Seeing increase in read timeouits when receiving XDR updates

Operations XDR (Cross Data Center Replication)
1

I am not sure if this is actually an XDR question per se as it has to do with the receiving data center. I know that XDR is only configured on the sending side and was not certain if the site being updated makes a distinction between updates from XDR and others.

When I enable XDR, on the receiving site I am seeing an increase in read latency leading to a spike in the occurrence of timeouts. I can mitigate the problem by throttling XDR but I have to throttle down very far to the point where the lag becomes a serious issue. The “high” time out rate is still on the order of 0.5%, so this is a problem affecting only the extremes, but I was wondering if there was a strategy to make sure that writes do not block reads in a situation where returning slightly stale data is preferable to timing out.

2

It is not possible to differentiate how reads are processed based on client writes vs XDR writes on the receiving cluster. If the reads on the receiving cluster are timing out due to the write activity, the receiving cluster needs to be sized accordingly: XDR writes must be taken into account along with client writes and client reads while selecting the SSDs. Throttling source XDR writes is not an option here as you stated since that causes XDR lag. Hope this helps.