source: trunk/third/moira/afssync/INSTRUCTIONS @ 23095

The executables are in /moira/bin/ on the moira server, with sources
in /mit/moiradev/src/afssync/.  Most of the commands are run on the
Moira server.

####    Set up a workspace                                              ####

mkdir -p /moira/sync
cd /moira/sync

####    This is preparation for the resync, to save non-Moira users.    ####
First, get a recent copy of the prdb, and extract non-Moira entries:

        /moira/bin/udebug prill -port 7002
        rcp -px root@prill:/usr/afs/db/prdb.DB0 prdb.old
        /moira/bin/udebug prill -port 7002
If the two udebugs show that the version changed, lather-rinse-repeat.
(udebug can be found in /usr/athena/bin; "prill" here and below is some 
DB server)
(Also check for "0 of them for write" at the end.  It might matter.)

        /moira/bin/pt_util -x -m -u -g -d prdb.extra -p prdb.old
        perl /moira/bin/pt_util.pl < prdb.extra > prdb.extra.sort
to extract and prepare the personal groups and special user entries in
the old prdb for being reincorporated into the new prdb.

        awk -F\| '$9 == 3 {print $1}' /backup/backup_1/users > /tmp/deactivated

and the following perl script:

#!/usr/athena/bin/perl -w

open(OUT, ">prdb.extra.trimmed");

for ( `cat /tmp/deactivated` ) {
    chop;
    $ex{$_} = 1;
}

$punt = 0;

foreach $L ( `cat prdb.extra.sort` ) {
    @w = split(/ /,$L);
    $_ = $w[0];
    if ( /:/ ) {
        @x = split(/:/,$w[0]);
        if ($ex{$x[0]}) {
            $punt=1;
        } else {
            $punt=0;
        }
    } else {
        # If we got here, we're either a user, a prefixless
        # group, or a group member.
        $punt = 0 if $w[0];
    }
    print OUT $L unless $punt == 1;
}

close(OUT);
exit 0;

to remove the personal groups for users who are deactivated

        awk '/^[^ ][^:]*@/ {printf "KERBEROS:%s\n",$1}' prdb.extra.trimmed \
                > foreign
        blanche afs-foreign-users -f foreign
Get a list of all the @andrew.cmu.edu type (non- athena.mit.edu cell)
users, and sync the Moira list afs-foreign-users to this list.
Moira then adds those entries to the group system:afs-foreign-users,
thus keeping them from being lost in the prdb resync.
Sanity checking the diffs before running the blanche command is recommended.

        awk '/^[^ 0-9][^:@]*$/ {printf "KERBEROS:%s@ATHENA.MIT.EDU\n",$1}' \
                prdb.extra.trimmed > oddities
        awk '/^[^ ][0-9.]* .*$/ {printf "KERBEROS:%s\n",$1}' prdb.extra.trimmed\
                 >> oddities
        echo "LIST:afs-foreign-users" >> oddities
        blanche afs-odd-entities -f oddities
Do the equivalent of afs-foreign-users for domestic users.  We make
the afs-foreign-users list a member of the more general afs-odd-entities.
Sanity checking the diffs before running the blanche command is recommended.

WAIT for the incremental updates from the `blanche` changes to complete.

####   Now the actual resync begins.  Incremental updates must stop.   ####

        touch /moira/afs/noafs
to disable AFS incremental updates during the synchronization.  The
afs.incr (?) will wait 30 minutes on an incremental update before
timing out, so the resync should complete in that time, or list
changes in Moira might need to be propagated by hand.

        /moira/bin/afssync prdb.moira
to dump the prdb data that is in Moira (users, groups, and group
memberships).  This step takes about ten minutes, but can be done
concurrently with the next few steps.

REPEAT the above commands, thus regenerating prdb.trimmed from a now
completely-up-to-date prdb.

*** Make sure the "afssync" command has completed ***

        cp prdb.moira prdb.new
        /moira/bin/pt_util -w -d prdb.extra.trimmed -p prdb.new \
                >& prdb.extra.err
This use of pt_util will presumably log errors about failed user
creations and list additions.  (To start over, do both the `cp` and
`pt_util` again.)  You can filter out the "User or group doesn't exist"
type of lines that were caused by a user deactivation with something
like:
        awk -F\| '$9 == 3 {print $1}' /backup/backup_1/users > /tmp/deactivated
        perl -e 'for(`cat /tmp/deactivated`){ chop; $ex{$_}=1;} \
                foreach $L (`cat prdb.extra.err`){ $f=0; \
                @w=split(/[ :]/,$L); for(@w){ $f=1 if $ex{$_}; } \
                next if $f; print $L; }'
Now, back to the resync.

The only remaining errors should be errors creating system:foo groups,
be cause they already exist.  These generally mean that that group has
an odd user on it (root instance, IP acl, etc.) and can safely be
ignored.

Errors of the form:
Error while creating dcctdw:foo: Badly formed name (group prefix doesn't match owner?)
are probably an indication that a user with personal groups had a
username change (in the past they have also meant that a user with
personal groups was deactivated and the uid was re-used (this was
becasue we didn't trim the prdb.extra.sort file in the past.))
Assuming htese errors are due to a username change, the groups should
be renamed, and you should regenerate prdb.extra.trimmed starting with
a fresh prdb from prill.  (You may want to abort and
rm /moira/afs/noafs and try again later.)

        pts listmax > prdb.listmax
        foreach i ( <db servers> )
            rsh $i -l root -x /bin/athena/detach -a     # detach packs
            rsh $i -l root -x rm -f /usr/afs/db/{prdb.new,pre-resync-prdb}
            rcp -px prdb.new root@${i}:/usr/afs/db/prdb.new
        end                                             # staging
        foreach i ( <db servers> )
            bos shutdown $i ptserver -wait
            bos exec $i "mv /usr/afs/db/prdb.DB0 /usr/afs/db/pre-resync-prdb; rm /usr/afs/db/prdb.DB*; mv /usr/afs/db/prdb.new /usr/afs/db/prdb.DB0"
        end
        foreach i ( <db servers> )
            bos restart $i ptserver
        end

        /moira/bin/udebug prill -port 7002
to watch the status of the servers to make sure things are going well,
where "prill" is preferred db server (the sync site).

Make sure the beacons are working, and that once quorum is established
(~90 seconds) that the servers are resynchronizing their notions of
the databases and that the "dbcurrent" and "up" fields all become set
and the state goes to "1f".  Also, if "sdi" isn't running, watch out
for large rx packet queues on port 7002 using rxdebug, as the
fileservers may get excessively backlogged, and restart servers, if
necessary, if the congestion remains excessive.

        pts listmax
        cat prdb.listmax
and if the id maxima are lower than the saved ones, reset them
appropriately to the saved ones using `pts setmax`.

        pts ex system:administrators
as a good spot check, especially since it has special people.
(also spot check one of the personal groups and perhaps, something like
the membership of rcmd.reynelda)

        rm /moira/afs/noafs
to remove the lock file and let Moira's afs incrementals continue.

        The afssync program doesn't deal with null instance KERBEROS
members of lists which are groups (example: if LIST zacheiss contains
KERBEROS zacheiss@ATHENA.MIT.EDU).  To get around this, run:

/moira/bin/sync.pl

Which will create /var/tmp/sync.out, which contains the pts commands
needed to add all the null instance KERBEROS members back to the pts
groups they belong in.  If it looks sane, run:

sh /var/tmp/sync.out

Any failed additions are probably from lists that contain both USER
username and KERBEROS username@ATHENA.MIT.EDU.

NOTES

1. Don't do this when you're tired...  There may be no cleanup procedure
available, with certain mistakes.

2. /moira/afs/noafs is only good for 30 minutes.  Keep track of the
critical log, and you may have to do some operations by hand when the
operation is complete.  Also, if requests depend on other requests, they
may be processed out of order, and fail, and may need to be done by hand.