Description
I'm not sure if this php-vips project is the most appropriate forum to writeup this bug report, but I'm not sure where else would be appropriate.
We're trying to get Vips to work with:
- Ubuntu 14.04.1 LTS
- Apache 2.4
- PHP 7.0
We tried to follow the guidelines as instructed in the main readme here:
https://github.com/jcupitt/php-vips
In doing so, we were able to get Apache to run appropriately with the Vips C executable, the PECL module, and your php-vips composer package:
- name: Stop Apache
shell: service apache2 stop
- name: VIPS Requirements
apt: name="{{ item }}" state=latest
with_items:
- build-essential
- pkg-config
- libglib2.0-dev
- libxml2-dev
- libtiff5-dev
- libjpeg-turbo8-dev
- libgsf-1-dev
- libvips-dev
- name: Download libvips package
command: aws s3 cp s3://our-3rd-party-software-repo/vips/vips-8.3.3.orig.tar.gz /tmp/vips-8.3.3.tar.gz
sudo: yes
sudo_user: "{{ s3_user }}"
- name: Untar libvips package
unarchive: src=/tmp/vips-8.3.3.tar.gz dest=/tmp/ copy=no
sudo: no
- name: Make libvips
shell: cd /tmp/vips-8.3.3 && ./configure && make
- name: Make install libvips
shell: cd /tmp/vips-8.3.3 && make install
sudo: yes
- file: path=/tmp/vips-8.3.3 state=absent
- file: path=/tmp/vips-8.3.3.tar.gz state=absent
- name: Install the binary PHP extension
shell: /usr/bin/yes '' | pecl install vips
sudo: yes
- name: Include extension in php.ini
lineinfile: dest={{ php_ini_apache_path }} regexp='^\[vips\]' line='[vips]'
- lineinfile: dest={{ php_ini_apache_path }} insertafter='^\[vips\]' regexp='^extension=vips.so' line='extension=vips.so'
- lineinfile: dest={{ php_ini_cli_path }} regexp='^\[vips\]' line='[vips]'
- lineinfile: dest={{ php_ini_cli_path }} insertafter='^\[vips\]' regexp='^extension=vips.so' line='extension=vips.so'
- name: Start Apache
shell: service apache2 start
Everything works beautifully, except when we run an apache2ctl graceful. As soon as we run a graceful, Apache no longer works and we end up with zombie processes.
For example, after provisioning completes, it works and looks healthy:
05:38 PM vagrant@dev:[~]: ps aux | grep apache
root 28121 0.2 0.6 581932 49412 ? Ss 17:38 0:00 /usr/sbin/apache2 -k start
www-data 28468 0.5 0.3 584504 29472 ? S 17:38 0:00 /usr/sbin/apache2 -k start
www-data 28469 0.3 0.3 583900 31360 ? S 17:38 0:00 /usr/sbin/apache2 -k start
www-data 28470 0.8 0.3 583764 29988 ? S 17:38 0:00 /usr/sbin/apache2 -k start
www-data 28471 0.0 0.1 581964 13344 ? S 17:38 0:00 /usr/sbin/apache2 -k start
www-data 28472 0.0 0.1 582164 16088 ? S 17:38 0:00 /usr/sbin/apache2 -k start
www-data 28542 0.0 0.1 581964 13344 ? S 17:38 0:00 /usr/sbin/apache2 -k start
www-data 28546 0.0 0.1 581964 13344 ? S 17:38 0:00 /usr/sbin/apache2 -k start
www-data 28549 0.0 0.1 581964 13344 ? S 17:38 0:00 /usr/sbin/apache2 -k start
www-data 28610 0.0 0.1 581964 13344 ? S 17:38 0:00 /usr/sbin/apache2 -k start
vagrant 28847 0.0 0.0 14120 2304 pts/1 R+ 17:38 0:00 grep --color=auto apache
Then when we run apache2ctl graceful, Apache becomes unresponsive and we end up with zombie processes:
05:38 PM vagrant@dev:[~]: sudo apache2ctl graceful
05:39 PM vagrant@dev:[~]: ps aux | grep apache
root 28121 0.2 0.4 483124 34008 ? Ss 17:38 0:00 /usr/sbin/apache2 -k start
www-data 28468 0.4 0.3 584504 29472 ? S 17:38 0:00 /usr/sbin/apache2 -k start
www-data 28469 0.2 0.3 583900 31360 ? S 17:38 0:00 /usr/sbin/apache2 -k start
www-data 28470 0.6 0.3 583764 29988 ? S 17:38 0:00 /usr/sbin/apache2 -k start
www-data 28471 0.0 0.0 0 0 ? Z 17:38 0:00 [apache2] <defunct>
www-data 28472 0.0 0.1 582164 16088 ? S 17:38 0:00 /usr/sbin/apache2 -k start
www-data 28542 0.0 0.0 0 0 ? Z 17:38 0:00 [apache2] <defunct>
www-data 28546 0.0 0.0 0 0 ? Z 17:38 0:00 [apache2] <defunct>
www-data 28549 0.0 0.0 0 0 ? Z 17:38 0:00 [apache2] <defunct>
www-data 28610 0.0 0.0 0 0 ? Z 17:38 0:00 [apache2] <defunct>
vagrant 29023 0.0 0.0 14120 2304 pts/1 R+ 17:39 0:00 grep --color=auto apache
What's stranger, is that if we simply run apache2ctl stop and then apache2ctl start, everything looks fine. However, when we deploy to production we need to run a graceful to make sure we don't kill live requests.
When we look at apache error logs, this is the error that appears:
05:39 PM vagrant@dev:[~]: tail -f /var/log/apache2/error.log
[Mon Dec 19 17:38:32.002616 2016] [core:notice] [pid 28121] AH00094: Command line: '/usr/sbin/apache2'
[Mon Dec 19 17:39:01.671918 2016] [mpm_prefork:notice] [pid 28121] AH00171: Graceful restart requested, doing restart
(banana:28121): GLib-GObject-WARNING **: cannot register existing type 'VipsObject'
(banana:28121): GLib-CRITICAL **: g_once_init_leave: assertion 'result != 0' failed
(banana:28121): GLib-GObject-CRITICAL **: g_type_register_static: assertion 'parent_type > 0' failed
(banana:28121): GLib-CRITICAL **: g_once_init_leave: assertion 'result != 0' failed
Any help would be appreciated. We would really like to use Vips with PHP 7 on Apache, but unless we can solve this problem it looks to be a deal breaker.
The only other thing we were confused by was whether or not we needed to install libvips by apt-get install libvips-dev and/or vips-8.3.3.tar.gz (from source). The Ansible playbook above does both.
Thanks in advance!