[PATCH datacenter-manager v2 0/4] add generic, per-remote (and global) cache for remote API responses

[PATCH datacenter-manager v2 0/4] add generic, per-remote (and global) cache for remote API responses

[Lukas Wagner]

The main intention is to avoid a sprawl of different caching approaches by
establishing a simple, easy to use cache implementation that can be used to
persistently cache API responses from remotes (and derived aggregations).

The `namespaced_cache` module is pretty generic and can be moved to proxmox.git
(maybe in proxmox-shared-cache) once it has sufficiently stabilized.

Changes since the RFC:
  - change storage location to /run/proxmox-datacenter-manager/api-cache
  - change name of pdm_cache to api_cache
  - use freestanding functions in api_cache
  - add async interface
  - minor code style improments
  - improve test coverage and documentation
  - add basic sanity checks for namespaces and keys (e.g. prohibiting ../)
  - add cleanup code for the old remote-updates cachefile

Thanks to Dominik C. and Wolfgang for their reviews of the RFC, highly appreciated!

Changes since v1:
  - The only changes happened in the last two patches:
     - Improved behavior in case the cache entry could not be read,
       mostly to ensure a self-healing behavior
     - Slightly restructured code to avoid some clone calls
     - When requesting the list of updates for a single remote, update the
       global summary cache asynchronously


proxmox-datacenter-manager:

Lukas Wagner (4):
  add persistent, generic, namespaced key-value cache implementation
  add api_cache as a specialized wrapper around the namespaced cache
  api: resources: subscriptions: switch over to api_cache
  remote-updates: switch over to new api_cache

 Cargo.toml                                    |   1 +
 debian/control                                |   1 +
 server/Cargo.toml                             |   4 +
 server/src/api/pve/mod.rs                     |   4 +-
 server/src/api/remotes/updates.rs             |   6 +-
 server/src/api/resources.rs                   |  84 +-
 server/src/api_cache.rs                       | 126 +++
 .../bin/proxmox-datacenter-privileged-api.rs  |   9 +-
 server/src/lib.rs                             |   2 +
 server/src/namespaced_cache.rs                | 742 ++++++++++++++++++
 server/src/remote_updates.rs                  | 120 +--
 11 files changed, 994 insertions(+), 105 deletions(-)
 create mode 100644 server/src/api_cache.rs
 create mode 100644 server/src/namespaced_cache.rs


Summary over all repositories:
  11 files changed, 994 insertions(+), 105 deletions(-)

-- 
Generated by murpp 0.12.0

[PATCH datacenter-manager v2 1/4] add persistent, generic, namespaced key-value cache implementation

[Lukas Wagner]

Namespaces map to directories inside a base directory. Cache contents
are stored as a single JSON-file per key inside the namespace directory.

Value expiry is implemented via a max_age parameter when retrieving
values. Callers might have different requirements to the freshness of
entries, so this seemed a better fit than statically setting an expiry
time when *setting* the value.

The cache offers both a blocking as well as an async interface. Once we
move this implementation to a shared crate, we should probably use
feature flags to gate these.

Signed-off-by: Lukas Wagner <l.wagner@proxmox.com>
---

Notes:
    This module might be well suited to be moved to proxmox.git when it has
    stabilized enough.
    
    Changes since the RFC:
      - In NamespacedCache::new, use Into<PathBuf> instead of AsRef<Path>
        (thx @ Wolfgang)
      - Improve test coverage
      - Basic sanity checks for namespaces and keys, for instance
        prohibiting "../"
      - Fix path generation for cases where there is already a
        file extension in the key (before, it would have been replaced by
        .json)
      - add async interface and rename the old read/write calls to
        read_blocking and write_blocking.
        read/write now return a distinct wrapper type that wraps all
        blocking operations in a `spawn_blocking`

 Cargo.toml                     |   1 +
 debian/control                 |   1 +
 server/Cargo.toml              |   4 +
 server/src/lib.rs              |   1 +
 server/src/namespaced_cache.rs | 742 +++++++++++++++++++++++++++++++++
 5 files changed, 749 insertions(+)
 create mode 100644 server/src/namespaced_cache.rs

diff --git a/Cargo.toml b/Cargo.toml
index 9806a4f0..5cf05b41 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -125,6 +125,7 @@ serde = { version = "1.0", features = ["derive"] }
 serde_cbor = "0.11.1"
 serde_json = "1.0"
 serde_plain = "1"
+tempfile = "3.15"
 thiserror = "1.0"
 tokio = "1.6"
 tracing = "0.1"
diff --git a/debian/control b/debian/control
index 61b9cb36..5955faa1 100644
--- a/debian/control
+++ b/debian/control
@@ -104,6 +104,7 @@ Build-Depends: debhelper-compat (= 13),
                librust-proxmox-syslog-api-1+default-dev,
                librust-proxmox-syslog-api-1+impl-dev,
                librust-proxmox-systemd-1+default-dev,
+               librust-tempfile-3+default-dev (>= 3.15.0),
                librust-proxmox-tfa-6+api-dev,
                librust-proxmox-tfa-6+api-types-dev,
                librust-proxmox-tfa-6+types-dev,
diff --git a/server/Cargo.toml b/server/Cargo.toml
index 3f185bbc..663f21cb 100644
--- a/server/Cargo.toml
+++ b/server/Cargo.toml
@@ -29,6 +29,7 @@ openssl.workspace = true
 percent-encoding.workspace = true
 serde.workspace = true
 serde_json.workspace = true
+thiserror.workspace = true
 tokio = { workspace = true, features = [ "fs", "io-util", "io-std", "macros", "net", "parking_lot", "process", "rt", "rt-multi-thread", "signal", "time" ] }
 tracing.workspace = true
 url.workspace = true
@@ -87,6 +88,9 @@ pdm-search.workspace = true
 pve-api-types = { workspace = true, features = [ "client" ] }
 pbs-api-types.workspace = true
 
+[dev-dependencies]
+tempfile.workspace = true
+
 [lints.rust.unexpected_cfgs]
 level = "warn"
 check-cfg = ['cfg(remote_config, values("faked"))']
diff --git a/server/src/lib.rs b/server/src/lib.rs
index 5ed10d69..0b7642ab 100644
--- a/server/src/lib.rs
+++ b/server/src/lib.rs
@@ -7,6 +7,7 @@ pub mod context;
 pub mod env;
 pub mod jobstate;
 pub mod metric_collection;
+pub mod namespaced_cache;
 pub mod parallel_fetcher;
 pub mod remote_cache;
 pub mod remote_tasks;
diff --git a/server/src/namespaced_cache.rs b/server/src/namespaced_cache.rs
new file mode 100644
index 00000000..9807fc54
--- /dev/null
+++ b/server/src/namespaced_cache.rs
@@ -0,0 +1,742 @@
+//! Generic namespaced cache implementation with optional value expiry.
+//!
+//! ```
+//! use std::time::Duration;
+//!
+//! use proxmox_sys::fs::CreateOptions;
+//! use server::namespaced_cache::NamespacedCache;
+//!
+//! let dir = tempfile::tempdir().unwrap();
+//! let cache = NamespacedCache::new(dir.as_ref(), CreateOptions::new(), CreateOptions::new());
+//! let write_guard = cache.write_blocking("remote-a", Duration::from_secs(1)).unwrap();
+//! write_guard.set("val1", 1).unwrap();
+//! assert_eq!(write_guard.get::<i32>("val1").unwrap().unwrap(), 1);
+//!
+//! ```
+
+use std::fs::File;
+use std::io::ErrorKind;
+use std::path::{Path, PathBuf};
+use std::sync::Arc;
+use std::time::Duration;
+
+use serde::{de::DeserializeOwned, Deserialize, Serialize};
+use tokio::task::JoinError;
+
+use proxmox_sys::fs::CreateOptions;
+
+/// Error type for [`NamespacedCache`].
+#[derive(thiserror::Error, Debug)]
+pub enum CacheError {
+    #[error("IO error: {0}")]
+    Io(#[from] std::io::Error),
+
+    #[error("serialization error: {0}")]
+    Serde(#[from] serde_json::Error),
+
+    #[error("error: {0}")]
+    Other(#[from] anyhow::Error),
+
+    #[error("invalid key: '{0}'")]
+    InvalidKey(String),
+
+    #[error("invalid namespace: '{0}'")]
+    InvalidNamespace(String),
+
+    #[error("join error: {0}")]
+    JoinError(#[from] JoinError),
+}
+
+/// A generic, namespaced cache with optional value expiration.
+pub struct NamespacedCache {
+    base_directory: PathBuf,
+    file_options: CreateOptions,
+    dir_options: CreateOptions,
+}
+
+impl NamespacedCache {
+    /// Create a new cache instance.
+    ///
+    /// Cache entries will be persisted in the provided `base_directory`.
+    /// `dir_options` are the [`CreateOptions`] used the namespace directories, while `file_options`
+    /// are the ones for persisted cache entries.
+    pub fn new<P: Into<PathBuf>>(
+        base_directory: P,
+        dir_options: CreateOptions,
+        file_options: CreateOptions,
+    ) -> Self {
+        Self {
+            base_directory: base_directory.into(),
+            dir_options,
+            file_options,
+        }
+    }
+
+    /// Lock a namespace for writing (blocking interface).
+    ///
+    /// This should *not* be called from async code. Use [`NamespacedCache::write`] instead.
+    pub fn write_blocking(
+        &self,
+        namespace: &str,
+        timeout: Duration,
+    ) -> Result<BlockingWritableCacheNamespace, CacheError> {
+        ensure_valid_namespace(namespace)?;
+        let lock = self.lock_namespace_impl_blocking(namespace, timeout, true)?;
+
+        Ok(BlockingWritableCacheNamespace {
+            inner: WriteableInner {
+                _lock: lock,
+                namespace: namespace.to_string(),
+                base_path: self.base_directory.clone(),
+                dir_options: self.dir_options,
+                file_options: self.file_options,
+            },
+        })
+    }
+
+    /// Lock a namespace for writing (async interface).
+    pub async fn write(
+        &self,
+        namespace: &str,
+        timeout: Duration,
+    ) -> Result<WritableCacheNamespace, CacheError> {
+        ensure_valid_namespace(namespace)?;
+        let lock = self.lock_namespace_impl(namespace, timeout, true).await?;
+
+        Ok(WritableCacheNamespace {
+            inner: Arc::new(WriteableInner {
+                _lock: lock,
+                namespace: namespace.to_string(),
+                base_path: self.base_directory.clone(),
+                dir_options: self.dir_options,
+                file_options: self.file_options,
+            }),
+        })
+    }
+
+    /// Lock a namespace for reading (blocking interface).
+    ///
+    /// This should *not* be called from async code. Use [`NamespacedCache::read`] instead.
+    pub fn read_blocking(
+        &self,
+        namespace: &str,
+        timeout: Duration,
+    ) -> Result<BlockingReadableCacheNamespace, CacheError> {
+        ensure_valid_namespace(namespace)?;
+        let lock = self.lock_namespace_impl_blocking(namespace, timeout, false)?;
+
+        Ok(BlockingReadableCacheNamespace {
+            inner: ReadableInner {
+                _lock: lock,
+                namespace: namespace.to_string(),
+                base_path: self.base_directory.clone(),
+            },
+        })
+    }
+
+    /// Lock a namespace for reading (async interface).
+    pub async fn read(
+        &self,
+        namespace: &str,
+        timeout: Duration,
+    ) -> Result<ReadableCacheNamespace, CacheError> {
+        ensure_valid_namespace(namespace)?;
+        let lock = self.lock_namespace_impl(namespace, timeout, false).await?;
+
+        Ok(ReadableCacheNamespace {
+            inner: Arc::new(ReadableInner {
+                _lock: lock,
+                namespace: namespace.to_string(),
+                base_path: self.base_directory.clone(),
+            }),
+        })
+    }
+
+    async fn lock_namespace_impl(
+        &self,
+        namespace: &str,
+        timeout: Duration,
+        exclusive: bool,
+    ) -> Result<File, CacheError> {
+        let path = get_lockfile(&self.base_directory, namespace);
+        let file_options = self.file_options;
+        let lock = tokio::task::spawn_blocking(move || {
+            proxmox_sys::fs::open_file_locked(&path, timeout, exclusive, file_options)
+        })
+        .await??;
+
+        Ok(lock)
+    }
+
+    fn lock_namespace_impl_blocking(
+        &self,
+        namespace: &str,
+        timeout: Duration,
+        exclusive: bool,
+    ) -> Result<File, CacheError> {
+        let path = get_lockfile(&self.base_directory, namespace);
+        let lock = proxmox_sys::fs::open_file_locked(&path, timeout, exclusive, self.file_options)?;
+
+        Ok(lock)
+    }
+}
+
+/// A readable cache namespace (blocking interface).
+pub struct BlockingReadableCacheNamespace {
+    inner: ReadableInner,
+}
+
+/// A readable cache namespace (async interface).
+pub struct ReadableCacheNamespace {
+    inner: Arc<ReadableInner>,
+}
+
+struct ReadableInner {
+    _lock: File,
+    namespace: String,
+    base_path: PathBuf,
+}
+
+impl BlockingReadableCacheNamespace {
+    /// Read a value from the cache.
+    ///
+    /// # Errors:
+    ///   - The file associated with this key could not be read
+    ///   - The file could not be deserialized (e.g. invalid format)
+    pub fn get<T: Serialize + DeserializeOwned>(&self, key: &str) -> Result<Option<T>, CacheError> {
+        get_impl(&self.inner.base_path, &self.inner.namespace, key, None)
+    }
+
+    /// Read a value from the cache, given a maximum age of the cache entry.
+    ///
+    /// # Errors:
+    ///   - The file associated with this key could not be read
+    ///   - The file could not be deserialized (e.g. invalid format)
+    pub fn get_with_max_age<T: Serialize + DeserializeOwned>(
+        &self,
+        key: &str,
+        max_age: i64,
+    ) -> Result<Option<T>, CacheError> {
+        get_impl(
+            &self.inner.base_path,
+            &self.inner.namespace,
+            key,
+            Some(max_age),
+        )
+    }
+}
+
+impl ReadableCacheNamespace {
+    /// Read a value from the cache.
+    ///
+    /// # Errors:
+    ///   - The file associated with this key could not be read
+    ///   - The file could not be deserialized (e.g. invalid format)
+    pub async fn get<T: Serialize + DeserializeOwned + Send + 'static>(
+        &self,
+        key: &str,
+    ) -> Result<Option<T>, CacheError> {
+        let key = key.to_string();
+        let inner = Arc::clone(&self.inner);
+
+        tokio::task::spawn_blocking(move || {
+            get_impl(&inner.base_path, &inner.namespace, &key, None)
+        })
+        .await?
+    }
+
+    /// Read a value from the cache, given a maximum age of the cache entry.
+    ///
+    /// # Errors:
+    ///   - The file associated with this key could not be read
+    ///   - The file could not be deserialized (e.g. invalid format)
+    pub async fn get_with_max_age<T: Serialize + DeserializeOwned + Send + 'static>(
+        &self,
+        key: &str,
+        max_age: i64,
+    ) -> Result<Option<T>, CacheError> {
+        let key = key.to_string();
+        let inner = Arc::clone(&self.inner);
+
+        tokio::task::spawn_blocking(move || {
+            get_impl(&inner.base_path, &inner.namespace, &key, Some(max_age))
+        })
+        .await?
+    }
+}
+
+/// A writable cache namespace (blocking interface).
+pub struct BlockingWritableCacheNamespace {
+    inner: WriteableInner,
+}
+
+/// A writable cache namespace (async interface).
+pub struct WritableCacheNamespace {
+    inner: Arc<WriteableInner>,
+}
+
+struct WriteableInner {
+    _lock: File,
+    namespace: String,
+    base_path: PathBuf,
+    dir_options: CreateOptions,
+    file_options: CreateOptions,
+}
+
+impl BlockingWritableCacheNamespace {
+    /// Remote a cache entry.
+    ///
+    /// This returns `Ok(())` if the key does not exist.
+    ///
+    /// # Errors:
+    ///   - The file could not be deleted due to insufficient privileges.
+    pub fn remove(&self, key: &str) -> Result<(), CacheError> {
+        remove_impl(&self.inner, key)
+    }
+
+    /// Set a cache entry.
+    ///
+    /// # Errors
+    ///   - `value` could not be serialized
+    ///   - The namespace directory could not be created
+    ///   - The cache file could not be written to or atomically replaced
+    pub fn set<T: Serialize + DeserializeOwned>(
+        &self,
+        key: &str,
+        value: T,
+    ) -> Result<(), CacheError> {
+        set_impl(&self.inner, key, value, proxmox_time::epoch_i64())
+    }
+
+    /// Set a cache entry with an explicitly provided timestamp.
+    ///
+    /// # Errors
+    ///   - `value` could not be serialized
+    ///   - The namespace directory could not be created
+    ///   - The cache file could not be written to or atomically replaced
+    pub fn set_with_timestamp<T: Serialize + DeserializeOwned>(
+        &self,
+        key: &str,
+        value: T,
+        timestamp: i64,
+    ) -> Result<(), CacheError> {
+        set_impl(&self.inner, key, value, timestamp)
+    }
+
+    /// Read a value from the cache.
+    ///
+    /// # Errors:
+    ///   - The file associated with this key could not be read
+    ///   - The file could not be deserialized (e.g. invalid format)
+    pub fn get<T: Serialize + DeserializeOwned>(&self, key: &str) -> Result<Option<T>, CacheError> {
+        get_impl(&self.inner.base_path, &self.inner.namespace, key, None)
+    }
+
+    /// Read a value from the cache, given a maximum age of the cache entry.
+    ///
+    /// # Errors:
+    ///   - The file associated with this key could not be read
+    ///   - The file could not be deserialized (e.g. invalid format)
+    pub fn get_with_max_age<T: Serialize + DeserializeOwned>(
+        &self,
+        key: &str,
+        max_age: i64,
+    ) -> Result<Option<T>, CacheError> {
+        get_impl(
+            &self.inner.base_path,
+            &self.inner.namespace,
+            key,
+            Some(max_age),
+        )
+    }
+}
+
+impl WritableCacheNamespace {
+    /// Remote a cache entry.
+    ///
+    /// This returns `Ok(())` if the key does not exist.
+    ///
+    /// # Errors:
+    ///   - The file could not be deleted due to insufficient privileges.
+    pub async fn remove(&self, key: &str) -> Result<(), CacheError> {
+        let inner = Arc::clone(&self.inner);
+        let key = key.to_string();
+
+        tokio::task::spawn_blocking(move || remove_impl(&inner, &key)).await?
+    }
+
+    /// Set a cache entry.
+    ///
+    /// # Errors
+    ///   - `value` could not be serialized
+    ///   - The namespace directory could not be created
+    ///   - The cache file could not be written to or atomically replaced
+    pub async fn set<T: Serialize + DeserializeOwned + Send + 'static>(
+        &self,
+        key: &str,
+        value: T,
+    ) -> Result<(), CacheError> {
+        let inner = Arc::clone(&self.inner);
+        let key = key.to_string();
+
+        tokio::task::spawn_blocking(move || {
+            set_impl(&inner, &key, value, proxmox_time::epoch_i64())
+        })
+        .await?
+    }
+
+    /// Set a cache entry with an explicitly provided timestamp.
+    ///
+    /// # Errors
+    ///   - `value` could not be serialized
+    ///   - The namespace directory could not be created
+    ///   - The cache file could not be written to or atomically replaced
+    pub async fn set_with_timestamp<T: Serialize + DeserializeOwned + Send + 'static>(
+        &self,
+        key: &str,
+        value: T,
+        timestamp: i64,
+    ) -> Result<(), CacheError> {
+        let inner = Arc::clone(&self.inner);
+        let key = key.to_string();
+        tokio::task::spawn_blocking(move || set_impl(&inner, &key, value, timestamp)).await?
+    }
+
+    /// Read a value from the cache.
+    ///
+    /// # Errors:
+    ///   - The file associated with this key could not be read
+    ///   - The file could not be deserialized (e.g. invalid format)
+    pub async fn get<T: Serialize + DeserializeOwned + Send + 'static>(
+        &self,
+        key: &str,
+    ) -> Result<Option<T>, CacheError> {
+        let key = key.to_string();
+        let inner = Arc::clone(&self.inner);
+
+        tokio::task::spawn_blocking(move || {
+            get_impl(&inner.base_path, &inner.namespace, &key, None)
+        })
+        .await?
+    }
+
+    /// Read a value from the cache, given a maximum age of the cache entry.
+    ///
+    /// # Errors:
+    ///   - The file associated with this key could not be read
+    ///   - The file could not be deserialized (e.g. invalid format)
+    pub async fn get_with_max_age<T: Serialize + DeserializeOwned + Send + 'static>(
+        &self,
+        key: &str,
+        max_age: i64,
+    ) -> Result<Option<T>, CacheError> {
+        let key = key.to_string();
+        let inner = Arc::clone(&self.inner);
+
+        tokio::task::spawn_blocking(move || {
+            get_impl(&inner.base_path, &inner.namespace, &key, Some(max_age))
+        })
+        .await?
+    }
+}
+
+#[derive(Serialize, Deserialize, Debug)]
+struct CacheEntry<T> {
+    timestamp: i64,
+    value: T,
+}
+
+impl<T> CacheEntry<T> {
+    fn is_expired(&self, now: i64, max_age: i64) -> bool {
+        if max_age == 0 {
+            return true;
+        }
+
+        let diff = now - self.timestamp;
+        diff >= max_age || diff < 0
+    }
+}
+
+fn get_impl<T: Serialize + DeserializeOwned>(
+    base: &Path,
+    namespace: &str,
+    key: &str,
+    max_age: Option<i64>,
+) -> Result<Option<T>, CacheError> {
+    // Namespace should already be verified at this point, no point in checking it again.
+    ensure_valid_key(key)?;
+
+    let path = get_path(base, namespace, key);
+    let content = proxmox_sys::fs::file_read_optional_string(path)?;
+
+    if let Some(content) = content {
+        let val = serde_json::from_str::<CacheEntry<T>>(&content)?;
+
+        if let Some(max_age) = max_age {
+            if val.is_expired(proxmox_time::epoch_i64(), max_age) {
+                return Ok(None);
+            }
+        }
+        return Ok(Some(val.value));
+    }
+
+    Ok(None)
+}
+
+fn set_impl<T: Serialize + DeserializeOwned>(
+    inner: &WriteableInner,
+    key: &str,
+    value: T,
+    timestamp: i64,
+) -> Result<(), CacheError> {
+    ensure_valid_key(key)?;
+    let path = get_path(&inner.base_path, &inner.namespace, key);
+
+    proxmox_sys::fs::create_path(
+        path.parent().unwrap(),
+        Some(inner.dir_options),
+        Some(inner.dir_options),
+    )?;
+
+    let entry = CacheEntry { timestamp, value };
+
+    let data = serde_json::to_vec(&entry)?;
+    proxmox_sys::fs::replace_file(path, &data, inner.file_options, true)?;
+
+    Ok(())
+}
+
+fn remove_impl(inner: &WriteableInner, key: &str) -> Result<(), CacheError> {
+    ensure_valid_key(key)?;
+    let path = get_path(&inner.base_path, &inner.namespace, key);
+
+    if let Err(err) = std::fs::remove_file(path) {
+        if err.kind() == ErrorKind::NotFound {
+            return Ok(());
+        }
+
+        return Err(err.into());
+    }
+
+    Ok(())
+}
+
+fn get_path(base: &Path, namespace: &str, key: &str) -> PathBuf {
+    let path = base.join(namespace).join(format!("{key}.json"));
+    path
+}
+
+fn get_lockfile(base: &Path, namespace: &str) -> PathBuf {
+    let path = base.join(format!(".{namespace}.lock"));
+    path
+}
+
+// Make sure that an identifier is safe to use as a cache key.
+//
+// At the moment, this only checks for '../', as to ensure that the base directory
+// is not escaped.
+fn ensure_valid_key(key: &str) -> Result<(), CacheError> {
+    if key.contains("../") {
+        return Err(CacheError::InvalidKey(key.into()));
+    }
+
+    Ok(())
+}
+
+// Make sure that an identifier is safe to use as a namespace.
+//
+// At the moment, this ensures that there is no '../' in the namespace, as well as that the
+// identifier does not start with '/'.
+fn ensure_valid_namespace(namespace: &str) -> Result<(), CacheError> {
+    if namespace.contains("../") || namespace.starts_with("/") {
+        return Err(CacheError::InvalidNamespace(namespace.into()));
+    }
+
+    Ok(())
+}
+
+#[cfg(test)]
+mod tests {
+    use tempfile::TempDir;
+
+    use super::*;
+
+    const TIMEOUT: Duration = Duration::from_secs(1);
+
+    fn make_cache() -> (TempDir, NamespacedCache) {
+        let dir = tempfile::tempdir().unwrap();
+
+        let cache = NamespacedCache::new(dir.as_ref(), CreateOptions::new(), CreateOptions::new());
+
+        (dir, cache)
+    }
+
+    #[test]
+    fn test_cache() {
+        let (_dir, cache) = make_cache();
+
+        let write_guard = cache.write_blocking("remote-a", TIMEOUT).unwrap();
+        write_guard.set("val1", 1).unwrap();
+        write_guard.set("val2", 1).unwrap();
+
+        assert_eq!(write_guard.get::<i32>("val1").unwrap().unwrap(), 1);
+
+        write_guard.remove("val1").unwrap();
+        assert!(write_guard.get::<String>("val1").unwrap().is_none());
+
+        drop(write_guard);
+
+        let read_guard = cache.read_blocking("remote-a", TIMEOUT).unwrap();
+
+        assert_eq!(read_guard.get::<i32>("val2").unwrap().unwrap(), 1);
+    }
+
+    #[test]
+    fn test_remove_nonexisting() {
+        let (_dir, cache) = make_cache();
+
+        let a = cache.write_blocking("remote-a", TIMEOUT).unwrap();
+
+        // Deleting a key that does not exist is okay and should not error.
+        assert!(a.remove("val").is_ok());
+    }
+
+    #[test]
+    fn test_remove_failure() {
+        let (dir, cache) = make_cache();
+
+        let a = cache.write_blocking("remote-a", TIMEOUT).unwrap();
+        // Triggering a general failure by generating a directory that conflicts with the cache key
+        std::fs::create_dir_all(dir.path().join("remote-a").join("val.json")).unwrap();
+        assert!(a.remove("val").is_err());
+    }
+
+    #[test]
+    fn test_get_with_max_age() {
+        let (_dir, cache) = make_cache();
+
+        let write_guard = cache.write_blocking("remote-a", TIMEOUT).unwrap();
+
+        let now = proxmox_time::epoch_i64();
+
+        write_guard
+            .set_with_timestamp("somekey", 1, now - 1000)
+            .unwrap();
+
+        assert!(write_guard
+            .get_with_max_age::<i32>("somekey", 999)
+            .unwrap()
+            .is_none());
+
+        drop(write_guard);
+        let read_guard = cache.read_blocking("remote-a", TIMEOUT).unwrap();
+        assert!(read_guard
+            .get_with_max_age::<i32>("somekey", 999)
+            .unwrap()
+            .is_none());
+    }
+
+    #[test]
+    fn test_expiration() {
+        let entry = CacheEntry {
+            value: (),
+            timestamp: 1000,
+        };
+
+        assert!(!entry.is_expired(1000, 100));
+        assert!(!entry.is_expired(1099, 100));
+        assert!(entry.is_expired(1100, 100));
+        assert!(entry.is_expired(1101, 100));
+
+        // if max-age is 0, the entry is never fresh
+        assert!(entry.is_expired(1000, 0));
+    }
+
+    #[test]
+    fn test_invalid_namespaces() {
+        let (_dir, cache) = make_cache();
+
+        for id in [
+            "../remote-a",
+            "remote-a/../something",
+            "remote-a/../",
+            "../",
+        ] {
+            assert!(matches!(
+                cache.write_blocking(id, TIMEOUT),
+                Err(CacheError::InvalidNamespace(_))
+            ));
+            assert!(matches!(
+                cache.read_blocking(id, TIMEOUT),
+                Err(CacheError::InvalidNamespace(_))
+            ));
+        }
+    }
+
+    #[test]
+    fn test_invalid_keys() {
+        let (_dir, cache) = make_cache();
+
+        let write_guard = cache.write_blocking("remote-a", TIMEOUT).unwrap();
+        let read_guard = cache.write_blocking("remote-b", TIMEOUT).unwrap();
+
+        for id in ["../somekey", "somekey/../something", "somekey/../", "../"] {
+            assert!(matches!(
+                write_guard.set(id, ()),
+                Err(CacheError::InvalidKey(_))
+            ));
+            assert!(matches!(
+                write_guard.get::<()>(id),
+                Err(CacheError::InvalidKey(_))
+            ));
+            assert!(matches!(
+                write_guard.get_with_max_age::<()>(id, 1000),
+                Err(CacheError::InvalidKey(_))
+            ));
+            assert!(matches!(
+                write_guard.remove(id),
+                Err(CacheError::InvalidKey(_))
+            ));
+            assert!(matches!(
+                read_guard.get::<()>(id),
+                Err(CacheError::InvalidKey(_))
+            ));
+            assert!(matches!(
+                read_guard.get_with_max_age::<()>(id, 1000),
+                Err(CacheError::InvalidKey(_))
+            ));
+        }
+    }
+
+    #[tokio::test]
+    async fn test_async() {
+        let (_dir, cache) = make_cache();
+
+        let lock = cache.write("some-remote", TIMEOUT).await.unwrap();
+
+        lock.set("somekey", 1234).await.unwrap();
+        assert_eq!(lock.get::<i32>("somekey").await.unwrap(), Some(1234));
+        lock.remove("somekey").await.unwrap();
+        assert!(lock.get::<i32>("somekey").await.unwrap().is_none());
+
+        lock.set_with_timestamp("somekey", 1234, proxmox_time::epoch_i64() - 1000)
+            .await
+            .unwrap();
+
+        assert!(lock
+            .get_with_max_age::<i32>("somekey", 900)
+            .await
+            .unwrap()
+            .is_none());
+
+        drop(lock);
+
+        let lock = cache.read("some-remote", TIMEOUT).await.unwrap();
+        assert_eq!(lock.get::<i32>("somekey").await.unwrap(), Some(1234));
+        assert!(lock
+            .get_with_max_age::<i32>("somekey", 900)
+            .await
+            .unwrap()
+            .is_none());
+    }
+}
-- 
2.47.3

[PATCH datacenter-manager v2 2/4] add api_cache as a specialized wrapper around the namespaced cache

[Lukas Wagner]

This is a thin wrapper around the previously introduced namespaced
key-value cache, but introducing PDM-specific concepts.

Instead of the higher-level read/write methods for locking a namespace,
this wrapper provides {read,write}_remote and {read,write}_global, for
accessing remote-specific and globally cached values. The
cache-namespaces are 'global' and 'remote-<remote-name>'.

The base directory for the cache is
/run/proxmox-datacenter-manager/api-cache

Signed-off-by: Lukas Wagner <l.wagner@proxmox.com>
---

Notes:
    Changes since the RFC:
    
    - Changed the cache location to
      /run/proxmox-datacenter-manager/api-cache
    
     - Renamed the module from pdm_cache to api_cache
    
     - removed the instance() accessor and use freestanding functions
       instead. While the instance-based approach might come in useful at a
       later stage when dependency injection is more widely used in our
       stack, it's a premature step that can be reintroduced later without
       big troubles
    
     - Some minor code cleanup

 server/src/api_cache.rs                       | 126 ++++++++++++++++++
 .../bin/proxmox-datacenter-privileged-api.rs  |   9 +-
 server/src/lib.rs                             |   1 +
 3 files changed, 135 insertions(+), 1 deletion(-)
 create mode 100644 server/src/api_cache.rs

diff --git a/server/src/api_cache.rs b/server/src/api_cache.rs
new file mode 100644
index 00000000..b20f0535
--- /dev/null
+++ b/server/src/api_cache.rs
@@ -0,0 +1,126 @@
+//! Cache for API responses from remotes.
+//!
+//! This cache is namespaced by remote and also offers a 'global' namespace
+//! for values that are valid across remotes (e.g. aggregations).
+//!
+//! The cache has both, a blocking, as well as an async interface that can be used to get or set
+//! cache entries.
+//!
+//! A namespace (so, either the remote one's or the global one) must be locked before it can be
+//! accessed. All locking functions use a 10 second timeout while waiting for the lock.
+//!
+//! ## Blocking interface
+//!   - [`read_remote_blocking`]
+//!   - [`write_remote_blocking`]
+//!   - [`read_global_blocking`]
+//!   - [`write_global_blocking`]
+//!
+//! These functions return [`BlockingReadableCacheNamespace`] and [`BlockingWritableCacheNamespace`], respectively.
+//! Both only offer blocking operations for interacting with the entries of the locked namespace.
+//!
+//! ## `async` interface
+//!   - [`read_remote`]
+//!   - [`write_remote`]
+//!   - [`read_global`]
+//!   - [`write_global`]
+//!
+//! These functions return [`ReadableCacheNamespace`] and [`WritableCacheNamespace`], respectively.
+//! Both offer an async wrapper for interacting with the entries of the locked namespace.
+//!
+//! ```no_run
+//! use server::api_cache;
+//!
+//! #[derive(serde::Serialize, serde::Deserialize)]
+//! struct CacheableData {
+//!     id: String,
+//! }
+//!
+//! let data = CacheableData {
+//!     id: "some-id".to_string(),
+//! };
+//!
+//! // Lock the cache namespace for 'some-remote' for write access
+//! let lock = api_cache::write_remote_blocking("some-remote").unwrap();
+//!
+//! // Set some value (must be Serialize + Deserialize)
+//! lock.set("some-key", data).unwrap();
+//!
+//! // Retrieve the cached entry
+//! let data: Option<CacheableData> = lock.get("some-key").unwrap();
+//!
+//! // Remove the cached entry
+//! lock.remove("some-key").unwrap();
+//!
+//! ```
+
+use std::path::PathBuf;
+use std::sync::LazyLock;
+use std::time::Duration;
+
+use nix::sys::stat::Mode;
+
+use crate::namespaced_cache::{
+    BlockingReadableCacheNamespace, BlockingWritableCacheNamespace, CacheError, NamespacedCache,
+    ReadableCacheNamespace, WritableCacheNamespace,
+};
+
+/// Path at which API responses are cached.
+pub const PDM_API_CACHE_PATH: &str = concat!(pdm_buildcfg::PDM_RUN_DIR_M!(), "/api-cache");
+
+const GLOBAL_NAMESPACE: &str = "global";
+const LOCK_TIMEOUT: Duration = Duration::from_secs(10);
+
+static CACHE: LazyLock<NamespacedCache> = LazyLock::new(|| {
+    let file_options = proxmox_product_config::default_create_options();
+    let dir_options = file_options.perm(Mode::from_bits_truncate(0o750));
+
+    NamespacedCache::new(PathBuf::from(PDM_API_CACHE_PATH), dir_options, file_options)
+});
+
+fn format_remote_namespace(remote: &str) -> String {
+    format!("remote-{remote}")
+}
+
+/// Lock the cache for reading remote-specific data (blocking interface).
+pub fn read_remote_blocking(remote: &str) -> Result<BlockingReadableCacheNamespace, CacheError> {
+    CACHE.read_blocking(&format_remote_namespace(remote), LOCK_TIMEOUT)
+}
+
+/// Lock the cache for writing remote-specific data (blocking interface).
+pub fn write_remote_blocking(remote: &str) -> Result<BlockingWritableCacheNamespace, CacheError> {
+    CACHE.write_blocking(&format_remote_namespace(remote), LOCK_TIMEOUT)
+}
+
+/// Lock the cache for reading global data (blocking interface).
+pub fn read_global_blocking() -> Result<BlockingReadableCacheNamespace, CacheError> {
+    CACHE.read_blocking(GLOBAL_NAMESPACE, LOCK_TIMEOUT)
+}
+
+/// Lock the cache for writing global data (blocking interface).
+pub fn write_global_blocking() -> Result<BlockingWritableCacheNamespace, CacheError> {
+    CACHE.write_blocking(GLOBAL_NAMESPACE, LOCK_TIMEOUT)
+}
+
+/// Lock the cache for reading remote-specific data (async interface).
+pub async fn read_remote(remote: &str) -> Result<ReadableCacheNamespace, CacheError> {
+    CACHE
+        .read(&format_remote_namespace(remote), LOCK_TIMEOUT)
+        .await
+}
+
+/// Lock the cache for writing remote-specific data (async interface).
+pub async fn write_remote(remote: &str) -> Result<WritableCacheNamespace, CacheError> {
+    CACHE
+        .write(&format_remote_namespace(remote), LOCK_TIMEOUT)
+        .await
+}
+
+/// Lock the cache for reading global data (async interface).
+pub async fn read_global() -> Result<ReadableCacheNamespace, CacheError> {
+    CACHE.read(GLOBAL_NAMESPACE, LOCK_TIMEOUT).await
+}
+
+/// Lock the cache for writing global data (async interface).
+pub async fn write_global() -> Result<WritableCacheNamespace, CacheError> {
+    CACHE.write(GLOBAL_NAMESPACE, LOCK_TIMEOUT).await
+}
diff --git a/server/src/bin/proxmox-datacenter-privileged-api.rs b/server/src/bin/proxmox-datacenter-privileged-api.rs
index 6b490f2b..a3c448cf 100644
--- a/server/src/bin/proxmox-datacenter-privileged-api.rs
+++ b/server/src/bin/proxmox-datacenter-privileged-api.rs
@@ -14,7 +14,7 @@ use proxmox_rest_server::{ApiConfig, RestServer};
 use proxmox_router::RpcEnvironmentType;
 use proxmox_sys::fs::CreateOptions;
 
-use server::auth;
+use server::{api_cache, auth};
 
 use pdm_buildcfg::configdir;
 
@@ -102,6 +102,13 @@ fn create_directories() -> Result<(), Error> {
         0o755,
     )?;
 
+    pdm_config::setup::mkdir_perms(
+        api_cache::PDM_API_CACHE_PATH,
+        api_user.uid,
+        api_user.gid,
+        0o755,
+    )?;
+
     server::jobstate::create_jobstate_dir()?;
 
     Ok(())
diff --git a/server/src/lib.rs b/server/src/lib.rs
index 0b7642ab..89ab3035 100644
--- a/server/src/lib.rs
+++ b/server/src/lib.rs
@@ -2,6 +2,7 @@
 
 pub mod acl;
 pub mod api;
+pub mod api_cache;
 pub mod auth;
 pub mod context;
 pub mod env;
-- 
2.47.3

[PATCH datacenter-manager v2 3/4] api: resources: subscriptions: switch over to api_cache

[Lukas Wagner]

Instead of storing the subscription state in a module-level HashMap, use
the newly introduced api_cache module to cache the status.

Signed-off-by: Lukas Wagner <l.wagner@proxmox.com>
---

Notes:
    Changes since the RFC:
      - Use new async cache interface
    
    Changes since v1:
      - Make cache self-healing in case an entry could not be read or
        deserialized. This ensure resiliency when the cached data type
        changes unexpectedly. In this case, we just log an error and
        return None

 server/src/api/resources.rs | 84 +++++++++++++++----------------------
 1 file changed, 33 insertions(+), 51 deletions(-)

diff --git a/server/src/api/resources.rs b/server/src/api/resources.rs
index 50315b11..c817f180 100644
--- a/server/src/api/resources.rs
+++ b/server/src/api/resources.rs
@@ -30,9 +30,10 @@ use proxmox_schema::{api, parse_boolean};
 use proxmox_sortable_macro::sortable;
 use proxmox_subscription::SubscriptionStatus;
 use pve_api_types::{ClusterResource, ClusterResourceNetworkType, ClusterResourceType};
+use serde::{Deserialize, Serialize};
 
 use crate::metric_collection::top_entities;
-use crate::{connection, views};
+use crate::{api_cache, connection, views};
 
 pub const ROUTER: Router = Router::new()
     .get(&list_subdirs_api_method!(SUBDIRS))
@@ -798,15 +799,11 @@ async fn get_top_entities(
     Ok(res)
 }
 
-#[derive(Clone)]
+#[derive(Clone, Serialize, Deserialize)]
 struct CachedSubscriptionState {
     node_info: HashMap<String, Option<NodeSubscriptionInfo>>,
-    timestamp: i64,
 }
 
-static SUBSCRIPTION_CACHE: LazyLock<RwLock<HashMap<String, CachedSubscriptionState>>> =
-    LazyLock::new(|| RwLock::new(HashMap::new()));
-
 /// Get the subscription state for a given remote.
 ///
 /// If recent enough cached data is available, it is returned
@@ -815,66 +812,51 @@ pub async fn get_subscription_info_for_remote(
     remote: &Remote,
     max_age: u64,
 ) -> Result<HashMap<String, Option<NodeSubscriptionInfo>>, Error> {
-    if let Some(cached_subscription) = get_cached_subscription_info(&remote.id, max_age) {
+    if let Some(cached_subscription) =
+        get_cached_subscription_info(remote.id.clone(), max_age).await?
+    {
         Ok(cached_subscription.node_info)
     } else {
         let node_info = fetch_remote_subscription_info(remote).await?;
-        let now = proxmox_time::epoch_i64();
-        update_cached_subscription_info(&remote.id, &node_info, now);
+        update_cached_subscription_info(remote.id.clone(), node_info.clone()).await?;
         Ok(node_info)
     }
 }
 
-fn get_cached_subscription_info(remote: &str, max_age: u64) -> Option<CachedSubscriptionState> {
-    let cache = SUBSCRIPTION_CACHE
-        .read()
-        .expect("subscription mutex poisoned");
+const SUBSCRIPTION_STATE_CACHE_KEY: &str = "subscription-state";
 
-    if max_age == 0 {
-        return None;
-    }
-    if let Some(cached_subscription) = cache.get(remote) {
-        let now = proxmox_time::epoch_i64();
-        let diff = now - cached_subscription.timestamp;
+async fn get_cached_subscription_info(
+    remote: String,
+    max_age: u64,
+) -> Result<Option<CachedSubscriptionState>, Error> {
+    let cache = api_cache::read_remote(&remote).await?;
+    let subscription_state = cache
+        .get_with_max_age(SUBSCRIPTION_STATE_CACHE_KEY, max_age as i64)
+        .await
+        .inspect_err(|err| log::error!("could not read subscription-state from API cache: {err}"))
+        .ok()
+        .flatten();
 
-        if diff >= max_age as i64 || diff < 0 {
-            // value is too old or from the future
-            None
-        } else {
-            Some(cached_subscription.clone())
-        }
-    } else {
-        None
-    }
+    Ok(subscription_state)
 }
 
 /// Update cached subscription data.
 ///
 /// If the cache already contains more recent data we don't insert the passed resources.
-fn update_cached_subscription_info(
-    remote: &str,
-    node_info: &HashMap<String, Option<NodeSubscriptionInfo>>,
-    now: i64,
-) {
-    // there is no good way to recover from this, so panicking should be fine
-    let mut cache = SUBSCRIPTION_CACHE
-        .write()
-        .expect("subscription mutex poisoned");
+async fn update_cached_subscription_info(
+    remote: String,
+    node_info: HashMap<String, Option<NodeSubscriptionInfo>>,
+) -> Result<(), Error> {
+    let cache = api_cache::write_remote(&remote).await?;
 
-    if let Some(cached_resource) = cache.get(remote) {
-        // skip updating if the data is new enough
-        if cached_resource.timestamp >= now {
-            return;
-        }
-    }
-
-    cache.insert(
-        remote.into(),
-        CachedSubscriptionState {
-            node_info: node_info.clone(),
-            timestamp: now,
-        },
-    );
+    Ok(cache
+        .set(
+            SUBSCRIPTION_STATE_CACHE_KEY,
+            CachedSubscriptionState {
+                node_info: node_info,
+            },
+        )
+        .await?)
 }
 
 /// Maps a list of node subscription infos into a single [`RemoteSubscriptionState`]
-- 
2.47.3

[PATCH datacenter-manager v2 4/4] remote-updates: switch over to new api_cache

[Lukas Wagner]

Use the new, centralized API cache for caching remote update summaries.
Add some cleanup logic to remove the old cachefile, which can be removed
at some point in the future.

Signed-off-by: Lukas Wagner <l.wagner@proxmox.com>
---

Notes:
    Changes since the RFC:
      - use new async cache interface
      - clean up old cachefile automatically
    
    Changes since v1:
      - Make cache self-healing in case an entry could not be read or
        deserialized. This ensure resiliency when the cached data type
        changes unexpectedly. In this case, we just log an error and
        return a default, empty summary.
      - Avoid some .clone() calls
      - When requesting the list of updates for a single remote, update the
        cache asynchronously

 server/src/api/pve/mod.rs         |   4 +-
 server/src/api/remotes/updates.rs |   6 +-
 server/src/remote_updates.rs      | 120 ++++++++++++++++++------------
 3 files changed, 77 insertions(+), 53 deletions(-)

diff --git a/server/src/api/pve/mod.rs b/server/src/api/pve/mod.rs
index 20892f38..649ab624 100644
--- a/server/src/api/pve/mod.rs
+++ b/server/src/api/pve/mod.rs
@@ -588,8 +588,8 @@ pub async fn get_options(remote: String) -> Result<serde_json::Value, Error> {
     },
 )]
 /// Return the cached update information about a remote.
-pub fn get_updates(remote: String) -> Result<RemoteUpdateSummary, Error> {
-    let update_summary = get_available_updates_for_remote(&remote)?;
+pub async fn get_updates(remote: String) -> Result<RemoteUpdateSummary, Error> {
+    let update_summary = get_available_updates_for_remote(&remote).await?;
 
     Ok(update_summary)
 }
diff --git a/server/src/api/remotes/updates.rs b/server/src/api/remotes/updates.rs
index 365ffc19..ea46ba0d 100644
--- a/server/src/api/remotes/updates.rs
+++ b/server/src/api/remotes/updates.rs
@@ -42,7 +42,7 @@ const SUBDIRS: SubdirMap = &sorted!([
     returns: { type: UpdateSummary }
 )]
 /// Return available update summary for managed remote nodes.
-pub fn update_summary(rpcenv: &mut dyn RpcEnvironment) -> Result<UpdateSummary, Error> {
+pub async fn update_summary(rpcenv: &mut dyn RpcEnvironment) -> Result<UpdateSummary, Error> {
     let auth_id = rpcenv
         .get_auth_id()
         .context("no authid available")?
@@ -53,7 +53,7 @@ pub fn update_summary(rpcenv: &mut dyn RpcEnvironment) -> Result<UpdateSummary,
         http_bail!(FORBIDDEN, "user has no access to resources");
     }
 
-    let mut update_summary = remote_updates::get_available_updates_summary()?;
+    let mut update_summary = remote_updates::get_available_updates_summary().await?;
 
     update_summary.remotes.retain(|remote_name, _| {
         user_info
@@ -136,7 +136,7 @@ async fn apt_update_available(remote: String, node: String) -> Result<Vec<APTUpd
     let (config, _digest) = pdm_config::remotes::config()?;
     let remote = get_remote(&config, &remote)?;
 
-    let updates = remote_updates::list_available_updates(remote.clone(), &node).await?;
+    let updates = remote_updates::list_available_updates(remote.clone(), node).await?;
 
     Ok(updates)
 }
diff --git a/server/src/remote_updates.rs b/server/src/remote_updates.rs
index 7aaacc46..824a12a6 100644
--- a/server/src/remote_updates.rs
+++ b/server/src/remote_updates.rs
@@ -1,6 +1,3 @@
-use std::fs::File;
-use std::io::ErrorKind;
-
 use anyhow::{bail, Error};
 use serde::{Deserialize, Serialize};
 
@@ -12,12 +9,13 @@ use pdm_api_types::remote_updates::{
 };
 use pdm_api_types::remotes::{Remote, RemoteType};
 use pdm_api_types::RemoteUpid;
-use pdm_buildcfg::PDM_CACHE_DIR_M;
 
-use crate::connection;
 use crate::parallel_fetcher::ParallelFetcher;
+use crate::{api_cache, connection};
 
-pub const UPDATE_CACHE: &str = concat!(PDM_CACHE_DIR_M!(), "/remote-updates.json");
+const OLD_CACHEFILE: &str = concat!(pdm_buildcfg::PDM_CACHE_DIR_M!(), "/remote-updates.json");
+
+const UPDATE_SUMMARY_CACHE_KEY: &str = "remote-updates";
 
 #[derive(Clone, Default, Debug, Deserialize, Serialize)]
 #[serde(rename_all = "kebab-case")]
@@ -28,14 +26,14 @@ struct NodeUpdateInfo {
     repository_status: ProductRepositoryStatus,
 }
 
-impl From<NodeUpdateInfo> for NodeUpdateSummary {
-    fn from(value: NodeUpdateInfo) -> Self {
+impl From<&NodeUpdateInfo> for NodeUpdateSummary {
+    fn from(value: &NodeUpdateInfo) -> Self {
         Self {
             number_of_updates: value.updates.len() as u32,
             last_refresh: value.last_refresh,
             status: NodeUpdateStatus::Success,
             status_message: None,
-            versions: value.versions,
+            versions: value.versions.clone(),
             repository_status: value.repository_status,
         }
     }
@@ -44,11 +42,20 @@ impl From<NodeUpdateInfo> for NodeUpdateSummary {
 /// Return a list of available updates for a given remote node.
 pub async fn list_available_updates(
     remote: Remote,
-    node: &str,
+    node: String,
 ) -> Result<Vec<APTUpdateInfo>, Error> {
-    let updates = fetch_available_updates((), remote.clone(), node.to_string()).await?;
+    let updates = fetch_available_updates((), remote.clone(), node.clone()).await?;
 
-    update_cached_summary_for_node(remote, node.into(), updates.clone().into()).await?;
+    let summary = (&updates).into();
+
+    // Update cache entry asynchronously, no need to wait for it.
+    tokio::task::spawn({
+        async move {
+            if let Err(err) = update_cached_summary_for_node(remote, node, summary).await {
+                log::error!("could not update 'remote-updates' API cache entry: {err}");
+            }
+        }
+    });
 
     Ok(updates.updates)
 }
@@ -106,10 +113,10 @@ pub async fn get_changelog(remote: &Remote, node: &str, package: String) -> Resu
 }
 
 /// Get update summary for all managed remotes.
-pub fn get_available_updates_summary() -> Result<UpdateSummary, Error> {
+pub async fn get_available_updates_summary() -> Result<UpdateSummary, Error> {
     let (config, _digest) = pdm_config::remotes::config()?;
 
-    let cache_content = get_cached_summary_or_default()?;
+    let cache_content = get_cached_summary_or_default().await?;
 
     let mut summary = UpdateSummary::default();
 
@@ -137,11 +144,11 @@ pub fn get_available_updates_summary() -> Result<UpdateSummary, Error> {
 }
 
 /// Return cached update information from specific remote
-pub fn get_available_updates_for_remote(remote: &str) -> Result<RemoteUpdateSummary, Error> {
+pub async fn get_available_updates_for_remote(remote: &str) -> Result<RemoteUpdateSummary, Error> {
     let (config, _digest) = pdm_config::remotes::config()?;
 
     if let Some(remote) = config.get(remote) {
-        let cache_content = get_cached_summary_or_default()?;
+        let cache_content = get_cached_summary_or_default().await?;
         Ok(cache_content
             .remotes
             .get(&remote.id)
@@ -156,22 +163,24 @@ pub fn get_available_updates_for_remote(remote: &str) -> Result<RemoteUpdateSumm
     }
 }
 
-fn get_cached_summary_or_default() -> Result<UpdateSummary, Error> {
-    match File::open(UPDATE_CACHE) {
-        Ok(file) => {
-            let content = match serde_json::from_reader(file) {
-                Ok(cache_content) => cache_content,
-                Err(err) => {
-                    log::error!("failed to deserialize remote update cache: {err:#}");
-                    Default::default()
-                }
-            };
+/// Read the cached summary from the API cache, or return a default, empty summary.
+///
+/// Note: This does not return an error if the cache entry could not be read (e.g. due to
+/// a deserialization error), but also returns the default, empty summary.
+/// This ensure that he cache self-heals if an entry got corrupted for some reason.
+async fn get_cached_summary_or_default() -> Result<UpdateSummary, Error> {
+    let guard = api_cache::read_global().await?;
 
-            Ok(content)
-        }
-        Err(err) if err.kind() == ErrorKind::NotFound => Ok(Default::default()),
-        Err(err) => Err(err.into()),
-    }
+    let summary = guard
+        .get::<UpdateSummary>(UPDATE_SUMMARY_CACHE_KEY)
+        .await
+        .inspect_err(|err| {
+            log::error!("could not read 'remote-updates' entry from API cache: {err}")
+        })
+        .unwrap_or_default()
+        .unwrap_or_default();
+
+    Ok(summary)
 }
 
 async fn update_cached_summary_for_node(
@@ -179,10 +188,11 @@ async fn update_cached_summary_for_node(
     node: String,
     node_data: NodeUpdateSummary,
 ) -> Result<(), Error> {
-    let mut file = File::open(UPDATE_CACHE)?;
-    let mut cache_content: UpdateSummary = serde_json::from_reader(&mut file)?;
-    let remote_entry =
-        cache_content
+    let cache = api_cache::write_global().await?;
+    let cache_content = cache.get::<UpdateSummary>(UPDATE_SUMMARY_CACHE_KEY).await?;
+
+    if let Some(mut entry) = cache_content {
+        let remote_entry = entry
             .remotes
             .entry(remote.id)
             .or_insert_with(|| RemoteUpdateSummary {
@@ -191,15 +201,9 @@ async fn update_cached_summary_for_node(
                 status: RemoteUpdateStatus::Success,
             });
 
-    remote_entry.nodes.insert(node, node_data);
-
-    let options = proxmox_product_config::default_create_options();
-    proxmox_sys::fs::replace_file(
-        UPDATE_CACHE,
-        &serde_json::to_vec(&cache_content)?,
-        options,
-        true,
-    )?;
+        remote_entry.nodes.insert(node, node_data);
+        cache.set(UPDATE_SUMMARY_CACHE_KEY, entry).await?;
+    }
 
     Ok(())
 }
@@ -212,7 +216,7 @@ pub async fn refresh_update_summary_cache(remotes: Vec<Remote>) -> Result<(), Er
         .do_for_all_remote_nodes(remotes.clone().into_iter(), fetch_available_updates)
         .await;
 
-    let mut content = get_cached_summary_or_default()?;
+    let mut content = get_cached_summary_or_default().await?;
 
     // Clean out any remotes that might have been removed from the remote config in the meanwhile.
     content
@@ -245,7 +249,7 @@ pub async fn refresh_update_summary_cache(remotes: Vec<Remote>) -> Result<(), Er
 
                     match node_response.data() {
                         Ok(update_info) => {
-                            entry.nodes.insert(node_name, update_info.clone().into());
+                            entry.nodes.insert(node_name, update_info.into());
                         }
                         Err(err) => {
                             // Could not fetch updates from node
@@ -275,8 +279,28 @@ pub async fn refresh_update_summary_cache(remotes: Vec<Remote>) -> Result<(), Er
         }
     }
 
-    let options = proxmox_product_config::default_create_options();
-    proxmox_sys::fs::replace_file(UPDATE_CACHE, &serde_json::to_vec(&content)?, options, true)?;
+    cleanup_old_cachefile().await?;
+
+    let cache = api_cache::write_global().await?;
+    cache.set(UPDATE_SUMMARY_CACHE_KEY, content).await?;
+
+    Ok(())
+}
+
+// FIXME: We can remove this pretty soon.
+async fn cleanup_old_cachefile() -> Result<(), Error> {
+    tokio::task::spawn_blocking(|| {
+        if let Err(err) = std::fs::remove_file(OLD_CACHEFILE) {
+            if err.kind() != std::io::ErrorKind::NotFound {
+                log::error!(
+                    "could not clean up old remote update cache file {OLD_CACHEFILE}: {err}"
+                );
+            }
+        } else {
+            log::info!("removed obsolete remote update cachefile {OLD_CACHEFILE}")
+        }
+    })
+    .await?;
 
     Ok(())
 }
-- 
2.47.3

Re: [PATCH datacenter-manager v2 0/4] add generic, per-remote (and global) cache for remote API responses

[Thomas Lamprecht]

On 15/05/2026 10:28, Lukas Wagner wrote:
> Changes since v1:
>   - The only changes happened in the last two patches:
>      - Improved behavior in case the cache entry could not be read,
>        mostly to ensure a self-healing behavior
>      - Slightly restructured code to avoid some clone calls
>      - When requesting the list of updates for a single remote, update the
>        global summary cache asynchronously

I was writing up a review for v1 and missed this before hitting send. Anyway,
some of my v1 comments should still apply.

superseded: [PATCH datacenter-manager v2 0/4] add generic, per-remote (and global) cache for remote API responses

[Lukas Wagner]

On Fri May 15, 2026 at 10:28 AM CEST, Lukas Wagner wrote:
> The main intention is to avoid a sprawl of different caching approaches by
> establishing a simple, easy to use cache implementation that can be used to
> persistently cache API responses from remotes (and derived aggregations).

superseded: https://lore.proxmox.com/all/20260515144927.427114-1-l.wagner@proxmox.com/T/#u