With the recent introduction of the RootNodes trait, there's nothing in
the fs module pulling in tvix-store dependencies, so it can live in
tvix-castore.
This allows other crates to make use of TvixStoreFS, without having to
pull in tvix-store.
For example, a tvix-build using a fuse mountpoint at /nix/store doesn't
need a PathInfoService to hold the root nodes that should be present,
but just a list.
tvix-store now has a pathinfoservice/fs module, which contains the
necessary glue logic to implement the RootNodes trait for a
PathInfoService.
To satisfy Rust orphan rules for trait implementations, we had to add a
small wrapper struct. It's mostly hidden away by the make_fs helper
function returning a TvixStoreFs.
It can't be entirely private, as its still leaking into the concrete
type of TvixStoreFS.
tvix-store still has `fuse` and `virtiofs` features, but they now simply
enable these features in the `tvix-castore` crate they depend on.
The tests for the fuse functionality stay in tvix-store for now, as
they populate the root nodes through a PathInfoService.
Once above mentioned "list of root nodes" implementation exists, we
might want to shuffle this around one more time.
Fixes b/341.
Change-Id: I989f664827a5a361b23b34368d242d10c157c756
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10378
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Reviewed-by: sterni <sternenseemann@systemli.org>
This makes PathInfoService::from_addr return a Box<dyn PathInfoService>,
rather than an Arc<dyn …>, and leaves it up to the consumers to rewrap
it into an Arc where needed.
This allows us to drop the Arc for the tvix-store daemon subcommand.
Change-Id: Ic83aa2ade6c51912281bd17c7eef7252e152b2d1
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10409
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Reviewed-by: sterni <sternenseemann@systemli.org>
These are still leftovers from before the traits being async, where we
had to clone before moving into an async closure.
Change-Id: I1b3937edf61ce3e23bb07803306622c37a3572c6
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10381
Tested-by: BuildkiteCI
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Autosubmit: flokli <flokli@flokli.de>
There's nothing store-path specific here anymore, it's just a name in
the mountpoint root.
Change-Id: I0f8004491baa03ba560d390053a42678ee81154a
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10377
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Reviewed-by: raitobezarius <tvl@lahfa.xyz>
To support tvix builds, we need to be able to use the `TvixStoreFs` to
materialize the sandbox's /nix/store filesystem with just the list of
inputs needed for the build. Currently we'd need to seed an in-memory
`PathInfoService`, which includes more functionality than what is
required for `TvixStoreFs`. Additionally, the `PathInfoService` is
specific to Nix. By decoupling `TvixStoreFs` and `PathInfoService`,
we allow for usage of `TvixStoreFs` with `tvix-castore` without needing
a `PathInfoService`.
This introduces a new `RootNodes` trait which provides a way for the
filesystem to look up CA nodes via their basename in the root directory
of the filesystem. We then implement `RootNodes` for any
`PathInfoService`. Additionally, the filesystem root inode tracker now
stores basenames rather than `StorePath`s since `StorePath`s are
specific to Nix.
As a followup we can rename `TvixStoreFs` to `TvixCaStoreFs` and move
it to the `castore` crate (or its own crate).
b/341
Change-Id: I928372955017c23b1bf2b37190cbc508a4ed10d5
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10363
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
For all these calls, the caller has enough context about what it did, so
it should be fine to use io::Result here.
We pretty much only constructed crate::Error::StorageError before
anyways, so this conveys *more* information.
Change-Id: I5cabb3769c9c2314bab926d34dda748fda9d3ccc
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10328
Reviewed-by: raitobezarius <tvl@lahfa.xyz>
Tested-by: BuildkiteCI
Autosubmit: flokli <flokli@flokli.de>
We don't really require the Path to be a PathBuf, we don't even require
it to be a Path, we only need it to be AsRef<Path>>.
This removes some conversion in the from_addr cases, which can just
reuse `url.path()` (a `&str`).
Change-Id: I38d536dbaf0b44421e41f211a9ad2b13605179e9
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10258
Autosubmit: flokli <flokli@flokli.de>
Reviewed-by: raitobezarius <tvl@lahfa.xyz>
Tested-by: BuildkiteCI
It's been a while since the last sled release, and that one binds to a
pretty old version of zstd, requiring workarounds like cl/10090.
Upstream sled main branch currently has zstd halfway patched out (it's
a no-op, but the feature flag and options are still there), and it's in
that state for a year.
Rather than maintaining our own fork of sled, let's just stop using the
compression feature in sled, dropping the version pin to zstd that way,
removing the need for cl/10090.
This doesn't mean we won't reintroduce per-blob compression - but we
probably just won't let sled take care of the compression, but do it
ourselves - which is necessary for more chunked blob storage anyways.
Even though we do drop the feature flag, we still need to explicitly use
use_compression(false).
Change-Id: I0e4892d29e41c76653272dc1a3625180da6fee12
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10257
Autosubmit: flokli <flokli@flokli.de>
Reviewed-by: raitobezarius <tvl@lahfa.xyz>
Tested-by: BuildkiteCI
Break up the go-bindings derivation. Keep "protos" containing all proto
files (well, and the buf config), and use it for a check phase running
linter and formatter, as well as the existing "go-bindings" attribute
Change-Id: I52cb9d08570bb76452acb831eb711c5b6c0eacfb
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10239
Reviewed-by: raitobezarius <tvl@lahfa.xyz>
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
This allows setitng the trusted-public-keys URL parameter to a
(whitespace-separated) list of public keys.
NARInfo files retrieved need to contain a valid signature.
Change-Id: Ifd6580b723cbae3182e9cadfa54f1ca2b41d6599
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10153
Autosubmit: flokli <flokli@flokli.de>
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
Introduce an Option<Vec<narinfo::PubKey>>, configurable with a
`set_public_keys` method.
If set, this configures NixHTTPPathInfoService to validate signatures.
Change-Id: I157c5e13c41fc9bfd40b0655381fb4cf33900868
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10152
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
Ensure the initially communicated NarHash/NarSize from the NarInfo
matches what we read, and don't return a PathInfo message if there's
a mismatch.
Also move the buffering layer around a bit.
Change-Id: I68c60ecfaf0f9cd5edacea648437ecb0c9729251
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10148
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
An error in the PathInfoService request can appear in case the
underlying request returns an error.
We shouldn't panic and bork the fuse mount, but instead return an IO
error.
Change-Id: I2daeae629e1627d06adcd7b82ddb76c50c602212
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10154
Tested-by: BuildkiteCI
Reviewed-by: Connor Brewster <cbrewster@hey.com>
This allows seeing a PathInfo as a nix_compat::narinfo::NarInfo<'_>.
It doesn't allocate any new data, but the NarInfo<'_> view allows us to
access things like signature verification, or rendering out
(alternations of this) as strings.
Change-Id: Id0d8d7feeb626ee02c3d8a4932f24ace77022619
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10108
Reviewed-by: edef <edef@edef.eu>
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
We can simply use .to_owned() on that thing afterwards if we want to
construct an owned StorePath.
Change-Id: I0f3e2e4434b99ee522f2a7dbfa391e13a987479c
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10105
Autosubmit: flokli <flokli@flokli.de>
Reviewed-by: edef <edef@edef.eu>
Tested-by: BuildkiteCI
We always know this needs to end with a .drv, and fail parsing if it
doesn't, so there's no need to hang onto these 4 bytes.
This will make it much easier to synthesize a NarInfo<'_> later on from
a PathInfo proto, because we don't have to make this ".drv" appear out
of thin air.
Change-Id: Id95e7fd937d7c9a420a39b5a4bab73985640ca3b
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10084
Tested-by: BuildkiteCI
Reviewed-by: edef <edef@edef.eu>
Reviewed-by: raitobezarius <tvl@lahfa.xyz>
Autosubmit: flokli <flokli@flokli.de>
NixHTTPPathInfoService acts as a bridge in between the Nix HTTP Binary
cache protocol provided by Nix binary caches such as cache.nixos.org,
and the Tvix Store Model.
It implements the [PathInfoService] trait in an interesting way: Every
[PathInfoService::get] fetches the .narinfo and referred NAR file,
inserting components into a [BlobService] and [DirectoryService], then
returning a [PathInfo] struct with the root.
Due to this being quite a costly operation, clients are expected to
layer this service with store composition, so they're only ingested
once.
The client is expected to be (indirectly) using the same [BlobService]
and [DirectoryService], so able to fetch referred Directories and Blobs.
[PathInfoService::put] and [PathInfoService::nar] are not implemented
and return an error if called.
This behaves very similar to the nar-bridge-pathinfo code in nar-bridge,
except it's now in Rust.
Change-Id: Ia03d4fed9d0657965d100299af97cd917a03f2f0
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10069
Tested-by: BuildkiteCI
Autosubmit: flokli <flokli@flokli.de>
Reviewed-by: raitobezarius <tvl@lahfa.xyz>
Allow taking advantage of the buffer of the underlying reader to avoid
unnecessary copies of file data.
We can't easily implement the methods of BufRead directly, since we
have some extra I/O to perform in the final consume() invocation.
That could be resolved at the cost of additional bookkeeping, but this
will suffice for now.
Change-Id: I8100cf0abd79e7469670b8596bd989be5db44a91
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10089
Reviewed-by: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
We rely on being able to make small reads cheaply, so this was already
an implicit practical requirement. Requiring it explicitly removes a
performance footgun, and makes further optimisations possible.
Change-Id: I7f65880a41b1d6b5e6bf2e52dfe47d4c49b34bcd
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10088
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
We can just use take(size) to restrict reading to that as a max.
Change-Id: I0fbda74e4fb98ffeababae86a325233416029acf
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10072
Reviewed-by: raitobezarius <tvl@lahfa.xyz>
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
This allows converting from the NarInfo falling out of the NarInfo
parser (which is a bit annoying to handle due to lifetimes) to the
PathInfo proto struct.
The narinfo field, containing most of the data from the original
NARInfo file, as well as the references (bytes) are populated.
The node field is not populated, because it requires ingesting the NAR
itself to describe the root node.
Change-Id: I9c04dd6ad4cae556b455188a4255e34b4f6443c5
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10067
Reviewed-by: raitobezarius <tvl@lahfa.xyz>
Tested-by: BuildkiteCI
Autosubmit: flokli <flokli@flokli.de>
This moves the sync `channel::from_url` to a async
`tonic::channel_from_url`. It now allows connecting non-lazily if `wait-
connect=1` is set in the URL params.
Also, make the pingpong tests for blobsvc and directorysvc use the wait-
connect=1 codepath.
Change-Id: Ibeea33117c8121814627e7f6aba0e943ae2e92ca
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10030
Tested-by: BuildkiteCI
Reviewed-by: Connor Brewster <cbrewster@hey.com>
All we do is constructing some strings, and checking if from_addr
succeeds or not.
This can be written in a much more concise way using test_case.
Use lazy_static to provide temporary directories.
Also add some more grpc-related test cases.
Change-Id: Ia310dd01f617f7628f1e7e21304ac70da2ab3534
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10027
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Tested-by: BuildkiteCI
These gRPC PathInfoService tests were actually not too useful in here,
what we're mostly testing is the channel construction, so move it to
there.
Change-Id: Ic8c07558a1b28b46f863d5c39bcaa3a79cea007a
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10024
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Tested-by: BuildkiteCI
We don't gain much from making this part of the trait, it's still up to
`tvix_store::pathinfoservice::from_addr` to do most of the construction.
Move it out of the trait and into the specific *Service impls directly.
This allows further refactorings in followup CLs.
Change-Id: I99b93ef4acd83637a2f4888a1e586f1ca96390dc
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10022
Tested-by: BuildkiteCI
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Do this upgrade whenever someone is actually interested in the children
of a directory, but that directory doesn't contain a more detailed
listing. This is much more predictable, and removes a bunch of confusing
code from the inode tracker itself.
Change-Id: Ib3a13694d6d5d22887d2d04ae429592137f39cb4
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9982
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Have it return libc::ENOENT errors rather than an Option<…>.
Also avoid having to traverse inode_data multiple times, by synthesizing
the Arc<…> on our own in the insert case. In that case, the data is
quite small, so cloning it is faster than traversing a second time.
Change-Id: I7ab14bac8bb23859ed8d166a12070d4f4749b6d4
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9981
Tested-by: BuildkiteCI
Reviewed-by: Connor Brewster <cbrewster@hey.com>
As already established in the two previous CLs, these two pieces of code
where doing the same.
Move to a get_directory_children helper.
Change-Id: Id6876f0c34f3f40a31a22d59a2cdbfef39e2d8de
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9980
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Tested-by: BuildkiteCI
Very similar to the previous CL
Change-Id: I0df07ddca742b7b9485d48771c8d295dc3aa7136
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9979
Tested-by: BuildkiteCI
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Code after this big match block only cares about parent_digest and
children, so there's no need to do another inode_tracker.get in there.
This also allows removing another if let block, right after, as we don't
need to destructure parent_data anymore.
Change-Id: I68fbbe3304194670caee5a453722369afa4e77ea
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9978
Tested-by: BuildkiteCI
Reviewed-by: Connor Brewster <cbrewster@hey.com>
This makes it much harder to keep the read lock around for too long, and
the code a bit easier to understand.
Change-Id: I7d99c85cadd433cad444b8edd34e2c43d7eaf5a8
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9977
Tested-by: BuildkiteCI
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Not a single call site actually makes use of the Vec.
Change-Id: I6cf31073c9f443d1702a21937a0c3938c2c643b8
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9988
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
This adds a function that can ingest a NAR file into a given
blob_service and directory_service, returning the root node.
Passing around the directory_putter is a bit annoying, but at least for
this linear "reading through a NAR" case, we don't need it to be clone
or sync.
Change-Id: Ia26d596f5dffc6dbd69aa39a82b59bba41685a9c
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9915
Reviewed-by: raitobezarius <tvl@lahfa.xyz>
Tested-by: BuildkiteCI
