Yeah, when that happens on LeaderLatch, you lose leadership. We use the listener mechanism to take and lose leadership, so if there is a prolonged lack of a connection, then the node just won't be the leader.
--Eric On Tue, Jul 2, 2013 at 9:37 AM, Jordan Zimmerman <[email protected] > wrote: > My general recommendation is to handle SUSPENDED and LOST in the same way. > In the case of LeaderLatch, your code should exit whatever critical section > it has for executing when leader. > > -JZ > > On Jul 2, 2013, at 12:29 PM, chao chu <[email protected]> wrote: > > sorry for the confusion. I knew that LeaderLatch will retry when it got > connection 'LOST' event and one of the participant should be elected as a > new leader when re-connected, however, I meant to ask *what if it stayed > in 'LOST'* for quite a long time (for example, the zk ensemble become > unavailable). I can understand that how to handle this situation should be > very application specific, I was just trying to know what's your reaction > for this in your code (want to see if there are any ideas we can borrow). > > not sure If i explained this clearly enough, thanks a lot for your reply > though. > > > On Tue, Jul 2, 2013 at 6:07 AM, Eric Tschetter <[email protected]> wrote: > >> My understanding is that LeaderLatch already handles those cases for >> you. The unit tests in TestLeaderLatch definitely have something that >> tries to test the LOST case. If there's a case that is not handled, it'd >> probably be best if you could provide a unit test that shows what's not >> handled to help shape the conversation. >> >> --Eric >> >> >> On Fri, Jun 28, 2013 at 8:03 AM, chao chu <[email protected]> wrote: >> >>> Hi Eric, >>> >>> Thanks for your sharing, by looking into your code, it's not very clear >>> to me that how do you handle the 'SUSPEND' or 'LOST' events of LeaderLatch? >>> Could you please shed some lights here? Thanks >>> >>> >>> On Wed, Jun 26, 2013 at 11:52 PM, Eric Tschetter <[email protected]>wrote: >>> >>>> ChuChao, >>>> >>>> We use it in the Druid project (http://www.github.com/metamx/druid/) >>>> >>>> You can see its use in the class com.metamx.druid.master.DruidMaster >>>> >>>> The class has a bunch of other stuff in it as well that is not specific >>>> to the LeaderLatch, but you can just ignore that and see how it handles the >>>> latch. >>>> >>>> --Eric >>>> >>>> >>>> On Wednesday, June 26, 2013, chao chu wrote: >>>> >>>>> Thanks a lot for your reply. Could you please name a few open source >>>>> projects that used LeaderLatch if you are aware of any? I'd like to take a >>>>> look at the code. >>>>> >>>>> btw, What about issues reported in the links I mentioned? are they >>>>> actual bugs or just used in an unexpected way? >>>>> >>>>> >>>>> >>>>> On Wed, Jun 26, 2013 at 7:29 AM, Jordan Zimmerman < >>>>> [email protected]> wrote: >>>>> >>>>>> Curator is being used at major companies (i.e. Netflix, eBay, etc.). >>>>>> Bugs are quickly fixed when reported. In particular, LeaderLatch is >>>>>> widely >>>>>> used. >>>>>> -JZ >>>>>> >>>>>> >>>>>> On Jun 25, 2013, at 11:03 AM, chao chu <[email protected]> wrote: >>>>>> >>>>>> Hi, >>>>>> >>>>>> I have been trying to use the LeaderLatch to implement Leader >>>>>> Election in my project and had written some scripts to simulate the >>>>>> situations when the zk ensemble become unstable due to network problems. >>>>>> It >>>>>> worked well and as expected so far. >>>>>> >>>>>> However, by digging into both zookeeper-users and curator-users >>>>>> mailing lists, there are indeed some bugs/edge cases reported, like >>>>>> LeaderLatch bug causing extra znodes appearing in >>>>>> Zookeeper<https://groups.google.com/forum/?fromgroups#!searchin/curator-users/LeaderLatch/curator-users/to8ViZp6p-E/xYbKbzqkZQYJ> >>>>>> and multiple participants thought they are >>>>>> leader<https://listserv.netflix.com/pipermail/curator-users/2012-October/000201.html> >>>>>> which >>>>>> worried me about the reliability of this. >>>>>> >>>>>> So, my question is that: are there any real world projects are using >>>>>> this recipe which have proved the quality of it, or are there any other >>>>>> known edge cases or open issues? >>>>>> >>>>>> >>>>>> Thanks & Regards, >>>>>> >>>>>> -- >>>>>> ChuChao >>>>>> >>>>>> >>>>>> >>>>> >>>>> >>>>> -- >>>>> ChuChao >>>>> >>>> >>> >>> >>> -- >>> ChuChao >>> >> >> > > > -- > ChuChao > > >
