There should be one manager of instance type PARTICIPANT per logical participant in the cluster (if each node is a participant, this is a limit of 1). There is no limit to how many managers of type ADMINISTRATOR or SPECTATOR you can create per node.
Also, there is no limit to HelixAdmin (i.e. ZKHelixAdmin) connections you have active. Kanak ---------------------------------------- > Date: Sun, 16 Feb 2014 11:26:18 -0800 > Subject: Re: Helix 0.6.2: ZKHelixManager v/s HelixManagerFactory > From: [email protected] > To: [email protected] > > Hi Kanak, > > Thanks for the quick responses I will follow your instructions but if I use > HelixManagerFactory when possible would I not be risking creating two > ZKHelixManager instances? Does that create any issues? What is the relation > between a node and HelixManager, should there be 1 manager for each node? > > Thanks, > > Sandeep > > > > > On Sun, Feb 16, 2014 at 11:21 AM, Kanak Biscuitwala > <[email protected]>wrote: > >> Hi Sandeep, >> >> Helix separates the administration of the cluster from the participation >> of it. For this, the easiest way is to use HelixAdmin to set up your >> cluster. There is a known issue in HelixManager (HELIX-369) where joining >> as ADMINISTRATOR without a setup cluster will not work correctly, >> unfortunately. >> >> Please use HelixManagerFactory when possible (instead of the >> ZKHelixManager constructor). In 0.7.x we're introducing a refactored >> connection class, and this abstraction will help ensure continued >> compatibility. >> >> Yes, we are exploring alternatives to ZooKeeper when consistency is not a >> strict system requirement. This is all in the extremely early stages. >> >> Kanak >> >> >> ---------------------------------------- >>> Date: Sun, 16 Feb 2014 11:06:17 -0800 >>> Subject: Helix 0.6.2: ZKHelixManager v/s HelixManagerFactory >>> From: [email protected] >>> To: [email protected] >>> >>> Hi, >>> >>> I was using HelixManagerFactory to create instances from InstanceType >> enum. >>> I found that the HelixManagerFactory fails because I do not have a >> cluster >>> created before using the API. I do see from examples that I need to >> create >>> ZKHelixManager and register instances, state machine etc. >>> >>> So one question is when do I use HelixManagerFactory and when do I go >>> straight to the underlying ZKHelixManager? Also I imagine I should not be >>> referencing/using both classes HelixManagerFactory and ZKHelixManager in >> my >>> codebase it appears like two instances of ZKHelixManager are created. The >>> API does not seem to safeguard that because both are public APIs. >>> >>> Also does the HelixManagerFactory indicate that in future there could be >>> other alternatives to HelixManager other than zookeeper? >>> >>> Thanks, >>> >>> Sandeep >>
