Now that V1 is no longer receiving Vector Set updates, we can starting working through all the things we need to do to get rid of the --enable-vector-set-preview flag.
Link PRs that accomplish tasks to this issue.
In no particular order:
VRANGE support has been removed from the checklist. VRANGE is a weird command that has a number of footguns for the large Vector Sets we expect to see, in particular - and + options could result in gigabytes of response data. Given that there's no trivial implementation path and the command is of limited utility, we'll wait until there's concrete demand to implement it.
Now that V1 is no longer receiving Vector Set updates, we can starting working through all the things we need to do to get rid of the
--enable-vector-set-previewflag.Link PRs that accomplish tasks to this issue.
In no particular order:
continue_searchfor too-big-for-initial-buffers searches (Update DiskANN FFI #1992)VADDcan update vectors correctlyRENAME(NX)support (see ImplementRENAME(NX)Support For Vector Sets #1913)WRONGTYPEon Vector Sets, or work as expected (Add tests (and fix bugs) for commands WRONGTYPE'ing appropriately against Vector Sets #1935)VADDupdates need to be correctly ordered in the AOF and replayed in correct order on replicasVSIM(see Implement RESP3 forVSIM(andVEMB) andRAWforVEMB#1916)RAWsupport forVEMB(see Implement RESP3 forVSIM(andVEMB) andRAWforVEMB#1916)VCARD(Add tests (and fix bugs) for commands WRONGTYPE'ing appropriately against Vector Sets #1935)VISMEMBER(Add tests (and fix bugs) for commands WRONGTYPE'ing appropriately against Vector Sets #1935)VLINKS(Update DiskANN FFI #1992)VRANDMEMBER(Update DiskANN FFI #1992)VSETATTR(ImplementVSETATTR#2001)KEYS,SCAN, etc. (Vector-set internal records are exposed/counted by DBSIZE, SCAN, KEYS, and INFO KEYSPACE #1921 ; see Exclude internal records from stats #1922)VRANGEsupport has been removed from the checklist.VRANGEis a weird command that has a number of footguns for the large Vector Sets we expect to see, in particular-and+options could result in gigabytes of response data. Given that there's no trivial implementation path and the command is of limited utility, we'll wait until there's concrete demand to implement it.