| 1 |
In order to exploit the full possibilities of VOMS/LCMAPS the use of |
| 2 |
LDAP as a source of user account information is recommended. |
| 3 |
A site can do without by going through the following procedure: |
| 4 |
|
| 5 |
The new system can be made entirely compatible with the |
| 6 |
current poolaccount setup, but then one whould not be |
| 7 |
surprised that you get only reduced functionality. In effect, |
| 8 |
when a site |
| 9 |
* pre-allocates a set of poolaccounts for each and every VO, |
| 10 |
* pre-assigns the proper group to each (set of accounts), and |
| 11 |
* does not use very-fine-grained group-allocations within a VO for |
| 12 |
job execution separation |
| 13 |
they can get away without LDAP. For many sites, this will be OK. |
| 14 |
|
| 15 |
For example a sites supports the ATLAS, Alice and LHCb VO's, |
| 16 |
does not want to use LDAP, but does use VOMS. That sites can |
| 17 |
create three three groups and three sets of poolaccounts (say |
| 18 |
"lhcb001-lhcb050", etc.) and put each poolaccount into the |
| 19 |
proper group. |
| 20 |
|
| 21 |
Now, a grid-mapfile on a per-VO basis will look like |
| 22 |
|
| 23 |
"/VO=lhcb/GROUP=lhcb/*" .lhcb |
| 24 |
"/VO=atlas/GROUP=atlas/*" .atlas |
| 25 |
"/VO=alice/GROUP=alice/*" .alice |
| 26 |
|
| 27 |
and a groupmapfile: |
| 28 |
|
| 29 |
"/VO=lhcb/GROUP=lhcb/*" lhcb |
| 30 |
"/VO=atlas/GROUP=atlas/*" atlas |
| 31 |
"/VO=alice/GROUP=alice/*" alice |
| 32 |
|
| 33 |
Now, the acquired credentials will (as if by chance) map |
| 34 |
exactly on to the credentials already present in the user |
| 35 |
directory (be it files, or NIS), and so updating of those |
| 36 |
credentials is not needed. So that site in that config can do |
| 37 |
without the need for updating a central userDB (like LDAP). |
| 38 |
|
| 39 |
You can even do it a bit more fine-grained, adding an LHCb MC |
| 40 |
prod manager like this: |
| 41 |
|
| 42 |
"/VO=lhcb/GROUP=lhcb/mcprod .lhcbmc |
| 43 |
"/VO=lhcb/GROUP=lhcb/* .lhcb |
| 44 |
|
| 45 |
and a group-mapfile: |
| 46 |
"/VO=lhcb/GROUP=lhcb/mcprod prodmc |
| 47 |
"/VO=lhcb/GROUP=lhcb/* lhcb |
| 48 |
|
| 49 |
|
| 50 |
and assign (statically) in /etc/passwd: |
| 51 |
|
| 52 |
lhcbmb001:*:501:100:Blah:/home/lhcbmc001:/bin/bash |
| 53 |
lhcbmb002:*:502:100:Blah:/home/lhcbmc002:/bin/bash |
| 54 |
lhcbmb003:*:503:100:Blah:/home/lhcbmc003:/bin/bash |
| 55 |
lhcb001:*:601:100:Blah:/home/lhcb001:/bin/bash |
| 56 |
lhcb002:*:602:100:Blah:/home/lhcb002:/bin/bash |
| 57 |
|
| 58 |
and in /etc/group |
| 59 |
|
| 60 |
lhcb:*:100: |
| 61 |
prodmc:*:200:lhcbmc001,lhcbmc002,lhcbmc003 |
| 62 |
|
| 63 |
This will also do the job of ensuring that the mapping |
| 64 |
created for the fork job manager (in-process, using |
| 65 |
posix_enf) that is used for doing the "qsub" is the same as |
| 66 |
the one used on job execuation on a worker node. |
| 67 |
|
| 68 |
And of course, the obvious solution for those sites that want |
| 69 |
fine-grained authorization but do not want to use LDAP: buy |
| 70 |
a true MPP systems with all your processors seeing the same |
| 71 |
file systems and memory. Cray/SGI will be very happy ;-) |