Segfault when resuming from xdebug breakpoint in PHPStorm (and maybe other places?)

[jaredkipe]

Here is the contents of the stacktrace:

Host: app1
ProcessID: 18253
ThreadID: 7f2172fff700
ThreadPID: 18257
Name: unknown program
Type: Segmentation fault
Runtime: hhvm
Version: tags/HHVM-3.4.0-0-g817b3a07fc4e509ce15635dbc87778e5b3496663
DebuggerCount: 0


#0  folly::IOBuf::countChainElements() const at /usr/local/bin/hhvm:0
#1  apache::thrift::async::TAsyncSocket::writeChain(apache::thrift::async::TAsyncTransport::WriteCallback*, std::unique_ptr<folly::IOBuf, std::default_delete<folly::IOBuf> >&&, apache::thrift::async::WriteFlags) at /usr/local/bin/hhvm:0
#2  HPHP::FastCGISession::writeStream(unsigned short, HPHP::FastCGISession::RecordType, std::unique_ptr<folly::IOBuf, std::default_delete<folly::IOBuf> >) at /usr/local/bin/hhvm:0
#3  HPHP::FastCGITransaction::onStdOut(std::unique_ptr<folly::IOBuf, std::default_delete<folly::IOBuf> >) at /usr/local/bin/hhvm:0
#4  std::_Function_handler<void (), HPHP::FastCGITransport::sendImpl(void const*, int, int, bool)::{lambda()#1}>::_M_invoke(std::_Any_data const&) at fastcgi-transport.cpp:0
#5  folly::EventBase::runFunctionPtr(std::function<void ()>*) at /usr/local/bin/hhvm:0
#6  folly::EventBase::FunctionRunner::messageAvailable(std::pair<void (*)(void*), void*>&&) at /usr/local/bin/hhvm:0
#7  folly::NotificationQueue<std::pair<void (*)(void*), void*> >::Consumer::handlerReady(unsigned short) at /usr/local/bin/hhvm:0
#8  event_base_loop at /usr/lib/x86_64-linux-gnu/libevent-2.0.so.5:0
#9  folly::EventBase::loopBody(int) at /usr/local/bin/hhvm:0
#10 folly::EventBase::loopForever() at /usr/local/bin/hhvm:0
#11 proxygen::WorkerThread::runLoop() at /usr/local/bin/hhvm:0
#12 std::thread::_Impl<std::_Bind_simple<proxygen::WorkerThread::start()::{lambda()#1} ()> >::_M_run() at WorkerThread.cpp:0
#13 std::this_thread::__sleep_for(std::chrono::duration<long, std::ratio<1l, 1l> >, std::chrono::duration<long, std::ratio<1l, 1000000000l> >) at /usr/lib/x86_64-linux-gnu/libstdc++.so.6:0
#14 start_thread at /lib/x86_64-linux-gnu/libpthread.so.0:0
#15 clone at /lib/x86_64-linux-gnu/libc.so.6:0

This MAY have been caused by an exception in a registered callback for __SystemLib\SessionForwardingHandler->close()

[Cholik]

Same here. Happens when I try to resume from a triggered xdebug breakpoint using PHPStorm.
Triggering the breakpoint works fine but only for the first time, hhvm hangs and terminates after some time with the same stacktrace. None of this happens if I use Codebug for xdebugging.

Host: developmentHWhhvm
ProcessID: 5656
ThreadID: 7f8b7bfff700
ThreadPID: 5664
Name: unknown program
Type: Segmentation fault
Runtime: hhvm
Version: tags/HHVM-3.4.0-0-g817b3a07fc4e509ce15635dbc87778e5b3496663
DebuggerCount: 0


# 0  folly::IOBuf::countChainElements() const at /usr/bin/hhvm:0
# 1  apache::thrift::async::TAsyncSocket::writeChain(apache::thrift::async::TAsyncTransport::WriteCallback*, std::unique_ptr<folly::IOBuf, std::default_delete<folly::IOBuf> >&&, $
# 2  HPHP::FastCGISession::writeStream(unsigned short, HPHP::FastCGISession::RecordType, std::unique_ptr<folly::IOBuf, std::default_delete<folly::IOBuf> >) at /usr/bin/hhvm:0
# 3  HPHP::FastCGITransaction::onStdOut(std::unique_ptr<folly::IOBuf, std::default_delete<folly::IOBuf> >) at /usr/bin/hhvm:0
# 4  std::_Function_handler<void (), HPHP::FastCGITransport::sendImpl(void const*, int, int, bool)::{lambda()#1}>::_M_invoke(std::_Any_data const&) at fastcgi-transport.cpp:0
# 5  folly::EventBase::runFunctionPtr(std::function<void ()>*) at /usr/bin/hhvm:0
# 6  folly::EventBase::FunctionRunner::messageAvailable(std::pair<void (*)(void*), void*>&&) at /usr/bin/hhvm:0
# 7  folly::NotificationQueue<std::pair<void (*)(void*), void*> >::Consumer::handlerReady(unsigned short) at /usr/bin/hhvm:0
# 8  event_base_loop at /usr/lib/x86_64-linux-gnu/libevent-2.0.so.5:0
# 9  folly::EventBase::loopBody(int) at /usr/bin/hhvm:0
# 10 folly::EventBase::loopForever() at /usr/bin/hhvm:0
# 11 proxygen::WorkerThread::runLoop() at /usr/bin/hhvm:0
# 12 std::thread::_Impl<std::_Bind_simple<proxygen::WorkerThread::start()::{lambda()#1} ()> >::_M_run() at WorkerThread.cpp:0
# 13 std::this_thread::__sleep_for(std::chrono::duration<long, std::ratio<1l, 1l> >, std::chrono::duration<long, std::ratio<1l, 1000000000l> >) at /usr/lib/x86_64-linux-gnu/libst$
# 14 start_thread at /lib/x86_64-linux-gnu/libpthread.so.0:0
# 15 clone at /lib/x86_64-linux-gnu/libc.so.6:0

[jaredkipe]

I got a second identical crash. I grow more certain that it is related to an exception (deadlock) in session callback.

[MaXal]

Same in HHVM 3.4.2.

[jaredkipe]

FWIW I 'moved' the sessions to Redis and this went away. (I didn't intend the sessions to be in the DB, it was a regression in a config file).

I'm more or less certain that this was caused by an exception (deadlock due to galera cluster) in the db based session implementation. Most likely an/the exception is not expected in a registered session callback handler.

[jwatzman]

• This happens with any breakpoint, in any project? (Trying to get a repro case.)
• Have you tried with the recently released 3.5?

[roman-shuhov]

happens in any project with any type of breakpoint
also happens in 3.5.0

[jwatzman]

Sounds like we have some consistent reports. The xdebug/phpstorm thing is worth looking into when we get some time -- the issue seems to happen with more than just that, but it sounds like a consistent repro.

[sobit]

happens to me also on remote debugging using phpstorm. my steps are:

1. catch a breakpoint
2. stop debugging in phpstorm
3. run the script again

at this point hhvm is stopped and not responding anymore.

[SvetlanaZem]

Hello, I'm working on debugger integration in PhpStorm and I'll be happy to provide you any information that will help to fix the issue. So if you need any assistance feel free to contact me.

[RoyHP]

I am also having this problem, with the same steps using phpstorm.

[jwatzman]

@paulbiss do you know if anyone is actively working on xdebug? Might be useful to take up @SvetlanaZem's offer if someone is, though I suspect we don't have anyone with the spare time right now?

[paulbiss]

@alexmalyshev is working on hphpd, and had been looking at Xdebug as a potential replacement. I think he handed off the Xdebug work to the IDE folks, but he may know who was looking at it most recently anyway.

[RoyHP]

If they are useful, I have attached HHVM's xdebug logs to the PhpStorm issue WI-26478 (https://youtrack.jetbrains.com/issue/WI-26478). I think @SvetlanaZem believes this is an issue with HHVM. Let me know if there's anything you want me to test.

[SvetlanaZem]

There are three responses for a single "breakpoint_set" command in xdebug.log files provided by @RoyHP Anyway it looks like another problem, please let me know if I should create separate issue.

[RoyHP]

@jwatzman Someone is reporting this issue is not only limited to PHPStorm, but also happening with MacGDBp. They are commenting on the issue I created with PHPStorm in my last comment.

[MaXal]

Issue is still there for HHVM 3.7.0.

[bheyser]

I can also confirm the issue:

• ubuntu 14.04
• hhvm 3.7.0 via nginx
• phpstorm 8.0.3
• ilias 5.0.1 - php project with db based session storage (https://github.com/ILIAS-eLearning)

happens after resuming from a breakpoint

[robertodormepoco]

same here with HHVM 3.7.0/Nginx with PhpStorm EAP or 8.0.*

the xdebug communication works quite well in Eclipse Luna 4.4, mainly because xdebug and the later talk with a different sequence of communication steps that was a misleading thought see my last comment

[robertodormepoco]

please see my colleague's comment about hhvm xdebug read buffer handling

[jwatzman]

Can someone who was having this problem try it again with the latest nightly build of HHVM? #4488 has been fixed and I wonder if this is too.

[RoyHP]

@jwatzman Someone on #4488 is having a new problem. I haven't tried it yet myself, and won't get the chance for a while.

[jwatzman]

Leaving this open until confirmed fixed, but won't be personally working on it right now.

[euphio]

The problem still exists for us when remote debugging to a Vagrant box:

• Ubuntu 14.04
• PhpStorm 9.0.2
• HHVM 3.9.2

Resuming from a breakpoint intermittently causes a segfault.

[paulbiss]

Is anyone still seeing this? The recent work on xdebug for nuclide debugging support fixed a lot of hard to find bugs.

[mofarrell]

Closing. Needs more info. (https://github.com/facebook/hhvm/wiki/Human-Timeouts)