Architecture of kdb+ systems

[k-mambo]

Architecture of kdb+ systems 

According to the official guide, the KDB system architecture describes the KDB Tick structure.

Which architecture is appropriate if you need to register historical time series data at the current time?

The reason why historical time series data is registered is to newly link existing IoT sites or modify already registered data points.

We understand that this is not appropriate for most time series databases.

[rocuinneagain]

There are some notes on late data in this blog post: https://kx.com/blog/partitioning-data-in-kdb/

It outlines the basics of using lookup tables per partition.

Another option involves taking copies of HDB partitions and upserting the late data to the copy.

Then this new partition with extra data replaces the old one. Coordination is needed to ensure any HDB process reading the data does not error during this process.

[k-mambo]

---

@rocuinneagain wrote:

There are some notes on late data in this blog post: https://kx.com/blog/partitioning-data-in-kdb/

It outlines the basics of using lookup tables per partition.

Another option involves taking copies of HDB partitions and upserting the late data to the copy.

Then this new partition with extra data replaces the old one. Coordination is needed to ensure any HDB process reading the data does not error during this process.

---

we already doing that. Is there any better way?

[rocuinneagain]

Which of the 2 have you implemented? 

What bottlenecks/issues are you finding?