Advice on building a different dedup algorithm

[y-nk]

i would like to build something similar to the deduplication feature, but i'd like to change the criteria for deduplication.

in current bullmq, criteria is "new job is discarded until matching dedup id job is marked as completed/failed". i'd like to change that for "marked as active" (as in: "currently processing" rather than "processed")

the usecase is indexing: i want to queue an indexing requests for an entity. the worker will fetch the entity, and proceed to indexing ; if the model was updated during the execution of the job, a new job would need to be added to the queue so that it can be processed a 2nd time, but there's no need to have more than 1 waiting job per entity in the queue.

the jobs would also need to have a global conccurency of 1 based on the dedup id.

any advice on how to build this? it seems I can't use the deduplication id nor can i use unique job ids.