IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an
email to Administrator. User accounts are meant only to access repo
and report issues and/or generate pull requests.
This is a purpose-specific Git hosting for
BaseALT
projects. Thank you for your understanding!
Только зарегистрированные пользователи имеют доступ к сервису!
Для получения аккаунта, обратитесь к администратору.
Redundant RELEASE_IPs gives nodes a preview of where an IP address
will move to. However, if the associated TAKEOVER_IP fails then the
node will actually be unhosted.
This is similar to commit 77a29b3733.
BUG: https://bugzilla.samba.org/show_bug.cgi?id=12470
Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
Autobuild-User(master): Amitay Isaacs <amitay@samba.org>
Autobuild-Date(master): Fri Dec 16 12:32:02 CET 2016 on sn-devel-144
Otherwise it prints 4294967295 for the PNN.
BUG: https://bugzilla.samba.org/show_bug.cgi?id=12470
Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
The GET_PUBLIC_IP_INFO control fails for unassigned addresses because
PNN is CTDB_UNKNOWN_PNN.
BUG: https://bugzilla.samba.org/show_bug.cgi?id=12470
Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
First argument to talloc_asprintf_append() is the string being
appended to, not a talloc context.
BUG: https://bugzilla.samba.org/show_bug.cgi?id=12470
Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
The new hash-table-based method of merging the IP information does not
sort, whereas the RB-tree method implicitly sorted. This probably
only really matters for the "all" case, but sort regardless to ensure
consistent output format.
Sorting has to be done here instead of when printing to ensure
consistency between ip[] and ipinfo[].
No longer reverse the sort order.
BUG: https://bugzilla.samba.org/show_bug.cgi?id=12470
Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
Add new function ctdb_sock_addr_cmp(), which returns a 3-way result
useful for qsort(3). Reimplent ctdb_sock_addr_same() using this.
In the process, make arguments const so that ctdb_sock_addr_cmp() can
be used with qsort().
BUG: https://bugzilla.samba.org/show_bug.cgi?id=12470
Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
The name of the addresses file to modify is based on the original
selection of a test node at the top of the test. Repeating the
selection a test node can result in a mismatch between the new test
node and the addresses file. This occurs on local daemons, because
the addresses file name has the original node number in it but the
test is being performed on the the newly selected node number.
For some reason this test has only occasionally failed. An upcoming
commit that stops the output of "ctdb ip" from being reversed causes
this test to fail (nearly?) every time.
BUG: https://bugzilla.samba.org/show_bug.cgi?id=12470
Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
When the test is over, the exit_hook will remove the temporary event
script directory and then CTDB is restarted. Explicitly shutting down
CTDB ensures that event script directory is not removed while CTDB is
still running.
Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>
This simplifies comparing the output to the expected output.
Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>
This gets rid of implicit check if a service needs to configured. As a
side effect, we also get rid of the monitor "replay" which was
introduced to avoid a collision between a script executed via event and
manually. Event scripts are not expected to be run by hand.
Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>
This will help get rid of implicit ctdb_service_check_reconfigure.
We still need to keep "reconfigure" event in 13.per_ip_routing, so that
the per ip routing can be refreshed if the configuration has changed.
The correct fix for this is to add caching of configuration and checking
of configuration changes in "ipreallocated" event.
Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>
These features are going away. There is nothing to reconfigure for
NFS anyway.
Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
This standalone test script performs the following:
- using ctdb_mutex_ceph_rados_helper, take a lock on the Ceph RADOS
object a CLUSTER/$POOL/$OBJECT using the Ceph keyring for $USER
+ confirm that lock is obtained, via ctdb_mutex_ceph_rados_helper "0"
output
- check RADOS object lock state, using the "rados lock info" command
- attempt to obtain the lock again, using ctdb_mutex_ceph_rados_helper
+ confirm that the lock is not successfully taken
- tell the first locker to drop the lock and exit, via SIGTERM
- once the first locker has exited, attempt to get the lock again
+ confirm that this attempt succeeds
Signed-off-by: David Disseldorp <ddiss@samba.org>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
Autobuild-User(master): Amitay Isaacs <amitay@samba.org>
Autobuild-Date(master): Fri Dec 9 07:59:33 CET 2016 on sn-devel-144
ctdb_mutex_ceph_rados_helper implements the cluster mutex helper API
atop Ceph using the librados rados_lock_exclusive()/rados_unlock()
functionality.
Once configured, split brain avoidance during CTDB recovery will be
handled using locks against an object located in a Ceph RADOS pool.
Signed-off-by: David Disseldorp <ddiss@samba.org>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
Disable generation/installation of the etcd cluster mutex helper by
default. Support can be explicitly enabled at configure time with
--enable-etcd-reclock.
Signed-off-by: David Disseldorp <ddiss@samba.org>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>
This introduces a mutex helper called ctdb_etcd_lock, which allows CTDB to
use an existing etcd cluster to provide the functionality of a recovery lock
using the API outlined in ctdb/doc/cluster_mutex_helper.txt.
Signed-off-by: Jose A. Rivera <jarrpa@samba.org>
Reviewed-by: David Disseldorp <ddiss@samba.org>
Autobuild-User(master): José A. Rivera <jarrpa@samba.org>
Autobuild-Date(master): Mon Dec 5 19:39:10 CET 2016 on sn-devel-144
The recovery helper does it's own logging, so there is no need to
pass logfd.
Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>
Autobuild-User(master): Martin Schwenke <martins@samba.org>
Autobuild-Date(master): Mon Dec 5 11:59:42 CET 2016 on sn-devel-144
This avoids the extra argument of logfd to ctdb_lock_helper. The log
messages from lock helper are captured by ctdbd.
Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>
- Use fprintf() before logging is initialized
- replace DEBUG_ALERT with DEBUG_ERR
Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>
This extracts the code from following files:
- server/ctdb_logging.c
- server/ctdb_logging_file.c
- server/ctdb_logging_syslog.c
This is in preparation for each daemon (and some processes) doing
their own loging instead of relying on CTDB.
Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>
The plan is to switch to Samba's (incompatible) debug levels, so just
expect CTDB users to use the symbolic name for the debug level.
Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
Use a "bitmap" of available IPs for each IP address instead of walking
the list of available IP addresses.
For ctdb/tests/takeover/lcp2.030.sh, this improves the time taken on
my laptop from:
real 0m11.997s
user 0m11.960s
sys 0m0.000s
to
real 0m8.571s
user 0m8.544s
sys 0m0.000s
So, when assigning all 900 IP addresses the improvement is about 25%.
For the no-op case (where all IPs are already assigned to nodes), the
extra setup adds a small fraction of a second for 900 IPs.
Intermediate cases result in intermediate improvements.
Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>