[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[LDM #HEQ-649192]: LDM fault tolerance



Geoffry,

> And if the downstream node has a failure for 30 minutes and then a new
> replacement comes up, will it receive notifications for the products that
> appeared during that 30 minutes? That is, do the downstream and upstream
> nodes coordinate acknowledgement of receipt?

Your mental model of the LDM is incorrect. For a connection in primary mode, 
the downstream LDM doesn't receive notifications of product availability; 
instead, the upstream LDM transmits data-products to the downstream LDM that 
match the subscription that the downstream LDM registered with the upstream LDM 
as soon as the products are available at the upstream site. The transmission of 
data-products in this mode is asynchronous and uni-directional (i.e., the 
downstream LDM doesn't reply). For a second connection in alternate mode with 
an identical subscription, the upstream LDM synchronously sends a notification 
of a newly-available data product to the downstream LDM and awaits a reply. 
Typically, the response by the downstream LDM will be that it already has the 
product (from the primary mode connection). Occasionally, however, the 
downstream LDM replies that it doesn't have the product, in which case the 
upstream LDM transmits it. If and when the downstream LDM determines th
 at most of its products are being received on the alternate mode connection, 
then it breaks that connection and re-establishes it in primary mode. The 
reverse is done by the downstream LDM if and when it determines that the 
minority of products are being received on the primary mode connection.

As long as the downstream site is offline for less than the minimum residency 
time of the upstream site's product-queue, the downstream LDM won't miss any 
products (providing the network bandwidth is sufficient, of course).

> What about if a download is started, the downstream node dies, a new node
> comes up—will the new node receive notice of that partially (but not
> successfully downloaded and processed) file?

No. An upstream LDM treats each connection as unique.

> If there's any more detailed documentation of the behavior I'd appreciate
> the pointer.

There were some state transition diagrams, but they were too confusing. You 
might find them in the online documentation for the older versions (e.g., 6.4).

> Cheers,
> Geoffry

Regards,
Steve Emmerson

Ticket Details
===================
Ticket ID: HEQ-649192
Department: Support LDM
Priority: Normal
Status: Closed