You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@logging.apache.org by "Ludvik Jerabek (JIRA)" <ji...@apache.org> on 2017/07/01 20:29:00 UTC
[jira] [Created] (LOG4PHP-238) Logging from class destructor fails
when from forked child processes
Ludvik Jerabek created LOG4PHP-238:
--------------------------------------
Summary: Logging from class destructor fails when from forked child processes
Key: LOG4PHP-238
URL: https://issues.apache.org/jira/browse/LOG4PHP-238
Project: Log4php
Issue Type: Bug
Components: Code
Affects Versions: 2.3.0
Environment: Linux virtual-machine 4.4.0-79-generic #100-Ubuntu SMP Wed May 17 19:58:14 UTC 2017 x86_64 x86_64 x86_64 GNU/Linux
Php Version:
PHP 7.0.18-0ubuntu0.16.04.1 (cli) ( NTS )
Copyright (c) 1997-2017 The PHP Group
Zend Engine v3.0.0, Copyright (c) 1998-2017 Zend Technologies
with Zend OPcache v7.0.18-0ubuntu0.16.04.1, Copyright (c) 1999-2017, by Zend Technologies
with Xdebug v2.4.0, Copyright (c) 2002-2016, by Derick Rethans
Reporter: Ludvik Jerabek
The issue appear to only manifest itself when implementing class level logging.
"$this->closed" will evaluate to true when called from a parent::__destruct() of a forked child processes. I assume the file handle is most likely closed by a forked child closing when it exits.
public function doAppend(LoggerLoggingEvent $event)
{
{color:red} if ($this->closed) {
return;
}{color}
if (!$this->isAsSevereAsThreshold($event->getLevel())) {
return;
}
$filter = $this->getFirstFilter();
while ($filter !== null) {
switch ($filter->decide($event)) {
case LoggerFilter::DENY:
return;
case LoggerFilter::ACCEPT:
return $this->append($event);
case LoggerFilter::NEUTRAL:
$filter = $filter->getNext();
}
}
$this->append($event);
}
To recreate the issue create a php daemon:
1. Create a parent object which performs one (or more) a pcntl_fork calls. Ensure the class object has a class level logger. Log from __construct() and __destruct().
2 The parent should handle SIG_TERM and signal the child processes to exit.
3. The parent uses pcntl_wait to wait for the child processes to exit cleanly. If possible use child objects with class level logging too to see how children react.
4. When the parent has completed waiting on the children to exit, the parent's "__destruct" will not write to the log.
Since every child gets a copy of the members variables and open file descriptors it seems that when forked children exit. It *somehow* impacts the ability of the parent to write to the logger object associated with the parent class. I eventually gave up troubleshooting due to time limitations, and resorted using direct syslog() calls throughout.
When I get some time I may be able to upload some sample code to demonstrate the bug.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)