MattW
Well-known member
I've been helping @bobster65 trouble shoot some issues with his server, and NAMED taking up all the CPU, causing his sites to stop loading, or take ~1 minute to load per page.
We are seeing load averages of 2.0 across the board, with namd using 160% of the available CPU on the dual core system he's got.
This is a fairly old box, but still decent enough to be running the sites at a reasonable speed.
However, named was causing some serious problems. The server was doing all the DNS lookups for the sites hosted on it, and you could see the DNS requests come in when you tried to access the sites, and named just sit there gobbling the CPU.
DNS zones were all configured correctly, nothing out of the ordinary with them, but WHM kept reporting one of the name servers configured on the box didn't have an IP address, and even when you set it, after a random time, it would vanish again.
The server was also reporting kernel errors for ip_conntrack, which tied up with named grabbing all the CPU.
I checked the setting in sysctl, and the there less entries than were available to handle, so it shouldn't be erroring like that.
So........I've moved his DNS to my DNSMadeEasy account, he's pointed his domains at the new DNS servers, and the sites are now loading instantly!
The errors have now stopped, and the load on the box has come right down again:
load average: 0.06, 0.06, 0.09
I've always been a great believer of having a specific service like this run by a specific service provider, such as DNSMadeEasy or Amazon's Route53. I'm now in my 4th year with DNSMadeEasy.
We are seeing load averages of 2.0 across the board, with namd using 160% of the available CPU on the dual core system he's got.
This is a fairly old box, but still decent enough to be running the sites at a reasonable speed.
However, named was causing some serious problems. The server was doing all the DNS lookups for the sites hosted on it, and you could see the DNS requests come in when you tried to access the sites, and named just sit there gobbling the CPU.
DNS zones were all configured correctly, nothing out of the ordinary with them, but WHM kept reporting one of the name servers configured on the box didn't have an IP address, and even when you set it, after a random time, it would vanish again.
The server was also reporting kernel errors for ip_conntrack, which tied up with named grabbing all the CPU.
Code:
Aug 3 05:28:55 server kernel: printk: 65358 messages suppressed.
Aug 3 05:28:55 server kernel: ip_conntrack: table full, dropping packet.
Aug 3 05:29:00 server kernel: printk: 65803 messages suppressed.
Aug 3 05:29:00 server kernel: ip_conntrack: table full, dropping packet.
Aug 3 05:29:05 server kernel: printk: 67396 messages suppressed.
Aug 3 05:29:05 server kernel: ip_conntrack: table full, dropping packet.
Aug 3 05:29:10 server kernel: printk: 63859 messages suppressed.
Aug 3 05:29:10 server kernel: ip_conntrack: table full, dropping packet.
Aug 3 05:29:15 server kernel: printk: 67713 messages suppressed.
Aug 3 05:29:15 server kernel: ip_conntrack: table full, dropping packet.
Aug 3 05:29:20 server kernel: printk: 64115 messages suppressed.
Aug 3 05:29:20 server kernel: ip_conntrack: table full, dropping packet.
I checked the setting in sysctl, and the there less entries than were available to handle, so it shouldn't be erroring like that.
So........I've moved his DNS to my DNSMadeEasy account, he's pointed his domains at the new DNS servers, and the sites are now loading instantly!
The errors have now stopped, and the load on the box has come right down again:
load average: 0.06, 0.06, 0.09
I've always been a great believer of having a specific service like this run by a specific service provider, such as DNSMadeEasy or Amazon's Route53. I'm now in my 4th year with DNSMadeEasy.