Bitcoin Core rpc_misc.py test failure #128

Activity

1. Sjors commented on Jan 17, 2025

   

   Sjors

   on Jan 17, 2025

   MemberAuthor

   More actions

   So far I tried on (M4) macOS 15.2 with commit 9a90f964a3c4fe75477926d34b7f1813b144449d:

   cmake -B build -DWITH_MULTIPROCESS=ON 
   cmake --build build
   BITCOIND=$(pwd)/build/src/bitcoin-node build/test/functional/rpc_misc.py

   Using multiprocess master @ 3b2617b
2. ryanofsky commented on Jan 17, 2025

   

   ryanofsky

   on Jan 17, 2025

   Collaborator

   More actions

   From the CI link above the error seems to be:

   node0 stderr libc++abi: terminating due to uncaught exception of type std::__1::system_error: mutex lock failed: Invalid argument

   which according to https://stackoverflow.com/questions/66773247/libcabi-dylib-terminating-with-uncaught-exception-of-type-std-1system-er "typically happens when .lock() is called on a mutex that is not yet constructed, or has already been destructed."

   In the test, the error happens when the node is shutting down after an echoipc call has been made:

   https://github.com/bitcoin/bitcoin/blob/90227f89411c1a1d4e62ed0e4de3099e1104bc19/test/functional/rpc_misc.py#L83
3. Sjors commented on Jan 21, 2025

   

   Sjors

   on Jan 21, 2025

   MemberAuthor

   More actions

   Another CI failure on the same test, this time on centOS. I'm unsure if it's the same issue:

   https://cirrus-ci.com/task/6027377219731456?logs=ci#L4105

   raise AssertionError("Unexpected stderr {} != {}".format(stderr, expected_stderr))
   AssertionError: Unexpected stderr terminate called after throwing an instance of 'kj::ExceptionImpl'
   what():  mp/proxy.cpp:242: disconnected: write(m_post_fd, &buffer, 1): Broken pipe
   
4. ryanofsky commented on Jan 21, 2025

   

   ryanofsky

   on Jan 21, 2025

   Collaborator

   More actions

   The failure on centos https://cirrus-ci.com/task/6027377219731456?logs=ci#L4111 also reported bitcoin/bitcoin#30975 (comment) is happening in nearly the same place as originally reported failure on macos, when the test tries to stop the node on line https://github.com/bitcoin/bitcoin/blob/11b293a9e7e89f25acc74d0bda5cc68cd164b551/test/functional/rpc_misc.py#L83 after an echoipc call is made.

   However, there are some differences between the two cases:

   • Macos case is failing earlier because the node process exiting with a nonzero status. https://github.com/bitcoin/bitcoin/blob/90227f89411c1a1d4e62ed0e4de3099e1104bc19/test/functional/test_framework/test_node.py#L423
   • Centos case is failing later because node process exits with status 0 but has unexpected stderr output
     https://github.com/bitcoin/bitcoin/blob/11b293a9e7e89f25acc74d0bda5cc68cd164b551/test/functional/test_framework/test_node.py#L429
   • In macos case, the parent process seems to be crashing immediately after the IPC event loop exits with stderr output libc++abi: terminating due to uncaught exception of type std::__1::system_error: mutex lock failed: Invalid argument. The crash happens very late though, after Shutdown: done is logged and the IPC event loop exits when the Init object is being destroyed
   • In centos case, the parent process does not seem to be crashing at all since it is exiting with status code 0. The stderr output terminate called after throwing an instance of 'kj::ExceptionImpl... probably comes from the child process not the parent process.

   In both cases there seems to be a race condition and the bug could have the same root cause. I think my next step will be to try to run the test in loop to see if I could reproduce locally.
5. ryanofsky commented on Jan 21, 2025

   

   ryanofsky

   on Jan 21, 2025

   Collaborator

   More actions

   The "Unexpected stderr" "terminate called after throwing an instance of 'kj::ExceptionImpl" "mp/proxy.cpp:242: disconnected: write(m_post_fd, &buffer, 1): Broken pipe" error that happens on CentOS was actually seen before and has an issue open bitcoin/bitcoin#31151

   👍React with 👍1Reacted by Sjors Provoost
6. ryanofsky commented on Jan 23, 2025

   

   ryanofsky

   on Jan 23, 2025

   Collaborator

   More actions

   Looking into this more it's not clear if this "mutex lock failed: Invalid argument" issue is the same as the other "disconnected: write(m_post_fd, &buffer, 1): Broken pipe" issue reported in bitcoin/bitcoin#31151, or if the fix in #129 will affect it.

   The other "Broken pipe" issue happens the when the child process in the echoipc test does not shut down cleanly due to a race condition in the EventLoop shutdown sequence that happens there. But the "mutex lock failed" here is happening in the parent process not the child process, and even though the last log messages are about the event loop shutting down, they could just indicate that the event loop shut down successfully, and the error is happening at some point after that.

   The echoipc parent process starts its event loop with the CapnpProtocol::connect function instead of CapnpProtocol::serve method running the event loop in the child process. And instead of shutting down when the last clean disconnects, it shuts down when the main() process exits and MakeNodeInit() destructor is called to destroy the interfaces::Init object which owns the interfaces::Ipc object which owns the CapnpProtocol object which calls removeClient in its destructor to stop the event loop.

   But even though the event loop shutdown sequence is different, it could have the race conditions described in bitcoin/bitcoin#31151 as long as there are any ProxyServerBase::m_impl objects that need to be destroyed asynchronously and could cause EventLoop::startAsyncThread to spawn a separate thread.

   The echoipc method in the parent process isn't passing any interface pointers to the child process so it isn't actually creating any ProxyServer objects of its own. However the IPC framework is creating ProxyServer<Thread> and
   ProxyServer<ThreadMap> objects and while the former is a custom class that doesn't inherit from ProxyServerBase, the latter does, so it's possible a similar race condition could exist that causes this issue and would be fixed by #129.

   But it's also possible #129 does not fix this, so will probably just need to see if it happens again.
7. ryanofsky commented on Jan 27, 2025

   

   ryanofsky

   on Jan 27, 2025

   Collaborator

   More actions

   I'm going to close this issue assuming #129 fixes it, but if it happens again, this can be reopened.
8. ryanofsky

   closed this as completedon Jan 27, 2025
9. ryanofsky commented on Feb 7, 2025

   

   ryanofsky

   on Feb 7, 2025

   Collaborator

   More actions

   But it's also possible #129 does not fix this, so will probably just need to see if it happens again.

   This issue did turn out to be different than #31151 and not fixed by #129. A similar macos "mutex lock failed" error was reported in #154. Both this issue and #154 should be fixed by bitcoin/bitcoin#31815, but this has not been confirmed yet.
10. bitcoin-core

    locked and limited conversation to collaborators on Feb 7, 2026