Skip to content

Commit 991d59e

Browse files
jstrachanJenkins CI
authored andcommitted
reorder the FAQ to make our recommendations a bit more obvious
1 parent b1712fa commit 991d59e

File tree

1 file changed

+5
-3
lines changed

1 file changed

+5
-3
lines changed

FAQ.md

Lines changed: 5 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -19,10 +19,12 @@ A typical fabric8 installation has a ZooKeeper ensemble (a number of ZooKeeper s
1919

2020
If there are only exactly 2 data centres and either is allowed to fail and the other takes over; then there's no real way to solve this the single-ZooKeeper ensemble route automatically; since there's no way to achieve quorum in either data centre. (Its trivial to have, say, 2 ZK servers in one DC X and 1 in the DC Y; but then if DC X fails; you can't achieve quorum automatically.
2121

22-
There’s a few ways this could work; it depends on trade-offs really. Here's a few options
23-
24-
* pick one DC as the master and run more ZK servers there. If that DC fails, you have to manually decide its down (and make sure the ZK servers are manually taken down) then run some more in the remaining DC. i.e. a manual failover process if the master fails (though automatic if the slave fails).
22+
There’s a few ways this could work; it depends on trade-offs really. Our recommended approach is:
2523

2624
* run 2 ZK ensembles in each DC. This then means things work fine; either DC can fail at any time. It does mean that you cannot have one global master service any more; you can only have DC-local masters (e.g. a master broker in each DC). Then you store/forward between the DCs. There's a [pending issue](https://github.com/fabric8io/fabric8/issues/622) to try make a 2-data centre and 2-fabric deployment appear more like a single logical fabric (by having a form of store/foward bridging at the ZK level).
2725

26+
If you want to favour one DC so that its the master and have some manual recovery mechanism you could consider this approach (though its not really recommended):
27+
28+
* pick one DC as the master and run more ZK servers there. If that DC fails, you have to manually decide its down (and make sure the ZK servers are manually taken down) then run some more in the remaining DC. i.e. a manual failover process if the master fails (though automatic if the slave fails).
29+
2830

0 commit comments

Comments
 (0)