|
|
-
order of async requests to zookeeper
Ted Yu 2011-12-23, 02:09
Hi, The following originated from discussion on a feature in HBase which relies on zookeeper for coordination.
A task is removed from the tasks map. An async ZK request is pending to remove the task's zk node.
Another request arrives for this to-be-deleted task. A new in-memory task structure is created. A ZK request is scheduled to create the same task node.
Problem will only occur if create-node executes before delete-node in ZK. (If create-node returns NODEEXISTS then we treat that as success).
Does ZK re-order async requests?
Thanks
-
Re: order of async requests to zookeeper
Benjamin Reed 2011-12-23, 02:45
are all of these requests being issued from the same zookeeper handle?
ben
On Thu, Dec 22, 2011 at 6:09 PM, Ted Yu <[EMAIL PROTECTED]> wrote: > Hi, > The following originated from discussion on a feature in HBase which relies > on zookeeper for coordination. > > A task is removed from the tasks map. An async ZK request is pending to > remove the task's zk node. > > Another request arrives for this to-be-deleted task. A new in-memory task > structure is created. A ZK request is scheduled to create the same task > node. > > Problem will only occur if create-node executes before delete-node in ZK. > (If create-node returns NODEEXISTS then we treat that as success). > > Does ZK re-order async requests? > > Thanks
-
Re: order of async requests to zookeeper
Ted Yu 2011-12-23, 18:43
The requests come from same zhandle.
Cheers
On Thu, Dec 22, 2011 at 6:09 PM, Ted Yu <[EMAIL PROTECTED]> wrote:
> Hi, > The following originated from discussion on a feature in HBase which > relies on zookeeper for coordination. > > A task is removed from the tasks map. An async ZK request is pending to > remove the task's zk node. > > Another request arrives for this to-be-deleted task. A new in-memory > task structure is created. A ZK request is scheduled to create the same > task node. > > Problem will only occur if create-node executes before delete-node in > ZK. (If create-node returns NODEEXISTS then we treat that as success). > > Does ZK re-order async requests? > > Thanks >
-
Re: order of async requests to zookeeper
Mahadev Konar 2011-12-27, 00:23
Ted, ZK does not re order async requests. They will be executed in the same order as you call them from the same zk handle. Does that help?
thanks mahadev
On Fri, Dec 23, 2011 at 10:43 AM, Ted Yu <[EMAIL PROTECTED]> wrote: > The requests come from same zhandle. > > Cheers > > On Thu, Dec 22, 2011 at 6:09 PM, Ted Yu <[EMAIL PROTECTED]> wrote: > >> Hi, >> The following originated from discussion on a feature in HBase which >> relies on zookeeper for coordination. >> >> A task is removed from the tasks map. An async ZK request is pending to >> remove the task's zk node. >> >> Another request arrives for this to-be-deleted task. A new in-memory >> task structure is created. A ZK request is scheduled to create the same >> task node. >> >> Problem will only occur if create-node executes before delete-node in >> ZK. (If create-node returns NODEEXISTS then we treat that as success). >> >> Does ZK re-order async requests? >> >> Thanks >>
-
Re: order of async requests to zookeeper
Ted Yu 2011-12-27, 01:53
Sure. That helps, Mahadev.
Can I assume that async requests from different zk handles don't have guaranteed order of execution ?
On Mon, Dec 26, 2011 at 4:23 PM, Mahadev Konar <[EMAIL PROTECTED]>wrote:
> Ted, > ZK does not re order async requests. They will be executed in the > same order as you call them from the same zk handle. Does that help? > > thanks > mahadev > > On Fri, Dec 23, 2011 at 10:43 AM, Ted Yu <[EMAIL PROTECTED]> wrote: > > The requests come from same zhandle. > > > > Cheers > > > > On Thu, Dec 22, 2011 at 6:09 PM, Ted Yu <[EMAIL PROTECTED]> wrote: > > > >> Hi, > >> The following originated from discussion on a feature in HBase which > >> relies on zookeeper for coordination. > >> > >> A task is removed from the tasks map. An async ZK request is pending to > >> remove the task's zk node. > >> > >> Another request arrives for this to-be-deleted task. A new in-memory > >> task structure is created. A ZK request is scheduled to create the same > >> task node. > >> > >> Problem will only occur if create-node executes before delete-node in > >> ZK. (If create-node returns NODEEXISTS then we treat that as success). > >> > >> Does ZK re-order async requests? > >> > >> Thanks > >> >
-
Re: order of async requests to zookeeper
Ted Dunning 2011-12-27, 05:23
Well, you have a constraint on the ordering from the ordering for requests from each separate handle.
Other than that, it is hard to see how you could get a stronger ordering.
On Mon, Dec 26, 2011 at 5:53 PM, Ted Yu <[EMAIL PROTECTED]> wrote:
> Sure. > That helps, Mahadev. > > Can I assume that async requests from different zk handles don't have > guaranteed order of execution ? > > On Mon, Dec 26, 2011 at 4:23 PM, Mahadev Konar <[EMAIL PROTECTED] > >wrote: > > > Ted, > > ZK does not re order async requests. They will be executed in the > > same order as you call them from the same zk handle. Does that help? > > > > thanks > > mahadev > > > > On Fri, Dec 23, 2011 at 10:43 AM, Ted Yu <[EMAIL PROTECTED]> wrote: > > > The requests come from same zhandle. > > > > > > Cheers > > > > > > On Thu, Dec 22, 2011 at 6:09 PM, Ted Yu <[EMAIL PROTECTED]> wrote: > > > > > >> Hi, > > >> The following originated from discussion on a feature in HBase which > > >> relies on zookeeper for coordination. > > >> > > >> A task is removed from the tasks map. An async ZK request is pending > to > > >> remove the task's zk node. > > >> > > >> Another request arrives for this to-be-deleted task. A new in-memory > > >> task structure is created. A ZK request is scheduled to create the > same > > >> task node. > > >> > > >> Problem will only occur if create-node executes before delete-node in > > >> ZK. (If create-node returns NODEEXISTS then we treat that as success). > > >> > > >> Does ZK re-order async requests? > > >> > > >> Thanks > > >> > > >
-
Re: order of async requests to zookeeper
Patrick Hunt 2011-12-27, 19:05
On Mon, Dec 26, 2011 at 5:53 PM, Ted Yu <[EMAIL PROTECTED]> wrote: > Can I assume that async requests from different zk handles don't have > guaranteed order of execution ?
Not sure what you mean. There will be _an_ order, and you're guaranteed that all sessions will see the same order. The order will be the order in which the ZK Leader sees the requests. Subsequent to that everything is strongly ordered, regardless if it's the same session or not.
Patrick
|
|