Error 1102 requires the failed invocation’s resource outcome to guide the repair. Check whether the evidence points to CPU work or memory retained in the Worker, separate from PHP on the host.
Then compare the failing branch and input with a normal request. A larger allowance alone does not explain a loop, repeated parsing or unnecessary data copies.
Guide reviewed: October 8, 2026. Interface labels may vary by version, account and device.
Confirm whether CPU or memory is exhausted
| Outcome or pattern | Inspect | Verification |
|---|---|---|
| Exceeded CPU | Loops, repeated scans and parsing | Same branch and representative larger input |
| Exceeded memory | Body copies, parsed objects and retained state | Memory outcome and complete output |
| Expanded route | New resource types selected | Intended and excluded paths |
| Evidence incomplete | Invocation metrics and reproduction | Confirm the resource before selecting a fix |
| Origin limit changed | Worker failure unchanged | Review the runtime enforcing this allowance |
Find the resource outcome for the failed request
- Record the failing URL, action and time zone.
- Locate the corresponding invocation or metrics.
- Identify the resource outcome instead of relying only on 1102.
- Compare one working input with the failure.
- Review the operation consuming that resource.
The Worker owner can compare the error time with the relevant invocation information and metrics. Cloudflare distinguishes exceeded CPU and exceeded memory outcomes. That distinction gives the developer a more precise problem to investigate.
Keep the exact hostname, path and action with the time. A Worker may process only certain requests, and a larger input or a different branch may use substantially more resources than a normal homepage visit.
If the evidence is incomplete, reproduce the request in a suitable diagnostic environment. A general statement that the site is slow does not identify whether computation, retained data or another component needs attention.
If CPU work is the issue
Review the computation done for the request. Repeated parsing, loops and multiple passes over the same data can increase processing work without changing the feature’s intended result.
For example, a response-processing feature may scan a large document several times to make small substitutions. Comparing that design with a simpler transformation can reveal unnecessary work. The example suggests what to review; it is not a universal cause of 1102.
CPU time and total elapsed time are different. Waiting for a dependency is not the same as actively processing data. The developer should use the relevant measurements rather than assume that every long request needs a larger computation budget.
If memory is the issue
Look at what data is retained during the request. Complete response bodies, parsed objects and additional copies can occupy memory at the same time. The size of the original download may therefore underestimate the amount used while processing it.
A feature that only modifies headers may not need to read the body at all. A content transformation may benefit from a suitable streaming design instead of retaining the complete response. The chosen correction still needs to produce the required output.
Shared or long-lived state also deserves a review. If data is retained beyond the request that needed it, repeated traffic can affect memory differently from a single isolated test. Profiling can help identify what remains allocated.
Compare the feature with its intended scope
A Worker designed for a narrow path may use more resources after its coverage expands to the whole site. Review the route and the requests that match it, especially if the error began after a configuration change.
Large downloads, exports and unusual API responses may not need the same processing as public HTML. A targeted design can keep the feature where it belongs without applying costly work to unrelated resources.
Any reduced scope must preserve required behavior. If the Worker supplies an access check or necessary routing, excluding a path has consequences beyond performance. The administrator should understand those consequences before making a change.
Choose a correction based on the measured bottleneck
Keep a test matrix with input, selected route, code version, resource outcome and intended response. Change one workload or design element, then repeat the same matrix. This makes a necessary architectural change distinguishable from extra capacity masking an unexplained operation.
The developer may optimize the operation, reduce retained data or reassess the workload’s architecture. An available limit change should be evaluated against the actual Worker plan and requirement, rather than assumed to repair every resource failure.
A larger allowance can accommodate necessary work, but it does not explain a loop or an unnecessary body copy. Understanding the cause also helps prevent the next increase in content size or traffic from producing the same symptom.
After the correction, repeat the failing request and compare the resource outcome. Then test a normal request and a representative larger input. Check the feature’s result as well as the absence of an error.
Cloudflare’s Workers limits guide distinguishes CPU and memory outcomes for 1102. Its guides to CPU profiling and memory profiling provide tools for identifying the actual resource use.
Verify the result and supported workload
Check the original failure, a normal request and a larger supported input. Confirm the output, access and routing requirements as well as resource outcomes. Keep the tested version and scope in the deployment record.