This code reads all pack headers from all packs and rebuilds the index
from scratch. Afterwards, all indexes are removed. This is needed
because in #434 the command `optimize` produced a broken index that
did not contain a blob any more. Running `rebuild-index` should fix
this.
This struct implements an io.ReadSeeker on top of a backend. This is the
easiest way to make the packer compatible to the new backend without
loading a complete pack into a bytes.Buffer.
I had the suspicion that one of my repositories contained redundant
data blobs that is not recorder in the index. In order to check this I
needed to dump information about the pack files without consulting the
index. The code here iterates over all packs and dumps the headers, and
dumps information about them in JSON (so we can use other tools to parse
that information again).
This allows iterating over the output channel without having to start
another Goroutine outside of the worker pool. This also removes the need
for calling Wait().
A worker pool is needed whenever something should be done concurrently.
This small library makes it easy to create a worker pool by specifying
channels, concurrency and a function that should be executed for each
job and returns a result and an error.
Before, the restic-server read the whole blob (up to 8MB) into memory
prior to writing it to disk. Concurrent writes consumed a lot
of memory. This change writes the blob to a tmp file directly and
renames it afterwards in case there where no errors.
- Removed external dependencies for test
- Prevent building restic-server w/ Go 1.3
Go versions 1.0, 1.1., and 1.2 are going to fail as well, but they
are "excluded" by README.md already.
