efa-project.org

Hi All,

I experience a huge memory and cpu load on my EFA box. It is the cleanup process of postfix that is responsible for this. I also notice the MailScanner process taking sometime a full core (99% cpu). I have no clue why, it is a rather greenfield deployment. Can someone please help me out? I can post my main.cf and MailScanner.conf if this helps. I am running only 2 child processes, 8GB mem and 4 vCPUs (on VMware). The iron is a HP DL380Gen8 High Performance. I increased the memory to 16GB, but this does not help. If I look at the VMware mem load of the VM, I see it takes all the mem it can get. So with 8GB it was to the roof, and with 16GB as well. It seems the performance deteriorates over time, like the first say 15 minutes it runs allright, but then it starts to become ugly slow.

Any ideas??

Thanks for the help in advance.

Grtz,
Ronald

if anyone can do a sanity check on these snippets of my main.cf:

# tarpit control
smtpd_error_sleep_time = 10
smtpd_soft_error_limit = 2
smtpd_hard_error_limit = 2
smtpd_junk_command_limit = 2
smtpd_recipient_overshoot_limit = 100

# smtpd restrictions
smtpd_data_restrictions = reject_unauth_pipelining
smtpd_delay_reject = yes
smtpd_helo_required = yes
smtpd_helo_restrictions =
permit_mynetworks,
permit_sasl_authenticated,
check_helo_access hash:/etc/postfix/helo_access,
reject_invalid_hostname,
reject_non_fqdn_hostname
smtpd_sender_restrictions =
permit_mynetworks,
permit_sasl_authenticated,
check_sender_access hash:/etc/postfix/sender_access,
reject_non_fqdn_sender,
reject_unknown_sender_domain,
reject_unknown_address
smtpd_client_restrictions =
permit_mynetworks,
permit_sasl_authenticated,
check_client_access hash:/etc/postfix/client_access,
reject_rbl_client zen.spamhaus.org,
# reject_rbl_client dnsbl.sorbs.net,
reject_rbl_client cbl.abuseat.org
smtpd_recipient_restrictions =
permit_sasl_authenticated,
permit_mynetworks,
check_sender_access hash:/etc/postfix/restricted_senders,
reject_unauth_destination
reject_invalid_hostname,
reject_non_fqdn_sender,
reject_unknown_sender_domain,
reject_non_fqdn_recipient,
reject_unknown_recipient_domain,
check_recipient_access hash:/etc/postfix/recipient_access
# check_policy_service inet:127.0.0.1:2501 # disable SQLGrey for now

# body restrictions
body_checks = regexp:/etc/postfix/mbl-body-deny # malware Patrol

how is your build... did you follow the method for "build on your hardware" exactly (with the ks.cff script), or did you do what i used to do, and use my own OS with the portions of the build script that were (i thought) relevant?

This was 100% on my end, I added an exhaustive body_check in postfix and that screwed it up. It works fine now. Only 2GB max usage and max 20% CPU load (on 4 CPUs). So this is good.

Just out of curiosity, what was the body_check you were doing?