Re: Memory Management using Tomcat

From: Joe Baldwin (jfbaldwi..arthlink.net)
Date: Wed Sep 16 2009 - 16:39:34 EDT

  • Next message: Mike Kienenberger: "Re: Memory Management using Tomcat"

    Mike,

    I looked through your code for the filter and have a few questions.

    1. How does this filter differ from the default Cayenne filter?
    2. It appears that your filter is doing a similar task to Cayenne
    filter. The docs say:
            A Servlet Filter that binds session DataContext to the current
    request thread. During the request application code
            without any knowledge of the servlet environment can access DataContext
            viaDataContext.getThreadDataContext() method.

            Is this correct?
    3. I have never written a filter and want to make sure that this is
    the problem before attempting this kind of change.

    It appears that your strategy is to create a new DataContext (or
    BaseContext) for each session. If so, then I am confused because I
    thought that the new Memory Management strategy for Cayenne 3.0 was to
    avoid having to do this.

    Am I missing something about the basic usage of Cayenne?

    Thanks,
    Joe

    On Sep 16, 2009, at 3:37 PM, Mike Kienenberger wrote:

    > I think there's a default filter provided by Cayenne you can specify
    > in your config file, but it really comes down to something as simple
    > as this to make it per request:
    >
    > public void doFilter(ServletRequest servletRequest,
    > ServletResponse servletResponse, FilterChain chain)
    > throws IOException, ServletException
    > {
    > // set base context on thread
    >
    > chain.doFilter(servletRequest, servletResponse);
    >
    > // remove base context from thread
    > }
    >
    > Attaching a simple one I wrote a long time ago that perserves the
    > DataContext across a session (but insures it's in a clean state at the
    > end of each request).
    >
    >
    > On Wed, Sep 16, 2009 at 3:29 PM, Joe Baldwin
    > <jfbaldwi..arthlink.net> wrote:
    >> Mike,
    >>
    >> RE BaseContext.getThreadObjectContext()
    >>
    >> Of course, this could be my problem. I was using DataContext until
    >> 3.0 then
    >> converted over to BaseContext.
    >>
    >>
    >>> 2.) BaseContext.getThreadObjectContext() just tells how you're
    >>> getting a context. It doesn't tell how it's managed. Do you
    >>> have a
    >>> servlet filter that creates a new ObjectContext at the start of a
    >>> request and then clears it out at the end? Or does it do this per
    >>> session?
    >>
    >> I do not know how it is being initialized in the webapp (in my
    >> experimental
    >> non-web apps I explicitly intitialize it, but in the webapp it is
    >> already
    >> initialized).
    >>
    >>
    >>> I've never looked at the method, but it might default to
    >>> creating one permanent data context per thread if you don't do
    >>> anything else (like set up a servlet filter). That could be part
    >>> of
    >>> the problem.
    >>
    >>
    >> That sounds plausible. I could not find an example of how to
    >> initialize
    >> this and manage it as you suggest in a web app. I have not created a
    >> servlet filter but have follow the Cayenne docs for configuration
    >> of the
    >> web.xml.
    >>
    >> If this is insufficient then I agree, this could be the problem.
    >> Unfortunately, I have not found docs on how to accomplish what you
    >> are
    >> recommending.
    >>
    >> Joe
    >>
    >>
    >>
    >>
    >> On Sep 16, 2009, at 3:10 PM, Mike Kienenberger wrote:
    >>
    >>> 1.) 128 still seems small to me. I don't think I run anything at
    >>> less than 256.
    >>> On the other hand, We have an app with 1000s of customers that uses
    >>> 512Mb, I think. So 1500 seems excessive.
    >>>
    >>> 2.) BaseContext.getThreadObjectContext() just tells how you're
    >>> getting a context. It doesn't tell how it's managed. Do you
    >>> have a
    >>> servlet filter that creates a new ObjectContext at the start of a
    >>> request and then clears it out at the end? Or does it do this per
    >>> session? I've never looked at the method, but it might default to
    >>> creating one permanent data context per thread if you don't do
    >>> anything else (like set up a servlet filter). That could be part
    >>> of
    >>> the problem.
    >>>
    >>> On Wed, Sep 16, 2009 at 2:49 PM, Joe Baldwin <jfbaldwi..arthlink.net
    >>> >
    >>> wrote:
    >>>>
    >>>> Caveat: Apparently I am not as well. :)
    >>>>
    >>>> 1.) I looked at the 65M issue. On my development box (OSX) I set
    >>>> it to
    >>>> -Xms128m -Xmx128m (basically arbitrary). So when I went to the
    >>>> remote
    >>>> hosting company, I purchase a similar amount.
    >>>> a. However, we are doing *very* little work (lots of product
    >>>> fetches
    >>>> and only a few product inserts and updates) and it runs out of
    >>>> memory
    >>>> *very*
    >>>> fast which I *assume* means it is my code, but I don't know.
    >>>> b. I am doing research and here is a "web
    >>>> recommendation" (for all
    >>>> that is worth)
    >>>> Whenever possible, Unidata recommends
    >>>> -Xmx1500m for 32-bit systems, and -Xmx2048m --Xmx4096m for
    >>>> 64-bit
    >>>> systems.
    >>>> c. Since the host is 64-bit, I am wondering whether my
    >>>> assumptions
    >>>> may be off for 64-bit systems.
    >>>>
    >>>> 2. DataContext: Sorry, but I am getting confused on this one. I
    >>>> am using
    >>>> BaseContext.getThreadObjectContext() based on recommendations (I
    >>>> converted
    >>>> all the old DataContext refs to BaseContext, but I don't really
    >>>> understand
    >>>> it from reading the docs) and am *not* releasing it at the end of
    >>>> session.
    >>>> Not quite sure of how to do this properly.
    >>>>
    >>>> 3. Don't know how to set the cache to retain N number of objects. I
    >>>> experimented with
    >>>> query.setPageSize(RowsPerPage);
    >>>> query.setCacheStrategy(QueryCacheStrategy.SHARED_CACHE);
    >>>> query.setCacheGroups("product", "ProductList");
    >>>>
    >>>> This seemed to help quite a bit but I still eventually ran out of
    >>>> memory.
    >>>> I
    >>>> recently removed *all* the SHARED_CACHE and it ran out of memory
    >>>> very
    >>>> fast.
    >>>>
    >>>> Thanks for your input
    >>>> Joe
    >>>>
    >>>>
    >>>>
    >>>> On Sep 16, 2009, at 2:25 PM, Mike Kienenberger wrote:
    >>>>
    >>>>> Caveat: I'm not really an expert on Cayenne memory management.
    >>>>>
    >>>>> 1) Are you allocating enough heap memory to the app server to
    >>>>> start
    >>>>> with? I don't know what the default is these days, but in the
    >>>>> old
    >>>>> days, an application by default only gets 64Mb of memory -- that's
    >>>>> pretty small.
    >>>>>
    >>>>> 2) Are you using a new DataContext per request? Or at least per
    >>>>> session?
    >>>>>
    >>>>> 3) I seem to remember that the cache strategy is configurable.
    >>>>> Have
    >>>>> you configured a cache that only retains N number of objects for a
    >>>>> suitable value of N?
    >>>>>
    >>>>> On Wed, Sep 16, 2009 at 2:11 PM, Joe Baldwin <jfbaldwi..arthlink.net
    >>>>> >
    >>>>> wrote:
    >>>>>>
    >>>>>> Hi,
    >>>>>>
    >>>>>> I have asked this question a number of ways but I still have a
    >>>>>> very
    >>>>>> serious
    >>>>>> problem with Cayenne-specific memory management configuration
    >>>>>> associated
    >>>>>> with Tomcat.
    >>>>>>
    >>>>>> The problem with debugging is that given that I have very little
    >>>>>> visibility
    >>>>>> into the Cayenne memory management it is extremely difficult to
    >>>>>> debug
    >>>>>> this
    >>>>>> using conventional strategies. Also I am not requesting
    >>>>>> anything from
    >>>>>> the
    >>>>>> system that is terribly exceptional so I am attempting to use
    >>>>>> default
    >>>>>> settings as much as possible.
    >>>>>>
    >>>>>> My strategy is to ask the experts for a Cayenne configuration and
    >>>>>> standard
    >>>>>> memory management steps I should take to conform to the new
    >>>>>> Cayenne
    >>>>>> memory
    >>>>>> management design intentions.
    >>>>>>
    >>>>>> Problem:
    >>>>>> 1. I have essentially a webstore, three tier design with Tomcat,
    >>>>>> Cayenne
    >>>>>> and
    >>>>>> MySQL.
    >>>>>> 2. When after only a few queries of products, tomcat freezes up
    >>>>>> and
    >>>>>> reports
    >>>>>> out of memory errors.
    >>>>>>
    >>>>>> I have attempted to configure the caching strategy ask best as
    >>>>>> I can
    >>>>>> understand from the docks but this only gets me a few more
    >>>>>> hours of
    >>>>>> usage
    >>>>>> before the out of memory errors. (I tried the SHARED_CACHE). The
    >>>>>> NO_CACHE
    >>>>>> strategy is worse.
    >>>>>>
    >>>>>> I would appreciate a set of steps (aka a primer) that should
    >>>>>> handle a
    >>>>>> website with a lot of fetches of hundreds of data objects (i.e.
    >>>>>> products)
    >>>>>> and very few updates.
    >>>>>>
    >>>>>> Note: My gut feeling is that I am not properly managing the
    >>>>>> data object
    >>>>>> array properly and it is leaking memory.
    >>>>>>
    >>>>>> I would appreciate any input, but I would first like to know
    >>>>>> what the
    >>>>>> minimum require steps are for managing at data object result set
    >>>>>> ArrayList
    >>>>>> so as to properly cache and then properly free the memory after
    >>>>>> it is
    >>>>>> no
    >>>>>> longer needed.
    >>>>>>
    >>>>>> Context: Tomcat, MySQL, Cayenne 3.0M6
    >>>>>>
    >>>>>> Thanks,
    >>>>>> Joe Baldwin
    >>>>>>
    >>>>>>
    >>>>
    >>>>
    >>
    >>
    > <DataContextManagerFilter.java>



    This archive was generated by hypermail 2.0.0 : Wed Sep 16 2009 - 16:40:22 EDT