7 * 2 new configuration variables for 60.nfs eventscript:
9 - CTDB_MONITOR_NFS_THREAD_COUNT
10 - CTDB_NFS_DUMP_STUCK_THREADS
12 See ctdb.sysconfig for details.
14 * Removed DeadlockTimeout tunable. To enable debug of locking issues set
16 CTDB_DEBUG_LOCKS=/etc/ctdb/debug_locks.sh
18 * In overall statistics and database statistics, lock buckets have been
19 updated to use following timings:
21 < 1ms, < 10ms, < 100ms, < 1s, < 2s, < 4s, < 8s, < 16s, < 32s, < 64s, >= 64s
23 * Initscript is now simplified with most CTDB-specific functionality
24 split out to ctdbd_wrapper, which is used to start and stop ctdbd.
26 * Add systemd support.
28 * CTDB subprocesses are now given informative names to allow them to
29 be easily distinguished when using programs like "top" or "perf".
34 * ctdb tool should not exit from a retry loop if a control times out
35 (e.g. under high load). This simple fix will stop an exit from the
36 retry loop on any error.
38 * When updating flags on all nodes, use the correct updated flags. This
39 should avoid wrong flag change messages in the logs.
41 * The recovery daemon will not ban other nodes if the current node
44 * ctdb dbstatistics command now correctly outputs database statistics.
46 * Fixed a panic with overlapping shutdowns (regression in 2.2).
48 * Fixed 60.ganesha "monitor" event (regression in 2.2).
50 * Fixed a buffer overflow in the "reloadips" implementation.
52 * Fixed segmentation faults in ping_pong (called with incorrect
53 argument) and test binaries (called when ctdbd not running).
55 Important internal changes
56 --------------------------
58 * The recovery daemon on stopped or banned node will stop participating in any
61 * Improve cluster wide database traverse by sending the records directly from
62 traverse child process to requesting node.
64 * TDB checking and dropping of all IPs moved from initscript to "init"
67 * To avoid "rogue IPs" the release IP callback now fails if the
68 released IP is still present on an interface.
77 * The "stopped" event has been removed.
79 The "ipreallocated" event is now run when a node is stopped. Use
80 this instead of "stopped".
82 * New --pidfile option for ctdbd, used by initscript
84 * The 60.nfs eventscript now uses configuration files in
85 /etc/ctdb/nfs-rpc-checks.d/ for timeouts and actions instead of
86 hardcoding them into the script.
88 * Notification handler scripts can now be dropped into /etc/ctdb/notify.d/.
90 * The NoIPTakeoverOnDisabled tunable has been renamed to
91 NoIPHostOnAllDisabled and now works properly when set on individual
94 * New ctdb subcommand "runstate" prints the current internal runstate.
95 Runstates are used for serialising startup.
100 * The Unix domain socket is now set to non-blocking after the
101 connection succeeds. This avoids connections failing with EAGAIN
102 and not being retried.
104 * Fetching from the log ringbuffer now succeeds if the buffer is full.
106 * Fix a severe recovery bug that can lead to data corruption for SMB clients.
108 * The statd-callout script now runs as root via sudo.
110 * "ctdb delip" no longer fails if it is unable to move the IP.
112 * A race in the ctdb tool's ipreallocate code was fixed. This fixes
113 potential bugs in the "disable", "enable", "stop", "continue",
114 "ban", "unban", "ipreallocate" and "sync" commands.
116 * The monitor cancellation code could sometimes hang indefinitely.
117 This could cause "ctdb stop" and "ctdb shutdown" to fail.
119 Important internal changes
120 --------------------------
122 * The socket I/O handling has been optimised to improve performance.
124 * IPs will not be assigned to nodes during CTDB initialisation. They
125 will only be assigned to nodes that are in the "running" runstate.
127 * Improved database locking code. One improvement is to use a
128 standalone locking helper executable - the avoids creating many
129 forked copies of ctdbd and potentially running a node out of memory.
131 * New control CTDB_CONTROL_IPREALLOCATED is now used to generate
132 "ipreallocated" events.
134 * Message handlers are now indexed, providing a significant
135 performance improvement.