forked from corosync/corosync
-
Notifications
You must be signed in to change notification settings - Fork 0
/
README.recovery
116 lines (95 loc) · 4.42 KB
/
README.recovery
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
SYNCHRONIZATION ALGORITHM:
-------------------------
The synchronization algorithm is used for every service in corosync to
synchronize state of he system.
There are 4 events of the synchronization algorithm. These events are in fact
functions that are registered in the service handler data structure. They
are called by the synchronization system whenever a network partitions or
merges.
init:
Within the init event a service handler should record temporary state variables
used by the process event.
process:
The process event is responsible for executing synchronization. This event
will return a state as to whether it has completed or not. This allows for
synchronization to be interrupted and recontinue when the message queue buffer
is full. The process event will be called again by the synchronization service
if requesed to do so by the return variable returned in process.
abort:
The abort event occurs when during synchronization a processor failure occurs.
activate:
The activate event occurs when process has returned no more processing is
necessary for any node in the cluster and all messages originated by process
have completed.
CHECKPOINT SYNCHRONIZATION ALGORITHM:
------------------------------------
The purpose of the checkpoint syncrhonization algorithm is to synchronize
checkpoints after a paritition or merge of two or more partitions. The
secondary purpose of the algorithm is to determine the cluster-wide reference
count for every checkpoint.
Every cluster contains a group of checkpoints. Each checkpoint has a
checkpoint name and checkpoint number. The number is used to uniquely reference
an unlinked but still open checkpoint in the cluser.
Every checkpoint contains a reference count which is used to determine when
that checkpoint may be released. The algorithm rebuilds the reference count
information each time a partition or merge occurs.
local variables
my_sync_state may have the values SYNC_CHECKPOINT, SYNC_REFCOUNT
my_current_iteration_state contains any data used to iterate the checkpoints
and sections.
checkpoint data
refcount_set contains reference count for every node consisting of
number of opened connections to checkpoint and node identifier
refcount contains a summation of every reference count in the refcount_set
pseudocode executed by a processor when the syncrhonization service calls
the init event
call process_checkpoints_enter
pseudocode executed by a processor when the synchronization service calls
the process event in the SYNC_CHECKPOINT state
if lowest processor identifier of old ring in new ring
transmit checkpoints or sections starting from my_current_iteration_state
if all checkpoints and sections could be queued
call sync_refcounts_enter
else
record my_current_iteration_state
require process to continue
pseudocode executed by a processor when the synchronization service calls
the process event in the SYNC_REFCOUNT state
if lowest processor identifier of old ring in new ring
transmit checkpoint reference counts
if all checkpoint reference counts could be queued
require process to not continue
else
record my_current_iteration_state for checkpoint reference counts
sync_checkpoints_enter:
my_sync_state = SYNC_CHECKPOINT
my_current_iteration_state set to start of checkpont list
sync_refcounts_enter:
my_sync_state = SYNC_REFCOUNT
on event receipt of foreign ring id message
ignore message
pseudocode executed on event receipt of checkpoint update
if checkpoint exists in temporary storage
ignore message
else
create checkpoint
reset checkpoint refcount array
pseudocode executed on event receipt of checkpoint section update
if checkpoint section exists in temporary storage
ignore message
else
create checkpoint section
pseudocode executed on event receipt of reference count update
update temporary checkpoint data storage reference count set by adding
any reference counts in the temporary message set to those from the
event
update that checkpoint's reference count
set the global checkpoint id to the current checkpoint id + 1 if it
would increase the global checkpoint id
pseudocode called when the synchronization service calls the activate event:
for all checkpoints
free all previously committed checkpoints and sections
convert temporary checkpoints and sections to regular sections
copy my_saved_ring_id to my_old_ring_id
pseudocode called when the synchronization service calls the abort event:
free all temporary checkpoints and temporary sections