From c83841d3a13a5d5b3fdaa7f9beecd743f1283691 Mon Sep 17 00:00:00 2001 From: Florian Klink Date: Mon, 13 Nov 2023 13:48:30 +0200 Subject: refactor(tvix/castore): remove DirectoryService::from_url Make directoryservice::from_addr use the more specific constructors. Change-Id: I9fee2afed77692505988d631d9fe246d9843d25a Reviewed-on: https://cl.tvl.fyi/c/depot/+/10029 Tested-by: BuildkiteCI Reviewed-by: Connor Brewster --- tvix/castore/src/directoryservice/from_addr.rs | 97 ++++++++++++++++++++++- tvix/castore/src/directoryservice/grpc.rs | 18 +---- tvix/castore/src/directoryservice/memory.rs | 62 --------------- tvix/castore/src/directoryservice/mod.rs | 6 -- tvix/castore/src/directoryservice/sled.rs | 102 ------------------------- 5 files changed, 98 insertions(+), 187 deletions(-) (limited to 'tvix') diff --git a/tvix/castore/src/directoryservice/from_addr.rs b/tvix/castore/src/directoryservice/from_addr.rs index 776cf06109..6082a8b49f 100644 --- a/tvix/castore/src/directoryservice/from_addr.rs +++ b/tvix/castore/src/directoryservice/from_addr.rs @@ -1,6 +1,8 @@ use std::sync::Arc; use url::Url; +use crate::{proto::directory_service_client::DirectoryServiceClient, Error}; + use super::{DirectoryService, GRPCDirectoryService, MemoryDirectoryService, SledDirectoryService}; /// Constructs a new instance of a [DirectoryService] from an URI. @@ -22,11 +24,44 @@ pub fn from_addr(uri: &str) -> Result, crate::Error> { .map_err(|e| crate::Error::StorageError(format!("unable to parse url: {}", e)))?; Ok(if url.scheme() == "memory" { - Arc::new(MemoryDirectoryService::from_url(&url)?) + // memory doesn't support host or path in the URL. + if url.has_host() || !url.path().is_empty() { + return Err(Error::StorageError("invalid url".to_string())); + } + Arc::new(MemoryDirectoryService::default()) } else if url.scheme() == "sled" { - Arc::new(SledDirectoryService::from_url(&url)?) + // sled doesn't support host, and a path can be provided (otherwise + // it'll live in memory only). + if url.has_host() { + return Err(Error::StorageError("no host allowed".to_string())); + } + + if url.path() == "/" { + return Err(Error::StorageError( + "cowardly refusing to open / with sled".to_string(), + )); + } + + // TODO: expose compression and other parameters as URL parameters? + + if url.path().is_empty() { + return Ok(Arc::new( + SledDirectoryService::new_temporary() + .map_err(|e| Error::StorageError(e.to_string()))?, + )); + } + return Ok(Arc::new( + SledDirectoryService::new(url.path().into()) + .map_err(|e| Error::StorageError(e.to_string()))?, + )); } else if url.scheme().starts_with("grpc+") { - Arc::new(GRPCDirectoryService::from_url(&url)?) + // schemes starting with grpc+ go to the GRPCPathInfoService. + // That's normally grpc+unix for unix sockets, and grpc+http(s) for the HTTP counterparts. + // - In the case of unix sockets, there must be a path, but may not be a host. + // - In the case of non-unix sockets, there must be a host, but no path. + // Constructing the channel is handled by tvix_castore::channel::from_url. + let client = DirectoryServiceClient::new(crate::channel::from_url(&url)?); + Arc::new(GRPCDirectoryService::from_client(client)) } else { Err(crate::Error::StorageError(format!( "unknown scheme: {}", @@ -34,3 +69,59 @@ pub fn from_addr(uri: &str) -> Result, crate::Error> { )))? }) } + +#[cfg(test)] +mod tests { + use super::from_addr; + use lazy_static::lazy_static; + use tempfile::TempDir; + use test_case::test_case; + + lazy_static! { + static ref TMPDIR_SLED_1: TempDir = TempDir::new().unwrap(); + static ref TMPDIR_SLED_2: TempDir = TempDir::new().unwrap(); + } + + /// This uses an unsupported scheme. + #[test_case("http://foo.example/test", false; "unsupported scheme")] + /// This configures sled in temporary mode. + #[test_case("sled://", true; "sled valid temporary")] + /// This configures sled with /, which should fail. + #[test_case("sled:///", false; "sled invalid root")] + /// This configures sled with a host, not path, which should fail. + #[test_case("sled://foo.example", false; "sled invalid host")] + /// This configures sled with a valid path path, which should succeed. + #[test_case(&format!("sled://{}", &TMPDIR_SLED_1.path().to_str().unwrap()), true; "sled valid path")] + /// This configures sled with a host, and a valid path path, which should fail. + #[test_case(&format!("sled://foo.example{}", &TMPDIR_SLED_2.path().to_str().unwrap()), false; "sled invalid host with valid path")] + /// This correctly sets the scheme, and doesn't set a path. + #[test_case("memory://", true; "memory valid")] + /// This sets a memory url host to `foo` + #[test_case("memory://foo", false; "memory invalid host")] + /// This sets a memory url path to "/", which is invalid. + #[test_case("memory:///", false; "memory invalid root path")] + /// This sets a memory url path to "/foo", which is invalid. + #[test_case("memory:///foo", false; "memory invalid root path foo")] + fn test_from_addr(uri_str: &str, is_ok: bool) { + assert_eq!(from_addr(uri_str).is_ok(), is_ok) + } + + // the gRPC tests below don't fail, because we connect lazily. + + /// Correct scheme to connect to a unix socket. + #[test_case("grpc+unix:///path/to/somewhere", true; "grpc valid unix socket")] + /// Correct scheme for unix socket, but setting a host too, which is invalid. + #[test_case("grpc+unix://host.example/path/to/somewhere", false; "grpc invalid unix socket and host")] + /// Correct scheme to connect to localhost, with port 12345 + #[test_case("grpc+http://[::1]:12345", true; "grpc valid IPv6 localhost port 12345")] + /// Correct scheme to connect to localhost over http, without specifying a port. + #[test_case("grpc+http://localhost", true; "grpc valid http host without port")] + /// Correct scheme to connect to localhost over http, without specifying a port. + #[test_case("grpc+https://localhost", true; "grpc valid https host without port")] + /// Correct scheme to connect to localhost over http, but with additional path, which is invalid. + #[test_case("grpc+http://localhost/some-path", false; "grpc valid invalid host and path")] + #[tokio::test] + async fn test_from_addr_tokio(uri_str: &str, is_ok: bool) { + assert_eq!(from_addr(uri_str).is_ok(), is_ok) + } +} diff --git a/tvix/castore/src/directoryservice/grpc.rs b/tvix/castore/src/directoryservice/grpc.rs index 43d460fae6..f0569ac878 100644 --- a/tvix/castore/src/directoryservice/grpc.rs +++ b/tvix/castore/src/directoryservice/grpc.rs @@ -35,18 +35,6 @@ impl GRPCDirectoryService { #[async_trait] impl DirectoryService for GRPCDirectoryService { - /// Constructs a [GRPCDirectoryService] from the passed [url::Url]: - /// - scheme has to match `grpc+*://`. - /// That's normally grpc+unix for unix sockets, and grpc+http(s) for the HTTP counterparts. - /// - In the case of unix sockets, there must be a path, but may not be a host. - /// - In the case of non-unix sockets, there must be a host, but no path. - fn from_url(url: &url::Url) -> Result { - let channel = crate::channel::from_url(url)?; - Ok(Self::from_client( - proto::directory_service_client::DirectoryServiceClient::new(channel), - )) - } - async fn get( &self, digest: &B3Digest, @@ -307,7 +295,7 @@ mod tests { use crate::{ directoryservice::{DirectoryService, GRPCDirectoryService, MemoryDirectoryService}, fixtures::{self, DIRECTORY_A, DIRECTORY_B}, - proto::GRPCDirectoryServiceWrapper, + proto::{directory_service_client::DirectoryServiceClient, GRPCDirectoryServiceWrapper}, utils::gen_directorysvc_grpc_client, }; @@ -476,7 +464,9 @@ mod tests { let grpc_client = { let url = url::Url::parse(&format!("grpc+unix://{}", socket_path.display())) .expect("must parse"); - GRPCDirectoryService::from_url(&url).expect("must succeed") + let client = + DirectoryServiceClient::new(crate::channel::from_url(&url).expect("must succeed")); + GRPCDirectoryService::from_client(client) }; assert!(grpc_client diff --git a/tvix/castore/src/directoryservice/memory.rs b/tvix/castore/src/directoryservice/memory.rs index ac67c999d0..3ba309f829 100644 --- a/tvix/castore/src/directoryservice/memory.rs +++ b/tvix/castore/src/directoryservice/memory.rs @@ -16,22 +16,6 @@ pub struct MemoryDirectoryService { #[async_trait] impl DirectoryService for MemoryDirectoryService { - /// Constructs a [MemoryDirectoryService] from the passed [url::Url]: - /// - scheme has to be `memory://` - /// - there may not be a host. - /// - there may not be a path. - fn from_url(url: &url::Url) -> Result { - if url.scheme() != "memory" { - return Err(crate::Error::StorageError("invalid scheme".to_string())); - } - - if url.has_host() || !url.path().is_empty() { - return Err(crate::Error::StorageError("invalid url".to_string())); - } - - Ok(Self::default()) - } - #[instrument(skip(self, digest), fields(directory.digest = %digest))] async fn get(&self, digest: &B3Digest) -> Result, Error> { let db = self.db.read()?; @@ -101,49 +85,3 @@ impl DirectoryService for MemoryDirectoryService { Box::new(SimplePutter::new(self.clone())) } } - -#[cfg(test)] -mod tests { - use super::DirectoryService; - use super::MemoryDirectoryService; - - /// This uses a wrong scheme. - #[test] - fn test_invalid_scheme() { - let url = url::Url::parse("http://foo.example/test").expect("must parse"); - - assert!(MemoryDirectoryService::from_url(&url).is_err()); - } - - /// This correctly sets the scheme, and doesn't set a path. - #[test] - fn test_valid_scheme() { - let url = url::Url::parse("memory://").expect("must parse"); - - assert!(MemoryDirectoryService::from_url(&url).is_ok()); - } - - /// This sets the host to `foo` - #[test] - fn test_invalid_host() { - let url = url::Url::parse("memory://foo").expect("must parse"); - - assert!(MemoryDirectoryService::from_url(&url).is_err()); - } - - /// This has the path "/", which is invalid. - #[test] - fn test_invalid_has_path() { - let url = url::Url::parse("memory:///").expect("must parse"); - - assert!(MemoryDirectoryService::from_url(&url).is_err()); - } - - /// This has the path "/foo", which is invalid. - #[test] - fn test_invalid_path2() { - let url = url::Url::parse("memory:///foo").expect("must parse"); - - assert!(MemoryDirectoryService::from_url(&url).is_err()); - } -} diff --git a/tvix/castore/src/directoryservice/mod.rs b/tvix/castore/src/directoryservice/mod.rs index 65beeef4a2..41d44078ca 100644 --- a/tvix/castore/src/directoryservice/mod.rs +++ b/tvix/castore/src/directoryservice/mod.rs @@ -21,12 +21,6 @@ pub use self::traverse::descend_to; /// digest. #[async_trait] pub trait DirectoryService: Send + Sync { - /// Create a new instance by passing in a connection URL. - /// TODO: check if we want to make this async, instead of lazily connecting - fn from_url(url: &url::Url) -> Result - where - Self: Sized; - /// Get looks up a single Directory message by its digest. /// In case the directory is not found, Ok(None) is returned. async fn get(&self, digest: &B3Digest) -> Result, Error>; diff --git a/tvix/castore/src/directoryservice/sled.rs b/tvix/castore/src/directoryservice/sled.rs index 041908f9f9..400d2fed59 100644 --- a/tvix/castore/src/directoryservice/sled.rs +++ b/tvix/castore/src/directoryservice/sled.rs @@ -34,34 +34,6 @@ impl SledDirectoryService { #[async_trait] impl DirectoryService for SledDirectoryService { - /// Constructs a [SledDirectoryService] from the passed [url::Url]: - /// - scheme has to be `sled://` - /// - there may not be a host. - /// - a path to the sled needs to be provided (which may not be `/`). - fn from_url(url: &url::Url) -> Result { - if url.scheme() != "sled" { - return Err(crate::Error::StorageError("invalid scheme".to_string())); - } - - if url.has_host() { - return Err(crate::Error::StorageError(format!( - "invalid host: {}", - url.host().unwrap() - ))); - } - - // TODO: expose compression and other parameters as URL parameters, drop new and new_temporary? - if url.path().is_empty() { - Self::new_temporary().map_err(|e| Error::StorageError(e.to_string())) - } else if url.path() == "/" { - Err(crate::Error::StorageError( - "cowardly refusing to open / with sled".to_string(), - )) - } else { - Self::new(url.path().into()).map_err(|e| Error::StorageError(e.to_string())) - } - } - #[instrument(skip(self, digest), fields(directory.digest = %digest))] async fn get(&self, digest: &B3Digest) -> Result, Error> { match self.db.get(digest.as_slice()) { @@ -137,77 +109,3 @@ impl DirectoryService for SledDirectoryService { Box::new(SimplePutter::new(self.clone())) } } - -#[cfg(test)] -mod tests { - use tempfile::TempDir; - - use super::DirectoryService; - use super::SledDirectoryService; - - /// This uses a wrong scheme. - #[test] - fn test_invalid_scheme() { - let url = url::Url::parse("http://foo.example/test").expect("must parse"); - - assert!(SledDirectoryService::from_url(&url).is_err()); - } - - /// This uses the correct scheme, and doesn't specify a path (temporary sled). - #[test] - fn test_valid_scheme_temporary() { - let url = url::Url::parse("sled://").expect("must parse"); - - assert!(SledDirectoryService::from_url(&url).is_ok()); - } - - /// This sets the path to a location that doesn't exist, which should fail (as sled doesn't mkdir -p) - #[test] - fn test_nonexistent_path() { - let tmpdir = TempDir::new().unwrap(); - - let mut url = url::Url::parse("sled://foo.example").expect("must parse"); - url.set_path(tmpdir.path().join("foo").join("bar").to_str().unwrap()); - - assert!(SledDirectoryService::from_url(&url).is_err()); - } - - /// This uses the correct scheme, and specifies / as path (which should fail - // for obvious reasons) - #[test] - fn test_invalid_path_root() { - let url = url::Url::parse("sled:///").expect("must parse"); - - assert!(SledDirectoryService::from_url(&url).is_err()); - } - - /// This uses the correct scheme, and sets a tempdir as location. - #[test] - fn test_valid_scheme_path() { - let tmpdir = TempDir::new().unwrap(); - - let mut url = url::Url::parse("sled://").expect("must parse"); - url.set_path(tmpdir.path().to_str().unwrap()); - - assert!(SledDirectoryService::from_url(&url).is_ok()); - } - - /// This sets a host, rather than a path, which should fail. - #[test] - fn test_invalid_host() { - let url = url::Url::parse("sled://foo.example").expect("must parse"); - - assert!(SledDirectoryService::from_url(&url).is_err()); - } - - /// This sets a host AND a valid path, which should fail - #[test] - fn test_invalid_host_and_path() { - let tmpdir = TempDir::new().unwrap(); - - let mut url = url::Url::parse("sled://foo.example").expect("must parse"); - url.set_path(tmpdir.path().to_str().unwrap()); - - assert!(SledDirectoryService::from_url(&url).is_err()); - } -} -- cgit 1.4.1