SYSTEM Cited by 1 source
Panda (Dropbox metadata stack)¶
Definition¶
Panda is Dropbox's petabyte-scale transactional key-value store — the metadata stack behind Magic Pocket. Linked from the 2026-04-02 storage-efficiency post as Panda metadata stack.
Stub page. This wiki's substantive coverage of Panda comes only from the 2026-04-02 source; when the dedicated Panda post is ingested, this page should be expanded to cover the KV model, sharding, transaction semantics, replication shape, and the consistency contract Magic Pocket depends on.
Why Panda shows up in the storage-efficiency story¶
Magic Pocket's compaction strategies differ in their metadata cost — i.e. the rate at which they generate Panda writes:
- L1 / L2 keep many blobs under the same volume identity after compaction; only donor blobs need location rewrites → moderate Panda write pressure.
- L3 (patterns/streaming-re-encoding-reclamation) writes every live blob into a brand-new volume → every reclaimed blob generates a new Panda location entry.
The practical consequence is that Panda write capacity is the binding constraint for how aggressively L3 can be run. The multi-strategy compaction design routes the sparsest volumes through L3 (low blob count per reclaimed volume → bounded per-reclaimed-volume metadata cost) and rate-limits each strategy path to prevent Panda saturation.
Seen in¶
- sources/2026-04-02-dropbox-magic-pocket-storage-efficiency-compaction — Panda as the metadata backplane whose write budget bounds L3's aggressiveness; explicit "load on storage and metadata systems" framing.