Multi Node Questions

Hey,
I have a couple of questions regarding multi-node timescale-db:

  1. In case of native replication of chunks, if one Data Node is down then the inserts would fail until the node is detached manually. Is there a way to automate it and also shouldn’t hamper new chunk creations?

  2. All the views are stored in Access Node and this can become a concern for storage when using lots of data like Stock prices. What’s the way to move ahead? And also are we planning for storing the views in the data nodes itself with native replication?

Thanks

1 Like

Hi Subham,

  1. In case of native replication of chunks, if one Data Node is down then the inserts would fail until the node is detached manually. Is there a way to automate it and also shouldn’t hamper new chunk creations?

We will be focusing on HA in multi-node soon and we expect to address this and other similar scenarios when we tackle that next.

  1. All the views are stored in Access Node and this can become a concern for storage when using lots of data like Stock prices. What’s the way to move ahead? And also are we planning for storing the views in the data nodes itself with native replication?

The current design/architecture has the views on the access node for now.

Hey Nikhil,
Thanks for your quick response. A couple more questions:

  1. What’s the ETA that you’re looking to tackle HA in multi-node?
  2. If a DN is down, then new chunk creation would also be routed to that node and eventually would fail. Is there a way which automates it now?
  3. Are there indexes on views as the data in views would grow substantially?
  4. If I use another partition apart from time then will the views also be partitioned on the same?
  5. Also I have installed the latest timescale db and trying to do a perf test using psycopg2 thread pools with locust. And the read latency is around 40 to 50ms and not able to scale as expected. Any suggestions on what parameters can be tweaked?
  1. What’s the ETA that you’re looking to tackle HA in multi-node?

We will start tackling HA in the next couple of quarters

  1. If a DN is down, then new chunk creation would also be routed to that node and eventually would fail. Is there a way which automates it now?

Automates what? As I said earlier, most of this comes under HA and it will be tackled soon…

  1. Are there indexes on views as the data in views would grow substantially?

Yes, indexes are created on the data in the views for faster access

  1. If I use another partition apart from time then will the views also be partitioned on the same?

The underlying data is actually a hypertable, so yes, it will have many chunks internally.

  1. Also I have installed the latest timescale db and trying to do a perf test using psycopg2 thread pools with locust. And the read latency is around 40 to 50ms and not able to scale as expected. Any suggestions on what parameters can be tweaked?

Might be a good idea to share more details about this on a separate thread…