Slow refresh - delay or timeout - source Amazon Redshift

[nandic]

Hi,

In the last few weeks sometimes dataset refresh takes much more than it should.
For 2 years it was perfect, everyday it took the same time.
And now, one days it takes 2 minutes, then it takes 30 minutes, then 5 days normal (2 minutes), then again some issues.

From AWS Redshift side everything looks fine. There are no issues, no throtling, no networks issues etc.
When we refresh dataset from local machine, it always works fine, alway. We never saw the issue when refreshing from local machine.

We only see this issue on datasets refreshed via Power BI Service.
And we use F128 capacity, there are no issues when checking monitoring.

Ticket was raised, IDs provided, screenshots provided, the answer "they don't know what is the issue, recommendation: we should schedule refreshe to another time".

Do you have any suggestions, does anyone has similar experience?

Notes: it has impact on almost all semantic models on our tenant. Data structure hasn't changed for years.
No new tables, calculations or something like that.

So, since November, someting wierd is happening. Just to see if it is only within my company? 🙂

Two dataset refreshes below:

Thanks in advance,
Nemanja

[nandic]

@R1k91 it is public network, no gateway is used.
@rohit1991 thank you for the information. This is also my opinion, but the issue is that when i look at capacity performance, everything looks ok, normal, there is no increase in usage or something like that. 
So i can't find the reason, the proof what is the issue 🙂 
And it appears for around 2 weeks now..

Thank you!

[v-nmadadi-msft]

Hi @nandic ,

Verify whether your capacity is being impacted by surge protection or throttling. When the workload exceeds the available capacity limits, Power BI may temporarily slow down or block certain operations to stabilize performance. Review your capacity metrics such as CPU utilization, memory usage, and refresh queue length  in the Capacity Metrics app or the Fabric Admin Portal. If you notice spikes, throttling events, or sustained high usage during the time of the issue, it may indicate that the capacity is overloaded. Addressing these bottlenecks by optimizing workloads, adjusting refresh schedules, or upgrading capacity can help prevent further interruptions

Understand your Fabric capacity throttling - Microsoft Fabric | Microsoft Learn

Surge protection - Microsoft Fabric | Microsoft Learn

I hope this information helps. Please do let us know if you have any further queries.
Thank you

[R1k91]

are you connecting on the public network to redshift or through Gateway?

[rohit1991 · Accepted solution]

Hii @nandic 

This doesn’t look like a Redshift or network issue because your Desktop refresh is always fast. Since the delays started only in November and happen only in the Power BI Service, it’s most likely a Service-side capacity/load issue on your F128 backend. We’ve seen similar cases where refresh time jumps from 2–3 minutes to 30+ minutes without any model changes.
Try checking capacity metrics, moving the refresh to a quieter time, or running Diagnose Refresh to confirm where the slowdown occurs  but the root cause is almost certainly on the Power BI Service side, not your data source.