Bash loses track of background children sometimes

[Ben Harris]

Under some circumstances (I've not worked out precisely what), bash can forget about background child processes. For instance, if the file "simpletest" contains:

#! /bin/bash

count=0

while [ $count -lt 51 ]; do
    count=$((count+1))
    /nonexist
    true &
    bgpids="$bgpids $!"
done

for pid in $bgpids; do
    wait $pid
done

then running "bash simpletest", seems to reliably report "simpletest: line 13: wait: pid 28850 is not a child of this shell" (with different PIDs each time, obviously). Increasing the "51" in the script increases the number of errors. Curiously, removing the call to a nonexistent program causes the problem to stop manifesting.

"getconf CHILD_MAX" returns "undefined", which I think means that Bash should support an unlimited number of unwaited children.

I've observed this on two different amd64 boxes running Ubuntu 10.04 LTS, and it persisted after I upgraded the bash package to 4.2-0ubuntu4. I couldn't reproduce it on an i386 box running Ubuntu 11.04.

ProblemType: Bug
DistroRelease: Ubuntu 10.04
Package: bash 4.1-2ubuntu3
ProcVersionSignature: Ubuntu 2.6.32-33.72-server 2.6.32.41+drm33.18
Uname: Linux 2.6.32-33-server x86_64
Architecture: amd64
Date: Thu Aug 18 17:04:23 2011
InstallationMedia: Ubuntu-Server 10.04.3 LTS "Lucid Lynx" - Release amd64 (20110719.2)
ProcEnviron:
 LANG=en_GB.UTF-8
 SHELL=/bin/bash
SourcePackage: bash

[Timothy Mayoh (timothy-m-deactivatedaccount)]

I can confirm this also occurs on Ubuntu 11.04 AMD64.

[Ben Harris (bjh21)]

Just a quick note: having moved some affected systems back to Ubuntu, this bug has just manifested in production on Ubuntu 16.04 LTS:

/opt/dump/bin/dump_multi: line 11: wait: pid 56652 is not a child of this shell

My test script above no longer demonstrates it though. I'll see if I can come up with one that does.