We're working on a large backfill of patient/careteam data. We've been following the recommendations in the Migrating to Medplum guide here: https://www.medplum.com/docs/migration We noticed a lot of slowness and inability to parallelize without transaction/serialization conflicts when trying to throttle up the migration of data. We do a lot of fetching/upserting by external identifier. Patient/CareTeam updates were taking 1.5 seconds. I came across the GIN index page and found that turning off

fastupdate

for the Patient and CareTeam tables really made a big difference in performance, bringing the time down to about 90ms per Patient/Care team. There isn't much documentation around the GIN index setup here. Is this backfill use case the reason it exists? Do you have any guidance around the use of these settings or tuning of the

gin_pending_list_limit

value for different types of workloads?

The guidance here is turning off

fastupdate

. we've toyed with the idea of having it off by default here