[MOBY-l] Re: [moby] Re: [MOBY-dev] lease versus agent for registry updating

[Phillip Lord]

>>>>> "Tom" == Tom Oinn <tmo at ebi.ac.uk> writes:

  Tom> Phillip Lord wrote:
  >>>>>>> "Mark" == Mark Wilkinson <markw at illuminae.com> writes:
  Mark> We're certainly of like-mind at this end, though certain
  Mark> people ;-) at myGrid think that the agent is the wrong way to
  Mark> go... I just don't see it.  IMO a leasing model would be akin
  Mark> to leasing space in Google!
  >> The lease is the more common idiom, I think. I don't think that
  >> there is a huge difference. My main criticism of the agent is
  >> that there does not seem to be a different connection between the
  >> agent poll and the service being available.

  Tom> Much as I hate* to agree with Phil I, um, do :) Seems to me
  Tom> that the agent approach is more part of a monitoring framework

It might be if it monitored the service rather than something else!

  Tom> than a registry, I don't know offhand whether we're planning to
  Tom> include metrics such as service liveness and response times as
  Tom> transient metadata in the registry but that's where I'd see
  Tom> agents fitting in rather than in the initial registration.


The current approach that feta takes to the "registry" is that it is a
very dumb thing. It really only uses it for storing files (which is
why it uses a webdav backend by default). 

It would be lovely to replace or augment this with something
richer, which understood liveness and response. Or, alternatively,
have a modular search system which could combine responses from feta
and some arbitrary service search system based on information about
liveness. 

It's all very feasible. Even a simple minded response time data model
would help a lot. 

Phil