Ioana Cristescu
Ioana Cristescu
In the dispatcher we have two types of reads - in the prefix or in the suffix. We should add stats to track those. Also in the pack_store, we can...
The traverse_pack_files on v2 stores are anyway broken https://github.com/mirage/irmin/pull/1903. But even more so on gced stores: the traversal of the suffix file needs to read gced objects, without knowing their...
for the case where the node is of type key (see https://github.com/mirage/irmin/pull/1795#discussion_r881642946)
A corrupted store was obtained while boostrapping with `hangzhounet` (available on comanche at `/bench/ioana/corrupt_hnet/hangzhounet-backup.tgz`) and it fails when validating block `643736` (https://hangzhounet.tzkt.io/BM8ZkiT9TVe7XZaEqUkJWZqCjT96TQSEzBYfgqsPug3xnX33eSt): ``` node_1 | 2022-03-15T14:25:21.901850755Z Mar 15 14:25:21.902 -...
See https://github.com/mirage/irmin/pull/1757#discussion_r827842208 and https://github.com/mirage/irmin/pull/1757#discussion_r827939666
When producing Merkle proofs incrementally, we need to regularly commit (flush to disk) the tree. We should provide a way to produce proofs from in-memory trees. See https://gitlab.com/tezos/tezos/-/issues/2780.
We can only have a single RW instance and a single RO instance in irmin-pack. This is because RW (or RO) instances share the same index. It is a problem...
During `unsafe_append` we only add direct keys to the store, hence to staging. There is maybe a way to simplify the code that handles both types of keys. See also...
Part of #2039 is to remove lwt from low level start (already done) and finalise (this PR). Rebased on #2063.
Calling `fsync` on newly created files is not enough - the corresponding entry in the containing directory might not make to disk. An explicit `fsync()` on a file descriptor for...