ninja-build/ninja

ninja hanging with zombie child process

Open

#2052 opened on Nov 18, 2021

View on GitHub
 (4 comments) (6 reactions) (0 assignees)C++ (10,194 stars) (1,544 forks)batch import
bughelp wanted

Description

Occasionally, ninja is hanging after running a long-running build step, with a zombie child process present:

$  ps ax | grep ninja
32692 pts/1    S+     0:00 ninja -C build/GccReleaseASan/
$ pstree 32692 -p
ninja(32692)───sh(32702)
$ gdb -p 32702
[...]
Attaching to process 32702
warning: process 32702 is a zombie - the process has already terminated
$ gdb -p 32692
[...]
0x00007f2839c2ed39 in __GI_ppoll (fds=0x1813c60, nfds=1, timeout=<optimized out>, sigmask=0x1807b40) at ../sysdeps/unix/sysv/linux/ppoll.c:51
51          result = INLINE_SYSCALL (ppoll, 5, fds, nfds, timeout, sigmask, _NSIG / 8);
(gdb) bt
#0  0x00007f2839c2ed39 in __GI_ppoll (fds=0x1813c60, nfds=1, timeout=<optimized out>, sigmask=0x1807b40) at ../sysdeps/unix/sysv/linux/ppoll.c:51
#1  0x000000000042e588 in SubprocessSet::DoWork() ()
#2  0x00000000004095c8 in RealCommandRunner::WaitForCommand(CommandRunner::Result*) ()
#3  0x000000000040edbb in Builder::Build(std::string*) ()
#4  0x0000000000407c1b in (anonymous namespace)::real_main(int, char**) ()
#5  0x0000000000403199 in main ()
$ ninja --version
1.10.2

Contributor guide

Tech stack
cpp
Domain
build system
Issue type
bug
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
4
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
3-5 days
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
stale
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
mostly clear
Prerequisites
C++ knowledgeUnderstanding of subprocess managementFamiliarity with ninja's architecture
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
25
Research direction
Investigate the SubprocessSet::DoWork() function in src/subprocess.cc. The issue is that the main thread is blocked in ppoll waiting for a file descriptor, while the child process is a zombie. This suggests that the SIGCHLD handler may not be properly reaping the child, or the pselect/ppoll loop is not checking for terminated processes after being woken by SIGCHLD. Look at how the child's exit status is collected and ensure that the ppoll call handles EINTR correctly.