Event delivery time in Forge vs Connect

I am currently migrating a Confluence app from Connect to Forge. For the benefit of the community, I wanted to share my findings regarding event delivery latency across the two platforms.

These stats were collected for one specific event type over a period of 48 hours starting on October 25 2026. I had the same Connect-on-Forge application configured to receive both Connect webhooks and Forge triggers for the same events, and it tracked the difference between the event creation time and the local time of receipt. This was measured with production workloads.

Connect is in green and Forge is in blue. Note that the mean and P95 charts are linear scale, but the P99 and max charts are log scale.

Atlassian is presently working very hard to reduce or prevent excess delays (ECO-1848). Just be aware that, when designing features, Atlassian has indicated that the expected delivery time on Forge is “up to 3 minutes”.




Thank your collecting these data!

Forge seems really slow when compared to Connect.

Thanks for sharing Scott.

Like you mentioned, we’re working hard on bringing that delivery time down as a top priority and folks can follow ECO-1848 for updates.

Did the P99 and max outliers cluster in particular hours, or were they spread evenly over the 48 hours? If they cluster, it would be worth lining those hours up against the Forge status page, since an “up to 3 minutes” target says nothing about the tail.

Hi all,

Providing an update here to say that overall we have seen a significant improvement in the time taken to process events through the Forge events pipeline over the last 3 weeks - thanks to the hard-working efforts of the engineering team :slight_smile:

Observing last week (Monday to Friday), p95 delivery across all regions was 1.17ms and p99 was 183.57ms (over 1 billion events processed).

We’re continuing to make efforts to bring down the p99 - the delivery spikes are mostly caused by individual noisy tenants.

Hey @HeyJoe

That’s great news! For the 1.17 ms mean, I imagine that this timing refers to an internal queueing delay that Atlassian has optimized—but do you know what vendors should expect from a whole-system approach?

I looked again at event delivery timing and I see definite improvement over the last month, but I am still seeing around a 2,000 ms average versus the single-digit milliseconds that you mentioned, so I assume I am measuring something different.

For what it’s worth, this is the timing as received on a Forge Remote, comparing the event reception time with the generation timestamp in the event itself. This is the same methodology used for the Connect events, which definitely arrive first, so I do not suspect significant clock skew. Although I don’t have your billions of events, I do have thousands of events, and my gut feel is that this data matches what I see when testing on dev.