A step that is expected to not use much memory uses more.
The step is not the only one using memory, parallel steps and/or services will be using memory at the same time.
Confirm that all parallel steps or services are relevant to this step and disable the ones that aren't, or if there are any pipeline-level services that don't need to be running for the duration of the whole pipeline.
Consider increasing the size of your instance, if all steps are required. Consider splitting these steps into multiple smaller, sequential groups of parallel steps; or splitting them into child pipelines with their own separate sets of resources from the parent.
The yellow warning banner does not necessarily mean there's an issue with the build, and instead only serves as a warning.
The red "out of memory" error banner will display if your build has used >90% of allocated memory and failed. Since OOM failures can often result in unexpected error patterns, Codefresh will display this warning if the build failed on any error at >90% memory usage, not just on explicit memory errors. Because of this, it is possible in some cases that builds displaying this error banner actually failed from reasons other than memory.