[sheepdog-users] Crash if not all zookeeper nodes are active

[Valerio Pachera]

2014-10-24 5:06 GMT+02:00 Ruoyu <liangry at ucweb.com>:
> Do you mean there is no problem now?

The problem for me is that I can't start the claster if I can't
contact all zookeeper nodes.
And that problem is still there.
Consider a small cluster of 3 nodes with copies 3.
Each node is also a zookeer node.
For any reason I have to stop the cluster and start it temporarily
with a single node.
Well, it will fail.

> Or the crash cannot be reproduced?

Now it exits with 2 errors line

Oct 23 23:00:06  ERROR [main] zk_init(1394) failed to connect to zk
server 192.168.2.45:2181,192.168.2.46:2181,192.168.2.47:2181 after 300
retries
Oct 23 23:00:06  ERROR [main] main(877) failed to create sheepdog cluster

instead of

Oct 21 09:19:47  EMERG [main] sd_backtrace(833) sheep.c:270: crash_handler
and all the error lines I reported previously

So I'm in doubt if it's crashing or it's a design matter now.


2014-10-24 5:06 GMT+02:00 Ruoyu <liangry at ucweb.com>:
> Do you mean there is no problem now?

The problem for me is that I can't start the claster if I can't
contact all zookeeper nodes.
And that problem is still there.
Consider a small cluster of 3 nodes with copies 3.
Each node is also a zookeer node.
For any reason I have to stop the cluster and start it temporarily
with a single node.
Well, it will fail.

> Or the crash cannot be reproduced?

Now it exits with 2 errors line

Oct 23 23:00:06  ERROR [main] zk_init(1394) failed to connect to zk
server 192.168.2.45:2181,192.168.2.46:2181,192.168.2.47:2181 after 300
retries
Oct 23 23:00:06  ERROR [main] main(877) failed to create sheepdog cluster

instead of

Oct 21 09:19:47  EMERG [main] sd_backtrace(833) sheep.c:270: crash_handler
and all the error lines I reported previously

So I'm in doubt if it's crashing or it's a design matter now.

This is a bit a corner case but I'm curious to know if it was meant to
behave like that or it's in fact a bug.