I think it is not that hard.
SNO just put command to free up 1tb
Satellite check what pieces it can easily delete from that node, and send it bloom filter to that node for delete piece’s, that it comfortable to delete, compaction will make the rest.
but it is 7 days of wait.
I understant it is not idial and it not garantee freeup 1tb exactly, it can be more or less.
at least it will not take months to shrink.
This is exactly the old GE algorithm, which requires a lot of resources per node and why we switched to a much more simple algorithm to recover missing pieces with repair workers.
In this proposal it also adds a heavy BF generator, which should have already moved pieces. So, pieces should be not only grouped by the node, not only filtered to match the requested amount, but also physically moved to a prepared list of other nodes. Only then BF can be generated and sent.
So, generating BF for the one node doesn’t make any sense.
There is another, but more dangerous workaround, you might put offline your node roughly on 12 hours, this will trigger a repair, if some of the segments reach the repair threshold, they will be recovered to another nodes. So, with the next BF part of data will be removed from your node. The risk that for having frequent offline events your node could be disqualified.
I’m thinking about another idea: make GE cancelable from the node’s side (currently it requires a support request). So you call GE, then some time later withdraw it, so BF will clean some data, which was repaired for that period from your node.