Right now all blob uploads are performed synchronously, this means if a
NAR contains many small files, the import time is dominated by round
trip time to the blob service. For small files, we can buffer them in
memory and upload them asynchronously to the blob service. Before
returning we make sure to join all the uploads to make sure they
complete successfully before responding OK.
This reduces time to import a bash-interactive closure on my machine
from 1m19s to 7s.
Change-Id: Ica3695c159e6c8ad8769281ac20d037e3143e856
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10679
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
Autosubmit: Connor Brewster <cbrewster@hey.com>
Sets up OpenTelemetry integration for nar-bridge. Right now it will
export spans for HTTP server requests and all gRPC client requests.
Having the spans available will make performance work significantly
easier as it provides a high level overview of where time is being
spent.
In the future we can add application-specifc metrics and
integrate logrus.
Change-Id: Ie3860675d7ffc626a95673ba062c3c798d8bb2a7
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10678
Reviewed-by: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Autosubmit: Connor Brewster <cbrewster@hey.com>
This now exists in tvix-store directly, as NixHTTPPathInfoService, and
contrary to this version, also validates signatures.
Change-Id: Ib6ca161e40d627b7d9741839fc849f2392f422da
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10155
Tested-by: BuildkiteCI
Autosubmit: flokli <flokli@flokli.de>
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Bump code.tvl.fyi/tvix/store/protos past cl/9649, where Validate()
already ensures the NarSha256 has the correct size.
Change-Id: I774668822f4d9dbd4dea47dde6e4745dc95e8e7f
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9665
Reviewed-by: edef <edef@edef.eu>
Tested-by: BuildkiteCI
As correctly mentioned in
https://cl.tvl.fyi/c/depot/+/9652/comment/03b9b96e_bbb337fd/,
we shouldn't be using these magic constants, but pull them from where
they're defined.
This already is a dependency of go-nix, and pkg/pathinfosvc/server.go,
so no changes in go.mod.
Change-Id: I0cc41ce040fcbddf4b6171417bc9b0de55af4991
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9653
Tested-by: BuildkiteCI
Reviewed-by: Brian McGee <brian@bmcgee.ie>
We have nixhash.FromHashTypeAndDigest now.
Also, run Validate() on the PathInfo received from the remote
PathInfoService.
Change-Id: I14db0d9356c539c084afc9dd712314b56da2587e
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9652
Tested-by: BuildkiteCI
Reviewed-by: Brian McGee <brian@bmcgee.ie>
… and nar size / sha256 digest.
Instead of producing sparse PathInfo messages when NARs are sent to
nar-bridge, the nar-bridge http server now keeps a lookup table
(narsha256) -> (rootNode, narSize)
This removes a whole bunch of noise, because we don't need to keep
sparse fields around.
A convenience function
`GenPathInfo(rootNode *castorev1pb.Node, narInfo *narinfo.NarInfo)` is
added, which is used to produce PathInfo messages, either when receiving
a NAR file over http and uploading it to a remote PathInfoService, or to
synthesize the PathInfoMessage to return to the client, if nar-bridge is
acting as a PathInfoService for a remove Nix HTTP Binary cache.
Change-Id: Ibba1ab6238a050816c4fab29cb21ae88877d8613
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9651
Tested-by: BuildkiteCI
Reviewed-by: Brian McGee <brian@bmcgee.ie>
Bumps the go module past cl/9604 and update the consumer side.
Change-Id: Id44245017f1dc2f8aac28051cdbb45b83bdc5be3
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9650
Reviewed-by: Brian McGee <brian@bmcgee.ie>
Tested-by: BuildkiteCI
This removes the Export method in nar-bridge, and updates all users to
the version now in storev1pb.
It moves the roundtrip test to the importer crate, and some of the
utility functions into a separate util_test.go file.
Change-Id: I81d9e0b35dfd78ef1042bed307281eecd2aaa2a8
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9603
Reviewed-by: Brian McGee <brian@bmcgee.ie>
Tested-by: BuildkiteCI
Export will traverse a given PathInfo structure, and write the contents
in NAR format to the passed Writer.
It uses directoryLookupFn and blobLookupFn to resolve references.
This is being moved over from nar-bridge. We need to keep the code there
around until we can bump go.mod to storev1 with this merged, but the
tests can already be moved entirely.
Change-Id: Ie0de3077b09344cafa00ff1e2ddb8b52e9e631bc
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9602
Tested-by: BuildkiteCI
Reviewed-by: Brian McGee <brian@bmcgee.ie>
Autosubmit: flokli <flokli@flokli.de>
We can use the helper to rename the node.
Change-Id: Id8defea7e5ebbd43d7b7a9b2992c62084e1828ec
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9601
Autosubmit: flokli <flokli@flokli.de>
Reviewed-by: Brian McGee <brian@bmcgee.ie>
Tested-by: BuildkiteCI
Convenience function, moves all code converting from a PathInfo struct
to to go-nix's NarInfo.
Change-Id: Idf0dcc38675674563f2dfd3286a4a55fa2a24a82
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9593
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Reviewed-by: Brian McGee <brian@bmcgee.ie>
We already have validation tests for Rust, let's add the missing ones
for golang too.
Change-Id: Iaf3a3e1ee72d5647da3f2aa977d6e0d0379b2ce5
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9595
Reviewed-by: Brian McGee <brian@bmcgee.ie>
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
This should make it quite quick to spot writing code breaking some of
the assumptions we have on PathInfo messages ourselves.
Change-Id: I480caaec41f8ea5246c3c3081460c7ad12e78569
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9554
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Tested-by: BuildkiteCI
Autosubmit: flokli <flokli@flokli.de>
We run narInfo.Check to ensure this parses to a StorePath, not
nixpath.Check.
Change-Id: Id91183128df74a60d98fa2a31174cd879194c34d
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9550
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
This check makes more sense there, and gives stronger semantics - Done()
only succeeds if the other side successfully received everything, *and*
came up with the same hashes as we did.
Change-Id: I20b706961053fd00d22cc70e1c8cc859705587e0
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9542
Tested-by: BuildkiteCI
Autosubmit: flokli <flokli@flokli.de>
Reviewed-by: Connor Brewster <cbrewster@hey.com>
This adds an additional nar-bridge-pathinfo command.
It exposes a PathInfoService for a HTTP Binary Cache, ingesting data
into a BlobService/DirectoryService as it goes through the NAR file.
It does this whenever it receives a Get request for a specific output
path, and waits returning with the PathInfo response until it ingested
the data.
It does not do any sort of caching - this means it re-downloads NAR
files again whenever the PathInfo is requested again, so you most likely
do not want to use this currently.
It's one building component as soon as we have store composition (which
we currently don't, so don't use this).
It can be used as an alternative mechanism to ingest data (Blobs and
Directories) of a given store path from a binary cache into tvix-store.
```
❯ nix-build -A third_party.nixpkgs.hello
/nix/store/mdi7lvrn2mx7rfzv3fdq3v5yw8swiks6-hello-2.12.1
❯ nix hash to-sri --type sha1 mdi7lvrn2mx7rfzv3fdq3v5yw8swiks6
sha1-Rs/INeK+7IGbG/u7fHoVNm96Yqs=
❯ out=$(mg build //tvix/nar-bridge)
$out/bin/nar-bridge-pathinfo --log-level debug &
INFO[0000] Starting nar-bridge-pathinfosvc at [::]:8001
❯ mg run //tvix:store -- daemon &
[mg] building target //tvix:store
[mg] running target //tvix:store
2023-10-03T16:21:57.433739Z INFO tvix_store: tvix-store listening on [::]:8000
at src/bin/tvix-store.rs:229
❯ evans --host localhost --port 8001 -r repl
[…]
tvix.store.v1.PathInfoService@localhost:8001> call Get
✔ by_output_hash
by_output_hash (TYPE_BYTES) => Rs/INeK+7IGbG/u7fHoVNm96Yqs=
{
"narinfo": {
"narSha256": "sXrPtjqhSoc2u0YfM1HVZThknkSYuRuHdtKCB6wkDFo=",
"narSize": "226552",
"referenceNames": [
"aw2fw9ag10wr9pf0qk4nk5sxi0q0bn56-glibc-2.37-8",
"mdi7lvrn2mx7rfzv3fdq3v5yw8swiks6-hello-2.12.1"
],
"signatures": [
{
"data": "7guDbfaF2Q29HY0c5axhtuacfxN6uxuEqeUfncDiSvMSAWvfHVMppB89ILqV8FE58pEQ04tSbMnRhR3FGPV0AA==",
"name": "cache.nixos.org-1"
}
]
},
"node": {
"directory": {
"digest": "xvo6BYbYaDw76IibLu5sr+VZoj9iM0ET2RUuYSYLwKE=",
"name": "bWRpN2x2cm4ybXg3cmZ6djNmZHEzdjV5dzhzd2lrczYtaGVsbG8tMi4xMi4x",
"size": 141
}
},
"references": [
"ptgFMIhdl2nJxMDdlDkITyXuBFc=",
"Rs/INeK+7IGbG/u7fHoVNm96Yqs="
]
}
```
Change-Id: I50167d0ac081c91adf5cf2733bbc4dc0993bd46e
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9539
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Reviewed-by: Brian Olsen <me@griff.name>
Use a genNarHandler() function accepting a boolean to construct the
HTTP handler.
Change-Id: I17c054826d91a9dbed8b1f53945a51f27fa60ace
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9537
Tested-by: BuildkiteCI
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Autosubmit: flokli <flokli@flokli.de>
This is only dealing with the HTTP interface.
Change-Id: I011b624fd9f11ea96231b92fea1166c118a219f2
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9535
Tested-by: BuildkiteCI
Autosubmit: flokli <flokli@flokli.de>
Reviewed-by: Connor Brewster <cbrewster@hey.com>
We can drop most of Hasher if we use a MultiWriter writing to the hash
function and a minimal CountingWriter.
This should make things a bit more understandable.
Change-Id: I37ee72d9a5c73f253aecc1ad761cb723389b89fc
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9529
Autosubmit: flokli <flokli@flokli.de>
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Tested-by: BuildkiteCI
This aligns behaviour more with how it should be - it's the
responsibility of the callback functions to return digests of the things
they consume(d). It allows further cleaning up the hasher struct.
Change-Id: I9cbfc87e6abd4ff17fadf39eb6563ec3cb7fcc6f
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9528
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Make the import function usable on any reader.
Change-Id: I84d2004cb73cdd7a11fe8efb0f2efb6335d5e6b0
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9527
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Tested-by: BuildkiteCI
Autosubmit: flokli <flokli@flokli.de>
This is only called once.
Change-Id: I342443b8d04050929733fc84d5f36cd64060afe3
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9525
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
More aligned with how it's called in other places
Change-Id: I759ac7ca3b5b69c1101d2d51a569d76c183a6330
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9362
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Create a pipe, pass the read end, and have a goroutine write to the
write end.
Change-Id: I301c273355705e60113b018e7e84b76972200e8c
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9361
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Only keep the `serve` subcommand, and make it appear at the root.
Introduce a --log-level argument, and be a bit less noisy in normal
operation.
Change-Id: I86b8abde1869a5c0c947508bcc29f845222aac09
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9360
Autosubmit: flokli <flokli@flokli.de>
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Tested-by: BuildkiteCI
`nix copy` checks if NARs and NARInfo files are present, before
uploading them. That's not an error, but normal behaviour, so no need to
log with level info for these cases.
We only want to log if the error is not a 404, and log with Warn level.
Change-Id: I762de3b862d070a0f18bc62e324e94ca5c7c3693
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9359
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Tested-by: BuildkiteCI
Let's make sure we don't end up blocking a client too much when
inserting very small blobs.
Change-Id: I640dda92efae538c70d32a40e6e85a23e9749e20
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9358
Tested-by: BuildkiteCI
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Instead of creating one big BlobChunk containing all data, and creating
way too large proto messages, chunk blobs up to a reasonable (1MiB)
chunk size, and send them to the server like that.
Change-Id: Ia45a53956a6d7c0599cc59ac516ba37e9fb1b30e
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9357
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Tested-by: BuildkiteCI
This change got lost in the rebases in cl/9348. There's unnecessary
`break`/`continues` that can be replaced by moving the conditional into
the for loop condition.
Change-Id: I559e21087630b05e483f768ab59f8067961a2eae
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9352
Reviewed-by: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Previously, nar-bridge, had a couple of bugs with tracking the current
directory when traversing a NAR file.
The included test case looks like:
```
/ (dir)
/test (dir)
/test/tested (file)
/tested (file)
```
Previously, we would do a string prefix match between the current node
and the top of the directory stack to determine if the node is in the
directory. In this case `/test` is a substring of `/tested`; however,
`/tested` is not in the `/test` directory. The fix is to append a `/` to
the directory name when doing the prefix match, so `/test/` is not a
prefix of `/tested`.
Additionally, when popping the stack, we need to continuously pop the
stack until the new node is in the directory at the top of the stack
(stopping before we pop the root directory)
Example:
```
/ (dir)
/a (dir)
/a/b (dir)
/a/b/c (file)
/z (file)
```
Previously, `z` would end up in directory `/a` because we only the pop
the stack once.
The included test case requires both of these issues to be fixed for it
to pass, so I think it is sufficient.
Change-Id: I22f601babf04d39d85535ba7ad585d3970757211
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9348
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
Autosubmit: Connor Brewster <cbrewster@hey.com>
In some cases, Nix is not able to stream the NAR file fast enough. Bump the
timeouts for now. We might want to get a better understanding in what's
happening here long-term, and/or make the timeouts configurable.
Change-Id: Ieaa9c8f04bc73c6ce0679a058d07eaf87126634e
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9340
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
This provides a Nix HTTP Binary Cache interface in front of a tvix-store
that's reachable via gRPC.
TODOs:
- remove import command, move serve up to toplevel. We have nix-copy-
closure and tvix-store commands.
- loop into CI. We should be able to fetch the protos as a third-party
dependency.
- Check if we can test nar-bridge slightly easier in an integration
test.
- Ensure we support connecting to unix sockets and grpc+http at least,
using the same syntax as tvix-store.
- Don't buffer the entire blob when rendering NAR
Co-Authored-By: Connor Brewster <cbrewster@hey.com>
Co-Authored-By: Márton Boros <martonboros@gmail.com>
Co-Authored-By: Vo Minh Thu <noteed@gmail.com>
Change-Id: I6064474e49dfe78cea67676957462d9f28658d4a
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9339
Tested-by: BuildkiteCI
Reviewed-by: tazjin <tazjin@tvl.su>
