Hyper.Img.Db.Gc.Sweep (Hyper v0.1.0)

Copy Markdown View Source

Pure accounting core for the layer GC.

A Hyper.Img.Db.Gc.Sweep.State accumulates a single full pass over the blobs table. absorb/3 classifies a keyset page into present (file on the medium) vs missing (file gone) and returns the missing blobs for the GC to prune; record_prune/4 folds the prune outcome back into the tally once the GC has executed it against the database.

Summary

Types

blob()

A blob page row: content-addressed id and its DB-recorded byte size.

check_fun()
presence()

Injected shared-medium probe. :present = file is there; :missing = file is genuinely gone (safe to prune); :unknown = could not determine (I/O error), so the blob is counted but never pruned.

Functions

absorb(sweep, batch, check_fun)

Fold one keyset page into the sweep. Advances cursor to the last blob's id (so the next page starts after it), counts present / missing / unknown, and returns only the genuinely-missing blobs (file confirmed absent) as {id, size} for the GC to prune. :unknown rows (probe I/O error) are counted but never returned, so an unreachable medium can never drive a delete.

continue?(batch, limit)

Full page (length == limit) means more rows may remain; page again.

new()

A fresh sweep state: nil cursor (start of table), zeroed counters.

record_prune(sweep, pruned, pruned_bytes, dangling)

Fold the result of pruning one page back into the sweep: pruned rows deleted totalling pruned_bytes, and dangling rows left in place (file gone but the blob is still referenced by an image - a data-loss condition the GC reports but must not delete).

Types

blob()

@type blob() :: {String.t(), non_neg_integer()}

A blob page row: content-addressed id and its DB-recorded byte size.

check_fun()

@type check_fun() :: (String.t() -> presence())

presence()

@type presence() :: :present | :missing | :unknown

Injected shared-medium probe. :present = file is there; :missing = file is genuinely gone (safe to prune); :unknown = could not determine (I/O error), so the blob is counted but never pruned.

Functions

absorb(sweep, batch, check_fun)

@spec absorb(Hyper.Img.Db.Gc.Sweep.State.t(), [blob()], check_fun()) ::
  {Hyper.Img.Db.Gc.Sweep.State.t(), [blob()]}

Fold one keyset page into the sweep. Advances cursor to the last blob's id (so the next page starts after it), counts present / missing / unknown, and returns only the genuinely-missing blobs (file confirmed absent) as {id, size} for the GC to prune. :unknown rows (probe I/O error) are counted but never returned, so an unreachable medium can never drive a delete.

continue?(batch, limit)

@spec continue?([term()], pos_integer()) :: boolean()

Full page (length == limit) means more rows may remain; page again.

new()

@spec new() :: Hyper.Img.Db.Gc.Sweep.State.t()

A fresh sweep state: nil cursor (start of table), zeroed counters.

record_prune(sweep, pruned, pruned_bytes, dangling)

@spec record_prune(
  Hyper.Img.Db.Gc.Sweep.State.t(),
  non_neg_integer(),
  non_neg_integer(),
  non_neg_integer()
) :: Hyper.Img.Db.Gc.Sweep.State.t()

Fold the result of pruning one page back into the sweep: pruned rows deleted totalling pruned_bytes, and dangling rows left in place (file gone but the blob is still referenced by an image - a data-loss condition the GC reports but must not delete).