Rate limiting targets only external REST API requests, which means that requests made within Confluence aren’t limited in any way. When users move around Confluence, creating pages, commenting, and completing other actions, they won’t be affected by rate limiting, as we’re seeing this as a regular user experience that shouldn’t be limited.

Let’s use an example to better illustrate this:

• When a user visits a space in Confluence, a number of requests are sent in the background — these requests ask Confluence for the pages, blog posts, etc. Since this traffic is internal to Confluence, it won’t be limited.

• When the same user opens up the terminal on their laptop and sends a request (like the one below) to get the contents of a space, it will be rate limited because it’s made outside of Confluence.
  
  

  curl -u user:password http://localhost:8090/rest/api/space/SPACEKEY/content

Authentication mechanisms

To give you more details on how we recognize which requests should be limited, we’re targeting external HTTP requests with these authentication mechanisms:

• Basic auth
• OAuth
• JSESSIONID cookie

Out of the many available techniques for enforcing rate limits, we’ve chosen to use token bucket, which gives users a balance of tokens that can be exchanged for requests. Here’s a summary of how it works:

Users are given tokens that are exchanged for requests. One token equals one request.

Users get new tokens at a constant rate so they can keep making new requests. This is their Requests allowed, and can be, for example, 10 every 1 minute.

Tokens are added to a user’s personal bucket until it’s full. This is their Max requests and allows them to adjust the usage of tokens to their own frequency, for example 20 every 2 minutes instead of 10 every 1 minute, as specified in their usual rate.

When a user tries to send more requests than the number of tokens they have, only requests that can draw tokens from the bucket will be successful. The remaining ones will end in a 429 error message (too many requests). The user can retry those requests once they get new tokens.

Confluence tastes best when used with our other products like Jira. Technically, products like these are external to Confluence, so they should be limited. In this case, however, we’re treating them as belonging to the same user experience and don’t want to enforce any limits for requests coming from or to these products.

The way it is now:

• Server: Not limited in any way.
• Cloud: There’s a known issue that applies rate limits to requests coming from/to cloud products. We’re working hard to disable rate limits for cloud products and should make that happen soon. For now, if you’re integrating Confluence with Jira cloud, you should make rate limits higher than usual.

The general assumption is that Marketplace apps are installed on a Confluence instance, make internal requests from within Confluence, and shouldn’t be limited. But, as always, it depends on how an app works.

• Internal: If an app in fact works internally, enhancing the user experience, it won’t be limited. An example of such app would be a special banner that’s displayed in a Confluence space. Let’s say this banner checks all pages that were created and shows this space’s winner — a user who’s created the most pages in the last month. Traffic like that would be internal, not limited.
• External: Apps whose requests are external to Confluence are limited. Let’s say we have an app that displays a wallboard on TV. It asks Jira for details about boards, issues, assignees, etc. and then reshuffles and displays them in its own way as the earlier mentioned wallboard. An app like that sends external requests and behaves just like a user sending requests over a terminal.

It really depends on the app, but we’re assuming most of them shouldn’t be limited.

Rate limiting is available for Data Center, so you most likely have a cluster of nodes behind a load balancer. You should know that each of your users will have a separate limit on each node (rate limits are applied per node, not per cluster).

In other words, if they have used their Requests allowed on one node and were rate limited, they could theoretically send requests again if they started a new session on a different node. Switching between the nodes isn’t something users can do, but keep in mind that this can happen.

Whatever limit you’ve chosen (e.g. 100 requests every 1 hour), the same limit will apply to each node, you don’t have to set it separately. This means that each user’s ability to send requests will still be limited, and Confluence will remain stable regardless of which node their requests are routed to.

Setting the right limit depends on many factors, so we can’t give you a simple answer. We have some suggestions, though.

Finding the right limit

The first step is to understand the size of traffic that your instance receives. You can do this by parsing the access log and finding a user than made the most REST requests over a day. Since UI traffic is not rate limited, this number will be higher than what you need as your rate limit. Now, that’s a base number — you need to modify it further based on the following questions:

1. Can you afford to interrupt your users’ work? If your users’ integrations are mission-critical, consider upgrading your hardware instead. The more critical the integrations, the higher the limit should be — consider multiplying the number you found by two or three.
2. Is your instance already experiencing problems due to the amount of REST traffic? If yes, then choose a limit that’s close to the base number you found on a day when the instance didn’t struggle. And if you’re not experiencing significant problems, consider adding an extra 50% to the base number — this shouldn’t interrupt your users and you still keep some capacity.

In general, the limit you choose should keep your instance safe, not control individual users. Rate limiting is more about protecting Confluence from integrations and scripts going haywire, rather than stoping users from getting their work done.

Requests allowed: 10/hour | Max requests: 100

One of the developers is sending requests on a regular basis, 10 per hour, throughout the day. If they try sending 20 requests in a single burst, only 10 of them will be successful. They could retry the remaining 10 in the next hour when they’re allowed new requests.

Another developer hasn’t sent any requests for the past 10 hours, so their allowed requests kept accumulating until they reached 100, which is the max requests they can have. They can now send a burst of 100 requests and all of them will be successful. Once they used up all available requests, they have to wait for another hour, and they’ll only get the allowed 10 requests.

If this same developer sent only 50 out of their 100 requests, they could send another 50 right away, or start accumulating again in the next hour.

Requests allowed: 1/second | Max requests: 60

A developer can choose to send 1 request every second or 60 requests every minute (at any frequency).

Since they can use the available 60 requests at any frequency, they can also send all of them at once or in very short intervals. In such a case, they would be exceeding their usual rate of 1 request per second.

Setting the right limit depends on many factors, so we can’t give you a simple answer. We have some suggestions, though.

Finding the right limit

The first step is to understand the size of traffic that your instance receives. You can do this by parsing the access log and finding a user that made the most REST requests over a day. Since UI traffic is not rate limited, this number will be higher than what you need as your rate limit. Now, that’s a base number — you need to modify it further based on the following questions:

1. Can you afford to interrupt your users’ work? If your users’ integrations are mission-critical, consider upgrading your hardware instead. The more critical the integrations, the higher the limit should be — consider multiplying the number you found by two or three.
2. Is your instance already experiencing problems due to the amount of REST traffic? If yes, then choose a limit that’s close to the base number you found on a day when the instance didn’t struggle. And if you’re not experiencing significant problems, consider adding an extra 50% to the base number — this shouldn’t interrupt your users and you still keep some capacity.

In general, the limit you choose should aim at keeping your instance safe, not to control individual users. Rate limiting is more about protecting Jira from integrations and scripts going haywire, rather than stoping users from getting their work done.

For example, a Marketplace app added some new API to Confluence. The app itself is used from the UI, so it shouldn’t be limited, but it might happen that Confluence sees this traffic as external and applies the rate limit. In this case, you could disable the app or increase the rate limit, but this brings additional complications.

To work around issues like this, you can allowlist the whole resource added by the app so it works without any limits.