7 * The locking code now correctly implements a per-database active
8 locks limit. Whole database lock requests can no longer be denied
9 because there are too many active locks - this is particularly
10 important for freezing databases during recovery.
12 * The debug_locks.sh script locks against itself. If it is already
13 running then subsequent invocations will exit immediately.
15 * ctdb tool commands that operate on databases now work correctly when
16 a database ID is given.
18 * Various code fixes for issues found by Coverity.
20 Important internal changes
21 --------------------------
23 * statd-callout has been updated so that statd client information is
24 always up-to-date across the cluster. This is implemented by
25 storing the client information in a persistent database using a new
26 "ctdb ptrans" command.
28 * The transaction code for persistent databases now retries until it
29 is able to take the transaction lock. This makes the transation
30 semantics compatible with Samba's implementation.
32 * Locking helpers are created with vfork(2) instead of fork(2),
33 providing a performance improvement.
35 * config.guess has been updated to the latest upstream version so CTDB
36 should build on more platforms.
45 * The default location of the ctdbd socket is now:
47 /var/run/ctdb/ctdbd.socket
49 If you currently set CTDB_SOCKET in configuration then unsetting it
50 will probably do what you want.
52 * The default location of CTDB TDB databases is now:
56 If you only set CTDB_DBDIR (to the old default of /var/ctdb) then
57 you probably want to move your databases to /var/lib/ctdb, drop your
58 setting of CTDB_DBDIR and just use the default.
60 To maintain the database files in /var/ctdb you will need to set
61 CTDB_DBDIR, CTDB_DBDIR_PERSISTENT and CTDB_DBDIR_STATE, since all of
64 * Use of CTDB_OPTIONS to set ctdbd command-line options is no longer
65 supported. Please use individual configuration variables instead.
67 * Obsolete tunables VacuumDefaultInterval, VacuumMinInterval and
68 VacuumMaxInterval have been removed. Setting them had no effect but
69 if you now try to set them in a configuration files via CTDB_SET_X=Y
70 then CTDB will not start.
72 * Much improved manual pages. Added new manpages ctdb(7),
73 ctdbd.conf(5), ctdb-tunables(7). Still some work to do.
75 * Most CTDB-specific configuration can now be set in
78 This avoids cluttering distribution-specific configuration files,
79 such as /etc/sysconfig/ctdb. It also means that we can say: see
80 ctdbd.conf(5) for more details. :-)
82 * Configuration variable NFS_SERVER_MODE is deprecated and has been
83 replaced by CTDB_NFS_SERVER_MODE. See ctdbd.conf(5) for more
86 * "ctdb reloadips" is much improved and should be used for reloading
87 the public IP configuration.
89 This commands attempts to yield much more predictable IP allocations
90 than using sequences of delip and addip commands. See ctdb(1) for
93 * Ability to pass comma-separated string to ctdb(1) tool commands via
94 the -n option is now documented and works for most commands. See
97 * "ctdb rebalancenode" is now a debugging command and should not be
98 used in normal operation. See ctdb(1) for details.
100 * "ctdb ban 0" is now invalid.
102 This was documented as causing a permanent ban. However, this was
103 not implemented and caused an "unban" instead. To avoid confusion,
104 0 is now an invalid ban duration. To administratively "ban" a node
105 use "ctdb stop" instead.
107 * The systemd configuration now puts the PID file in /run/ctdb (rather
108 than /run/ctdbd) for consistency with the initscript and other uses
114 * Traverse regression fixed.
116 * The default recovery method for persistent databases has been
117 changed to use database sequence numbers instead of doing
118 record-by-record recovery (using record sequence numbers). This
119 fixes issues including registry corruption.
121 * Banned nodes are no longer told to run the "ipreallocated" event
122 during a takeover run, when in fallback mode with nodes that don't
123 support the IPREALLOCATED control.
125 Important internal changes
126 --------------------------
128 * Persistent transactions are now compatible with Samba and work
131 * The recovery master role has been made more stable by resetting the
132 priority time each time a node becomes inactive. This means that
133 nodes that are active for a long time are more likely to retain the
134 recovery master role.
136 * The incomplete libctdb library has been removed.
138 * Test suite now starts ctdbd with the --sloppy-start option to speed
139 up startup. However, this should not be done in production.
148 * A missing network interface now causes monitoring to fail and the
149 node to become unhealthy.
151 * Changed ctdb command's default control timeout from 3s to 10s.
153 * debug-hung-script.sh now includes the output of "ctdb scriptstatus"
154 to provide more information.
159 * Starting CTDB daemon by running ctdbd directly should not remove
160 existing unix socket unconditionally.
162 * ctdbd once again successfully kills client processes on releasing
163 public IPs. It was checking for them as tracked child processes
164 and not finding them, so wasn't killing them.
166 * ctdbd_wrapper now exports CTDB_SOCKET so that child processes of
167 ctdbd (such as uses of ctdb in eventscripts) use the correct socket.
169 * Always use Jenkins hash when creating volatile databases. There
170 were a few places where TDBs would be attached with the wrong flags.
172 * Vacuuming code fixes in CTDB 2.2 introduced bugs in the new code
173 which led to header corruption for empty records. This resulted
174 in inconsistent headers on two nodes and a request for such a record
175 keeps bouncing between nodes indefinitely and logs "High hopcount"
176 messages in the log. This also caused performance degradation.
178 * ctdbd was losing log messages at shutdown because they weren't being
179 given time to flush. ctdbd now sleeps for a second during shutdown
180 to allow time to flush log messages.
182 * Improved socket handling introduced in CTDB 2.2 caused ctdbd to
183 process a large number of packets available on single FD before
184 polling other FDs. Use fixed size queue buffers to allow fair
185 scheduling across multiple FDs.
187 Important internal changes
188 --------------------------
190 * A node that fails to take/release multiple IPs will only incur a
191 single banning credit. This makes a brief failure less likely to
192 cause node to be banned.
194 * ctdb killtcp has been changed to read connections from stdin and
195 10.interface now uses this feature to improve the time taken to kill
198 * Improvements to hot records statistics in ctdb dbstatistics.
200 * Recovery daemon now assembles up-to-date node flags information
201 from remote nodes before checking if any flags are inconsistent and
204 * ctdbd no longer creates multiple lock sub-processes for the same
205 key. This reduces the number of lock sub-processes substantially.
207 * Changed the nfsd RPC check failure policy to failover quickly
208 instead of trying to repair a node first by restarting NFS. Such
209 restarts would often hang if the cause of the RPC check failure was
210 the cluster filesystem or storage.
212 * Logging improvements relating to high hopcounts and sticky records.
214 * Make sure lower level tdb messages are logged correctly.
216 * CTDB commands disable/enable/stop/continue are now resilient to
217 individual control failures and retry in case of failures.
226 * 2 new configuration variables for 60.nfs eventscript:
228 - CTDB_MONITOR_NFS_THREAD_COUNT
229 - CTDB_NFS_DUMP_STUCK_THREADS
231 See ctdb.sysconfig for details.
233 * Removed DeadlockTimeout tunable. To enable debug of locking issues set
235 CTDB_DEBUG_LOCKS=/etc/ctdb/debug_locks.sh
237 * In overall statistics and database statistics, lock buckets have been
238 updated to use following timings:
240 < 1ms, < 10ms, < 100ms, < 1s, < 2s, < 4s, < 8s, < 16s, < 32s, < 64s, >= 64s
242 * Initscript is now simplified with most CTDB-specific functionality
243 split out to ctdbd_wrapper, which is used to start and stop ctdbd.
245 * Add systemd support.
247 * CTDB subprocesses are now given informative names to allow them to
248 be easily distinguished when using programs like "top" or "perf".
253 * ctdb tool should not exit from a retry loop if a control times out
254 (e.g. under high load). This simple fix will stop an exit from the
255 retry loop on any error.
257 * When updating flags on all nodes, use the correct updated flags. This
258 should avoid wrong flag change messages in the logs.
260 * The recovery daemon will not ban other nodes if the current node
263 * ctdb dbstatistics command now correctly outputs database statistics.
265 * Fixed a panic with overlapping shutdowns (regression in 2.2).
267 * Fixed 60.ganesha "monitor" event (regression in 2.2).
269 * Fixed a buffer overflow in the "reloadips" implementation.
271 * Fixed segmentation faults in ping_pong (called with incorrect
272 argument) and test binaries (called when ctdbd not running).
274 Important internal changes
275 --------------------------
277 * The recovery daemon on stopped or banned node will stop participating in any
280 * Improve cluster wide database traverse by sending the records directly from
281 traverse child process to requesting node.
283 * TDB checking and dropping of all IPs moved from initscript to "init"
286 * To avoid "rogue IPs" the release IP callback now fails if the
287 released IP is still present on an interface.
296 * The "stopped" event has been removed.
298 The "ipreallocated" event is now run when a node is stopped. Use
299 this instead of "stopped".
301 * New --pidfile option for ctdbd, used by initscript
303 * The 60.nfs eventscript now uses configuration files in
304 /etc/ctdb/nfs-rpc-checks.d/ for timeouts and actions instead of
305 hardcoding them into the script.
307 * Notification handler scripts can now be dropped into /etc/ctdb/notify.d/.
309 * The NoIPTakeoverOnDisabled tunable has been renamed to
310 NoIPHostOnAllDisabled and now works properly when set on individual
313 * New ctdb subcommand "runstate" prints the current internal runstate.
314 Runstates are used for serialising startup.
319 * The Unix domain socket is now set to non-blocking after the
320 connection succeeds. This avoids connections failing with EAGAIN
321 and not being retried.
323 * Fetching from the log ringbuffer now succeeds if the buffer is full.
325 * Fix a severe recovery bug that can lead to data corruption for SMB clients.
327 * The statd-callout script now runs as root via sudo.
329 * "ctdb delip" no longer fails if it is unable to move the IP.
331 * A race in the ctdb tool's ipreallocate code was fixed. This fixes
332 potential bugs in the "disable", "enable", "stop", "continue",
333 "ban", "unban", "ipreallocate" and "sync" commands.
335 * The monitor cancellation code could sometimes hang indefinitely.
336 This could cause "ctdb stop" and "ctdb shutdown" to fail.
338 Important internal changes
339 --------------------------
341 * The socket I/O handling has been optimised to improve performance.
343 * IPs will not be assigned to nodes during CTDB initialisation. They
344 will only be assigned to nodes that are in the "running" runstate.
346 * Improved database locking code. One improvement is to use a
347 standalone locking helper executable - the avoids creating many
348 forked copies of ctdbd and potentially running a node out of memory.
350 * New control CTDB_CONTROL_IPREALLOCATED is now used to generate
351 "ipreallocated" events.
353 * Message handlers are now indexed, providing a significant
354 performance improvement.