Summary
Track the single-process root-set SST GC work.
This is intentionally scoped to the single-process Tonbo model. It is not a distributed GC plan, and it should not be read as the first incremental step toward distributed GC. When Tonbo needs distributed GC, that should be specified in a separate RFC with a different mental model.
Retention policy is also intentionally out of scope for this first milestone. The current GC work should not introduce max_ttl, max_versions, or broader time-travel semantics yet.
Child Issues
Notes
- Baseline-first benchmarking and observability are part of the definition of done for each item.
- WAL GC stays standalone and floor-driven in this effort.
- Benchmarks should follow the adjacent-repo pattern: explicit baselines like
aisle, benchmark-integrated counters like fusio-manifest.
- Retention policy should be added later as a separate follow-up issue, not folded into the first GC delivery.
Summary
Track the single-process root-set SST GC work.
This is intentionally scoped to the single-process Tonbo model. It is not a distributed GC plan, and it should not be read as the first incremental step toward distributed GC. When Tonbo needs distributed GC, that should be specified in a separate RFC with a different mental model.
Retention policy is also intentionally out of scope for this first milestone. The current GC work should not introduce
max_ttl,max_versions, or broader time-travel semantics yet.Child Issues
Notes
aisle, benchmark-integrated counters likefusio-manifest.