I have noticed some unique behavior with regards to where vpopmail
decides to place a new user folder.

We have a domain with over 5000 mailboxes and vpopmail is using folders
under the domain to hold the maildirs.

For example:

[EMAIL PROTECTED] has the maildir location of:


There are subfolders ranging from 0-9 as well as A-Z and the letter a.
The break down of maildirs per directory is:

0      61
1      59
2      55
3      54
4      48
5      50
6      49
7      54
8      48
9      48
A      52
B      52
C      50
D      57
E      56
F      49
G      54
H      53
I      52
J      57
K      50
L      40
M      56
N      47
O      52
P      51
Q      52
R      39
S      44
T      48
U      40
V      35
W      36
X      30
Y      41
Z      36
a    3485

(Generated with: for i in `ls -la | grep drw| awk '{print $9}'`; do echo
-n $i; ls -la $i | grep drw| wc -l; done)

It seems that a preponderance of maildirs are being created in the
directory a. To the point where I think the directory hash is suffering
(ufs based filesystems).

I have also had a problem with vpopmail removing deactivated accounts
completely. Sometimes when a users account is deactivated I can go back
to the users directory and find some dir (tmp, cur, or new) still
remaining. This causes a reactivation to fail with "User's directory
already exists?". I then delete the users maildir remnants and the
reactivation occurs as it should.

I am also running NFS on the backend... So that might have something to
do with the removal problem... But is certainly not the problem with the
directory selection. Seems like more entropy is needed in the directory
selection algorithm.

The system is FreeBSD 4.8 running vpopmail 5.3.20 with an NFS backend
for the ~vpopmail/domains/.

Thoughts or comments?

Tom Walsh
Network Administrator

Reply via email to