I've just replaced our Dell R210 II servers, which were struggling with poor disk IO, with lovely Dell R310s with a RAID10 disk subsystem. Things are now flying and everything is running smoothly, just as they should.
But in the process of checking everything in as much detail as possible, I noticed that qmail-send *seems* to be writing massive amounts of data to disk on one particular system.
I'm basing this only on what atop tells me when I start it up:
Code: Select all
*** system and process activity since boot ***
PID RUID EUID THR SYSCPU USRCPU VGROW RGROW RDDSK WRDSK ST EXC S CPUNR CPU CMD
49017 2522 2522 1 62m06s 14m42s 4124K 560K 13496K 660.5G N- - S 1 1% qmail-send
How can that be? Why would that be? Is it even accurate?
There aren't zillions of messages passing through this server. The number of messages actually sent is not, by any standard, large. It is a Plesk server with 250-ish sites and no significant number of outgoing messages. It doesn't even get masses of incoming messages - nothing out of the ordinary anyway. There has not been a spam outbreak from this system.
Load average on the server is perfectly OK. qmail-send does seem to use a lot of CPU compared to some things. But that's it.
Most significantly, every time I check, iotop does not show qmail-send as being a disk hog in any way, be it read or write, and iowaits are insignificant for it.
Does anybody have any ideas? Am I reading the numbers incorrectly? Is atop reporting numbers incorrectly? Does qmail-send do something other than put messages in the local or remote queues?