Fix bad check in compact query

[brandond]

• Fix issue where compact_rev_key was protected from deletion, instead of the row it points to. Honestly we should have stored the compact_rev marker in the value instead of the prev_revision column. (#61)

Test:

1. Run k3s with kine backend (any SQL backend; doesn't matter): k3s --debug server
2. Wait for compaction message to appear in log output - something like level=debug msg="COMPACT deleted 0 rows from 3 revisions in 2.484763ms - compacted to 3/1745"
3. Stop k3s+kine
4. Run query: SELECT id, name FROM kine WHERE name = "/registry/health";
5. Run query: UPDATE kine SET prev_revision = 1 WHERE name = 'compact_rev_key'; - use the ID from the previous step instead of 1 if it was different.
6. Start k3s; wait for compaction message to appear in log output
7. Run query: SELECT id, name FROM kine WHERE name = "/registry/health"
8. Confirm that /registry/health entry still present with the same ID

Signed-off-by: Brad Davidson brad.davidson@rancher.com

[ibuildthecloud]

LGTM

[brandond]

Can tackle additional query optimizations later, for now how about we just stop deleting things we shouldn't. Patch is updated.