Straight into the service manual page
WrenchQuarry
More

5 Atlassian flow limits to check for a dispatch queue

WrenchQuarry Editorial Team · Published · 11 min read · How these guides are written

Check five Atlassian Cloud automation limits for a dispatch queue: advanced flow size, flow groups, site labels, scheduled search results, and scheduled run time.

Key takeaways

  • Count conditions, branches, and actions in advanced flows and flow groups.
  • Review labels across the whole cloud site.
  • Test scheduled searches against real data.
  • Measure scheduled run time before tightening the interval.

If you run a dispatch queue on it, those ceilings decide whether your rules keep working. This guide covers five of them and shows how to check each one before it causes trouble.

Most dispatch logic lives in automation rules. A ticket arrives, and a rule assigns it. Another rule raises the priority when the clock runs down. A scheduled search sweeps up anything that slipped through. It works well for months, and then one morning it quietly stops. A rule that used to finish in seconds starts failing. A label list grows past what anyone remembers creating. A search that once returned a handful of items now returns thousands.

You usually find out when a run fails, or when a team lead asks why a ticket sat untouched for a day. The goal here is to check the numbers before they turn into an incident, and to check them again after any large change to the queue. For each limit, I'll explain what it is, why it matters for a queue in particular, and what to look at in your own setup.

A quick note on sources. The Atlassian figures come from its automation service limits page. Atlassian can change those limits, so treat the page as the reference and re-check it on a regular schedule. Nothing in this post replaces testing your own rules against your own volumes.

1. Advanced flows: 500 conditions, branches, and actions

Start with the size of a single advanced flow. The limits page treats an advanced flow that contains more than 500 conditions, branches, and actions as over the limit.

Dispatch logic tends to grow by accretion. Someone adds a branch for a new region. Someone else adds one for a weekend rota, then another for a vendor that needs a different status value. Each change makes sense on its own. Over a year, one flow can carry dozens of branches, and the total climbs without anyone tracking it.

To check it, open each advanced flow and count what it holds. The limit counts conditions, branches, and actions together, so a flow full of small steps can be heavier than it looks. A long run of simple checks can use up the budget faster than a few complicated ones.

Don't wait for a flow to cross the line. When one gets close, split it. A common pattern is one flow for each queue, team, or region, with a small routing flow at the front that decides which specialist flow should handle a given ticket. The router stays short, and each specialist flow stays readable. A second person can also follow the logic during an incident, which matters more than most teams expect.

The count is about structure, not traffic. A flow can process thousands of tickets and still sit well under the limit. The reverse is also true. A flow that touches only a handful of tickets can be too large if someone built it as one long chain of checks.

2. Flow groups: 65 conditions, branches, and actions

Flow groups sit one level up in how you organize automation. The same page says that a flow group containing more than 65 conditions, branches, and actions is over the limit. Read that wording carefully. The limit applies to everything the group holds, so the logic inside it adds up across all of its flows.

For a dispatch team, this is a tighter budget than the flow-level limit. Teams often put every rule for one queue into a single group, because it keeps related work together and makes it easy to find. With 65 as the ceiling, a busy queue can reach that number well before any single flow looks crowded. A group that felt fine last spring may be near the line today.

Check it the same way you checked the flows. Open each group, add up the conditions, branches, and actions across its flows, and compare the total with 65. If the number is creeping upward, decide which logic belongs in its own group and which logic can go. Retired branches are the easiest win. Many queues still carry steps for categories, statuses, or vendors that no longer exist, and every one of those steps counts toward the total. Use the work order status checks when reviewing which statuses still belong in dispatch logic.

One habit helps here. When you add a new branch, note in the same sitting what you could remove. If a change adds three steps, look for three steps that are no longer needed. The total stays stable, and the team gets a regular conversation about what the queue actually does. It also gives you a clear answer when someone asks why a new rule went in and an old one came out.

3. Flow labels: 600 across the cloud site

Labels are the tags you attach to flows so you can find and group them. They look harmless, which is exactly why they pile up. The limits page says that a total number of flow labels for a cloud site exceeding 600 is over the limit.

The important word there is site. This limit covers the whole cloud site, not one queue or one team. Every team sharing the site draws from the same pool. A dispatch team that tags each rule with a region, a skill, a priority, a customer, and a status can use up its share quickly. Review label use when a queue is copied for a new client or a new region.

To check it, review the labels in use across your site and sort them. Look for duplicates, near-duplicates such as "Urgent" and "urgent-ticket", and labels that belong to flows nobody runs anymore. Merging a near-duplicate into one label is usually better than adding a new label beside it.

Going forward, agree on a short naming scheme before anyone adds a label. Label only what someone will actually use to search, filter, or report. A label for "maybe useful later" is the kind that never gets removed. On a large or shared site, name an owner for the label list. Otherwise no one feels responsible for the cleanup. A monthly look at what was added is enough for most sites, even if the review is only a few minutes long.

4. Scheduled JQL searches: 999 work items

Scheduled searches are the safety net of most dispatch queues. A scheduled JQL search finds tickets that went unassigned, sat too long in one status, or missed a change, and then a flow acts on what it finds. The limit to check here is the result count. The page treats a scheduled JQL search that returns more than 999 work items, and is not streamlined enough, as over the limit.

Notice that the page ties the problem to how streamlined the search is, not only to the count. A search that returns a large number of items usually asks for too much. In a dispatch queue, this often happens because the query asks for "all open tickets in the queue." That query was fine when the queue held a small set of tickets. As the backlog grows, the same query grows with it, until it crosses the line.

The fix is usually in the query, not the schedule. Narrow the search so it asks only for what the flow needs to act on. The page offers a useful pattern: check that the updated time of a work item falls within a certain range, with the example updated > -1w, which it describes as including only last week's work items. Applied to a queue, that means looking at recently changed tickets rather than every ticket the queue has ever held.

A narrow query also limits the damage when something goes wrong. If a scheduled search has a bug, a tight search touches a small set of tickets. A broad one can touch many more before anyone notices. Treat a count approaching 999 as a warning sign, not a number to live with. Ask what the flow will do with each result, and whether a person would want all of those items handled in one run.

Test the query against real data before you schedule it. Run it manually, read the results, and confirm they match what you meant. Then check the count on a busy day, not only a quiet one. The backlog that matters is the one you have at peak.

5. Scheduled flows: 5 minutes to finish

The last limit is about time. The page treats a scheduled flow that takes longer than 5 minutes to execute, and is scheduled every 5 minutes, as over the limit. That pairing is the trap. A flow that runs comfortably once an hour can become a problem when someone reschedules it to run every five minutes, because the team wants faster assignment.

Queue owners often make this change for good reasons. A customer waits longer than they should, a supervisor sees tickets sitting unassigned, and the obvious response is to run the assignment flow more often. But the schedule and the run time have to be considered together. Tightening the schedule without measuring run time first is a common way teams end up over the line.

Time windows like this show up in other developer tools too. Stripe's webhook documentation notes that its libraries have a default tolerance of 5 minutes between the timestamp and the current time. That kind of window is a deliberate choice. Your queue faces the same question: how stale is too stale to act on? Answering it before you set the schedule makes the rest of the decisions easier.

To check it, list every scheduled flow with its interval and its typical run time. Measure run time on a busy day, when the queue is full, not on a quiet morning. A flow that finishes in a minute on a Tuesday may take far longer when the backlog peaks. Use whatever run history your site provides for each rule, and record the numbers next to the interval.

If a flow regularly approaches five minutes, you have three main options. You can shorten the flow by removing steps or simplifying conditions. You can widen the interval so there is room for a run to finish. Or you can split the work so each run handles a smaller slice of the queue, for example by region or by ticket age. Each option has a cost. Shorter intervals are better for response time, while wider intervals and smaller slices are easier on the flow. Choose based on what the queue needs, and write the decision down. Review customer arrival updates alongside assignment timing so the team also considers what customers need to hear.

Illustrative example: a scheduled flow

Consider a hypothetical queue for a field-service team. In testing on a busy day, a full run takes longer than five minutes while the flow is scheduled every five minutes. That matches the case the page describes. The team weighs its options: widen the interval so a run has room to finish, or split the flow by region so each run handles a smaller part of the queue. They pick the split, since it keeps response time close to what the team wanted. They time it again on the next busy day and write the result in their sheet.

A routine for staying inside the limits

Checking once is not enough. Limits are easy to forget and easy to exceed slowly, so build a short routine around them. Nothing here is complicated, but it has to happen regularly to work.

Before any change to a queue, count the advanced flows, flow groups, and labels that the change touches. If the change adds steps, name the steps you will remove. If it adds a label, check whether an existing one already covers the purpose. This takes a few minutes and prevents most of the slow growth described above.

Once a quarter, review every scheduled search for result counts, and every scheduled flow for run time against its interval. Quarterly is a sensible starting point for most teams, though a very busy queue may need a closer look. Note the date you checked, so the next review has a baseline to compare against.

After any incident, ask two questions. Did a rule run late or not at all? If so, did a search or flow hit one of these limits? Incidents are often the first sign that a queue has grown past its budget, so treat them as a prompt to check the numbers, not only as a fault to patch.

Keep the results in a simple sheet. A row for each flow with its name, owner, count of conditions, branches, and actions, scheduled interval, typical run time, and last checked date is enough. A sheet is cheap to maintain, and it makes it obvious which flow is closest to a ceiling. A named owner means someone notices when a flow starts to bloat.

Give the routine a home as well. Put the quarterly review on the same calendar as the queue's other maintenance, and assign it to a named person. A check with no owner tends to slip, and the slip usually shows up as a late ticket rather than a missed meeting.

Where to start

None of these limits is exotic. They are the ordinary ceilings of an automation service that runs real work every day. The dispatch queues that hold up are not the ones with the simplest rules. They are the ones where someone checks the numbers before a problem shows up in the ticket queue.

Pick the five checks above and run them on your busiest queue today. Write down what you find, and fix the worst item first. Then re-read the Atlassian limits page, since the figures can change and it is the reference for the Atlassian limits discussed here.

What should you check?

Should you test on a quiet day?

Measure run time and search results on a busy day. Quiet-day results may miss the backlog that matters at peak.

What should you record?

Record each flow's name, owner, step count, scheduled interval, typical run time, and last checked date. Keep the results together for the next review.

Sources

More field guides