Re: [libmicrohttpd] epoll and memory leaks

[Erik Smith]

Update: using jemalloc stats may not be useful without more configuration as I think the increase might just be some dynamic pre-allocation within jemalloc.   When I remove the jemaloc dependency and just monitor it with top, using top -p $(pidof demo_server), I see no increase in RES but some increase in VIRT.   There is something about the MHD thread pool mode that is problematic for us that requires us to restart our production servers after a few days.  It does not appear to happen when the thread pool is not used or potentially it is some interaction with jemalloc.

Erik

On Fri, Dec 10, 2021 at 6:29 AM Erik Smith <cruisercoder@gmail.com> wrote:

This is happening with the latest version (0.9.73) and on the master branch.

erik

On Thu, Dec 9, 2021 at 11:06 PM Evgeny Grin <k2k@yandex.ru> wrote:

Hi Erik,

It's hard to move forward without knowing exact MHD version used.

Please share information about your MHD version.

I suspect that your "Reply-To" header may confuse mailing list system.
Do not use "Reply-To" headers with mailing list.

--
Evgeny

-------- Original Message --------
From: Erik Smith <cruisercoder@gmail.com>
Sent: Friday, December 10, 2021, 03:29 UTC+3
Subject: [libmicrohttpd] epoll and memory leaks

> I've been able to reproduce this with a modified program from the
> examples to show the memory consumption I'm seeing.  I'm using jemalloc
> to capture memory consumption and here's what it looks like for the
> program below when repeatedly hitting the endpoint:
>
> allocated: 120072, active: 163840, resident: 9150464
> allocated: 150536, active: 196608, resident: 9228288
> allocated: 181000, active: 229376, resident: 9306112
> allocated: 211464, active: 262144, resident: 9383936
> allocated: 211464, active: 262144, resident: 9383936
> allocated: 241928, active: 294912, resident: 9461760
> allocated: 272392, active: 327680, resident: 9539584
> allocated: 272392, active: 327680, resident: 9539584
> allocated: 302856, active: 360448, resident: 9617408
>
> The delay in the handler and the use of ASAN tend to inflate the memory
> growth.   The key factor here seems to be the use of the thread poll
> with either poll or epoll.  Without the thread pool, there is no memory
> growth at all.  The growth happens on low connection rates (manual
> refreshing in the browser).   I haven't yet tried compiling MHD with ASAN.
>
> I'm also not getting responses to my threads in email for some reason
> but I'm checking the archive.
>
> #include <cstring>
> #include <iostream>
> #include <jemalloc/jemalloc.h>
> #include <microhttpd.h>
> #include <sstream>
> #include <thread>
>
> static enum MHD_Result handler(void *, struct MHD_Connection *connection,
>                                 const char *url, const char *method,
>                                 const char *, const char *, size_t *,
>                                 void **ptr) {
>    static int aptr;
>
>    if (&aptr != *ptr) {
>      *ptr = &aptr;
>      return MHD_YES;
>    }
>    *ptr = NULL;
>
>    std::this_thread::sleep_for(std::chrono::milliseconds(40));
>
>    size_t sz = sizeof(size_t);
>    uint64_t epoch = 1;
>    mallctl("thread.tcache.flush", NULL, NULL, NULL, 0);
>    mallctl("epoch", &epoch, &sz, &epoch, sz);
>
>    std::size_t allocated, active, metadata, resident, mapped;
>    mallctl("stats.allocated", &allocated, &sz, NULL, 0);
>    mallctl("stats.active", &active, &sz, NULL, 0);
>    mallctl("stats.resident", &resident, &sz, NULL, 0);
>
>    std::stringstream s;
>    s << "allocated: " << allocated << ", active: " << active
>      << ", resident: " << resident << "\n";
>    auto msg = s.str();
>
>    std::cout << msg;
>
>    struct MHD_Response *response = MHD_create_response_from_buffer(
>        msg.size(), msg.data(), MHD_RESPMEM_MUST_COPY);
>    MHD_Result ret = MHD_queue_response(connection, MHD_HTTP_OK, response);
>    MHD_destroy_response(response);
>    return ret;
> }
>
> int main(int argc, char *argv[]) {
>    struct MHD_Daemon *d;
>
>    int port = argc > 1 ? atoi(argv[1]) : 10000;
>
>    // epoll mode with thread pool
>    unsigned int concurrency = std::thread::hardware_concurrency();
>    std::cout << "concurrency: " << concurrency << "\n";
>
>    d = MHD_start_daemon(MHD_USE_EPOLL_INTERNAL_THREAD |
> MHD_USE_ERROR_LOG, port,
>                         NULL, NULL, handler, NULL,
> MHD_OPTION_CONNECTION_TIMEOUT,
>                         (unsigned int)120, MHD_OPTION_STRICT_FOR_CLIENT,
> (int)1,
>                         MHD_OPTION_THREAD_POOL_SIZE, concurrency, NULL,
>                         MHD_OPTION_END);
>
>    if (d == NULL)
>      return 1;
>    std::cout << "listening on port: " << port << "\n";
>    std::cout << "hit key to stop"
>              << "\n";
>
>    // type a key to end
>    (void)getc(stdin);
>    MHD_stop_daemon(d);
>    return 0;
> }
>
>
>     Hi Erik,
>     Which MHD version are you using?
>     Some problems with externally added connections with epoll mode were
>     fixed in v0.9.72.
>     If you have any blocking calls, make sure that you use connection
>     suspend/resume. Alternatively, you can you use thread-per-connection
>     mode, this is less efficient, but simpler to implement.
>     epoll mode does not have special memory allocation, connections are
>     processed in the same way, like in other modes. MHD typically does
>     not allocate memory during connection processing, except when new
>     connection is started.
>     Do you use postprosessor or authentication functions? MHD has some
>     memory allocs in these functions.
>     The issue is not connected with quoted comment definitely. It is
>     just a bad wording. Actually nothing is leaked, but may be locked
>     until end of sending of response. Moreover, MHD does not use memory
>     pool in the way where such lock is possible. Memory pool is reset
>     after each request-reply cycle. Memory pool size for each connection
>     is fixed and cannot grow.
>     A few suggestions:
>     * make sure that you are using the latest MHD version (0.9.73 at the
>     moment), * check whether you destroy responses and free all
>     resources connected to responses, * if you are testing your code
>     with ASAN, make sure that leak detector is enabled. You can build
>     static MHD lib with ASAN and link it with our application compiled
>     with ASAN,
>     * use Valgrind or simpler tools like memstat or memprof.
>     --
>     Wishes,
>     Evgeny
>     On 22.11.2021 22:56, Erik Smith wrote:
>     /* Reallocate a block of memory obtained from the pool.
>     * This is particularly efficient when growing or
>     * shrinking the block that was last (re)allocated.
>     * If the given block is not the most recently
>     * (re)allocated block, the memory of the previous
>     * allocation may be leaked until the pool is
>     * destroyed or reset. */
>     Can anyone confirm whether this might be related?
>     ASAN does not seem to detect any issues in our code presently (not
>     sure about MHD)
>
>     We have started to experiment with running MHD with epoll + thread
>     pool as we do the FD limit in certain situations.  We understand
>     that there are caveats to this given that we have some
>     blocking database calls. This seems to get us past the FD limit
>     errors and the performance is similar.   However, we are running
>     into growing memory consumption in our server over time
>     running epoll+threads that require a restart frequently.   This does
>     not seem to occur with just epoll (without the thread pool).   We
>     are running jemalloc, but it does not seem to be related to the leak
>     when it is disabled.  There is the following comment in the MHD code
>     for the MHD_pool_reallocate function that might be connected to this
>     issue:
>