    <slpz> antrik: at which point do you think pages get duplicated?
    <antrik> slpz: it was a question. I don't remember whether you proposed
      something or not :-)
    <antrik> slpz: basically, whenever free memory goes below the paging
      threshold (which seems to be around 6 MiB) while there is other I/O
      happening, swap usage begins to increase continuously; and only gets
      freed again when the filesystem translator in question exits
    <antrik> so it sounds *very* much like pages go to swap because the
      filesystem isn't quick enough to properly page them out
    <slpz> antrik: I see
    <slpz> antrik: I didn't addressed this problem directly, but when I've
      modified the pageout mechanism to provide a special treatment for
      external pages, I also removed the possibility of sending them to the
      default pager
    <slpz> antrik: this was in my experimental environment, of course
    <antrik> slpz: oh, nice... so it may fix the issues I'm seeing? :-)
    <antrik> anything testable yet?
    <slpz> antrik: yes, only anonymous memory could be swapped with that
    <slpz> antrik: it works, but is ugly as hell
    <antrik> tschwinge: these is also your observation about compilations
      getting slower on further runs, and my followups... I *suspect* it's the
      same issue

[[performance/degradation]].

    <slpz> antrik: I'm thinking about establishing a repository for these
      experimental versions, so they don't get lost with the time
    <antrik> slpz: please do :-)
    <slpz> antrik: perhaps in savannah's HARD project
    <antrik> even if it's not ready for upstream, it would be nice if I could
      test it -- right now it's bothering me more than any other Hurd issues I
      think...
    <slpz> also, there's another problem which causes performance degradation
      with the simple use of the system
    <tschwinge> slpz: Please just push to Savannah Hurd.  Under your
      slpz/... or similar.
    <tschwinge> antrik: Might very well be, yes.
    <slpz> and I almost sure it is the fragmentation of the task map
    <slpz> tschwinge: ok
    <slpz> after playing a bit with a translator, it can easily get more than
      3000 entries in its map
    <antrik> slpz: yeah, other issues might play a role here as well. I
      observed that terminating the problematic FS servers does free most of
      the memory and remove most of the performance degradation, but in some
      cases it's still very slow
    <slpz> that makes vm_map_lookup a lot slower
    <antrik> on a related note: any idea what can cause paging errors and a
      system hang even when there is plenty of free swap?
    <antrik> (I'm not entirely sure, but my impression is that it *might* be
      related to the swap usage and performance degradation problems)
    <slpz> I think this degree of fragmentation has something to do with the
      reiterative mapping of memory objects which is done in pager-memcpy.c
    <slpz> antrik: which kind of paging errors?
    <antrik> hm... I don't think I ever noted down the exact message; but I
      think it's the same you get when actually running out of swap
    <slpz> antrik: that could be the default pager dying for some internal bug
    <antrik> well, but it *seems* to go along with the performance degradation
      and/or swap usage
    <slpz> I also have the impression that we're using memory objects the wrong
      way
    <antrik> basically, once I get to a certain level of swap use and slowness
      (after about a month of use), the system eventually dies
    <slpz> antrik: I never had a system running for that time, so it could be a
      completely different problem from what I've seen before :-/
    <slpz> Anybody has experience with block-level caches on microkernel
      environments?
    <antrik> slpz: yeah, it typically happens after about a month of my normal
      use... but I can significantly accellerate it by putting some problematic
      load on it, such as large apt-get runs...
    <slpz> I wonder if it would be better to put them in kernel or in user
      space. And in the latter, if it would be better to have one per-device
      shared for all accesing translators, or just each task should have its
      own cache...
    <antrik> slpz:
      http://lists.gnu.org/archive/html/bug-hurd/2011-09/msg00041.html is where
      I described the issue(s)
    <antrik> (should send another update for the most recent findings I
      guess...)
    <antrik> slpz: well, if we move to userspace drivers, the kernel part of
      the question is already answered ;-)
    <antrik> but I'm not sure about per-device cache vs. caching in FS server