Arve Knudsen c65189cc44 tsdb/seriesmetadata: sorted-slice inverted index with copy-on-write ... Replace map[uint64]struct{} inner maps in the resource attribute inverted index with sorted []uint64 slices using copy-on-write semantics. This reduces memory ~4x (8 bytes/entry vs ~50 bytes/entry for hash map overhead) and eliminates snapshot copies on read. sortedInsert/sortedRemove create new slices on mutation, so readers holding old slices are safe from concurrent changes. LookupResourceAttr returns the live slice directly (zero-copy). The layered reader uses two-pointer sorted merge for union operations. The web API uses two-pointer sorted intersection for filter intersection. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> Signed-off-by: Arve Knudsen <arve.knudsen@gmail.com>		2026-03-20 17:56:16 +01:00
..
testhelpers	tsdb: remove metric metadata persistence, refactor to generic kind framework	2026-03-20 17:56:16 +01:00
v1	tsdb/seriesmetadata: sorted-slice inverted index with copy-on-write	2026-03-20 17:56:16 +01:00