Cache /getQuarantineConfig on disk with short TTL for MutTestReport

constants/src/lib.rs

@@ -19,6 +19,8 @@ pub const EXIT_FAILURE: i32 = 1;
 pub const CODEOWNERS_LOCATIONS: &[&str] = &[".github", ".bitbucket", ".", "docs", ".gitlab"];
 
 pub const DEFAULT_ORIGIN: &str = "https://api.trunk.io";
+pub const DEFAULT_QUARANTINED_TESTS_DISK_CACHE_TTL_SECS: u64 = 5 * 60; // 5 minutes
+pub const CACHE_DIR: &str = "trunk-flaky-tests";
 pub const TRUNK_PUBLIC_API_ADDRESS_ENV: &str = "TRUNK_PUBLIC_API_ADDRESS";
 pub const TRUNK_API_CLIENT_RETRY_COUNT_ENV: &str = "TRUNK_API_CLIENT_RETRY_COUNT";
 
@@ -31,6 +33,8 @@ pub const TRUNK_REPO_HEAD_SHA_ENV: &str = "TRUNK_REPO_HEAD_SHA";
 pub const TRUNK_REPO_HEAD_BRANCH_ENV: &str = "TRUNK_REPO_HEAD_BRANCH";
 pub const TRUNK_REPO_HEAD_COMMIT_EPOCH_ENV: &str = "TRUNK_REPO_HEAD_COMMIT_EPOCH";
 pub const TRUNK_REPO_HEAD_AUTHOR_NAME_ENV: &str = "TRUNK_REPO_HEAD_AUTHOR_NAME";
+pub const TRUNK_QUARANTINED_TESTS_DISK_CACHE_TTL_SECS_ENV: &str =
+    "TRUNK_QUARANTINED_TESTS_DISK_CACHE_TTL_SECS";
 pub const TRUNK_CODEOWNERS_PATH_ENV: &str = "TRUNK_CODEOWNERS_PATH";
 pub const TRUNK_VARIANT_ENV: &str = "TRUNK_VARIANT";
 pub const TRUNK_USE_UNCLONED_REPO_ENV: &str = "TRUNK_USE_UNCLONED_REPO";

rspec-trunk-flaky-tests/README.md

@@ -146,7 +146,7 @@ Understanding the quarantining flow helps you know what to expect when using thi
 
    - Tests run one by one as normal.
    - When a test **fails**, the gem intercepts the failure through RSpec's `set_exception` hook.
-   - **On the first failure**, the gem makes an API call to fetch the list of quarantined tests from Trunk servers. This list is then **cached in memory** for the remainder of the test run to minimize API calls.
+   - **On the first failure**, the gem makes an API call to fetch the list of quarantined tests from Trunk servers. This list is then **cached in memory** for the remainder of the test run to minimize API calls. The list is also cached on disk with a TTL, configurable via the `TRUNK_QUARANTINED_TESTS_DISK_CACHE_TTL_SECS` environment variable (default 300s = 5m).
    - For each subsequent failure, the gem checks the cached quarantine list (no additional API calls).
 
 3. **Quarantine Check and Exception Override**

rspec-trunk-flaky-tests/lib/trunk_spec_helper.rb

@@ -25,6 +25,7 @@
 #   TRUNK_DRY_RUN - Set to 'true' to save bundle locally instead of uploading
 #   TRUNK_USE_UNCLONED_REPO - Set to 'true' for uncloned repo mode
 #   TRUNK_LOCAL_UPLOAD_DIR - Directory to save test results locally (disables upload)
+#   TRUNK_QUARANTINED_TESTS_DISK_CACHE_TTL_SECS - Time to cache quarantined tests on disk (in seconds)
 #   DISABLE_RSPEC_TRUNK_FLAKY_TESTS - Set to 'true' to completely disable Trunk
 #
 require 'rspec/core'
@@ -59,7 +60,7 @@ def trunk_disabled
   ENV['DISABLE_RSPEC_TRUNK_FLAKY_TESTS'] == 'true' || ENV['TRUNK_ORG_URL_SLUG'].nil? || ENV['TRUNK_API_TOKEN'].nil?
 end
 
-# we want to cache the test report so we can add to it as we go and reduce the number of API calls
+# we want to cache the test report in memory so we can add to it as we go and reduce the number of API calls
 $test_report = TestReport.new('rspec', "#{$PROGRAM_NAME} #{ARGV.join(' ')}", nil)
 
 module RSpec

test_report/Cargo.toml

@@ -16,6 +16,7 @@ prost-wkt-types = { version = "0.5.1", features = ["vendored-protox"] }
 prost = "0.12.6"
 tempfile = "3.2.0"
 chrono = "0.4.33"
+serde = { version = "1.0", features = ["derive"] }
 serde_json = "1.0.133"
 js-sys = { version = "0.3.70", optional = true }
 tokio = "1.43.1"
@@ -28,6 +29,7 @@ api = { version = "0.1.0", path = "../api" }
 context = { version = "0.1.0", path = "../context" }
 codeowners = { version = "0.1.3", path = "../codeowners" }
 constants = { path = "../constants" }
+uuid = { version = "1.10.0", features = ["v5"] }
 
 [dev-dependencies]
 assert_matches = "1.5.0"

test_report/src/report.rs

@@ -3,7 +3,7 @@ use std::{
     collections::HashMap,
     env, fs,
     path::{Path, PathBuf},
-    time::SystemTime,
+    time::{Duration, SystemTime},
 };
 
 use api::{client::ApiClient, message};
@@ -19,18 +19,27 @@ use proto::test_context::test_run::{
     CodeOwner, TestCaseRun, TestCaseRunStatus, TestReport as TestReportProto, TestResult,
     UploaderMetadata,
 };
+use serde::{Deserialize, Serialize};
 use third_party::sentry;
 use tracing_subscriber::{filter::FilterFn, prelude::*};
 use trunk_analytics_cli::{context::gather_initial_test_context, upload_command::run_upload};
+use uuid::Uuid;
 #[cfg(feature = "wasm")]
 use wasm_bindgen::prelude::wasm_bindgen;
 
+#[derive(Debug, Clone, Serialize, Deserialize)]
+struct QuarantinedTestsDiskCacheEntry {
+    quarantined_tests: HashMap<String, bool>,
+    cached_at_secs: u64,
+}
+
 #[derive(Debug, Clone, PartialEq)]
 pub struct TestReport {
     test_report: TestReportProto,
     command: String,
     started_at: SystemTime,
     quarantined_tests: Option<HashMap<String, bool>>,
+    quarantined_tests_disk_cache_ttl: Duration,
     codeowners: Option<CodeOwners>,
     variant: Option<String>,
     repo: Option<BundleRepo>,
@@ -143,11 +152,18 @@ impl MutTestReport {
                         .as_deref(),
                 )
             });
+        let quarantined_tests_disk_cache_ttl = Duration::from_secs(
+            env::var(constants::TRUNK_QUARANTINED_TESTS_DISK_CACHE_TTL_SECS_ENV)
+                .ok()
+                .and_then(|v| v.parse().ok())
+                .unwrap_or(constants::DEFAULT_QUARANTINED_TESTS_DISK_CACHE_TTL_SECS),
+        );
         Self(RefCell::new(TestReport {
             test_report,
             command,
             started_at,
             quarantined_tests: None,
+            quarantined_tests_disk_cache_ttl,
             codeowners,
             repo,
             variant: variant.clone(),
@@ -298,22 +314,127 @@ impl MutTestReport {
         }
     }
 
+    fn get_quarantined_tests_cache_file_path(&self, org_url_slug: &str, repo_url: &str) -> PathBuf {
+        let cache_key = Uuid::new_v5(
+            &Uuid::NAMESPACE_URL,
+            format!("{org_url_slug}#{repo_url}").as_bytes(),
+        )
+        .to_string();
+        let quarantined_tests_cache_file_name = format!("quarantined_tests_{cache_key}.json");
+
+        env::temp_dir()
+            .join(constants::CACHE_DIR)
+            .join(quarantined_tests_cache_file_name)
+    }
+
+    fn load_quarantined_tests_from_disk_cache(
+        &self,
+        org_url_slug: &str,
+        repo_url: &str,
+    ) -> Option<HashMap<String, bool>> {
+        let cache_path = self.get_quarantined_tests_cache_file_path(org_url_slug, repo_url);
+
+        let cache_data = match fs::read_to_string(&cache_path) {
+            Ok(data) => data,
+            Err(err) => {
+                tracing::warn!("Failed to read quarantined tests cache file: {:?}", err);
+                return None;
+            }
+        };
+
+        let cache_entry: QuarantinedTestsDiskCacheEntry = match serde_json::from_str(&cache_data) {
+            Ok(entry) => entry,
+            Err(err) => {
+                tracing::warn!("Failed to parse quarantined tests cache file: {:?}", err);
+                return None;
+            }
+        };
+
+        let now = match SystemTime::now().duration_since(SystemTime::UNIX_EPOCH) {
+            Ok(duration) => duration.as_secs(),
+            Err(_) => {
+                tracing::warn!("Failed to get current time");
+                return None;
+            }
+        };
+        let cache_age = now.saturating_sub(cache_entry.cached_at_secs);
+
+        if cache_age < self.0.borrow().quarantined_tests_disk_cache_ttl.as_secs() {
+            Some(cache_entry.quarantined_tests)
+        } else {
+            let _ = fs::remove_file(&cache_path);
+            None
+        }
+    }
+
+    fn save_quarantined_tests_to_disk_cache(
+        &self,
+        org_url_slug: &str,
+        repo_url: &str,
+        quarantined_tests: &HashMap<String, bool>,
+    ) {
+        let cache_path = self.get_quarantined_tests_cache_file_path(org_url_slug, repo_url);
+
+        let now = match SystemTime::now().duration_since(SystemTime::UNIX_EPOCH) {
+            Ok(duration) => duration.as_secs(),
+            Err(_) => {
+                tracing::warn!("Failed to get current time");
+                return;
+            }
+        };
+
+        let cache_entry = QuarantinedTestsDiskCacheEntry {
+            quarantined_tests: quarantined_tests.clone(),
+            cached_at_secs: now,
+        };
+
+        // create cache directory if it doesn't exist
+        let cache_dir = match cache_path.parent() {
+            Some(dir) => dir,
+            None => {
+                tracing::warn!("Failed to get cache directory");
+                return;
+            }
+        };
+        if let Err(err) = fs::create_dir_all(cache_dir) {
+            tracing::warn!("Failed to create cache directory: {:?}", err);
+            return;
+        }
+
+        if let Ok(json) = serde_json::to_string(&cache_entry) {
+            if let Err(err) = fs::write(&cache_path, json) {
+                tracing::warn!("Failed to write quarantined tests cache file: {:?}", err);
+            }
+        }
+    }
+
     fn populate_quarantined_tests(
         &self,
         api_client: &ApiClient,
         repo: &RepoUrlParts,
         repo_url: String,
         org_url_slug: String,
     ) {
+        // first check in-memory cache
         if self.0.borrow().quarantined_tests.as_ref().is_some() {
-            // already fetched
             return;
         }
+
+        // then check disk cache
+        if let Some(quarantined_tests) =
+            self.load_quarantined_tests_from_disk_cache(&org_url_slug, &repo_url)
+        {
+            // update in-memory cache
+            self.0.borrow_mut().quarantined_tests = Some(quarantined_tests);
+            return;
+        }
+
+        // cache miss - make API call
         let mut quarantined_tests = HashMap::new();
         let request = message::GetQuarantineConfigRequest {
-            org_url_slug,
+            org_url_slug: org_url_slug.clone(),
             test_identifiers: vec![],
-            remote_urls: vec![repo_url],
+            remote_urls: vec![repo_url.clone()],
             repo: repo.clone(),
         };
         let response = tokio::runtime::Builder::new_multi_thread()
@@ -336,7 +457,10 @@ impl MutTestReport {
                 );
             }
         }
-        self.0.borrow_mut().quarantined_tests = Some(quarantined_tests);
+
+        // update both in-memory and disk cache
+        self.0.borrow_mut().quarantined_tests = Some(quarantined_tests.clone());
+        self.save_quarantined_tests_to_disk_cache(&org_url_slug, &repo_url, &quarantined_tests);
     }
 
     fn get_org_url_slug(&self) -> String {