package mz_persist_client.internal.diff

Get desktop application:
View/edit binary Protocol Buffers messages

• string applier_version = 1

• uint64 seqno_from = 2

• uint64 seqno_to = 3

• uint64 walltime_ms = 6

• string latest_rollup_key = 4

• optional ProtoStateFieldDiffs field_diffs = 5

Used in: ProtoStateFieldDiffs

• DEPRECATED_ROLLUPS = 0

  Proto doesn't let us reorder this down, sadly

• ROLLUPS = 8

• HOSTNAME = 7

• LAST_GC_REQ = 1

• LEASED_READERS = 2

• CRITICAL_READERS = 6

• WRITERS = 3

• SCHEMAS = 12

• SINCE = 4

• LEGACY_BATCHES = 5

• HOLLOW_BATCHES = 9

• SPINE_BATCHES = 10

• SPINE_MERGES = 11

• ACTIVE_ROLLUP = 13

• ACTIVE_GC = 14

Used in: ProtoStateFieldDiffs

• INSERT = 0

• UPDATE = 1

• DELETE = 2

A columnar encoding of Vec<StateFieldDiff<K, V>> with heterogeneous K and V. - The number of diffs (len) == fields.len() == diff_types.len() - Each diff is encoded into 1 data slice for the key and 1 (Insert, Delete) or 2 (Update) data slices for the val - These data slices are concatenated together in data_bytes and the corresponding lengths of each slice in data_lens. (So number of data slices == data_lens.len().) - We store the length of each data_slice, not e.g. an offset into data_bytes. This makes random access slower, but we only ever iterate (random access by diff idx would be tricky anyway because each diff gets a variable number of data slices based on it's type.) Storing the lengths, OTOH, compresses much better with protobuf varints.

Used in: ProtoStateDiff

• repeated ProtoStateField fields = 1

• repeated ProtoStateFieldDiffType diff_types = 2

• repeated uint64 data_lens = 3

• bytes data_bytes = 4