[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[LDM #OQD-608743]: ldm failover question
- Subject: [LDM #OQD-608743]: ldm failover question
- Date: Wed, 28 Oct 2009 12:24:13 -0600
Sue,
> Can you tell me how the downstream
> LDM makes the determination of which upstream host is PRIMARY
> and which is ALTERNATE ? Does this determination happen
> on a product by product basis ? Is it traceable by the LDM user?
A downstream LDM process keeps track of its success in adding data-products to
the product-queue and also knows how many other downstream LDM peer-processes
have requested the same data.
A downstream LDM process in alternate mode will disconnect and reconnect in
primary mode if and only if it has been more successful in adding data-products
to the product-queue than any other peer-process over the last minute.
Similarly, a downstream LDM process in primary mode will disconnect and
reconnect in alternate mode if and only if it has not been more successful in
adding data-products to the product-queue than any other peer-process over the
last minute.
Downstream LDM processes log when they switch modes.
The rtstats(1) web-pages can be used to determine the relative efficiency of
routes.
> I am unclear on what it means to "ensure that the data-product signatures
> (i.e., MD5 checksums) [are] computed identically. "
> How does the LDM user do this?
The product-signature is computed by the LDM ingester responsible for creating
the data-product and adding it to the product-queue.
> Certainly the script which creates product ids, feed-type,
> and the insert command are indentical on upstream hosts.
If the data-products are inserted using the pqinsert(1) utility in identical
ways, then the product-signatures will be identical.
> It is possible that the underlying data for a particular
> product id could differ if one host is experiencing
> some data latency issues that effect product generation.
> Is that an issue that could cause duplicate products( product ids)
> in the queue?
The only way I can think of that this would happen is if the pqinsert(1)
utility was told to use the product-identifier rather than the product data as
the basis for the product-signature and the identifier encoded the time.
Regards,
Steve Emmerson
Ticket Details
===================
Ticket ID: OQD-608743
Department: Support LDM
Priority: Normal
Status: Closed