FAQ How to change a userid

[John Wade]

If anyone is interested, we have a perl script that does the following:

The script will:
#1) obtain list of mailboxes under the old UI via direct IMAP calls
#2) create the new mailboxes via direct IMAP calls
#3) do the recursive copy of mail files from old UID to new UID
#4) reconstruct new UID mailbox
#5) get the seen file for the old UID and prep for the new UID
#6) get the contents of all cyrus.header files for old & new UIDs into 
arrays
#7) hula-hoop thru seen file stuff; do tricky search and replace
#9) handle search and replace for old UID to new UID in <newUID>.sub file
#10) move any sieve scripts to the new UID

I can't say it is a perfect implementation because it does not try to 
address items 2 or 3 below, but these are not used extensively in our 
environment.

If you would like a copy, email me directly and I will be happy to send it,

John Wade


Rob Siemborski wrote:

>On Mon, 30 Jun 2003, twk wrote:
>
>  
>
>>>Then Why Cant I similarly  rename
>>>     'user/username ----> user/newusername '
>>>      
>>>
>>Because you can't. It's an architectural limitation of Cyrus.
>>    
>>
>
>This isn't true.  Its certainly possible, its just quite laborous to do
>correctly.
>
>  
>
>>Here is what we do :
>>
>>- Create a new account (new)
>>- cp -r /imap/user/o/old/* /imap/user/n/new
>>- reconstruct -r user.new
>>- delete old account
>>
>>I wouldn't consider that particularly laborious.
>>    
>>
>
>Except you haven't renamed the user, you've just renamed the mailbox.
>
>To rename the user you need to also do the following:
>
>1. move the subscriptions, seen state & sieve scripts (ok, this isn't
>hard)
>2. change the ACLs on all the mailboxes so that the old username is
>replaced by the new username (this is quite intensive to do on a large
>mail store)
>3. update the group memberships (since cyrus doesn't control its own
>group memberships, this involves the use of an outside process somehow).
>
>There is an 'allowusermoves' option in 2.2 that does some of this, but (2)
>and (3) just can't be done in any efficient way using the current scheme
>(We have thought about using numberic uids for users, but that would be a
>significant architechural shift).
>
>-Rob
>
>-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
>Rob Siemborski * Andrew Systems Group * Cyert Hall 207 * 412-268-7456
>Research Systems Programmer * /usr/contributed Gatekeeper
>
>
>  
>