Skip to Content.
Sympa Menu

perfsonar-dev - Re: [pS-dev] RFC: basic description of tops interactions

Subject: perfsonar development work

List archive

Re: [pS-dev] RFC: basic description of tops interactions


Chronological Thread 
  • From: Jason Zurawski <>
  • To: ulisses <>
  • Cc:
  • Subject: Re: [pS-dev] RFC: basic description of tops interactions
  • Date: Sun, 24 Sep 2006 21:19:27 -0400

Ulisses;




+TOPSDataRequest (client requests some or all of TOPS db info)
+ TOPSDataResponse
+
+

Ok

+TOPSClientRegisterRequest (client 'registers' with the TOPS,
please note that what I tried to explain in the description I sent is that
the TOPSDataRequest also acts as a registration by including a time to live
field
for the data request, this way the server will push the data each time the
requested
data changes until the ttl expires

+ TOPS provides
+ some sort of 'key' back [client responsible
for
+ updating own key so it doesn't expire? Will
it
+ expire?]. Will data push to registered
clients
+ when changes occur [triggers], the list of
+ registered clients will exist on the TOPS
+ somwhere?)


I understand that you want to save the amount of messages that need to be defined and recognized, but it seems a little complex to roll two functionalities (requesting data directly, registering for push data at a later time) into a single message. Perhaps you can explain why you think a single message accomplishes this goal better?



+ TOPSClientRegisterResponse

IMHO, not needed

Absolutely this is needed, how would it be possible to tell if this succeeded or failed? Even if its just a simple return of an error code, it counts as a message type.

+
+
+
+ [We may need submessages in this category, imitating the LS:]
+
+ ClientRegisterWithTOPS
+ TOPSClientResponseFromTOPS
+
+ ClientUpdateRegistrationWithTOPS
+ TOPSClientResponseFromTOPS
+
+ ClientKeepaliveWithTOPS
+ TOPSClientResponseFromTOPS
+
+ ClientDe-registerWithTOPS
+ TOPSClientResponseFromTOPS
+

I would avoid that much amount of messages, IMHO I think that what i propose
is just enough for now and implements the required funcionality.



Right, you are proposing enough to get by for now, but these are all valid use cases (if we are to follow the general template of how the LS manages data/client interaction) and will eventually need to be dealt with. If your system will periodically 'clean up' the client registrations by TTL it may be nice to have the ability to either 're-request' with the same key (thus updating the TTL) either through the standard registration mechanism or through some sort of keepalive. I suppose de-registration is not required but may be nice if a client wishes to be polite and not continue to utilize system resources when it knows it will be going offline before a TTL is exceeded.

-jason





Archive powered by MHonArc 2.6.16.

Top of Page