I wanted to verify a thought experiment. I have been dealing with excessive lag (measured in hours) for a system where latency is measured in milliseconds and I had a hypothesis as to what went wrong. I have been experimenting with XDR on and off for weeks and for an extended period of time I had disabled XDR by means of setting max-throughput to 0 as opposed to some of the other means of disabling documented in FAQ - How to disable XDR. While it does not state so specifically, my hypothesis was that since I did not explicitly rewind when I restarted, that Aerospike queued up the updates during the time that the max-throughput had been set to 0 and when the max-throughput was restored to its default attempted to ship all the intervening changes. My question is whether this sounds plausible and whether I would get better results by disabling XDR temporarily by disassociating the namespace from the datacenter and later adding it again?
1.Please specify the XDR version you are using. 2. Assuming you are using XDR 5.0, The scenario you paint is totally plausible. If you are using XDR 5.0, you must have EE license, so you must have access to Aerospike Academy. Please review XDR 5.0 course module and it should explain all you ask for.
Yes, 5.3. Also the servers were arranged as a hub and spokes. And yes we are Enterprise customers, thank you (by the way enjoyed your talk in Boston). I had raised this issue in. support ticket but I figured that since my follow-up question was more theoretical it wouldn’t hurt to try the forums! Thanks.
Thank you for your compliment about the Boston Roadshow presentation.