Bug #9115
closedA large number of VLANs causes PHP issues when making an interface change
100%
Description
I generated a configuration with 250 VLANs (assigned, enabled, with DHCP active) based on a user complaint of problems with a similar number of VLANs.
When booting up most things are OK, but slow. All the VLANs show on the console, in the widgets, menus, etc, but making a change to any one of the VLANs seems to get it stuck in a very slow cycle of touching every interface, which is disruptive and also makes the GUI sluggish and occasionally non-responsive.
See also: https://forum.netgate.com/topic/136946/dhcp-fails-silently-but-works-on-reboot-of-pfsense/
I made a change to only one VLAN, and then tried to apply the changes. As you can see from the logs, it's somehow taking almost two minutes to process each VLAN:
Nov 13 13:46:49 missy check_reload_status: Reloading filter Nov 13 13:50:21 missy check_reload_status: Syncing firewall Nov 13 13:50:23 missy check_reload_status: Restarting ipsec tunnels Nov 13 13:51:38 missy check_reload_status: updating dyndns opt5 Nov 13 13:51:38 missy check_reload_status: Restarting ipsec tunnels Nov 13 13:51:38 missy kernel: vlan0: changing name to 'ix3.1' Nov 13 13:52:51 missy check_reload_status: updating dyndns opt5 Nov 13 13:52:51 missy kernel: vlan1: changing name to 'ix3.2' Nov 13 13:52:51 missy check_reload_status: Restarting ipsec tunnels Nov 13 13:53:22 missy missy.lab.jimp.pw nginx: 2018/11/13 13:53:22 [error] 16591#100204: *1 upstream timed out (60: Operation timed out) while reading response header from upstream, client: 198.51.100.108, server: , request: "POST /interfaces.php?if=opt5 HTTP/2.0", upstream: "fastcgi://unix:/var/run/php-fpm.socket", host: "missy.lab.jimp.pw", referrer: "https://missy.lab.jimp.pw/interfaces.php?if=opt5" Nov 13 13:54:05 missy check_reload_status: updating dyndns opt6 Nov 13 13:54:05 missy check_reload_status: Restarting ipsec tunnels Nov 13 13:54:05 missy kernel: vlan2: changing name to 'ix3.3' Nov 13 13:55:43 missy check_reload_status: updating dyndns opt7 Nov 13 13:55:44 missy kernel: vlan3: changing name to 'ix3.4' Nov 13 13:55:44 missy check_reload_status: Restarting ipsec tunnels Nov 13 13:57:41 missy check_reload_status: updating dyndns opt8 Nov 13 13:57:41 missy check_reload_status: Restarting ipsec tunnels Nov 13 13:57:41 missy kernel: vlan4: changing name to 'ix3.5' Nov 13 13:59:39 missy check_reload_status: updating dyndns opt9 Nov 13 13:59:40 missy kernel: vlan5: changing name to 'ix3.6' Nov 13 13:59:40 missy check_reload_status: Restarting ipsec tunnels Nov 13 14:01:36 missy check_reload_status: updating dyndns opt10 Nov 13 14:01:37 missy kernel: vlan6: changing name to 'ix3.7' Nov 13 14:01:37 missy check_reload_status: Restarting ipsec tunnels Nov 13 14:03:09 missy check_reload_status: updating dyndns opt11 Nov 13 14:03:09 missy check_reload_status: Restarting ipsec tunnels Nov 13 14:03:09 missy kernel: vlan7: changing name to 'ix3.8'
The contents of /tmp/.interfaces.apply
also agree that only one interface was touched:
a:1:{s:4:"opt5";a:2:{s:5:"ifcfg";a:7:{s:5:"descr";s:3:"VL1";s:2:"if";s:5:"ix3.1";s:6:"enable";s:0:"";s:6:"ipaddr";s:10:"198.18.1.1";s:6:"subnet";s:2:"24";s:8:"spoofmac";s:0:"";s:6:"realif";s:5:"ix3.1";}s:4:"ppps";a:0:{}}}
Occasionally a PHP thread will go up to 100% CPU most most of the time is spent idle/waiting.
This is on an SG-5100, but the customer in the thread above saw it on an XG-1541.
Updated by Jim Pingle about 6 years ago
- Subject changed from A large number of VLANs causes PHP to spin at 100% when making an interface change to A large number of VLANs causes PHP issues when making an interface change
Updated by Renato Botelho almost 6 years ago
- Assignee set to Luiz Souza
- Target version changed from 48 to 2.4.4-p1
Updated by Renato Botelho almost 6 years ago
- Status changed from New to Feedback
Updated by Luiz Souza almost 6 years ago
- % Done changed from 0 to 100
This regression is now fixed and only when really necessary the VLANs will be recreated.
Updated by Jim Pingle almost 6 years ago
Looks a lot better here with the new method. Editing the parent is still a problem, however, but that can be split off into its own issue.
Updated by Jim Pingle almost 6 years ago
- Status changed from Feedback to Resolved
I split the parent interface issue off to #9154 -- this one can be closed.