On 2014年12月10日 20:19, Hitoshi Mitake wrote:
At Wed, 10 Dec 2014 19:24:04 +0800,
wzy8...@139.com wrote:
[1 <multipart/alternative (7bit)>]
[1.1 <text/plain; us-ascii (base64)>]
Hi, All
On the issue, in my production environment, it is very easy to
happen(there is 20 gateway nodes and 10 sheep nodes in my production
environment )
The only way to slove the problem is useing zk_control tools, when the zk node is
too huge, use the tools purge periodically , but I think the tools can not solve the
issue essentially, why not when the zk node in " /sheepdog/queue " is in no
longer use(maybe in the watcher function after watch the event), we delete the it
immediately.
Several months ago, I found the problem in our production environment,
too. At that time, I could not stop and start sheep services without
further permission.
Fortunately, it was zookeeper eating more and more memory, not sheep
itself. Therefore, I wrote the purge feature for zk_control. The
solution is ugly, simple, safe and useful.
It is appreciated if anyone could fix this problem gracefully. In my
opinion, deleting queue node might be danger because we cannot confirm
whether it is in use or will be in use by other sheep nodes.
I'm cc-ing Ruoyu, the maintainer of zookeeper subsystem.
Ruoyu, how do you think about the idea? Is it difficult to implement?
Thanks,
Hitoshi
Best Regards!
Zhengyong Wang
[1.2 <text/html; us-ascii (quoted-printable)>]
[2 <text/plain; us-ascii (7bit)>]
--
sheepdog mailing list
sheepdog@lists.wpkg.org
http://lists.wpkg.org/mailman/listinfo/sheepdog
--
sheepdog mailing list
sheepdog@lists.wpkg.org
http://lists.wpkg.org/mailman/listinfo/sheepdog