blob: 3e805ebab55b6fbed699a3c23413f8f791104a8a (
plain)
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
|
--- Start phase 1 - Qmgr -------------------------------------------
1) Set timer 1 - TimeToWaitAlive
2) Send CM_REGREQ to all connected(and connecting) nodes
3) Wait until -
a) The precident answers CM_REGCONF
b) All nodes has answered and I'm the candidate -> election won
c) 30s has passed and I'm the candidate -> election won
d) TimeToWaitAlive has passed -> Failure to start
When receiving CM_REGCONF
4) Send CM_NODEINFOREQ to all connected(and connecting) nodes
reported in CM_REGCONF
5) Wait until -
a) All CM_NODEINFO_CONF has arrived
b) TimeToWaitAlive has passed -> Failure to start
6) Send CM_ACKADD to president
7) Wait until -
a) Receive CM_ADD(CommitNew) from president -> I'm in the qmgr cluster
b) TimeToWaitAlive has passed -> Failure to start
NOTE:
30s is hardcoded in 3c.
TimeToWaitAlive should be atleast X sec greater than 30s. i.e. 30+X sec
to support "partial starts"
NOTE:
In 3b, a more correct number (instead of all) would be
N-NG+1 where N is #nodes and NG is #node groups = (N/R where R is # replicas)
But Qmgr has no notion about node groups or replicas
--- Start phase X - Qmgr -------------------------------------------
President - When accepting a CM_REGREQ
1) Send CM_REGCONF to starting node
2) Send CM_ADD(Prepare) to all started nodes + starting node
3) Send CM_ADD(AddCommit) to all started nodes
4) Send CM_ADD(CommitNew) to starting node
Cluster participant -
1) Wait for both CM_NODEINFOREQ from starting and CM_ADD(Prepare) from pres.
2) Send CM_ACKADD(Prepare)
3) Wait for CM_ADD(AddCommit) from president
4) Send CM_ACKADD(AddCommit)
--- Start phase 2 - NdbCntr ----------------------------------------
- Use same TimeToWaitAliveTimer
1) Check sysfile (DIH_RESTART_REQ)
2) Read nodes (from Qmgr) P = qmgr president
3) Send CNTR_MASTER_REQ to cntr(P)
including info in DIH_RESTART_REF/CONF
4) Wait until -
b) Receiving CNTR_START_CONF -> continue
b) Receiving CNTR_START_REF -> P = node specified in REF, goto 3
c) TimeToWaitAlive has passed -> Failure to start
4) Run ndb-startphase 1
--
Initial start/System restart NdbCntr (on qmgr president node)
1) Wait until -
a) Receiving CNTR_START_REQ with GCI > than own GCI
send CNTR_START_REF to all waiting nodes
b) Receiving all CNTR_START_REQ (for all defined nodes)
c) TimeToWait has passed and partition win
d) TimeToWait has passed and partitioning
and configuration "start with partition" = true
2) Send CNTR_START_CONF to all nodes "with filesystem"
3) Wait until -
Receiving CNTR_START_REP for all starting nodes
4) Start waiting nodes (if any)
NOTE:
1c) Partition win = 1 node in each node group and 1 full node group
1d) Pattitioning = at least 1 node in each node group
--
Running NdbCntr
When receiving CNTR_MASTER_REQ
1) If I'm not master send CNTR_MASTER_REF (including master node id)
2) If I'm master
Coordinate parallell node restarts
send CNTR_MASTER_CONF (node restart)
|
