1
00:00:00,867 --> 00:00:02,934
Can Swarm handle failure?

2
00:00:03,967 --> 00:00:07,501
One word answer is, yes, it can.

3
00:00:07,867 --> 00:00:10,934
But more interesting part is, how?

4
00:00:11,767 --> 00:00:14,801
Let's take the previous example of the service running

5
00:00:14,801 --> 00:00:18,234
three replicas of Nginx, each hosted on one worker

6
00:00:18,601 --> 00:00:22,767
or master and our workers are healthy and running.

7
00:00:23,034 --> 00:00:25,701
What if one of the workers go down?

8
00:00:25,701 --> 00:00:29,634
Let's say in this case, worker 3 went down. If that

9
00:00:29,634 --> 00:00:32,933
happens, the task 3 will be sheduled on one

10
00:00:32,933 --> 00:00:36,034
of the other workers. Once worker 3 is back to

11
00:00:36,034 --> 00:00:39,534
its running state, task three might get moved back to it

12
00:00:39,534 --> 00:00:42,834
and if it's not causing any overload on worker 2,

13
00:00:42,834 --> 00:00:45,667
it may just stay there and worker 3

14
00:00:45,667 --> 00:00:47,834
might be ready to host other tasks when they

15
00:00:47,834 --> 00:00:51,067
arrive in future. In a nutshell, if one of the

16
00:00:51,067 --> 00:00:54,501
nodes go down, the other nodes can handle its load.

17
00:00:54,501 --> 00:00:57,067
If the master goes down, though, the workers

18
00:00:57,067 --> 00:00:59,867
perform a mutual election, where one of the

19
00:00:59,867 --> 00:01:02,667
workers gets promoted and the cluster starts

20
00:01:02,667 --> 00:01:06,733
working again. The next question would be how many

21
00:01:06,733 --> 00:01:09,801
nodes can go down without affecting Swarm?

22
00:01:09,967 --> 00:01:13,634
Well, to make sure that the Swarm cluster functions

23
00:01:13,634 --> 00:01:17,967
properly, at least more than half of the nodes should be working.

24
00:01:17,967 --> 00:01:20,867
Minimum number of required working nodes

25
00:01:20,867 --> 00:01:23,534
for a happy Swarm cluster is equal

26
00:01:23,534 --> 00:01:27,967
to the number of total nodes divided by 2 plus 1,

27
00:01:27,967 --> 00:01:30,334
which again, means more than half.

28
00:01:30,334 --> 00:01:31,701
[No Audio]