hhvm crashes under heavy load

[marksamman]

The latest hhvm-nightly from the Debian 7 64-bit apt repository is crashing under heavy load.

Running hhvm with this command:
hhvm -m server -c config.hdf

config.hdf only contains:

Server {
        Port = 8080
}

The PHP script I'm running is as simple as:

<?php
echo 'Hello World';

The command I use to reproduce this:
ab -n 5000 -c 4 http://localhost:8080/helloworld.php

gdb bt on the core dump shows:

(gdb) bt
#0  __strcasecmp_l_sse2 () at ../sysdeps/x86_64/multiarch/../strcmp.S:213
#1  0x00007fad2347cb3c in evhttp_find_header (
    headers=headers@entry=0x7fad0fe97a90, 
    key=key@entry=0x7fad23486f86 "Transfer-Encoding") at http.c:1373
#2  0x00007fad2347d00a in evhttp_maybe_add_content_length_header (
    content_length=11, headers=0x7fad0fe97a90) at http.c:455
#3  evhttp_make_header_response (req=0x7fad135b29c0, evcon=<optimized out>)
    at http.c:493
#4  evhttp_make_header (evcon=evcon@entry=0x7fad138a4f80, 
    req=req@entry=0x7fad135b29c0) at http.c:528
#5  0x00007fad2347de03 in evhttp_send_reply_sync_begin (req=0x7fad135b29c0, 
    code=<optimized out>, reason=<optimized out>, databuf=0x0) at http.c:2022
#6  0x0000000000f4ebf2 in HPHP::LibEventServer::onResponse(int, evhttp_request*, int, HPHP::LibEventTransport*) ()
#7  0x0000000000f39701 in HPHP::LibEventTransport::sendImpl(void const*, int, int, bool) ()
#8  0x0000000000f71c02 in HPHP::Transport::sendRawLocked(void*, int, int, bool, bool, char const*) ()
#9  0x0000000000f3b1a2 in HPHP::HttpRequestHandler::executePHPRequest(HPHP::Transport*, HPHP::RequestURI&, HPHP::SourceRootInfo&, bool) ()
#10 0x0000000000f3c3a0 in HPHP::HttpRequestHandler::handleRequest(HPHP::Transport*) ()
#11 0x0000000000f223a8 in HPHP::ServerWorker<std::shared_ptr<HPHP::LibEventJob>, HPHP::LibEventTransportTraits>::doJobImpl(std::shared_ptr<HPHP::LibEventJob>, bool) ()
#12 0x0000000000f22648 in HPHP::ServerWorker<std::shared_ptr<HPHP::LibEventJob>, HPHP::LibEventTransportTraits>::doJob(std::shared_ptr<HPHP::LibEventJob>) ()
#13 0x0000000000f21ae7 in HPHP::JobQueueWorker<std::shared_ptr<HPHP::LibEventJob>, HPHP::Server*, true, false, HPHP::JobQueueDropVMStack>::start() ()
#14 0x00000000015db3a7 in HPHP::AsyncFuncImpl::threadFuncImpl() ()
#15 0x00000000015db54b in HPHP::AsyncFuncImpl::ThreadFunc(void*) ()
#16 0x00007fad220b7b50 in start_thread (arg=<optimized out>)
    at pthread_create.c:304
#17 0x00007fad216630ed in clone ()
    at ../sysdeps/unix/sysv/linux/x86_64/clone.S:112
#18 0x0000000000000000 in ?? ()

Sometimes it stops responding to requests (0% cpu usage) instead of crashing, so I'm guessing there could be a deadlock somewhere too.

[scannell]

Thanks for reporting this. Feel free to submit a PR if you (or someone else reading this) figures this out before we do.

[marksamman]

I've updated to the latest nightly and this appears to have been fixed, #1937 might have been a duplicate report of this too (I had similar backtraces). Closing this issue as I'm no longer able to reproduce it.