Removes imbl::OrdMap in favour of an Rc over the standard library's BTreeMap,
which allows us to drop the imbl dependency completely.
In my local tests this is actually slightly faster for `hello` and `firefox`.
Change-Id: Ic9597ead4e98bf9530f290c6a94a3c5c3efd0acc
Reviewed-on: https://cl.tvl.fyi/c/depot/+/12201
Reviewed-by: aspen <root@gws.fyi>
Tested-by: BuildkiteCI
This replaces the OpCode enum with a new Op enum which is guaranteed to fit in a
single byte. Instead of carrying enum variants with data, every variant that has
runtime data encodes it into the `Vec<u8>` that a `Chunk` now carries.
This has several advantages:
* Less stack space is required at runtime, and fewer allocations are required
while compiling.
* The OpCode doesn't need to carry "weird" special-cased data variants anymore.
* It is faster (albeit, not by much). On my laptop, results consistently look
approximately like this:
Benchmark 1: ./before -E '(import <nixpkgs> {}).firefox.outPath' --log-level ERROR --no-warnings
Time (mean ± σ): 8.224 s ± 0.272 s [User: 7.149 s, System: 0.688 s]
Range (min … max): 7.759 s … 8.583 s 10 runs
Benchmark 2: ./after -E '(import <nixpkgs> {}).firefox.outPath' --log-level ERROR --no-warnings
Time (mean ± σ): 8.000 s ± 0.198 s [User: 7.036 s, System: 0.633 s]
Range (min … max): 7.718 s … 8.334 s 10 runs
See notes below for why the performance impact might be less than expected.
* It is faster while at the same time dropping some optimisations we previously
performed.
This has several disadvantages:
* The code is closer to how one would write it in C or Go.
* Bit shifting!
* There is (for now) slightly more code than before.
On performance I have the following thoughts at the moment:
In order to prepare for adding GC, there's a couple of places in Tvix where I'd
like to fence off certain kinds of complexity (such as mutating bytecode, which,
for various reaons, also has to be part of data that is subject to GC). With
this change, we can drop optimisations like retroactively modifying existing
bytecode and *still* achieve better performance than before.
I believe that this is currently worth it to pave the way for changes that are
more significant for performance.
In general this also opens other avenues of optimisation: For example, we can
profile which argument sizes actually exist and remove the copy overhead of
varint decoding (which does show up in profiles) by using more adequately sized
types for, e.g., constant indices.
Known regressions:
* Op::Constant is no longer printing its values in disassembly (this can be
fixed, I just didn't get around to it, will do separately).
Change-Id: Id9b3a4254623a45de03069dbdb70b8349e976743
Reviewed-on: https://cl.tvl.fyi/c/depot/+/12191
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
This splits the existing ReferenceScanner into a ReferenceScanner and
ReferencePattern as well as adds an AsyncRead implementation that can
do a scan while you read from it.
The reason to split the scanner in two is that generating the pattern
is expensive and when ingesting build results with multiple outputs you
want to do several independant scans that look for the same pattern.
The reader is for scanning files without having to load the entire file
into memory.
Change-Id: I993f5a32308c12d9035840f8e04fe82e8dc1d962
Reviewed-on: https://cl.tvl.fyi/c/depot/+/12052
Autosubmit: Brian Olsen <me@griff.name>
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
Rather than storing the leaked allocation for the string as the key in
the interner, store the hash (using NoHashHashBuilder). I thought this
would improve performance, but it doesn't:
hello outpath time: [736.85 ms 748.42 ms 760.42 ms]
change: [-2.0754% +0.4798% +2.7096%] (p = 0.72 > 0.05)
No change in performance detected.
but it at least doesn't *hurt* performance, and it *does* avoid an
`unsafe`, so it's probably net good.
Change-Id: Ie413955bdb6f04b1f468f511e5ebce56e329fa37
Reviewed-on: https://cl.tvl.fyi/c/depot/+/12049
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
Autosubmit: aspen <root@gws.fyi>
Use the faster and newer MiMalloc memory allocator for all endpoints in
the workspace.
Change-Id: Ic60237284ed168e46ec6e8f28e2710bae4385c6f
Reviewed-on: https://cl.tvl.fyi/c/depot/+/12149
Tested-by: BuildkiteCI
Reviewed-by: aspen <root@gws.fyi>
Per https://nnethercote.github.io/perf-book/hashing.html, we have
basically no reason to use the default hasher over a faster,
non-DoS-resistant hasher. This gives a nice perf boost basically for
free:
hello outpath time: [704.76 ms 714.91 ms 725.63 ms]
change: [-7.2391% -6.1018% -4.9189%] (p = 0.00 < 0.05)
Performance has improved.
Change-Id: If5587f444ed3af69f8af4eead6af3ea303b4ae68
Reviewed-on: https://cl.tvl.fyi/c/depot/+/12046
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
Reviewed-by: Ilan Joselevich <personal@ilanjoselevich.com>
Autosubmit: aspen <root@gws.fyi>
This provides a PathInfoService implementation using redb
(https://github.com/cberner/redb) as the underlying storage engine.
Both an in-memory variant, as well as a filesystem one is provided,
similar how it's done with the sled implementation.
Supersedes: https://cl.tvl.fyi/c/depot/+/11692
Change-Id: I744619c51bf2efd0fb63659b12a27cbe0b2fd6fc
Signed-off-by: Ilan Joselevich <personal@ilanjoselevich.com>
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11995
Reviewed-by: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
It introduces a new accept_trace function for axum0.7 which can be used
to accept a header trace from a received request. This function can be
used for tonic 0.12 once that version is released, and the specific
`accept_trace` function within `tvix_tracing::propagate::tonic` can then
be removed.
This also integrates http propagation into the nar_bridge crate.
Change-Id: I46dcc797d494bb3977c2633753e7060d88d29129
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11925
Reviewed-by: Brian Olsen <me@griff.name>
Tested-by: BuildkiteCI
Reviewed-by: Simon Hauser <simon.hauser@helsinki-systems.de>
Reviewed-by: flokli <flokli@flokli.de>
This ingests NAR files into the {Blob,Directory}Service, which are
already part of the AppState.
As we then need to correlate the root node to the uploaded PathInfo, we
need to keep a (short-lived) lookup table from NARHash to root node
around. We insert it into a `LruCache` after the NAR is uploaded, and
use `peek()` to do the lookup, which doesn't update the LRU list.
Change-Id: I48a4c6246bacf76559c5a4ccad2a0bc25c1b7900
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11986
Tested-by: BuildkiteCI
Reviewed-by: Brian Olsen <me@griff.name>
This adds an implementation of nar-bridge in Rust.
Currently, only the GET parts are implemented.
Contrary to the Go variant, this doesn't try to keep a mapping from nar
hashes to root node in memory, it simply encodes the root node itself
(stripped by its basename) into the URL.
This pulls in a more recent version of axum than what we use in
tonic, causing two versions of http and hyper, however dealing with
`Body::from_stream` in axum 0.6 is much more annoying, and
https://github.com/hyperium/tonic/pull/1740 suggests this will be fixed
soon.
Change-Id: Ia4c2dbda7cd3fdbe47a75f3e33544d19eac6e44e
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11898
Autosubmit: flokli <flokli@flokli.de>
Reviewed-by: Brian Olsen <me@griff.name>
Tested-by: BuildkiteCI
Juggle around the internals of the tvix-cli crate so that we expose the
Repl as a public type with a `send` method, that sends a string to the
repl and *captures all output* so that it can be subsequently asserted
on in tests. Then, demonstrate that this works with a single (for now)
REPL test using expect-test to assert on the output of a single command
sent to the REPL.
As the REPL gets more complicated, this will allow us to make tests that
cover that complex behavior.
Change-Id: I88175bd72d8760c79faade95ebb1d956f08a7b83
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11958
Autosubmit: aspen <root@gws.fyi>
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
Allow binding variables at the REPL's toplevel in the same way the Nix
REPL does, using the syntax <ident> = <expr>. This fully, strictly
evaluates the value and sets it in the repl's "env", which gets passed
in at the toplevel when evaluating expressions.
The laziness behavior differs from Nix's, but I think this is good:
❯ nix repl
Welcome to Nix version 2.3.18. Type :? for help.
nix-repl> x = builtins.trace "x" 1
nix-repl> x
trace: x
1
nix-repl> x
1
vs tvix:
tvix-repl> x = builtins.trace "x" 1
trace: "x" :: string
tvix-repl> x
=> 1 :: int
tvix-repl> x
=> 1 :: int
Bug: https://b.tvl.fyi/issues/371
Change-Id: Ieb2d626b7195fa87be638c9a4dae2eee45eb9ab1
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11954
Reviewed-by: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Autosubmit: aspen <root@gws.fyi>
Introduces a helper function within tvix-tracing that returns a reqwest
tracing middleware that will ingest the traceparent if otlp is enabled.
It is feature flagged in tvix-tracing so not every consumer of that
library automatically has reqwest in its dependencies.
Tested using netcat to verify that the `traceparent` header is there if
otlp is enabled and missing if otlp feature is disabled.
Change-Id: I5abccae777b725f5ff7382e3686165383c477a39
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11886
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
This allows binding on unix sockets, as well as systemd socket
activation.
Change-Id: Icf648c4fd0895468c52607deb6397b8b5928102b
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11901
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Reviewed-by: Connor Brewster <cbrewster@hey.com>
This introduces optional helper function in tvix/tracing for trace
propagation and uses these helper in the `tvix-store`.
The GRPCBlobService, GRPCDirectoryService and GRPCPathInfoService now
accept a generic client, meaning the client can be generated with either
`::new` or `::with_interceptor`.
This was tested and validated by starting a `tvix-store daemon` and
`tvix-store import`.
Change-Id: I4b194483bf09266820104b4b56e4a135dca2b77a
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11863
Reviewed-by: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Don't show an empty spinner for daemon commands.
Move the bar to the right, so the text is better aligned between spinner
progress and bar progress styles.
Generally, push progress bars a bit more down to the place where we can
track progress. This includes adding one in the upload_blob span.
Introduce another progress style template for transfers, which
interprets the counter as bytes (not just a plain integer), and also a data rate.
Use it for here and in the fetching code, and also make the progress bar
itself a bit less wide.
Change-Id: I15c2ea3d2b24b5186cec19cd3dbd706638497f40
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11845
Tested-by: BuildkiteCI
Autosubmit: flokli <flokli@flokli.de>
Reviewed-by: Simon Hauser <simon.hauser@helsinki-systems.de>
Both umounts happening from another process, as well as tvix-store
itself calling umount() on FuseDaemon will cause the FUSE worker threads
to terminate.
So far there was no nice way to wait on these threads to be terminated
from multiple places, causing the `tvix-store mount` command to only be
terminated if interrupted via ctrl-c, not via an external umount.
Update FuseDaemon to use a ThreadPool, which gives us a join primitive
over all threads, that can also be called from multiple places.
Await on a join() from there to end the program, not the ctrl-c signal
handler as it was before.
Using FuseDaemon from multiple tasks requires Arc<>-ing both the
ThreadPool as well as the inner FuseSession (which also needs to be
inside a Mutex if we want to unmount), but now we can clone FuseDaemon
around and use it in two places. We could probably also have used an
Option and drop the FuseSession after the first umount, but this looks
cleaner.
Change-Id: Id635ef59b560c111db52ad0b3ca3d12bc7ae28ca
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11825
Reviewed-by: Brian Olsen <me@griff.name>
Tested-by: BuildkiteCI
This introduces another feature flag, "tracy" to the `tvix-tracing` crate.
If enabled (not enabled by default), it'll add an additional layer
emitting packets in a format that https://github.com/wolfpld/tracy can
display.
I had to be a bit tricky with the combinatorial complexity when adding
this, but the resulting code still seems manageable.
Change-Id: Ica824496728fa276ceae3f7a9754be0166e6558f
Reviewed-on: https://cl.tvl.fyi/c/depot/+/10952
Tested-by: BuildkiteCI
Reviewed-by: Simon Hauser <simon.hauser@helsinki-systems.de>
Reviewed-by: flokli <flokli@flokli.de>
Provide a new interface for forcing a flush of otlp traces and use this
interface to shutdown otlp prior to exiting tvix-store, either if the
tool was stopped with a SIGTERM or ended regularly.
This also fixes an issue where traces were not even exported if for
example we just imported 10 paths and never even emitted more than 256
traces. The implementation uses a mpsc channel so a flush can be done
without having to wait for it to complete. If you want to wait for a
flush to complete you can provide a oneshot channel which will receive a
message once flushing is complete.
Because of a otlp bug `force_flush` as well as
`shutdown_tracer_provider` need to be executed using `spawn_blocking`
otherwise the function will deadlock. See
https://github.com/open-telemetry/opentelemetry-rust/issues/1395#issuecomment-1953280335
Change-Id: I0a828391adfb1f72dc8305f62ced8cba0515847c
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11803
Reviewed-by: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Autosubmit: Simon Hauser <simon.hauser@helsinki-systems.de>
This gives some better feedback something is going on, gives a (named)
root for all the progress children we're drawing, and also counts the
time we are in eval.
Change-Id: Ibe81dcebf0a2b59bb0680da62e206eb5270f9d3c
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11798
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Reviewed-by: aspen <root@gws.fyi>
In `store_path_to_node`, in case we need to build or fetch something,
render a progress bar, using the spinner for now.
We can upgrade this to a progress *bar* later.
Change-Id: I4a7cf5ef8f639076f176af9b39d276be3f37c8ff
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11793
Autosubmit: flokli <flokli@flokli.de>
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Tested-by: BuildkiteCI
Introduce a new common crate that contains tracing boilerplate which then
can be used in the cli, tvix-store and tvix-build crates.
It has otlp as an optional feature, which is currently only used by
tvix-store.
Change-Id: I41468ac4d9c65174515d721513b96fea463d6ed2
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11758
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
Autosubmit: Simon Hauser <simon.hauser@helsinki-systems.de>
We don't use this in tvix-glue currently. A previous version of cl/11747
did, and I forgot to remove it afterwards.
Change-Id: I6191ffc60450f3e79f22a3aca55246e3956ee9cc
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11756
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
Reviewed-by: Simon Hauser <simon.hauser@helsinki-systems.de>
Autosubmit: flokli <flokli@flokli.de>
This adds the tracing-indicatif crate, and configures it as a layer in
our tracing_subscriber pipeline to emit progress for every span that's
configured so.
It also moves from using std::io::stderr to write logs to using their
writer, to avoid clobbering output.
Progress bar styles are defined in a lazy_static, moving this into a
general tracing is left for later.
This adds some usage of this to the `imports` and `copy` commands.
The output can still be improved a bit - we should probably split each
task up into a smaller (instrumented) helper functions, so we can create
a progress bar for each task.
Change-Id: I59a1915aa4e0caa89c911632dec59c4cbeba1b89
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11747
Reviewed-by: flokli <flokli@flokli.de>
Reviewed-by: Simon Hauser <simon.hauser@helsinki-systems.de>
Tested-by: BuildkiteCI
Autosubmit: flokli <flokli@flokli.de>
Transparently support multiline input in the Tvix REPL, by handling the
UnexpectedEOF error returned by the parser and using it to progressively
build up an input expr over multiple iterations of the REPL's outer
loop.
This works quite nicely:
❯ cargo r --bin tvix
Compiling tvix-cli v0.1.0 (/home/aspen/code/depot/tvix/cli)
Finished dev [unoptimized + debuginfo] target(s) in 1.72s
Running `target/debug/tvix`
tvix-repl> { foo
> =
> 1;
> }
=> { foo = 1; } :: set
<press up arrow>
tvix-repl> { foo
=
1;
}
=> { foo = 1; } :: set
Change-Id: Ib0ed4766b13e8231d696cdc27281ac158e20a777
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11732
Reviewed-by: tazjin <tazjin@tvl.su>
Autosubmit: aspen <root@gws.fyi>
Reviewed-by: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
In order to be compatible with the nix XML generator, it’s easier to
generate the XML directly, instead of going through a library which we
have to bend to do what we need.
Removes dependency on `xml-rs`, which came with a full XML parser that
we didn’t use. Only takes a tiny bit of code for the XML escaping,
somewhat simplified.
I add a little escaping value, to make sure we have the same behaviour
as nix proper.
Interestingly enough, we never need to escape XML attribute names,
because the `builtins.toXML` format encodes user-defined values as
attribute keys only. So we only escape attribute values.
Fixes: https://b.tvl.fyi/issues/399
Change-Id: If4d407d324864b3bb9aa3160e2ec6889f7727127
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11697
Tested-by: BuildkiteCI
Reviewed-by: flokli <flokli@flokli.de>
Autosubmit: Profpatsch <mail@profpatsch.de>
Drop the (unused) feature flags here, and get rid of some crate
dependencies.
Change-Id: Id64852b498725467d56abb12b548301cfba6a760
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11678
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
This gives multi-listener support, as well as listening on named socket
FDs.
Change-Id: I92b441f854e2faf80074463d0ca6bdc23cbd890a
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11663
Tested-by: BuildkiteCI
Reviewed-by: yuka <yuka@yuka.dev>
Autosubmit: flokli <flokli@flokli.de>
This aligns the tonic version we use directly in Tvix to the one pulled
in by the OTLP stack.
Change-Id: I658528c8dabb7cd6948f1207ddcdef1984e82037
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11666
Autosubmit: flokli <flokli@flokli.de>
Reviewed-by: yuka <yuka@yuka.dev>
Tested-by: BuildkiteCI
This one doesn't require us to deal with poisoning, is upgradeable and
the right thing to use when locking access to data, not IO resources.
Change-Id: I78634953a73404500d28f51f1d93a87e215c8149
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11612
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
Reviewed-by: Connor Brewster <cbrewster@hey.com>
This provides an implementation of PathInfoService storing PathInfo in
memory up to a certain capacity, then evicting these that have been used
the least recently.
Change-Id: I9d738687caf4f181a957f72245f26b92832313cd
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11622
Tested-by: BuildkiteCI
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Autosubmit: flokli <flokli@flokli.de>
Rust 1.77 supports async recursion as long as there is some form of
indirection (ie. `Box::pin`). This removes the need to use the
async-recursion crate.
Change-Id: Ic9613ab7f32016f0103032a861edff92e2fb8b41
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11596
Reviewed-by: flokli <flokli@flokli.de>
Autosubmit: Connor Brewster <cbrewster@hey.com>
Tested-by: BuildkiteCI
Rename read_nar to ingest_nar, and have it use the async nar reader
version, and the ingest_entries machinery.
This means we can now drop all code dealing with manually assembling
castore nodes.
Update our consumer, NixHTTPPathInfoService to use the new API.
As we now accept an AsyncRead, we don't need to do any blocking here
anymore, and can use the same async-compression crate as in the fetching
logic (and support some more compression formats out of the box).
Change-Id: I8646d20bd8603f8da47b5c84bc9e4ac236eb7f1a
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11580
Tested-by: BuildkiteCI
Autosubmit: flokli <flokli@flokli.de>
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Nix supports file:// - URLs for `fetchurl` and `fetchTarball`.
Convert the enums and function arguments to hold a URL type.
reqwest::Url is a re-export of the url crate, but they don't re-export
the parsing errors, and as we want to hold these in our Error types, add
it to Cargo.toml explicitly.
The Fetcher::download function now checks on the scheme, and either
opens the file locally, or does do a HTTP request as before.
Fetch gets its custom debug impl, removing potentially sensitive
username and password out of URLs.
Change-Id: I777db1fe487370e822cbfec4624034aca5e08045
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11504
Autosubmit: flokli <flokli@flokli.de>
Reviewed-by: raitobezarius <tvl@lahfa.xyz>
Tested-by: BuildkiteCI
We actually want to delay fetching until we actually need the file. A
simple evaluation asking for `.outPath` or `.drvPath` should work even
in a pure offline environment.
Before this CL, the fetching logic was quite distributed between
tvix_store_io, and builtins/fetchers.rs.
Rather than having various functions and conversions between structs,
describe a Fetch as an enum type, with the fields describing the fetch.
Define a store_path() function on top of `Fetch` which can be used to
ask for the calculated store path (if the digest has been provided
upfront).
Have a `Fetcher` struct, and give it a `fetch_and_persist` function,
taking a `Fetch` as well as a desired name, and have it deal with all
the logic of persisting the PathInfos. It also returns a StorePathRef,
similar to the `.store_path()` method on a `Fetch` struct.
In a followup CL, we can extend KnownPaths to track fetches AND
derivations, and then use `Fetcher` when we need to do IO into that
store path.
Change-Id: Ib39a96baeb661750a8706b461f8ba4abb342e777
Reviewed-on: https://cl.tvl.fyi/c/depot/+/11500
Reviewed-by: raitobezarius <tvl@lahfa.xyz>
Autosubmit: flokli <flokli@flokli.de>
Tested-by: BuildkiteCI
